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> | |
1da177e4 LT |
42 | |
43 | #include <asm/kmap_types.h> | |
44 | #include <asm/uaccess.h> | |
1da177e4 | 45 | |
68d70d03 AV |
46 | #include "internal.h" |
47 | ||
4e179bca KO |
48 | #define AIO_RING_MAGIC 0xa10a10a1 |
49 | #define AIO_RING_COMPAT_FEATURES 1 | |
50 | #define AIO_RING_INCOMPAT_FEATURES 0 | |
51 | struct aio_ring { | |
52 | unsigned id; /* kernel internal index number */ | |
53 | unsigned nr; /* number of io_events */ | |
54 | unsigned head; | |
55 | unsigned tail; | |
56 | ||
57 | unsigned magic; | |
58 | unsigned compat_features; | |
59 | unsigned incompat_features; | |
60 | unsigned header_length; /* size of aio_ring */ | |
61 | ||
62 | ||
63 | struct io_event io_events[0]; | |
64 | }; /* 128 bytes + ring size */ | |
65 | ||
66 | #define AIO_RING_PAGES 8 | |
4e179bca | 67 | |
e1bdd5f2 KO |
68 | struct kioctx_cpu { |
69 | unsigned reqs_available; | |
70 | }; | |
71 | ||
4e179bca KO |
72 | struct kioctx { |
73 | atomic_t users; | |
36f55889 | 74 | atomic_t dead; |
4e179bca KO |
75 | |
76 | /* This needs improving */ | |
77 | unsigned long user_id; | |
78 | struct hlist_node list; | |
79 | ||
e1bdd5f2 KO |
80 | struct __percpu kioctx_cpu *cpu; |
81 | ||
82 | /* | |
83 | * For percpu reqs_available, number of slots we move to/from global | |
84 | * counter at a time: | |
85 | */ | |
86 | unsigned req_batch; | |
3e845ce0 KO |
87 | /* |
88 | * This is what userspace passed to io_setup(), it's not used for | |
89 | * anything but counting against the global max_reqs quota. | |
90 | * | |
58c85dc2 | 91 | * The real limit is nr_events - 1, which will be larger (see |
3e845ce0 KO |
92 | * aio_setup_ring()) |
93 | */ | |
4e179bca KO |
94 | unsigned max_reqs; |
95 | ||
58c85dc2 KO |
96 | /* Size of ringbuffer, in units of struct io_event */ |
97 | unsigned nr_events; | |
4e179bca | 98 | |
58c85dc2 KO |
99 | unsigned long mmap_base; |
100 | unsigned long mmap_size; | |
101 | ||
102 | struct page **ring_pages; | |
103 | long nr_pages; | |
104 | ||
4e23bcae KO |
105 | struct rcu_head rcu_head; |
106 | struct work_struct rcu_work; | |
107 | ||
108 | struct { | |
34e83fc6 KO |
109 | /* |
110 | * This counts the number of available slots in the ringbuffer, | |
111 | * so we avoid overflowing it: it's decremented (if positive) | |
112 | * when allocating a kiocb and incremented when the resulting | |
113 | * io_event is pulled off the ringbuffer. | |
e1bdd5f2 KO |
114 | * |
115 | * We batch accesses to it with a percpu version. | |
34e83fc6 KO |
116 | */ |
117 | atomic_t reqs_available; | |
4e23bcae KO |
118 | } ____cacheline_aligned_in_smp; |
119 | ||
120 | struct { | |
121 | spinlock_t ctx_lock; | |
122 | struct list_head active_reqs; /* used for cancellation */ | |
123 | } ____cacheline_aligned_in_smp; | |
124 | ||
58c85dc2 KO |
125 | struct { |
126 | struct mutex ring_lock; | |
4e23bcae KO |
127 | wait_queue_head_t wait; |
128 | } ____cacheline_aligned_in_smp; | |
58c85dc2 KO |
129 | |
130 | struct { | |
131 | unsigned tail; | |
132 | spinlock_t completion_lock; | |
4e23bcae | 133 | } ____cacheline_aligned_in_smp; |
58c85dc2 KO |
134 | |
135 | struct page *internal_pages[AIO_RING_PAGES]; | |
36bc08cc | 136 | struct file *aio_ring_file; |
4e179bca KO |
137 | }; |
138 | ||
1da177e4 | 139 | /*------ sysctl variables----*/ |
d55b5fda ZB |
140 | static DEFINE_SPINLOCK(aio_nr_lock); |
141 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
142 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
143 | /*----end sysctl variables---*/ |
144 | ||
e18b890b CL |
145 | static struct kmem_cache *kiocb_cachep; |
146 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 147 | |
1da177e4 LT |
148 | /* aio_setup |
149 | * Creates the slab caches used by the aio routines, panic on | |
150 | * failure as this is done early during the boot sequence. | |
151 | */ | |
152 | static int __init aio_setup(void) | |
153 | { | |
0a31bd5f CL |
154 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
155 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 156 | |
caf4167a | 157 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
158 | |
159 | return 0; | |
160 | } | |
385773e0 | 161 | __initcall(aio_setup); |
1da177e4 LT |
162 | |
163 | static void aio_free_ring(struct kioctx *ctx) | |
164 | { | |
36bc08cc GZ |
165 | int i; |
166 | struct file *aio_ring_file = ctx->aio_ring_file; | |
1da177e4 | 167 | |
36bc08cc GZ |
168 | for (i = 0; i < ctx->nr_pages; i++) { |
169 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, | |
170 | page_count(ctx->ring_pages[i])); | |
58c85dc2 | 171 | put_page(ctx->ring_pages[i]); |
36bc08cc | 172 | } |
1da177e4 | 173 | |
58c85dc2 KO |
174 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) |
175 | kfree(ctx->ring_pages); | |
36bc08cc GZ |
176 | |
177 | if (aio_ring_file) { | |
178 | truncate_setsize(aio_ring_file->f_inode, 0); | |
179 | pr_debug("pid(%d) i_nlink=%u d_count=%d d_unhashed=%d i_count=%d\n", | |
180 | current->pid, aio_ring_file->f_inode->i_nlink, | |
181 | aio_ring_file->f_path.dentry->d_count, | |
182 | d_unhashed(aio_ring_file->f_path.dentry), | |
183 | atomic_read(&aio_ring_file->f_inode->i_count)); | |
184 | fput(aio_ring_file); | |
185 | ctx->aio_ring_file = NULL; | |
186 | } | |
187 | } | |
188 | ||
189 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) | |
190 | { | |
191 | vma->vm_ops = &generic_file_vm_ops; | |
192 | return 0; | |
193 | } | |
194 | ||
