]>
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 | |
e34ecee2 KO |
83 | struct percpu_ref reqs; |
84 | ||
4e179bca | 85 | unsigned long user_id; |
4e179bca | 86 | |
e1bdd5f2 KO |
87 | struct __percpu kioctx_cpu *cpu; |
88 | ||
89 | /* | |
90 | * For percpu reqs_available, number of slots we move to/from global | |
91 | * counter at a time: | |
92 | */ | |
93 | unsigned req_batch; | |
3e845ce0 KO |
94 | /* |
95 | * This is what userspace passed to io_setup(), it's not used for | |
96 | * anything but counting against the global max_reqs quota. | |
97 | * | |
58c85dc2 | 98 | * The real limit is nr_events - 1, which will be larger (see |
3e845ce0 KO |
99 | * aio_setup_ring()) |
100 | */ | |
4e179bca KO |
101 | unsigned max_reqs; |
102 | ||
58c85dc2 KO |
103 | /* Size of ringbuffer, in units of struct io_event */ |
104 | unsigned nr_events; | |
4e179bca | 105 | |
58c85dc2 KO |
106 | unsigned long mmap_base; |
107 | unsigned long mmap_size; | |
108 | ||
109 | struct page **ring_pages; | |
110 | long nr_pages; | |
111 | ||
723be6e3 | 112 | struct work_struct free_work; |
4e23bcae KO |
113 | |
114 | struct { | |
34e83fc6 KO |
115 | /* |
116 | * This counts the number of available slots in the ringbuffer, | |
117 | * so we avoid overflowing it: it's decremented (if positive) | |
118 | * when allocating a kiocb and incremented when the resulting | |
119 | * io_event is pulled off the ringbuffer. | |
e1bdd5f2 KO |
120 | * |
121 | * We batch accesses to it with a percpu version. | |
34e83fc6 KO |
122 | */ |
123 | atomic_t reqs_available; | |
4e23bcae KO |
124 | } ____cacheline_aligned_in_smp; |
125 | ||
126 | struct { | |
127 | spinlock_t ctx_lock; | |
128 | struct list_head active_reqs; /* used for cancellation */ | |
129 | } ____cacheline_aligned_in_smp; | |
130 | ||
58c85dc2 KO |
131 | struct { |
132 | struct mutex ring_lock; | |
4e23bcae KO |
133 | wait_queue_head_t wait; |
134 | } ____cacheline_aligned_in_smp; | |
58c85dc2 KO |
135 | |
136 | struct { | |
137 | unsigned tail; | |
138 | spinlock_t completion_lock; | |
4e23bcae | 139 | } ____cacheline_aligned_in_smp; |
58c85dc2 KO |
140 | |
141 | struct page *internal_pages[AIO_RING_PAGES]; | |
36bc08cc | 142 | struct file *aio_ring_file; |
db446a08 BL |
143 | |
144 | unsigned id; | |
4e179bca KO |
145 | }; |
146 | ||
1da177e4 | 147 | /*------ sysctl variables----*/ |
d55b5fda ZB |
148 | static DEFINE_SPINLOCK(aio_nr_lock); |
149 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
150 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
151 | /*----end sysctl variables---*/ |
152 | ||
e18b890b CL |
153 | static struct kmem_cache *kiocb_cachep; |
154 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 155 | |
1da177e4 LT |
156 | /* aio_setup |
157 | * Creates the slab caches used by the aio routines, panic on | |
158 | * failure as this is done early during the boot sequence. | |
159 | */ | |
160 | static int __init aio_setup(void) | |
161 | { | |
0a31bd5f CL |
162 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
163 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 164 | |
caf4167a | 165 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
166 | |
167 | return 0; | |
168 | } | |
385773e0 | 169 | __initcall(aio_setup); |
1da177e4 | 170 | |
5e9ae2e5 BL |
171 | static void put_aio_ring_file(struct kioctx *ctx) |
172 | { | |
173 | struct file *aio_ring_file = ctx->aio_ring_file; | |
174 | if (aio_ring_file) { | |
175 | truncate_setsize(aio_ring_file->f_inode, 0); | |
176 | ||
177 | /* Prevent further access to the kioctx from migratepages */ | |
178 | spin_lock(&aio_ring_file->f_inode->i_mapping->private_lock); | |
179 | aio_ring_file->f_inode->i_mapping->private_data = NULL; | |
180 | ctx->aio_ring_file = NULL; | |
181 | spin_unlock(&aio_ring_file->f_inode->i_mapping->private_lock); | |
182 | ||
183 | fput(aio_ring_file); | |
184 | } | |
185 | } | |
186 | ||
1da177e4 LT |
187 | static void aio_free_ring(struct kioctx *ctx) |
188 | { | |
36bc08cc | 189 | int i; |
1da177e4 | 190 | |
36bc08cc GZ |
191 | for (i = 0; i < ctx->nr_pages; i++) { |
192 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, | |
193 | page_count(ctx->ring_pages[i])); | |
58c85dc2 | 194 | put_page(ctx->ring_pages[i]); |
36bc08cc | 195 | } |
1da177e4 | 196 | |
5e9ae2e5 BL |
197 | put_aio_ring_file(ctx); |
198 | ||
58c85dc2 KO |
199 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) |
200 | kfree(ctx->ring_pages); | |
36bc08cc GZ |
201 | } |
202 | ||
203 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) | |
204 | { | |
205 | vma->vm_ops = &generic_file_vm_ops; | |
206 | return 0; | |
207 | } | |
208 | ||
209 | static const struct file_operations aio_ring_fops = { | |
210 | .mmap = aio_ring_mmap, | |
211 | }; | |
212 | ||
213 | static int aio_set_page_dirty(struct page *page) | |
214 | { | |
215 | return 0; | |
216 | } | |
217 | ||
0c45355f | 218 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc GZ |
219 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
220 | struct page *old, enum migrate_mode mode) | |
221 | { | |
5e9ae2e5 | 222 | struct kioctx *ctx; |
36bc08cc | 223 | unsigned long flags; |
36bc08cc GZ |
224 | int rc; |
225 | ||
226 | /* Writeback must be complete */ | |
227 | BUG_ON(PageWriteback(old)); | |
228 | put_page(old); | |
229 | ||
230 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode); | |
231 | if (rc != MIGRATEPAGE_SUCCESS) { | |
232 | get_page(old); | |
233 | return rc; | |
234 | } | |
235 | ||
236 | get_page(new); | |
237 | ||
5e9ae2e5 BL |
238 | /* We can potentially race against kioctx teardown here. Use the |
239 | * address_space's private data lock to protect the mapping's | |
240 | * private_data. | |
241 | */ | |
242 | spin_lock(&mapping->private_lock); | |
243 | ctx = mapping->private_data; | |
244 | if (ctx) { | |
245 | pgoff_t idx; | |
246 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
247 | migrate_page_copy(new, old); | |
248 | idx = old->index; | |
249 | if (idx < (pgoff_t)ctx->nr_pages) | |
250 | ctx->ring_pages[idx] = new; | |
251 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
252 | } else | |
253 | rc = -EBUSY; | |
254 | spin_unlock(&mapping->private_lock); | |
36bc08cc GZ |
255 | |
256 | return rc; | |
1da177e4 | 257 | } |
0c45355f | 258 | #endif |
1da177e4 | 259 | |
36bc08cc GZ |
260 | static const struct address_space_operations aio_ctx_aops = { |
261 | .set_page_dirty = aio_set_page_dirty, | |
0c45355f | 262 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc | 263 | .migratepage = aio_migratepage, |
0c45355f | 264 | #endif |