195 | static const struct file_operations aio_ring_fops = { | |
196 | .mmap = aio_ring_mmap, | |
197 | }; | |
198 | ||
199 | static int aio_set_page_dirty(struct page *page) | |
200 | { | |
201 | return 0; | |
202 | } | |
203 | ||
0c45355f | 204 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc GZ |
205 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
206 | struct page *old, enum migrate_mode mode) | |
207 | { | |
208 | struct kioctx *ctx = mapping->private_data; | |
209 | unsigned long flags; | |
210 | unsigned idx = old->index; | |
211 | int rc; | |
212 | ||
213 | /* Writeback must be complete */ | |
214 | BUG_ON(PageWriteback(old)); | |
215 | put_page(old); | |
216 | ||
217 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode); | |
218 | if (rc != MIGRATEPAGE_SUCCESS) { | |
219 | get_page(old); | |
220 | return rc; | |
221 | } | |
222 | ||
223 | get_page(new); | |
224 | ||
225 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
226 | migrate_page_copy(new, old); | |
227 | ctx->ring_pages[idx] = new; | |
228 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
229 | ||
230 | return rc; | |
1da177e4 | 231 | } |
0c45355f | 232 | #endif |
1da177e4 | 233 | |
36bc08cc GZ |
234 | static const struct address_space_operations aio_ctx_aops = { |
235 | .set_page_dirty = aio_set_page_dirty, | |
0c45355f | 236 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc | 237 | .migratepage = aio_migratepage, |
0c45355f | 238 | #endif |
36bc08cc GZ |
239 | }; |
240 | ||
1da177e4 LT |
241 | static int aio_setup_ring(struct kioctx *ctx) |
242 | { | |
243 | struct aio_ring *ring; | |
1da177e4 | 244 | unsigned nr_events = ctx->max_reqs; |
41003a7b | 245 | struct mm_struct *mm = current->mm; |
41badc15 | 246 | unsigned long size, populate; |
1da177e4 | 247 | int nr_pages; |
36bc08cc GZ |
248 | int i; |
249 | struct file *file; | |
1da177e4 LT |
250 | |
251 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
252 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
253 | ||
254 | size = sizeof(struct aio_ring); | |
255 | size += sizeof(struct io_event) * nr_events; | |
1da177e4 | 256 | |
36bc08cc | 257 | nr_pages = PFN_UP(size); |
1da177e4 LT |
258 | if (nr_pages < 0) |
259 | return -EINVAL; | |
260 | ||
36bc08cc GZ |
261 | file = anon_inode_getfile_private("[aio]", &aio_ring_fops, ctx, O_RDWR); |
262 | if (IS_ERR(file)) { | |
263 | ctx->aio_ring_file = NULL; | |
264 | return -EAGAIN; | |
265 | } | |
266 | ||
267 | file->f_inode->i_mapping->a_ops = &aio_ctx_aops; | |
268 | file->f_inode->i_mapping->private_data = ctx; | |
269 | file->f_inode->i_size = PAGE_SIZE * (loff_t)nr_pages; | |
270 | ||
271 | for (i = 0; i < nr_pages; i++) { | |
272 | struct page *page; | |
273 | page = find_or_create_page(file->f_inode->i_mapping, | |
274 | i, GFP_HIGHUSER | __GFP_ZERO); | |
275 | if (!page) | |
276 | break; | |
277 | pr_debug("pid(%d) page[%d]->count=%d\n", | |
278 | current->pid, i, page_count(page)); | |
279 | SetPageUptodate(page); | |
280 | SetPageDirty(page); | |
281 | unlock_page(page); | |
282 | } | |
283 | ctx->aio_ring_file = file; | |
284 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) | |
285 | / sizeof(struct io_event); | |
1da177e4 | 286 | |
58c85dc2 | 287 | ctx->ring_pages = ctx->internal_pages; |
1da177e4 | 288 | if (nr_pages > AIO_RING_PAGES) { |
58c85dc2 KO |
289 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
290 | GFP_KERNEL); | |
291 | if (!ctx->ring_pages) | |
1da177e4 | 292 | return -ENOMEM; |
1da177e4 LT |
293 | } |
294 | ||
58c85dc2 KO |
295 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
296 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
36bc08cc | 297 | |
41003a7b | 298 | down_write(&mm->mmap_sem); |
36bc08cc GZ |
299 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
300 | PROT_READ | PROT_WRITE, | |
301 | MAP_SHARED | MAP_POPULATE, 0, &populate); | |
58c85dc2 | 302 | if (IS_ERR((void *)ctx->mmap_base)) { |
41003a7b | 303 | up_write(&mm->mmap_sem); |
58c85dc2 | 304 | ctx->mmap_size = 0; |
1da177e4 LT |
305 | aio_free_ring(ctx); |
306 | return -EAGAIN; | |
307 | } | |
36bc08cc GZ |
308 | up_write(&mm->mmap_sem); |
309 | ||
310 | mm_populate(ctx->mmap_base, populate); | |
1da177e4 | 311 | |
58c85dc2 KO |
312 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
313 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, | |
314 | 1, 0, ctx->ring_pages, NULL); | |
36bc08cc GZ |
315 | for (i = 0; i < ctx->nr_pages; i++) |
316 | put_page(ctx->ring_pages[i]); | |
1da177e4 | 317 | |
58c85dc2 | 318 | if (unlikely(ctx->nr_pages != nr_pages)) { |
1da177e4 LT |
319 | aio_free_ring(ctx); |
320 | return -EAGAIN; | |
321 | } | |
322 | ||
58c85dc2 KO |
323 | ctx->user_id = ctx->mmap_base; |
324 | ctx->nr_events = nr_events; /* trusted copy */ | |
1da177e4 | 325 | |
58c85dc2 | 326 | ring = kmap_atomic(ctx->ring_pages[0]); |
1da177e4 LT |
327 | ring->nr = nr_events; /* user copy */ |
328 | ring->id = ctx->user_id; | |
329 | ring->head = ring->tail = 0; | |
330 | ring->magic = AIO_RING_MAGIC; | |
331 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
332 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
333 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 334 | kunmap_atomic(ring); |
58c85dc2 | 335 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 LT |
336 | |
337 | return 0; | |
338 | } | |
339 | ||
1da177e4 LT |
340 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
341 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
342 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
343 | ||
0460fef2 KO |
344 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
345 | { | |
346 | struct kioctx *ctx = req->ki_ctx; | |
347 | unsigned long flags; | |
348 | ||
349 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
350 | ||
351 | if (!req->ki_list.next) | |
352 | list_add(&req->ki_list, &ctx->active_reqs); | |
353 | ||
354 | req->ki_cancel = cancel; | |
355 | ||
356 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
357 | } | |
358 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
359 | ||
906b973c KO |
360 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, |
361 | struct io_event *res) | |
362 | { | |
0460fef2 | 363 | kiocb_cancel_fn *old, *cancel; |
906b973c KO |
364 | int ret = -EINVAL; |
365 | ||
0460fef2 KO |
366 | /* |
367 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
368 | * actually has a cancel function, hence the cmpxchg() | |
369 | */ | |
370 | ||
371 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
372 | do { | |
373 | if (!cancel || cancel == KIOCB_CANCELLED) | |
374 | return ret; | |
906b973c | 375 | |
0460fef2 KO |
376 | old = cancel; |
377 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
378 | } while (cancel != old); | |
906b973c | 379 | |
0460fef2 KO |
380 | atomic_inc(&kiocb->ki_users); |
381 | spin_unlock_irq(&ctx->ctx_lock); | |
382 | ||
383 | memset(res, 0, sizeof(*res)); | |
384 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; | |
385 | res->data = kiocb->ki_user_data; | |
386 | ret = cancel(kiocb, res); | |
387 | ||
388 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c KO |
389 | |
390 | return ret; | |
391 | } | |
392 | ||
36f55889 KO |
393 | static void free_ioctx_rcu(struct rcu_head *head) |
394 | { | |
395 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
e1bdd5f2 KO |
396 | |
397 | free_percpu(ctx->cpu); | |
36f55889 KO |
398 | kmem_cache_free(kioctx_cachep, ctx); |
399 | } | |
400 | ||
401 | /* | |
402 | * When this function runs, the kioctx has been removed from the "hash table" | |
403 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
404 | * now it's safe to cancel any that need to be. | |
405 | */ | |
406 | static void free_ioctx(struct kioctx *ctx) | |
407 | { | |
3e845ce0 | 408 | struct aio_ring *ring; |
36f55889 KO |
409 | struct io_event res; |
410 | struct kiocb *req; | |
e1bdd5f2 | 411 | unsigned cpu, head, avail; |
36f55889 KO |
412 | |
413 | spin_lock_irq(&ctx->ctx_lock); | |
414 | ||
415 | while (!list_empty(&ctx->active_reqs)) { | |
416 | req = list_first_entry(&ctx->active_reqs, | |
417 | struct kiocb, ki_list); | |
418 | ||
419 | list_del_init(&req->ki_list); | |
420 | kiocb_cancel(ctx, req, &res); | |
421 | } | |
422 | ||
423 | spin_unlock_irq(&ctx->ctx_lock); | |
424 | ||
e1bdd5f2 KO |
425 | for_each_possible_cpu(cpu) { |
426 | struct kioctx_cpu *kcpu = per_cpu_ptr(ctx->cpu, cpu); | |
427 | ||
428 | atomic_add(kcpu->reqs_available, &ctx->reqs_available); | |
429 | kcpu->reqs_available = 0; | |
430 | } | |
431 | ||
58c85dc2 | 432 | ring = kmap_atomic(ctx->ring_pages[0]); |
3e845ce0 KO |
433 | head = ring->head; |
434 | kunmap_atomic(ring); | |
435 | ||
34e83fc6 | 436 | while (atomic_read(&ctx->reqs_available) < ctx->nr_events - 1) { |
03e04f04 | 437 | wait_event(ctx->wait, |
34e83fc6 KO |
438 | (head != ctx->tail) || |
439 | (atomic_read(&ctx->reqs_available) >= | |
440 | ctx->nr_events - 1)); | |
3e845ce0 | 441 | |
58c85dc2 | 442 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; |
3e845ce0 | 443 | |
34e83fc6 | 444 | atomic_add(avail, &ctx->reqs_available); |
3e845ce0 | 445 | head += avail; |
58c85dc2 | 446 | head %= ctx->nr_events; |
3e845ce0 KO |
447 | } |
448 | ||
34e83fc6 | 449 | WARN_ON(atomic_read(&ctx->reqs_available) > ctx->nr_events - 1); |
36f55889 KO |
450 | |
451 | aio_free_ring(ctx); | |
452 | ||
36f55889 KO |
453 | pr_debug("freeing %p\n", ctx); |
454 | ||
455 | /* | |
456 | * Here the call_rcu() is between the wait_event() for reqs_active to | |
457 | * hit 0, and freeing the ioctx. | |
458 | * | |
459 | * aio_complete() decrements reqs_active, but it has to touch the ioctx | |
460 | * after to issue a wakeup so we use rcu. | |
461 | */ | |
462 | call_rcu(&ctx->rcu_head, free_ioctx_rcu); | |
463 | } | |
464 | ||
465 | static void put_ioctx(struct kioctx *ctx) | |
466 | { | |
467 | if (unlikely(atomic_dec_and_test(&ctx->users))) | |
468 | free_ioctx(ctx); | |
469 | } | |
470 | ||
1da177e4 LT |
471 | /* ioctx_alloc |
472 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
473 | */ | |
474 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
475 | { | |
41003a7b | 476 | struct mm_struct *mm = current->mm; |
1da177e4 | 477 | struct kioctx *ctx; |
e23754f8 | 478 | int err = -ENOMEM; |
1da177e4 | 479 | |
e1bdd5f2 KO |
480 | /* |
481 | * We keep track of the number of available ringbuffer slots, to prevent | |
482 | * overflow (reqs_available), and we also use percpu counters for this. | |
483 | * | |
484 | * So since up to half the slots might be on other cpu's percpu counters | |
485 | * and unavailable, double nr_events so userspace sees what they | |
486 | * expected: additionally, we move req_batch slots to/from percpu | |
487 | * counters at a time, so make sure that isn't 0: | |
488 | */ | |
489 | nr_events = max(nr_events, num_possible_cpus() * 4); | |
490 | nr_events *= 2; | |
491 | ||
1da177e4 LT |
492 | /* Prevent overflows */ |