36bc08cc GZ |
265 | }; |
266 | ||
1da177e4 LT |
267 | static int aio_setup_ring(struct kioctx *ctx) |
268 | { | |
269 | struct aio_ring *ring; | |
1da177e4 | 270 | unsigned nr_events = ctx->max_reqs; |
41003a7b | 271 | struct mm_struct *mm = current->mm; |
41badc15 | 272 | unsigned long size, populate; |
1da177e4 | 273 | int nr_pages; |
36bc08cc GZ |
274 | int i; |
275 | struct file *file; | |
1da177e4 LT |
276 | |
277 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
278 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
279 | ||
280 | size = sizeof(struct aio_ring); | |
281 | size += sizeof(struct io_event) * nr_events; | |
1da177e4 | 282 | |
36bc08cc | 283 | nr_pages = PFN_UP(size); |
1da177e4 LT |
284 | if (nr_pages < 0) |
285 | return -EINVAL; | |
286 | ||
36bc08cc GZ |
287 | file = anon_inode_getfile_private("[aio]", &aio_ring_fops, ctx, O_RDWR); |
288 | if (IS_ERR(file)) { | |
289 | ctx->aio_ring_file = NULL; | |
290 | return -EAGAIN; | |
291 | } | |
292 | ||
293 | file->f_inode->i_mapping->a_ops = &aio_ctx_aops; | |
294 | file->f_inode->i_mapping->private_data = ctx; | |
295 | file->f_inode->i_size = PAGE_SIZE * (loff_t)nr_pages; | |
296 | ||
297 | for (i = 0; i < nr_pages; i++) { | |
298 | struct page *page; | |
299 | page = find_or_create_page(file->f_inode->i_mapping, | |
300 | i, GFP_HIGHUSER | __GFP_ZERO); | |
301 | if (!page) | |
302 | break; | |
303 | pr_debug("pid(%d) page[%d]->count=%d\n", | |
304 | current->pid, i, page_count(page)); | |
305 | SetPageUptodate(page); | |
306 | SetPageDirty(page); | |
307 | unlock_page(page); | |
308 | } | |
309 | ctx->aio_ring_file = file; | |
310 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) | |
311 | / sizeof(struct io_event); | |
1da177e4 | 312 | |
58c85dc2 | 313 | ctx->ring_pages = ctx->internal_pages; |
1da177e4 | 314 | if (nr_pages > AIO_RING_PAGES) { |
58c85dc2 KO |
315 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
316 | GFP_KERNEL); | |
317 | if (!ctx->ring_pages) | |
1da177e4 | 318 | return -ENOMEM; |
1da177e4 LT |
319 | } |
320 | ||
58c85dc2 KO |
321 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
322 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
36bc08cc | 323 | |
41003a7b | 324 | down_write(&mm->mmap_sem); |
36bc08cc GZ |
325 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
326 | PROT_READ | PROT_WRITE, | |
327 | MAP_SHARED | MAP_POPULATE, 0, &populate); | |
58c85dc2 | 328 | if (IS_ERR((void *)ctx->mmap_base)) { |
41003a7b | 329 | up_write(&mm->mmap_sem); |
58c85dc2 | 330 | ctx->mmap_size = 0; |
1da177e4 LT |
331 | aio_free_ring(ctx); |
332 | return -EAGAIN; | |
333 | } | |
334 | ||
58c85dc2 | 335 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
d6c355c7 BL |
336 | |
337 | /* We must do this while still holding mmap_sem for write, as we | |
338 | * need to be protected against userspace attempting to mremap() | |
339 | * or munmap() the ring buffer. | |
340 | */ | |
58c85dc2 KO |
341 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, |
342 | 1, 0, ctx->ring_pages, NULL); | |
d6c355c7 BL |
343 | |
344 | /* Dropping the reference here is safe as the page cache will hold | |
345 | * onto the pages for us. It is also required so that page migration | |
346 | * can unmap the pages and get the right reference count. | |
347 | */ | |
36bc08cc GZ |
348 | for (i = 0; i < ctx->nr_pages; i++) |
349 | put_page(ctx->ring_pages[i]); | |
1da177e4 | 350 | |
d6c355c7 BL |
351 | up_write(&mm->mmap_sem); |
352 | ||
58c85dc2 | 353 | if (unlikely(ctx->nr_pages != nr_pages)) { |
1da177e4 LT |
354 | aio_free_ring(ctx); |
355 | return -EAGAIN; | |
356 | } | |
357 | ||
58c85dc2 KO |
358 | ctx->user_id = ctx->mmap_base; |
359 | ctx->nr_events = nr_events; /* trusted copy */ | |
1da177e4 | 360 | |
58c85dc2 | 361 | ring = kmap_atomic(ctx->ring_pages[0]); |
1da177e4 | 362 | ring->nr = nr_events; /* user copy */ |
db446a08 | 363 | ring->id = ~0U; |
1da177e4 LT |
364 | ring->head = ring->tail = 0; |
365 | ring->magic = AIO_RING_MAGIC; | |
366 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
367 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
368 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 369 | kunmap_atomic(ring); |
58c85dc2 | 370 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 LT |
371 | |
372 | return 0; | |
373 | } | |
374 | ||
1da177e4 LT |
375 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
376 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
377 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
378 | ||
0460fef2 KO |
379 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
380 | { | |
381 | struct kioctx *ctx = req->ki_ctx; | |
382 | unsigned long flags; | |
383 | ||
384 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
385 | ||
386 | if (!req->ki_list.next) | |
387 | list_add(&req->ki_list, &ctx->active_reqs); | |
388 | ||
389 | req->ki_cancel = cancel; | |
390 | ||
391 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
392 | } | |
393 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
394 | ||
bec68faa | 395 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb) |
906b973c | 396 | { |
0460fef2 | 397 | kiocb_cancel_fn *old, *cancel; |
906b973c | 398 | |
0460fef2 KO |
399 | /* |
400 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
401 | * actually has a cancel function, hence the cmpxchg() | |
402 | */ | |
403 | ||
404 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
405 | do { | |
406 | if (!cancel || cancel == KIOCB_CANCELLED) | |
57282d8f | 407 | return -EINVAL; |
906b973c | 408 | |
0460fef2 KO |
409 | old = cancel; |
410 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
411 | } while (cancel != old); | |
906b973c | 412 | |
57282d8f | 413 | return cancel(kiocb); |
906b973c KO |
414 | } |
415 | ||
e34ecee2 | 416 | static void free_ioctx(struct work_struct *work) |
36f55889 | 417 | { |
e34ecee2 KO |
418 | struct kioctx *ctx = container_of(work, struct kioctx, free_work); |
419 | ||
420 | pr_debug("freeing %p\n", ctx); | |
e1bdd5f2 | 421 | |
e34ecee2 | 422 | aio_free_ring(ctx); |
e1bdd5f2 | 423 | free_percpu(ctx->cpu); |
36f55889 KO |
424 | kmem_cache_free(kioctx_cachep, ctx); |
425 | } | |
426 | ||
e34ecee2 KO |
427 | static void free_ioctx_reqs(struct percpu_ref *ref) |
428 | { | |
429 | struct kioctx *ctx = container_of(ref, struct kioctx, reqs); | |
430 | ||
431 | INIT_WORK(&ctx->free_work, free_ioctx); | |
432 | schedule_work(&ctx->free_work); | |
433 | } | |
434 | ||
36f55889 KO |
435 | /* |
436 | * When this function runs, the kioctx has been removed from the "hash table" | |