493 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
494 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
495 | pr_debug("ENOMEM: nr_events too high\n"); | |
496 | return ERR_PTR(-EINVAL); | |
497 | } | |
498 | ||
2dd542b7 | 499 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
1da177e4 LT |
500 | return ERR_PTR(-EAGAIN); |
501 | ||
c3762229 | 502 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
503 | if (!ctx) |
504 | return ERR_PTR(-ENOMEM); | |
505 | ||
1da177e4 | 506 | ctx->max_reqs = nr_events; |
1da177e4 | 507 | |
86b62a2c | 508 | atomic_set(&ctx->users, 2); |
36f55889 | 509 | atomic_set(&ctx->dead, 0); |
1da177e4 | 510 | spin_lock_init(&ctx->ctx_lock); |
0460fef2 | 511 | spin_lock_init(&ctx->completion_lock); |
58c85dc2 | 512 | mutex_init(&ctx->ring_lock); |
1da177e4 LT |
513 | init_waitqueue_head(&ctx->wait); |
514 | ||
515 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 | 516 | |
e1bdd5f2 KO |
517 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
518 | if (!ctx->cpu) | |
1da177e4 LT |
519 | goto out_freectx; |
520 | ||
e1bdd5f2 KO |
521 | if (aio_setup_ring(ctx) < 0) |
522 | goto out_freepcpu; | |
523 | ||
34e83fc6 | 524 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
e1bdd5f2 KO |
525 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
526 | BUG_ON(!ctx->req_batch); | |
34e83fc6 | 527 | |
1da177e4 | 528 | /* limit the number of system wide aios */ |
9fa1cb39 | 529 | spin_lock(&aio_nr_lock); |
2dd542b7 AV |
530 | if (aio_nr + nr_events > aio_max_nr || |
531 | aio_nr + nr_events < aio_nr) { | |
9fa1cb39 | 532 | spin_unlock(&aio_nr_lock); |
1da177e4 | 533 | goto out_cleanup; |
2dd542b7 AV |
534 | } |
535 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 536 | spin_unlock(&aio_nr_lock); |
1da177e4 | 537 | |
39fa0031 | 538 | /* now link into global list. */ |
abf137dd JA |
539 | spin_lock(&mm->ioctx_lock); |
540 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); | |
541 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 | 542 | |
caf4167a | 543 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
58c85dc2 | 544 | ctx, ctx->user_id, mm, ctx->nr_events); |
1da177e4 LT |
545 | return ctx; |
546 | ||
547 | out_cleanup: | |
e23754f8 AV |
548 | err = -EAGAIN; |
549 | aio_free_ring(ctx); | |
e1bdd5f2 KO |
550 | out_freepcpu: |
551 | free_percpu(ctx->cpu); | |
1da177e4 | 552 | out_freectx: |
36bc08cc GZ |
553 | if (ctx->aio_ring_file) |
554 | fput(ctx->aio_ring_file); | |
1da177e4 | 555 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 556 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 557 | return ERR_PTR(err); |
1da177e4 LT |
558 | } |
559 | ||
36f55889 | 560 | static void kill_ioctx_work(struct work_struct *work) |
1da177e4 | 561 | { |
36f55889 | 562 | struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); |
06af121e | 563 | |
36f55889 KO |
564 | wake_up_all(&ctx->wait); |
565 | put_ioctx(ctx); | |
566 | } | |
906b973c | 567 | |
36f55889 KO |
568 | static void kill_ioctx_rcu(struct rcu_head *head) |
569 | { | |
570 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
1da177e4 | 571 | |
36f55889 KO |
572 | INIT_WORK(&ctx->rcu_work, kill_ioctx_work); |
573 | schedule_work(&ctx->rcu_work); | |
574 | } | |
1da177e4 | 575 | |
36f55889 KO |
576 | /* kill_ioctx |
577 | * Cancels all outstanding aio requests on an aio context. Used | |
578 | * when the processes owning a context have all exited to encourage | |
579 | * the rapid destruction of the kioctx. | |
580 | */ | |
581 | static void kill_ioctx(struct kioctx *ctx) | |
582 | { | |
583 | if (!atomic_xchg(&ctx->dead, 1)) { | |
584 | hlist_del_rcu(&ctx->list); | |
dee11c23 | 585 | |
36f55889 | 586 | /* |
4fcc712f KO |
587 | * It'd be more correct to do this in free_ioctx(), after all |
588 | * the outstanding kiocbs have finished - but by then io_destroy | |
589 | * has already returned, so io_setup() could potentially return | |
590 | * -EAGAIN with no ioctxs actually in use (as far as userspace | |
591 | * could tell). | |
36f55889 | 592 | */ |
4fcc712f KO |
593 | spin_lock(&aio_nr_lock); |
594 | BUG_ON(aio_nr - ctx->max_reqs > aio_nr); | |
595 | aio_nr -= ctx->max_reqs; | |
596 | spin_unlock(&aio_nr_lock); | |
597 | ||
598 | if (ctx->mmap_size) | |
599 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
600 | ||
601 | /* Between hlist_del_rcu() and dropping the initial ref */ | |
602 | call_rcu(&ctx->rcu_head, kill_ioctx_rcu); | |
36f55889 | 603 | } |
1da177e4 LT |
604 | } |
605 | ||
606 | /* wait_on_sync_kiocb: | |
607 | * Waits on the given sync kiocb to complete. | |
608 | */ | |
fc9b52cd | 609 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
1da177e4 | 610 | { |
11599eba | 611 | while (atomic_read(&iocb->ki_users)) { |
1da177e4 | 612 | set_current_state(TASK_UNINTERRUPTIBLE); |
11599eba | 613 | if (!atomic_read(&iocb->ki_users)) |
1da177e4 | 614 | break; |
41d10da3 | 615 | io_schedule(); |
1da177e4 LT |
616 | } |
617 | __set_current_state(TASK_RUNNING); | |
618 | return iocb->ki_user_data; | |
619 | } | |
385773e0 | 620 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 621 | |
36f55889 KO |
622 | /* |
623 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
624 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
625 | * called on the context. | |
626 | * | |
627 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
628 | * them. | |
1da177e4 | 629 | */ |
fc9b52cd | 630 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 631 | { |
abf137dd | 632 | struct kioctx *ctx; |
36f55889 | 633 | struct hlist_node *n; |
abf137dd | 634 | |
36f55889 | 635 | hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { |
1da177e4 LT |
636 | if (1 != atomic_read(&ctx->users)) |
637 | printk(KERN_DEBUG | |
638 | "exit_aio:ioctx still alive: %d %d %d\n", | |
36f55889 KO |
639 | atomic_read(&ctx->users), |
640 | atomic_read(&ctx->dead), | |
34e83fc6 | 641 | atomic_read(&ctx->reqs_available)); |
936af157 AV |
642 | /* |
643 | * We don't need to bother with munmap() here - | |
644 | * exit_mmap(mm) is coming and it'll unmap everything. | |
645 | * Since aio_free_ring() uses non-zero ->mmap_size | |
646 | * as indicator that it needs to unmap the area, | |
647 | * just set it to 0; aio_free_ring() is the only | |
648 | * place that uses ->mmap_size, so it's safe. | |
936af157 | 649 | */ |
58c85dc2 | 650 | ctx->mmap_size = 0; |
36f55889 | 651 | |
4fcc712f | 652 | kill_ioctx(ctx); |
1da177e4 LT |
653 | } |
654 | } | |
655 | ||
e1bdd5f2 KO |
656 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
657 | { | |
658 | struct kioctx_cpu *kcpu; | |
659 | ||
660 | preempt_disable(); | |
661 | kcpu = this_cpu_ptr(ctx->cpu); | |
662 | ||
663 | kcpu->reqs_available += nr; | |
664 | while (kcpu->reqs_available >= ctx->req_batch * 2) { | |
665 | kcpu->reqs_available -= ctx->req_batch; | |
666 | atomic_add(ctx->req_batch, &ctx->reqs_available); | |
667 | } | |
668 | ||
669 | preempt_enable(); | |
670 | } | |
671 | ||
672 | static bool get_reqs_available(struct kioctx *ctx) | |
673 | { | |
674 | struct kioctx_cpu *kcpu; | |
675 | bool ret = false; | |
676 | ||
677 | preempt_disable(); | |
678 | kcpu = this_cpu_ptr(ctx->cpu); | |
679 | ||
680 | if (!kcpu->reqs_available) { | |
681 | int old, avail = atomic_read(&ctx->reqs_available); | |
682 | ||
683 | do { | |
684 | if (avail < ctx->req_batch) | |
685 | goto out; | |
686 | ||
687 | old = avail; | |
688 | avail = atomic_cmpxchg(&ctx->reqs_available, | |
689 | avail, avail - ctx->req_batch); | |
690 | } while (avail != old); | |
691 | ||
692 | kcpu->reqs_available += ctx->req_batch; | |
693 | } | |
694 | ||
695 | ret = true; | |
696 | kcpu->reqs_available--; | |
697 | out: | |
698 | preempt_enable(); | |
699 | return ret; | |
700 | } | |
701 | ||
1da177e4 | 702 | /* aio_get_req |
11599eba | 703 | * Allocate a slot for an aio request. Increments the ki_users count |
1da177e4 LT |
704 | * of the kioctx so that the kioctx stays around until all requests are |
705 | * complete. Returns NULL if no requests are free. | |
706 | * | |
11599eba | 707 | * Returns with kiocb->ki_users set to 2. The io submit code path holds |
1da177e4 LT |
708 | * an extra reference while submitting the i/o. |
709 | * This prevents races between the aio code path referencing the | |
710 | * req (after submitting it) and aio_complete() freeing the req. | |
711 | */ | |
a1c8eae7 | 712 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
1da177e4 | 713 | { |
a1c8eae7 KO |
714 | struct kiocb *req; |
715 | ||
e1bdd5f2 | 716 | if (!get_reqs_available(ctx)) |
a1c8eae7 KO |
717 | return NULL; |
718 | ||
0460fef2 | 719 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
1da177e4 | 720 | if (unlikely(!req)) |
a1c8eae7 | 721 | goto out_put; |
1da177e4 | 722 | |
11599eba | 723 | atomic_set(&req->ki_users, 2); |
1da177e4 | 724 | req->ki_ctx = ctx; |
080d676d | 725 | return req; |
a1c8eae7 | 726 | out_put: |
e1bdd5f2 | 727 | put_reqs_available(ctx, 1); |
a1c8eae7 | 728 | return NULL; |
1da177e4 LT |
729 | } |
730 | ||
11599eba | 731 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 732 | { |
1d98ebfc KO |
733 | if (req->ki_filp) |
734 | fput(req->ki_filp); | |
13389010 DL |
735 | if (req->ki_eventfd != NULL) |
736 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 LT |
737 | if (req->ki_dtor) |
738 | req->ki_dtor(req); | |
eed4e51f BP |
739 | if (req->ki_iovec != &req->ki_inline_vec) |
740 | kfree(req->ki_iovec); | |
1da177e4 | 741 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
742 | } |
743 | ||
2d68449e | 744 | void aio_put_req(struct kiocb *req) |
1da177e4 | 745 | { |
11599eba KO |
746 | if (atomic_dec_and_test(&req->ki_users)) |
747 | kiocb_free(req); | |
1da177e4 | 748 | } |
385773e0 | 749 | EXPORT_SYMBOL(aio_put_req); |
1da177e4 | 750 | |
d5470b59 | 751 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 752 | { |
abf137dd | 753 | struct mm_struct *mm = current->mm; |
65c24491 | 754 | struct kioctx *ctx, *ret = NULL; |
1da177e4 | 755 | |
abf137dd JA |
756 | rcu_read_lock(); |
757 | ||
b67bfe0d | 758 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
36f55889 KO |
759 | if (ctx->user_id == ctx_id) { |
760 | atomic_inc(&ctx->users); | |
65c24491 | 761 | ret = ctx; |
1da177e4 LT |
762 | break; |
763 | } | |
abf137dd | 764 | } |
1da177e4 | 765 | |
abf137dd | 766 | rcu_read_unlock(); |
65c24491 | 767 | return ret; |
1da177e4 LT |
768 | } |
769 | ||
1da177e4 LT |
770 | /* aio_complete |
771 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 772 | */ |
2d68449e | 773 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
774 | { |
775 | struct kioctx *ctx = iocb->ki_ctx; | |
1da177e4 | 776 | struct aio_ring *ring; |
21b40200 | 777 | struct io_event *ev_page, *event; |
1da177e4 | 778 | unsigned long flags; |
21b40200 | 779 | unsigned tail, pos; |
1da177e4 | 780 | |
20dcae32 ZB |
781 | /* |
782 | * Special case handling for sync iocbs: | |