437 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
438 | * now it's safe to cancel any that need to be. | |
439 | */ | |
e34ecee2 | 440 | static void free_ioctx_users(struct percpu_ref *ref) |
36f55889 | 441 | { |
e34ecee2 | 442 | struct kioctx *ctx = container_of(ref, struct kioctx, users); |
36f55889 KO |
443 | struct kiocb *req; |
444 | ||
445 | spin_lock_irq(&ctx->ctx_lock); | |
446 | ||
447 | while (!list_empty(&ctx->active_reqs)) { | |
448 | req = list_first_entry(&ctx->active_reqs, | |
449 | struct kiocb, ki_list); | |
450 | ||
451 | list_del_init(&req->ki_list); | |
bec68faa | 452 | kiocb_cancel(ctx, req); |
36f55889 KO |
453 | } |
454 | ||
455 | spin_unlock_irq(&ctx->ctx_lock); | |
456 | ||
e34ecee2 KO |
457 | percpu_ref_kill(&ctx->reqs); |
458 | percpu_ref_put(&ctx->reqs); | |
36f55889 KO |
459 | } |
460 | ||
db446a08 BL |
461 | static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) |
462 | { | |
463 | unsigned i, new_nr; | |
464 | struct kioctx_table *table, *old; | |
465 | struct aio_ring *ring; | |
466 | ||
467 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 468 | rcu_read_lock(); |
77d30b14 | 469 | table = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
470 | |
471 | while (1) { | |
472 | if (table) | |
473 | for (i = 0; i < table->nr; i++) | |
474 | if (!table->table[i]) { | |
475 | ctx->id = i; | |
476 | table->table[i] = ctx; | |
d9b2c871 | 477 | rcu_read_unlock(); |
db446a08 BL |
478 | spin_unlock(&mm->ioctx_lock); |
479 | ||
480 | ring = kmap_atomic(ctx->ring_pages[0]); | |
481 | ring->id = ctx->id; | |
482 | kunmap_atomic(ring); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | new_nr = (table ? table->nr : 1) * 4; | |
487 | ||
d9b2c871 | 488 | rcu_read_unlock(); |
db446a08 BL |
489 | spin_unlock(&mm->ioctx_lock); |
490 | ||
491 | table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * | |
492 | new_nr, GFP_KERNEL); | |
493 | if (!table) | |
494 | return -ENOMEM; | |
495 | ||
496 | table->nr = new_nr; | |
497 | ||
498 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 499 | rcu_read_lock(); |
77d30b14 | 500 | old = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
501 | |
502 | if (!old) { | |
503 | rcu_assign_pointer(mm->ioctx_table, table); | |
504 | } else if (table->nr > old->nr) { | |
505 | memcpy(table->table, old->table, | |
506 | old->nr * sizeof(struct kioctx *)); | |
507 | ||
508 | rcu_assign_pointer(mm->ioctx_table, table); | |
509 | kfree_rcu(old, rcu); | |
510 | } else { | |
511 | kfree(table); | |
512 | table = old; | |
513 | } | |
514 | } | |
515 | } | |
516 | ||
e34ecee2 KO |
517 | static void aio_nr_sub(unsigned nr) |
518 | { | |
519 | spin_lock(&aio_nr_lock); | |
520 | if (WARN_ON(aio_nr - nr > aio_nr)) | |
521 | aio_nr = 0; | |
522 | else | |
523 | aio_nr -= nr; | |
524 | spin_unlock(&aio_nr_lock); | |
525 | } | |
526 | ||
1da177e4 LT |
527 | /* ioctx_alloc |
528 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
529 | */ | |
530 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
531 | { | |
41003a7b | 532 | struct mm_struct *mm = current->mm; |
1da177e4 | 533 | struct kioctx *ctx; |
e23754f8 | 534 | int err = -ENOMEM; |
1da177e4 | 535 | |
e1bdd5f2 KO |
536 | /* |
537 | * We keep track of the number of available ringbuffer slots, to prevent | |
538 | * overflow (reqs_available), and we also use percpu counters for this. | |
539 | * | |
540 | * So since up to half the slots might be on other cpu's percpu counters | |
541 | * and unavailable, double nr_events so userspace sees what they | |
542 | * expected: additionally, we move req_batch slots to/from percpu | |
543 | * counters at a time, so make sure that isn't 0: | |
544 | */ | |
545 | nr_events = max(nr_events, num_possible_cpus() * 4); | |
546 | nr_events *= 2; | |
547 | ||
1da177e4 LT |
548 | /* Prevent overflows */ |
549 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
550 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
551 | pr_debug("ENOMEM: nr_events too high\n"); | |
552 | return ERR_PTR(-EINVAL); | |
553 | } | |
554 | ||
4cd81c3d | 555 | if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) |
1da177e4 LT |
556 | return ERR_PTR(-EAGAIN); |
557 | ||
c3762229 | 558 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
559 | if (!ctx) |
560 | return ERR_PTR(-ENOMEM); | |
561 | ||
1da177e4 | 562 | ctx->max_reqs = nr_events; |
1da177e4 | 563 | |
e34ecee2 KO |
564 | if (percpu_ref_init(&ctx->users, free_ioctx_users)) |
565 | goto err; | |
566 | ||
567 | if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs)) | |
568 | goto err; | |
723be6e3 | 569 | |
1da177e4 | 570 | spin_lock_init(&ctx->ctx_lock); |
0460fef2 | 571 | spin_lock_init(&ctx->completion_lock); |
58c85dc2 | 572 | mutex_init(&ctx->ring_lock); |
1da177e4 LT |
573 | init_waitqueue_head(&ctx->wait); |
574 | ||
575 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 | 576 | |
e1bdd5f2 KO |
577 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
578 | if (!ctx->cpu) | |
e34ecee2 | 579 | goto err; |
1da177e4 | 580 | |
e1bdd5f2 | 581 | if (aio_setup_ring(ctx) < 0) |
e34ecee2 | 582 | goto err; |
e1bdd5f2 | 583 | |
34e83fc6 | 584 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
e1bdd5f2 | 585 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
6878ea72 BL |
586 | if (ctx->req_batch < 1) |
587 | ctx->req_batch = 1; | |
34e83fc6 | 588 | |
1da177e4 | 589 | /* limit the number of system wide aios */ |
9fa1cb39 | 590 | spin_lock(&aio_nr_lock); |
4cd81c3d | 591 | if (aio_nr + nr_events > (aio_max_nr * 2UL) || |
2dd542b7 | 592 | aio_nr + nr_events < aio_nr) { |
9fa1cb39 | 593 | spin_unlock(&aio_nr_lock); |
e34ecee2 KO |
594 | err = -EAGAIN; |
595 | goto err; | |
2dd542b7 AV |
596 | } |
597 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 598 | spin_unlock(&aio_nr_lock); |
1da177e4 | 599 | |
723be6e3 KO |
600 | percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ |
601 | ||
da90382c BL |
602 | err = ioctx_add_table(ctx, mm); |
603 | if (err) | |
e34ecee2 | 604 | goto err_cleanup; |
da90382c | 605 | |
caf4167a | 606 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
58c85dc2 | 607 | ctx, ctx->user_id, mm, ctx->nr_events); |
1da177e4 LT |
608 | return ctx; |
609 | ||
e34ecee2 KO |
610 | err_cleanup: |
611 | aio_nr_sub(ctx->max_reqs); | |
612 | err: | |
e1bdd5f2 | 613 | free_percpu(ctx->cpu); |
e34ecee2 | 614 | free_percpu(ctx->reqs.pcpu_count); |
723be6e3 | 615 | free_percpu(ctx->users.pcpu_count); |
1da177e4 | 616 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 617 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 618 | return ERR_PTR(err); |