783 | * - events go directly into the iocb for fast handling | |
784 | * - the sync task with the iocb in its stack holds the single iocb | |
785 | * ref, no other paths have a way to get another ref | |
786 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
787 | */ |
788 | if (is_sync_kiocb(iocb)) { | |
11599eba | 789 | BUG_ON(atomic_read(&iocb->ki_users) != 1); |
1da177e4 | 790 | iocb->ki_user_data = res; |
11599eba | 791 | atomic_set(&iocb->ki_users, 0); |
1da177e4 | 792 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 793 | return; |
1da177e4 LT |
794 | } |
795 | ||
36f55889 | 796 | /* |
36f55889 | 797 | * Take rcu_read_lock() in case the kioctx is being destroyed, as we |
34e83fc6 | 798 | * need to issue a wakeup after incrementing reqs_available. |
1da177e4 | 799 | */ |
36f55889 | 800 | rcu_read_lock(); |
1da177e4 | 801 | |
0460fef2 KO |
802 | if (iocb->ki_list.next) { |
803 | unsigned long flags; | |
804 | ||
805 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
806 | list_del(&iocb->ki_list); | |
807 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
808 | } | |
11599eba | 809 | |
1da177e4 LT |
810 | /* |
811 | * cancelled requests don't get events, userland was given one | |
812 | * when the event got cancelled. | |
813 | */ | |
0460fef2 | 814 | if (unlikely(xchg(&iocb->ki_cancel, |
3e845ce0 | 815 | KIOCB_CANCELLED) == KIOCB_CANCELLED)) { |
e1bdd5f2 KO |
816 | /* |
817 | * Can't use the percpu reqs_available here - could race with | |
818 | * free_ioctx() | |
819 | */ | |
34e83fc6 | 820 | atomic_inc(&ctx->reqs_available); |
3e845ce0 | 821 | /* Still need the wake_up in case free_ioctx is waiting */ |
1da177e4 | 822 | goto put_rq; |
3e845ce0 | 823 | } |
1da177e4 | 824 | |
0460fef2 KO |
825 | /* |
826 | * Add a completion event to the ring buffer. Must be done holding | |
4b30f07e | 827 | * ctx->completion_lock to prevent other code from messing with the tail |
0460fef2 KO |
828 | * pointer since we might be called from irq context. |
829 | */ | |
830 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
831 | ||
58c85dc2 | 832 | tail = ctx->tail; |
21b40200 KO |
833 | pos = tail + AIO_EVENTS_OFFSET; |
834 | ||
58c85dc2 | 835 | if (++tail >= ctx->nr_events) |
4bf69b2a | 836 | tail = 0; |
1da177e4 | 837 | |
58c85dc2 | 838 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
839 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
840 | ||
1da177e4 LT |
841 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
842 | event->data = iocb->ki_user_data; | |
843 | event->res = res; | |
844 | event->res2 = res2; | |
845 | ||
21b40200 | 846 | kunmap_atomic(ev_page); |
58c85dc2 | 847 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
848 | |
849 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
caf4167a KO |
850 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
851 | res, res2); | |
1da177e4 LT |
852 | |
853 | /* after flagging the request as done, we | |
854 | * must never even look at it again | |
855 | */ | |
856 | smp_wmb(); /* make event visible before updating tail */ | |
857 | ||
58c85dc2 | 858 | ctx->tail = tail; |
1da177e4 | 859 | |
58c85dc2 | 860 | ring = kmap_atomic(ctx->ring_pages[0]); |
21b40200 | 861 | ring->tail = tail; |
e8e3c3d6 | 862 | kunmap_atomic(ring); |
58c85dc2 | 863 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 | 864 | |
0460fef2 KO |
865 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
866 | ||
21b40200 | 867 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
8d1c98b0 DL |
868 | |
869 | /* | |
870 | * Check if the user asked us to deliver the result through an | |
871 | * eventfd. The eventfd_signal() function is safe to be called | |
872 | * from IRQ context. | |
873 | */ | |
87c3a86e | 874 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
875 | eventfd_signal(iocb->ki_eventfd, 1); |
876 | ||
1da177e4 LT |
877 | put_rq: |
878 | /* everything turned out well, dispose of the aiocb. */ | |
11599eba | 879 | aio_put_req(iocb); |
1da177e4 | 880 | |
6cb2a210 QB |
881 | /* |
882 | * We have to order our ring_info tail store above and test | |
883 | * of the wait list below outside the wait lock. This is | |
884 | * like in wake_up_bit() where clearing a bit has to be | |
885 | * ordered with the unlocked test. | |
886 | */ | |
887 | smp_mb(); | |
888 | ||
1da177e4 LT |
889 | if (waitqueue_active(&ctx->wait)) |
890 | wake_up(&ctx->wait); | |
891 | ||
36f55889 | 892 | rcu_read_unlock(); |
1da177e4 | 893 | } |
385773e0 | 894 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 895 | |
a31ad380 KO |
896 | /* aio_read_events |
897 | * Pull an event off of the ioctx's event ring. Returns the number of | |
898 | * events fetched | |
1da177e4 | 899 | */ |
a31ad380 KO |
900 | static long aio_read_events_ring(struct kioctx *ctx, |
901 | struct io_event __user *event, long nr) | |
1da177e4 | 902 | { |
1da177e4 | 903 | struct aio_ring *ring; |
a31ad380 KO |
904 | unsigned head, pos; |
905 | long ret = 0; | |
906 | int copy_ret; | |
907 | ||
58c85dc2 | 908 | mutex_lock(&ctx->ring_lock); |
1da177e4 | 909 | |
58c85dc2 | 910 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 KO |
911 | head = ring->head; |
912 | kunmap_atomic(ring); | |
913 | ||
58c85dc2 | 914 | pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events); |
1da177e4 | 915 | |
58c85dc2 | 916 | if (head == ctx->tail) |
1da177e4 LT |
917 | goto out; |
918 | ||
a31ad380 KO |
919 | while (ret < nr) { |
920 | long avail; | |
921 | struct io_event *ev; | |
922 | struct page *page; | |
923 | ||
58c85dc2 KO |
924 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; |
925 | if (head == ctx->tail) | |
a31ad380 KO |
926 | break; |
927 | ||
928 | avail = min(avail, nr - ret); | |
929 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
930 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
931 | ||
932 | pos = head + AIO_EVENTS_OFFSET; | |
58c85dc2 | 933 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
a31ad380 KO |
934 | pos %= AIO_EVENTS_PER_PAGE; |
935 | ||
936 | ev = kmap(page); | |
937 | copy_ret = copy_to_user(event + ret, ev + pos, | |
938 | sizeof(*ev) * avail); | |
939 | kunmap(page); | |
940 | ||
941 | if (unlikely(copy_ret)) { | |
942 | ret = -EFAULT; | |
943 | goto out; | |
944 | } | |
945 | ||
946 | ret += avail; | |
947 | head += avail; | |
58c85dc2 | 948 | head %= ctx->nr_events; |
1da177e4 | 949 | } |
1da177e4 | 950 | |
58c85dc2 | 951 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 952 | ring->head = head; |
91d80a84 | 953 | kunmap_atomic(ring); |
58c85dc2 | 954 | flush_dcache_page(ctx->ring_pages[0]); |
a31ad380 | 955 | |
58c85dc2 | 956 | pr_debug("%li h%u t%u\n", ret, head, ctx->tail); |
3e845ce0 | 957 | |
e1bdd5f2 | 958 | put_reqs_available(ctx, ret); |
a31ad380 | 959 | out: |
58c85dc2 | 960 | mutex_unlock(&ctx->ring_lock); |
a31ad380 | 961 | |
1da177e4 LT |
962 | return ret; |
963 | } | |
964 | ||
a31ad380 KO |
965 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
966 | struct io_event __user *event, long *i) | |
1da177e4 | 967 | { |
a31ad380 | 968 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 969 | |
a31ad380 KO |
970 | if (ret > 0) |
971 | *i += ret; | |
1da177e4 | 972 | |
a31ad380 KO |
973 | if (unlikely(atomic_read(&ctx->dead))) |
974 | ret = -EINVAL; | |
1da177e4 | 975 | |
a31ad380 KO |
976 | if (!*i) |
977 | *i = ret; | |
1da177e4 | 978 | |
a31ad380 | 979 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
980 | } |
981 | ||
a31ad380 | 982 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
983 | struct io_event __user *event, |
984 | struct timespec __user *timeout) | |
985 | { | |
a31ad380 KO |
986 | ktime_t until = { .tv64 = KTIME_MAX }; |
987 | long ret = 0; | |
1da177e4 | 988 | |
1da177e4 LT |
989 | if (timeout) { |
990 | struct timespec ts; | |
a31ad380 | 991 | |
1da177e4 | 992 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 993 | return -EFAULT; |
1da177e4 | 994 | |
a31ad380 | 995 | until = timespec_to_ktime(ts); |
1da177e4 LT |
996 | } |
997 | ||
a31ad380 KO |
998 | /* |
999 | * Note that aio_read_events() is being called as the conditional - i.e. | |
1000 | * we're calling it after prepare_to_wait() has set task state to | |
1001 | * TASK_INTERRUPTIBLE. | |
1002 | * | |
1003 | * But aio_read_events() can block, and if it blocks it's going to flip | |
1004 | * the task state back to TASK_RUNNING. | |
1005 | * | |
1006 | * This should be ok, provided it doesn't flip the state back to | |
1007 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
1008 | * will only happen if the mutex_lock() call blocks, and we then find | |
1009 | * the ringbuffer empty. So in practice we should be ok, but it's | |
1010 | * something to be aware of when touching this code. | |
1011 | */ | |
1012 | wait_event_interruptible_hrtimeout(ctx->wait, | |
1013 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
1da177e4 | 1014 | |
a31ad380 KO |
1015 | if (!ret && signal_pending(current)) |
1016 | ret = -EINTR; | |
1da177e4 | 1017 | |
a31ad380 | 1018 | return ret; |
1da177e4 LT |
1019 | } |
1020 | ||
1da177e4 LT |
1021 | /* sys_io_setup: |
1022 | * Create an aio_context capable of receiving at least nr_events. | |
1023 | * ctxp must not point to an aio_context that already exists, and | |
1024 | * must be initialized to 0 prior to the call. On successful | |
1025 | * creation of the aio_context, *ctxp is filled in with the resulting | |
1026 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
1027 | * if the specified nr_events exceeds internal limits. May fail | |
1028 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
1029 | * of available events. May fail with -ENOMEM if insufficient kernel | |
1030 | * resources are available. May fail with -EFAULT if an invalid | |
1031 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
1032 | * implemented. | |
1033 | */ | |
002c8976 | 1034 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
1035 | { |
1036 | struct kioctx *ioctx = NULL; | |
1037 | unsigned long ctx; | |
1038 | long ret; | |
1039 | ||
1040 | ret = get_user(ctx, ctxp); | |
1041 | if (unlikely(ret)) | |
1042 | goto out; | |
1043 | ||
1044 | ret = -EINVAL; | |
d55b5fda ZB |
1045 | if (unlikely(ctx || nr_events == 0)) { |
1046 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
1047 | ctx, nr_events); | |
1da177e4 LT |
1048 | goto out; |
1049 | } | |
1050 | ||
1051 | ioctx = ioctx_alloc(nr_events); | |
1052 | ret = PTR_ERR(ioctx); | |
1053 | if (!IS_ERR(ioctx)) { | |
1054 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a | 1055 | if (ret) |
36f55889 | 1056 | kill_ioctx(ioctx); |
a2e1859a | 1057 | put_ioctx(ioctx); |
1da177e4 LT |
1058 | } |
1059 | ||
1060 | out: | |
1061 | return ret; | |
1062 | } | |
1063 | ||
1064 | /* sys_io_destroy: | |
1065 | * Destroy the aio_context specified. May cancel any outstanding | |
1066 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 1067 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