1da177e4 LT |
619 | } |
620 | ||
36f55889 KO |
621 | /* kill_ioctx |
622 | * Cancels all outstanding aio requests on an aio context. Used | |
623 | * when the processes owning a context have all exited to encourage | |
624 | * the rapid destruction of the kioctx. | |
625 | */ | |
db446a08 | 626 | static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx) |
36f55889 KO |
627 | { |
628 | if (!atomic_xchg(&ctx->dead, 1)) { | |
db446a08 BL |
629 | struct kioctx_table *table; |
630 | ||
631 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 632 | rcu_read_lock(); |
77d30b14 | 633 | table = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
634 | |
635 | WARN_ON(ctx != table->table[ctx->id]); | |
636 | table->table[ctx->id] = NULL; | |
d9b2c871 | 637 | rcu_read_unlock(); |
db446a08 BL |
638 | spin_unlock(&mm->ioctx_lock); |
639 | ||
723be6e3 KO |
640 | /* percpu_ref_kill() will do the necessary call_rcu() */ |
641 | wake_up_all(&ctx->wait); | |
dee11c23 | 642 | |
36f55889 | 643 | /* |
4fcc712f KO |
644 | * It'd be more correct to do this in free_ioctx(), after all |
645 | * the outstanding kiocbs have finished - but by then io_destroy | |
646 | * has already returned, so io_setup() could potentially return | |
647 | * -EAGAIN with no ioctxs actually in use (as far as userspace | |
648 | * could tell). | |
36f55889 | 649 | */ |
e34ecee2 | 650 | aio_nr_sub(ctx->max_reqs); |
4fcc712f KO |
651 | |
652 | if (ctx->mmap_size) | |
653 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
654 | ||
723be6e3 | 655 | percpu_ref_kill(&ctx->users); |
36f55889 | 656 | } |
1da177e4 LT |
657 | } |
658 | ||
659 | /* wait_on_sync_kiocb: | |
660 | * Waits on the given sync kiocb to complete. | |
661 | */ | |
57282d8f | 662 | ssize_t wait_on_sync_kiocb(struct kiocb *req) |
1da177e4 | 663 | { |
57282d8f | 664 | while (!req->ki_ctx) { |
1da177e4 | 665 | set_current_state(TASK_UNINTERRUPTIBLE); |
57282d8f | 666 | if (req->ki_ctx) |
1da177e4 | 667 | break; |
41d10da3 | 668 | io_schedule(); |
1da177e4 LT |
669 | } |
670 | __set_current_state(TASK_RUNNING); | |
57282d8f | 671 | return req->ki_user_data; |
1da177e4 | 672 | } |
385773e0 | 673 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 674 | |
36f55889 KO |
675 | /* |
676 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
677 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
678 | * called on the context. | |
679 | * | |
680 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
681 | * them. | |
1da177e4 | 682 | */ |
fc9b52cd | 683 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 684 | { |
db446a08 | 685 | struct kioctx_table *table; |
abf137dd | 686 | struct kioctx *ctx; |
db446a08 BL |
687 | unsigned i = 0; |
688 | ||
689 | while (1) { | |
690 | rcu_read_lock(); | |
691 | table = rcu_dereference(mm->ioctx_table); | |
692 | ||
693 | do { | |
694 | if (!table || i >= table->nr) { | |
695 | rcu_read_unlock(); | |
696 | rcu_assign_pointer(mm->ioctx_table, NULL); | |
697 | if (table) | |
698 | kfree(table); | |
699 | return; | |
700 | } | |
701 | ||
702 | ctx = table->table[i++]; | |
703 | } while (!ctx); | |
704 | ||
705 | rcu_read_unlock(); | |
abf137dd | 706 | |
936af157 AV |
707 | /* |
708 | * We don't need to bother with munmap() here - | |
709 | * exit_mmap(mm) is coming and it'll unmap everything. | |
710 | * Since aio_free_ring() uses non-zero ->mmap_size | |
711 | * as indicator that it needs to unmap the area, | |
712 | * just set it to 0; aio_free_ring() is the only | |
713 | * place that uses ->mmap_size, so it's safe. | |
936af157 | 714 | */ |
58c85dc2 | 715 | ctx->mmap_size = 0; |
36f55889 | 716 | |
db446a08 | 717 | kill_ioctx(mm, ctx); |
1da177e4 LT |
718 | } |
719 | } | |
720 | ||
e1bdd5f2 KO |
721 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
722 | { | |
723 | struct kioctx_cpu *kcpu; | |
724 | ||
725 | preempt_disable(); | |
726 | kcpu = this_cpu_ptr(ctx->cpu); | |
727 | ||
728 | kcpu->reqs_available += nr; | |
729 | while (kcpu->reqs_available >= ctx->req_batch * 2) { | |
730 | kcpu->reqs_available -= ctx->req_batch; | |
731 | atomic_add(ctx->req_batch, &ctx->reqs_available); | |
732 | } | |
733 | ||
734 | preempt_enable(); | |
735 | } | |
736 | ||
737 | static bool get_reqs_available(struct kioctx *ctx) | |
738 | { | |
739 | struct kioctx_cpu *kcpu; | |
740 | bool ret = false; | |
741 | ||
742 | preempt_disable(); | |
743 | kcpu = this_cpu_ptr(ctx->cpu); | |
744 | ||
745 | if (!kcpu->reqs_available) { | |
746 | int old, avail = atomic_read(&ctx->reqs_available); | |
747 | ||
748 | do { | |
749 | if (avail < ctx->req_batch) | |
750 | goto out; | |
751 | ||
752 | old = avail; | |
753 | avail = atomic_cmpxchg(&ctx->reqs_available, | |
754 | avail, avail - ctx->req_batch); | |
755 | } while (avail != old); | |
756 | ||
757 | kcpu->reqs_available += ctx->req_batch; | |
758 | } | |
759 | ||
760 | ret = true; | |
761 | kcpu->reqs_available--; | |
762 | out: | |
763 | preempt_enable(); | |
764 | return ret; | |
765 | } | |
766 | ||
1da177e4 | 767 | /* aio_get_req |
57282d8f KO |
768 | * Allocate a slot for an aio request. |
769 | * Returns NULL if no requests are free. | |
1da177e4 | 770 | */ |
a1c8eae7 | 771 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
1da177e4 | 772 | { |
a1c8eae7 KO |
773 | struct kiocb *req; |
774 | ||
e1bdd5f2 | 775 | if (!get_reqs_available(ctx)) |
a1c8eae7 KO |
776 | return NULL; |
777 | ||
0460fef2 | 778 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
1da177e4 | 779 | if (unlikely(!req)) |
a1c8eae7 | 780 | goto out_put; |
1da177e4 | 781 | |
e34ecee2 KO |
782 | percpu_ref_get(&ctx->reqs); |
783 | ||
1da177e4 | 784 | req->ki_ctx = ctx; |
080d676d | 785 | return req; |
a1c8eae7 | 786 | out_put: |
e1bdd5f2 | 787 | put_reqs_available(ctx, 1); |
a1c8eae7 | 788 | return NULL; |
1da177e4 LT |
789 | } |
790 | ||
11599eba | 791 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 792 | { |
1d98ebfc KO |
793 | if (req->ki_filp) |
794 | fput(req->ki_filp); | |
13389010 DL |
795 | if (req->ki_eventfd != NULL) |
796 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 | 797 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
798 | } |
799 | ||
d5470b59 | 800 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 801 | { |
db446a08 | 802 | struct aio_ring __user *ring = (void __user *)ctx_id; |
abf137dd | 803 | struct mm_struct *mm = current->mm; |
65c24491 | 804 | struct kioctx *ctx, *ret = NULL; |
db446a08 BL |
805 | struct kioctx_table *table; |
806 | unsigned id; | |
807 | ||
808 | if (get_user(id, &ring->id)) | |
809 | return NULL; | |
1da177e4 | 810 | |
abf137dd | 811 | rcu_read_lock(); |
db446a08 | 812 | table = rcu_dereference(mm->ioctx_table); |
abf137dd | 813 | |
db446a08 BL |
814 | if (!table || id >= table->nr) |
815 | goto out; | |
1da177e4 | 816 | |
db446a08 | 817 | ctx = table->table[id]; |
f30d704f | 818 | if (ctx && ctx->user_id == ctx_id) { |
db446a08 BL |
819 | percpu_ref_get(&ctx->users); |
820 | ret = ctx; | |
821 | } | |
822 | out: | |
abf137dd | 823 | rcu_read_unlock(); |
65c24491 | 824 | return ret; |
1da177e4 LT |
825 | } |
826 | ||
1da177e4 LT |
827 | /* aio_complete |
828 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 829 | */ |
2d68449e | 830 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
831 | { |
832 | struct kioctx *ctx = iocb->ki_ctx; | |
1da177e4 | 833 | struct aio_ring *ring; |
21b40200 | 834 | struct io_event *ev_page, *event; |
1da177e4 | 835 | unsigned long flags; |
21b40200 | 836 | unsigned tail, pos; |
1da177e4 | 837 | |
20dcae32 ZB |
838 | /* |
839 | * Special case handling for sync iocbs: | |
840 | * - events go directly into the iocb for fast handling | |
841 | * - the sync task with the iocb in its stack holds the single iocb | |
842 | * ref, no other paths have a way to get another ref | |
843 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
844 | */ |
845 | if (is_sync_kiocb(iocb)) { | |
1da177e4 | 846 | iocb->ki_user_data = res; |
57282d8f KO |
847 | smp_wmb(); |
848 | iocb->ki_ctx = ERR_PTR(-EXDEV); | |
1da177e4 | 849 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 850 | return; |
1da177e4 LT |
851 | } |
852 | ||
0460fef2 KO |
853 | if (iocb->ki_list.next) { |
854 | unsigned long flags; | |
855 | ||
856 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
857 | list_del(&iocb->ki_list); | |
858 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
859 | } | |
11599eba | 860 | |
0460fef2 KO |
861 | /* |
862 | * Add a completion event to the ring buffer. Must be done holding | |
4b30f07e | 863 | * ctx->completion_lock to prevent other code from messing with the tail |
0460fef2 KO |
864 | * pointer since we might be called from irq context. |
865 | */ | |
866 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
867 | ||
58c85dc2 | 868 | tail = ctx->tail; |
21b40200 KO |
869 | pos = tail + AIO_EVENTS_OFFSET; |
870 | ||
58c85dc2 | 871 | if (++tail >= ctx->nr_events) |
4bf69b2a | 872 | tail = 0; |
1da177e4 | 873 | |
58c85dc2 | 874 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
875 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
876 | ||
1da177e4 LT |
877 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
878 | event->data = iocb->ki_user_data; | |
879 | event->res = res; | |
880 | event->res2 = res2; | |
881 | ||
21b40200 | 882 | kunmap_atomic(ev_page); |
58c85dc2 | 883 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
884 | |
885 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
caf4167a KO |
886 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
887 | res, res2); | |
1da177e4 LT |
888 | |
889 | /* after flagging the request as done, we | |
890 | * must never even look at it again | |
891 | */ | |
892 | smp_wmb(); /* make event visible before updating tail */ | |
893 | ||
58c85dc2 | 894 | ctx->tail = tail; |
1da177e4 | 895 | |
58c85dc2 | 896 | ring = kmap_atomic(ctx->ring_pages[0]); |
21b40200 | 897 | ring->tail = tail; |
e8e3c3d6 | 898 | kunmap_atomic(ring); |
58c85dc2 | 899 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 | 900 | |
0460fef2 KO |
901 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
902 | ||
21b40200 | 903 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
8d1c98b0 DL |
904 | |
905 | /* | |
906 | * Check if the user asked us to deliver the result through an | |
907 | * eventfd. The eventfd_signal() function is safe to be called | |
908 | * from IRQ context. | |
909 | */ | |
87c3a86e | 910 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
911 | eventfd_signal(iocb->ki_eventfd, 1); |
912 | ||
1da177e4 | 913 | /* everything turned out well, dispose of the aiocb. */ |
57282d8f | 914 | kiocb_free(iocb); |
1da177e4 | 915 | |
6cb2a210 QB |
916 | /* |
917 | * We have to order our ring_info tail store above and test | |
918 | * of the wait list below outside the wait lock. This is | |
919 | * like in wake_up_bit() where clearing a bit has to be | |
920 | * ordered with the unlocked test. | |
921 | */ | |
922 | smp_mb(); | |
923 | ||
1da177e4 LT |
924 | if (waitqueue_active(&ctx->wait)) |
925 | wake_up(&ctx->wait); | |
926 | ||
e34ecee2 | 927 | percpu_ref_put(&ctx->reqs); |
1da177e4 | 928 | } |
385773e0 | 929 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 930 | |
a31ad380 KO |
931 | /* aio_read_events |
932 | * Pull an event off of the ioctx's event ring. Returns the number of | |
933 | * events fetched | |
1da177e4 | 934 | */ |
a31ad380 KO |
935 | static long aio_read_events_ring(struct kioctx *ctx, |
936 | struct io_event __user *event, long nr) | |
1da177e4 | 937 | { |
1da177e4 | 938 | struct aio_ring *ring; |
5ffac122 | 939 | unsigned head, tail, pos; |
a31ad380 KO |
940 | long ret = 0; |
941 | int copy_ret; | |
942 | ||
58c85dc2 | 943 | mutex_lock(&ctx->ring_lock); |
1da177e4 | 944 | |
58c85dc2 | 945 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 946 | head = ring->head; |
5ffac122 | 947 | tail = ring->tail; |
a31ad380 KO |
948 | kunmap_atomic(ring); |
949 | ||
5ffac122 | 950 | pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); |
1da177e4 | 951 | |
5ffac122 | 952 | if (head == tail) |
1da177e4 LT |
953 | goto out; |
954 | ||
a31ad380 KO |
955 | while (ret < nr) { |
956 | long avail; | |
957 | struct io_event *ev; | |
958 | struct page *page; | |
959 | ||
5ffac122 KO |
960 | avail = (head <= tail ? tail : ctx->nr_events) - head; |
961 | if (head == tail) | |
a31ad380 KO |
962 | break; |
963 | ||
964 | avail = min(avail, nr - ret); | |
965 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
966 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
967 | ||
968 | pos = head + AIO_EVENTS_OFFSET; | |
58c85dc2 | 969 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
a31ad380 KO |
970 | pos %= AIO_EVENTS_PER_PAGE; |
971 | ||
972 | ev = kmap(page); | |
973 | copy_ret = copy_to_user(event + ret, ev + pos, | |
974 | sizeof(*ev) * avail); | |
975 | kunmap(page); | |
976 | ||
977 | if (unlikely(copy_ret)) { | |
978 | ret = -EFAULT; | |
979 | goto out; | |
980 | } | |
981 | ||
982 | ret += avail; | |
983 | head += avail; | |
58c85dc2 | 984 | head %= ctx->nr_events; |
1da177e4 | 985 | } |
1da177e4 | 986 | |
58c85dc2 | 987 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 988 | ring->head = head; |
91d80a84 | 989 | kunmap_atomic(ring); |
58c85dc2 | 990 | flush_dcache_page(ctx->ring_pages[0]); |
a31ad380 | 991 | |
5ffac122 | 992 | pr_debug("%li h%u t%u\n", ret, head, tail); |
3e845ce0 | 993 | |
e1bdd5f2 | 994 | put_reqs_available(ctx, ret); |
a31ad380 | 995 | out: |
58c85dc2 | 996 | mutex_unlock(&ctx->ring_lock); |
a31ad380 | 997 | |
1da177e4 LT |
998 | return ret; |
999 | } | |
1000 | ||
a31ad380 KO |
1001 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
1002 | struct io_event __user *event, long *i) | |
1da177e4 | 1003 | { |
a31ad380 | 1004 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 1005 | |
a31ad380 KO |
1006 | if (ret > 0) |
1007 | *i += ret; | |
1da177e4 | 1008 | |
a31ad380 KO |
1009 | if (unlikely(atomic_read(&ctx->dead))) |
1010 | ret = -EINVAL; | |
1da177e4 | 1011 | |
a31ad380 KO |
1012 | if (!*i) |
1013 | *i = ret; | |
1da177e4 | 1014 | |
a31ad380 | 1015 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
1016 | } |
1017 | ||
a31ad380 | 1018 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
1019 | struct io_event __user *event, |
1020 | struct timespec __user *timeout) | |
1021 | { | |
a31ad380 KO |
1022 | ktime_t until = { .tv64 = KTIME_MAX }; |
1023 | long ret = 0; | |
1da177e4 | 1024 | |
1da177e4 LT |
1025 | if (timeout) { |
1026 | struct timespec ts; | |
a31ad380 | 1027 | |
1da177e4 | 1028 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 1029 | return -EFAULT; |
1da177e4 | 1030 | |
a31ad380 | 1031 | until = timespec_to_ktime(ts); |
1da177e4 LT |
1032 | } |
1033 | ||
a31ad380 KO |
1034 | /* |
1035 | * Note that aio_read_events() is being called as the conditional - i.e. | |
1036 | * we're calling it after prepare_to_wait() has set task state to | |
1037 | * TASK_INTERRUPTIBLE. | |
1038 | * | |
1039 | * But aio_read_events() can block, and if it blocks it's going to flip | |
1040 | * the task state back to TASK_RUNNING. | |
1041 | * | |
1042 | * This should be ok, provided it doesn't flip the state back to | |
1043 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
1044 | * will only happen if the mutex_lock() call blocks, and we then find | |
1045 | * the ringbuffer empty. So in practice we should be ok, but it's | |
1046 | * something to be aware of when touching this code. | |
1047 | */ | |
1048 | wait_event_interruptible_hrtimeout(ctx->wait, | |
1049 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
1da177e4 | 1050 | |
a31ad380 KO |
1051 | if (!ret && signal_pending(current)) |
1052 | ret = -EINTR; | |
1da177e4 | 1053 | |
a31ad380 | 1054 | return ret; |
1da177e4 LT |
1055 | } |
1056 | ||
1da177e4 LT |
1057 | /* sys_io_setup: |
1058 | * Create an aio_context capable of receiving at least nr_events. | |
1059 | * ctxp must not point to an aio_context that already exists, and | |
1060 | * must be initialized to 0 prior to the call. On successful | |
1061 | * creation of the aio_context, *ctxp is filled in with the resulting | |
1062 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
1063 | * if the specified nr_events exceeds internal limits. May fail | |
1064 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
1065 | * of available events. May fail with -ENOMEM if insufficient kernel | |
1066 | * resources are available. May fail with -EFAULT if an invalid | |
1067 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
1068 | * implemented. | |
1069 | */ | |
002c8976 | 1070 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
1071 | { |
1072 | struct kioctx *ioctx = NULL; | |
1073 | unsigned long ctx; | |
1074 | long ret; | |
1075 | ||
1076 | ret = get_user(ctx, ctxp); | |
1077 | if (unlikely(ret)) | |
1078 | goto out; | |
1079 | ||
1080 | ret = -EINVAL; | |
d55b5fda ZB |
1081 | if (unlikely(ctx || nr_events == 0)) { |
1082 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
1083 | ctx, nr_events); | |
1da177e4 LT |
1084 | goto out; |
1085 | } | |
1086 | ||
1087 | ioctx = ioctx_alloc(nr_events); | |
1088 | ret = PTR_ERR(ioctx); | |
1089 | if (!IS_ERR(ioctx)) { | |
1090 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a | 1091 | if (ret) |
db446a08 | 1092 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1093 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1094 | } |
1095 | ||
1096 | out: | |
1097 | return ret; | |
1098 | } | |
1099 | ||
1100 | /* sys_io_destroy: | |
1101 | * Destroy the aio_context specified. May cancel any outstanding | |
1102 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 1103 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
1104 | * is invalid. |
1105 | */ | |
002c8976 | 1106 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
1107 | { |
1108 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
1109 | if (likely(NULL != ioctx)) { | |
db446a08 | 1110 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1111 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1112 | return 0; |
1113 | } | |
1114 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
1115 | return -EINVAL; | |
1116 | } | |
1117 | ||
41ef4eb8 KO |
1118 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
1119 | unsigned long, loff_t); | |
1120 | ||
8bc92afc KO |
1121 | static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb, |
1122 | int rw, char __user *buf, | |
1123 | unsigned long *nr_segs, | |
1124 | struct iovec **iovec, | |
1125 | bool compat) | |
eed4e51f BP |
1126 | { |
1127 | ssize_t ret; | |
1128 | ||
8bc92afc | 1129 | *nr_segs = kiocb->ki_nbytes; |
41ef4eb8 | 1130 | |
9d85cba7 JM |
1131 | #ifdef CONFIG_COMPAT |
1132 | if (compat) | |
41ef4eb8 | 1133 | ret = compat_rw_copy_check_uvector(rw, |
8bc92afc KO |
1134 | (struct compat_iovec __user *)buf, |
1135 | *nr_segs, 1, *iovec, iovec); | |
9d85cba7 JM |
1136 | else |
1137 | #endif | |
41ef4eb8 | 1138 | ret = rw_copy_check_uvector(rw, |
8bc92afc KO |
1139 | (struct iovec __user *)buf, |
1140 | *nr_segs, 1, *iovec, iovec); | |
eed4e51f | 1141 | if (ret < 0) |
41ef4eb8 | 1142 | return ret; |
a70b52ec | 1143 | |
41ef4eb8 | 1144 | /* ki_nbytes now reflect bytes instead of segs */ |
eed4e51f | 1145 | kiocb->ki_nbytes = ret; |
41ef4eb8 | 1146 | return 0; |
eed4e51f BP |
1147 | } |
1148 | ||
8bc92afc KO |
1149 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb, |
1150 | int rw, char __user *buf, | |
1151 | unsigned long *nr_segs, | |
1152 | struct iovec *iovec) | |
eed4e51f | 1153 | { |
8bc92afc | 1154 | if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes))) |
41ef4eb8 | 1155 | return -EFAULT; |
a70b52ec | 1156 | |
8bc92afc KO |
1157 | iovec->iov_base = buf; |
1158 | iovec->iov_len = kiocb->ki_nbytes; | |
1159 | *nr_segs = 1; | |
eed4e51f BP |
1160 | return 0; |
1161 | } | |
1162 | ||
1da177e4 LT |
1163 | /* |
1164 | * aio_setup_iocb: | |
1165 | * Performs the initial checks and aio retry method | |
1166 | * setup for the kiocb at the time of io submission. | |
1167 | */ | |
8bc92afc KO |
1168 | static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode, |
1169 | char __user *buf, bool compat) | |
1da177e4 | 1170 | { |
41ef4eb8 KO |
1171 | struct file *file = req->ki_filp; |
1172 | ssize_t ret; | |
8bc92afc | 1173 | unsigned long nr_segs; |
41ef4eb8 KO |
1174 | int rw; |
1175 | fmode_t mode; | |
1176 | aio_rw_op *rw_op; | |
8bc92afc | 1177 | struct iovec inline_vec, *iovec = &inline_vec; |
1da177e4 | 1178 | |
8bc92afc | 1179 | switch (opcode) { |
1da177e4 | 1180 | case IOCB_CMD_PREAD: |
eed4e51f | 1181 | case IOCB_CMD_PREADV: |
41ef4eb8 KO |
1182 | mode = FMODE_READ; |
1183 | rw = READ; | |
1184 | rw_op = file->f_op->aio_read; | |
1185 | goto rw_common; | |
1186 | ||
1187 | case IOCB_CMD_PWRITE: | |
eed4e51f | 1188 | case IOCB_CMD_PWRITEV: |
41ef4eb8 KO |
1189 | mode = FMODE_WRITE; |
1190 | rw = WRITE; | |
1191 | rw_op = file->f_op->aio_write; | |
1192 | goto rw_common; | |
1193 | rw_common: | |
1194 | if (unlikely(!(file->f_mode & mode))) | |
1195 | return -EBADF; | |
1196 | ||
1197 | if (!rw_op) | |
1198 | return -EINVAL; | |
1199 | ||
8bc92afc KO |
1200 | ret = (opcode == IOCB_CMD_PREADV || |
1201 | opcode == IOCB_CMD_PWRITEV) | |
1202 | ? aio_setup_vectored_rw(req, rw, buf, &nr_segs, | |
1203 | &iovec, compat) | |
1204 | : aio_setup_single_vector(req, rw, buf, &nr_segs, | |
1205 | iovec); | |
eed4e51f | 1206 | if (ret) |
41ef4eb8 KO |
1207 | return ret; |
1208 | ||
1209 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); | |
8bc92afc KO |
1210 | if (ret < 0) { |
1211 | if (iovec != &inline_vec) | |
1212 | kfree(iovec); | |
41ef4eb8 | 1213 | return ret; |
8bc92afc | 1214 | } |
41ef4eb8 KO |
1215 | |
1216 | req->ki_nbytes = ret; | |
41ef4eb8 | 1217 | |
73a7075e KO |
1218 | /* XXX: move/kill - rw_verify_area()? */ |
1219 | /* This matches the pread()/pwrite() logic */ | |
1220 | if (req->ki_pos < 0) { | |
1221 | ret = -EINVAL; | |
1222 | break; | |
1223 | } | |
1224 | ||
1225 | if (rw == WRITE) | |
1226 | file_start_write(file); | |
1227 | ||
8bc92afc | 1228 | ret = rw_op(req, iovec, nr_segs, req->ki_pos); |
73a7075e KO |
1229 | |
1230 | if (rw == WRITE) | |
1231 | file_end_write(file); | |
1da177e4 | 1232 | break; |
41ef4eb8 | 1233 | |
1da177e4 | 1234 | case IOCB_CMD_FDSYNC: |
41ef4eb8 KO |
1235 | if (!file->f_op->aio_fsync) |
1236 | return -EINVAL; | |
1237 | ||
1238 | ret = file->f_op->aio_fsync(req, 1); | |
1da177e4 | 1239 | break; |
41ef4eb8 | 1240 | |
1da177e4 | 1241 | case IOCB_CMD_FSYNC: |
41ef4eb8 KO |
1242 | if (!file->f_op->aio_fsync) |
1243 | return -EINVAL; | |
1244 | ||
1245 | ret = file->f_op->aio_fsync(req, 0); | |
1da177e4 | 1246 | break; |
41ef4eb8 | 1247 | |
1da177e4 | 1248 | default: |
caf4167a | 1249 | pr_debug("EINVAL: no operation provided\n"); |
41ef4eb8 | 1250 | return -EINVAL; |
1da177e4 LT |
1251 | } |
1252 | ||
8bc92afc KO |
1253 | if (iovec != &inline_vec) |
1254 | kfree(iovec); | |
1255 | ||
41ef4eb8 KO |
1256 | if (ret != -EIOCBQUEUED) { |
1257 | /* | |
1258 | * There's no easy way to restart the syscall since other AIO's | |
1259 | * may be already running. Just fail this IO with EINTR. | |
1260 | */ | |
1261 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1262 | ret == -ERESTARTNOHAND || | |
1263 | ret == -ERESTART_RESTARTBLOCK)) | |
1264 | ret = -EINTR; | |
1265 | aio_complete(req, ret, 0); | |
1266 | } | |
1da177e4 LT |
1267 | |
1268 | return 0; | |
1269 | } | |
1270 | ||
d5470b59 | 1271 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
a1c8eae7 | 1272 | struct iocb *iocb, bool compat) |
1da177e4 LT |
1273 | { |
1274 | struct kiocb *req; | |
1da177e4 LT |
1275 | ssize_t ret; |
1276 | ||
1277 | /* enforce forwards compatibility on users */ | |
9c3060be | 1278 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1279 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1280 | return -EINVAL; |
1281 | } | |
1282 | ||
1283 | /* prevent overflows */ | |
1284 | if (unlikely( | |
1285 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1286 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1287 | ((ssize_t)iocb->aio_nbytes < 0) | |
1288 | )) { | |
1289 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1290 | return -EINVAL; | |
1291 | } | |
1292 | ||
41ef4eb8 | 1293 | req = aio_get_req(ctx); |
1d98ebfc | 1294 | if (unlikely(!req)) |
1da177e4 | 1295 | return -EAGAIN; |
1d98ebfc KO |
1296 | |
1297 | req->ki_filp = fget(iocb->aio_fildes); | |
1298 | if (unlikely(!req->ki_filp)) { | |
1299 | ret = -EBADF; | |
1300 | goto out_put_req; | |
1da177e4 | 1301 | } |
1d98ebfc | 1302 | |
9c3060be DL |
1303 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1304 | /* | |
1305 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1306 | * instance of the file* now. The file descriptor must be | |
1307 | * an eventfd() fd, and will be signaled for each completed | |
1308 | * event using the eventfd_signal() function. | |
1309 | */ | |
13389010 | 1310 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1311 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1312 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1313 | req->ki_eventfd = NULL; |
9c3060be DL |
1314 | goto out_put_req; |
1315 | } | |
1316 | } | |
1da177e4 | 1317 | |
8a660890 | 1318 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
1da177e4 | 1319 | if (unlikely(ret)) { |
caf4167a | 1320 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1321 | goto out_put_req; |
1322 | } | |
1323 | ||
1324 | req->ki_obj.user = user_iocb; | |
1325 | req->ki_user_data = iocb->aio_data; | |
1326 | req->ki_pos = iocb->aio_offset; | |
73a7075e | 1327 | req->ki_nbytes = iocb->aio_nbytes; |
1da177e4 | 1328 | |
8bc92afc KO |
1329 | ret = aio_run_iocb(req, iocb->aio_lio_opcode, |
1330 | (char __user *)(unsigned long)iocb->aio_buf, | |
1331 | compat); | |
41003a7b | 1332 | if (ret) |
7137c6bd | 1333 | goto out_put_req; |
41003a7b | 1334 | |
1da177e4 | 1335 | return 0; |
1da177e4 | 1336 | out_put_req: |
e1bdd5f2 | 1337 | put_reqs_available(ctx, 1); |
e34ecee2 | 1338 | percpu_ref_put(&ctx->reqs); |
57282d8f | 1339 | kiocb_free(req); |
1da177e4 LT |
1340 | return ret; |
1341 | } | |
1342 | ||
9d85cba7 JM |
1343 | long do_io_submit(aio_context_t ctx_id, long nr, |
1344 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1345 | { |
1346 | struct kioctx *ctx; | |
1347 | long ret = 0; | |
080d676d | 1348 | int i = 0; |
9f5b9425 | 1349 | struct blk_plug plug; |
1da177e4 LT |
1350 | |
1351 | if (unlikely(nr < 0)) | |
1352 | return -EINVAL; | |
1353 | ||
75e1c70f JM |
1354 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1355 | nr = LONG_MAX/sizeof(*iocbpp); | |
1356 | ||
1da177e4 LT |
1357 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1358 | return -EFAULT; | |
1359 | ||
1360 | ctx = lookup_ioctx(ctx_id); | |
1361 | if (unlikely(!ctx)) { | |
caf4167a | 1362 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1363 | return -EINVAL; |
1364 | } | |
1365 | ||
9f5b9425 SL |
1366 | blk_start_plug(&plug); |
1367 | ||
1da177e4 LT |
1368 | /* |
1369 | * AKPM: should this return a partial result if some of the IOs were | |
1370 | * successfully submitted? | |
1371 | */ | |
1372 | for (i=0; i<nr; i++) { | |
1373 | struct iocb __user *user_iocb; | |
1374 | struct iocb tmp; | |
1375 | ||
1376 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1377 | ret = -EFAULT; | |
1378 | break; | |
1379 | } | |
1380 | ||
1381 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1382 | ret = -EFAULT; | |
1383 | break; | |
1384 | } | |
1385 | ||
a1c8eae7 | 1386 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
1da177e4 LT |
1387 | if (ret) |
1388 | break; | |
1389 | } | |
9f5b9425 | 1390 | blk_finish_plug(&plug); |
1da177e4 | 1391 | |
723be6e3 | 1392 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1393 | return i ? i : ret; |
1394 | } | |
1395 | ||
9d85cba7 JM |
1396 | /* sys_io_submit: |
1397 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1398 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1399 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1400 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1401 | * is invalid for the file descriptor in the iocb. May fail with | |
1402 | * -EFAULT if any of the data structures point to invalid data. May | |
1403 | * fail with -EBADF if the file descriptor specified in the first | |
1404 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1405 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1406 | * fail with -ENOSYS if not implemented. | |
1407 | */ | |
1408 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1409 | struct iocb __user * __user *, iocbpp) | |
1410 | { | |
1411 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1412 | } | |
1413 | ||
1da177e4 LT |
1414 | /* lookup_kiocb |
1415 | * Finds a given iocb for cancellation. | |
1da177e4 | 1416 | */ |
25ee7e38 AB |
1417 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1418 | u32 key) | |
1da177e4 LT |
1419 | { |
1420 | struct list_head *pos; | |
d00689af ZB |
1421 | |
1422 | assert_spin_locked(&ctx->ctx_lock); | |
1423 | ||
8a660890 KO |
1424 | if (key != KIOCB_KEY) |
1425 | return NULL; | |
1426 | ||
1da177e4 LT |
1427 | /* TODO: use a hash or array, this sucks. */ |
1428 | list_for_each(pos, &ctx->active_reqs) { | |
1429 | struct kiocb *kiocb = list_kiocb(pos); | |
8a660890 | 1430 | if (kiocb->ki_obj.user == iocb) |
1da177e4 LT |
1431 | return kiocb; |
1432 | } | |
1433 | return NULL; | |
1434 | } | |
1435 | ||
1436 | /* sys_io_cancel: | |
1437 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1438 | * the operation is successfully cancelled, the resulting event is | |
1439 | * copied into the memory pointed to by result without being placed | |
1440 | * into the completion queue and 0 is returned. May fail with | |
1441 | * -EFAULT if any of the data structures pointed to are invalid. | |
1442 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1443 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1444 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1445 | */ | |
002c8976 HC |
1446 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1447 | struct io_event __user *, result) | |
1da177e4 | 1448 | { |
1da177e4 LT |
1449 | struct kioctx *ctx; |
1450 | struct kiocb *kiocb; | |
1451 | u32 key; | |
1452 | int ret; | |
1453 | ||
1454 | ret = get_user(key, &iocb->aio_key); | |
1455 | if (unlikely(ret)) | |
1456 | return -EFAULT; | |
1457 | ||
1458 | ctx = lookup_ioctx(ctx_id); | |
1459 | if (unlikely(!ctx)) | |
1460 | return -EINVAL; | |
1461 | ||
1462 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1463 | |
1da177e4 | 1464 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c | 1465 | if (kiocb) |
bec68faa | 1466 | ret = kiocb_cancel(ctx, kiocb); |
906b973c KO |
1467 | else |
1468 | ret = -EINVAL; | |
1469 | ||
1da177e4 LT |
1470 | spin_unlock_irq(&ctx->ctx_lock); |
1471 | ||
906b973c | 1472 | if (!ret) { |
bec68faa KO |
1473 | /* |
1474 | * The result argument is no longer used - the io_event is | |
1475 | * always delivered via the ring buffer. -EINPROGRESS indicates | |
1476 | * cancellation is progress: | |
906b973c | 1477 | */ |
bec68faa | 1478 | ret = -EINPROGRESS; |
906b973c | 1479 | } |
1da177e4 | 1480 | |
723be6e3 | 1481 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1482 | |
1483 | return ret; | |
1484 | } | |
1485 | ||
1486 | /* io_getevents: | |
1487 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1488 | * the completion queue for the aio_context specified by ctx_id. If |
1489 | * it succeeds, the number of read events is returned. May fail with | |
1490 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1491 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1492 | * if any of the memory specified is invalid. May return 0 or | |
1493 | * < min_nr if the timeout specified by timeout has elapsed | |
1494 | * before sufficient events are available, where timeout == NULL | |
1495 | * specifies an infinite timeout. Note that the timeout pointed to by | |
6900807c | 1496 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
1da177e4 | 1497 | */ |
002c8976 HC |
1498 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1499 | long, min_nr, | |
1500 | long, nr, | |
1501 | struct io_event __user *, events, | |
1502 | struct timespec __user *, timeout) | |
1da177e4 LT |
1503 | { |
1504 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1505 | long ret = -EINVAL; | |
1506 | ||
1507 | if (likely(ioctx)) { | |
2e410255 | 1508 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 | 1509 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
723be6e3 | 1510 | percpu_ref_put(&ioctx->users); |
1da177e4 | 1511 | } |
1da177e4 LT |
1512 | return ret; |
1513 | } |