1068 | * is invalid. |
1069 | */ | |
002c8976 | 1070 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
1071 | { |
1072 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
1073 | if (likely(NULL != ioctx)) { | |
36f55889 | 1074 | kill_ioctx(ioctx); |
a2e1859a | 1075 | put_ioctx(ioctx); |
1da177e4 LT |
1076 | return 0; |
1077 | } | |
1078 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
1079 | return -EINVAL; | |
1080 | } | |
1081 | ||
eed4e51f | 1082 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
1da177e4 | 1083 | { |
eed4e51f BP |
1084 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
1085 | ||
1086 | BUG_ON(ret <= 0); | |
1087 | ||
1088 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { | |
1089 | ssize_t this = min((ssize_t)iov->iov_len, ret); | |
1090 | iov->iov_base += this; | |
1091 | iov->iov_len -= this; | |
1092 | iocb->ki_left -= this; | |
1093 | ret -= this; | |
1094 | if (iov->iov_len == 0) { | |
1095 | iocb->ki_cur_seg++; | |
1096 | iov++; | |
897f15fb | 1097 | } |
eed4e51f | 1098 | } |
1da177e4 | 1099 | |
eed4e51f BP |
1100 | /* the caller should not have done more io than what fit in |
1101 | * the remaining iovecs */ | |
1102 | BUG_ON(ret > 0 && iocb->ki_left == 0); | |
1da177e4 LT |
1103 | } |
1104 | ||
41ef4eb8 KO |
1105 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
1106 | unsigned long, loff_t); | |
1107 | ||
1108 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb, int rw, aio_rw_op *rw_op) | |
1da177e4 LT |
1109 | { |
1110 | struct file *file = iocb->ki_filp; | |
eed4e51f BP |
1111 | struct address_space *mapping = file->f_mapping; |
1112 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1113 | ssize_t ret = 0; |
1114 | ||
c2ec6682 RR |
1115 | /* This matches the pread()/pwrite() logic */ |
1116 | if (iocb->ki_pos < 0) | |
1117 | return -EINVAL; | |
1118 | ||
41ef4eb8 | 1119 | if (rw == WRITE) |
8d71db4f | 1120 | file_start_write(file); |
897f15fb | 1121 | do { |
eed4e51f BP |
1122 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
1123 | iocb->ki_nr_segs - iocb->ki_cur_seg, | |
1124 | iocb->ki_pos); | |
1125 | if (ret > 0) | |
1126 | aio_advance_iovec(iocb, ret); | |
1127 | ||
1128 | /* retry all partial writes. retry partial reads as long as its a | |
1129 | * regular file. */ | |
1130 | } while (ret > 0 && iocb->ki_left > 0 && | |
41ef4eb8 | 1131 | (rw == WRITE || |
eed4e51f | 1132 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); |
41ef4eb8 | 1133 | if (rw == WRITE) |
8d71db4f | 1134 | file_end_write(file); |
1da177e4 | 1135 | |
eed4e51f BP |
1136 | /* This means we must have transferred all that we could */ |
1137 | /* No need to retry anymore */ | |
1da177e4 LT |
1138 | if ((ret == 0) || (iocb->ki_left == 0)) |
1139 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1140 | ||
7adfa2ff RR |
1141 | /* If we managed to write some out we return that, rather than |
1142 | * the eventual error. */ | |
41ef4eb8 | 1143 | if (rw == WRITE |
41003a7b | 1144 | && ret < 0 && ret != -EIOCBQUEUED |
7adfa2ff RR |
1145 | && iocb->ki_nbytes - iocb->ki_left) |
1146 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1147 | ||
1da177e4 LT |
1148 | return ret; |
1149 | } | |
1150 | ||
41ef4eb8 | 1151 | static ssize_t aio_setup_vectored_rw(int rw, struct kiocb *kiocb, bool compat) |
eed4e51f BP |
1152 | { |
1153 | ssize_t ret; | |
1154 | ||
41ef4eb8 KO |
1155 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
1156 | ||
9d85cba7 JM |
1157 | #ifdef CONFIG_COMPAT |
1158 | if (compat) | |
41ef4eb8 | 1159 | ret = compat_rw_copy_check_uvector(rw, |
9d85cba7 | 1160 | (struct compat_iovec __user *)kiocb->ki_buf, |
41ef4eb8 | 1161 | kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, |
ac34ebb3 | 1162 | &kiocb->ki_iovec); |
9d85cba7 JM |
1163 | else |
1164 | #endif | |
41ef4eb8 | 1165 | ret = rw_copy_check_uvector(rw, |
9d85cba7 | 1166 | (struct iovec __user *)kiocb->ki_buf, |
41ef4eb8 | 1167 | kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, |
ac34ebb3 | 1168 | &kiocb->ki_iovec); |
eed4e51f | 1169 | if (ret < 0) |
41ef4eb8 | 1170 | return ret; |
a70b52ec | 1171 | |
41ef4eb8 | 1172 | /* ki_nbytes now reflect bytes instead of segs */ |
eed4e51f | 1173 | kiocb->ki_nbytes = ret; |
41ef4eb8 | 1174 | return 0; |
eed4e51f BP |
1175 | } |
1176 | ||
41ef4eb8 | 1177 | static ssize_t aio_setup_single_vector(int rw, struct kiocb *kiocb) |
eed4e51f | 1178 | { |
41ef4eb8 KO |
1179 | if (unlikely(!access_ok(!rw, kiocb->ki_buf, kiocb->ki_nbytes))) |
1180 | return -EFAULT; | |
a70b52ec | 1181 | |
eed4e51f BP |
1182 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
1183 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; | |
41ef4eb8 | 1184 | kiocb->ki_iovec->iov_len = kiocb->ki_nbytes; |
eed4e51f | 1185 | kiocb->ki_nr_segs = 1; |
eed4e51f BP |
1186 | return 0; |
1187 | } | |
1188 | ||
1da177e4 LT |
1189 | /* |
1190 | * aio_setup_iocb: | |
1191 | * Performs the initial checks and aio retry method | |
1192 | * setup for the kiocb at the time of io submission. | |
1193 | */ | |
41ef4eb8 | 1194 | static ssize_t aio_run_iocb(struct kiocb *req, bool compat) |
1da177e4 | 1195 | { |
41ef4eb8 KO |
1196 | struct file *file = req->ki_filp; |
1197 | ssize_t ret; | |
1198 | int rw; | |
1199 | fmode_t mode; | |
1200 | aio_rw_op *rw_op; | |
1da177e4 | 1201 | |
41ef4eb8 | 1202 | switch (req->ki_opcode) { |
1da177e4 | 1203 | case IOCB_CMD_PREAD: |
eed4e51f | 1204 | case IOCB_CMD_PREADV: |
41ef4eb8 KO |
1205 | mode = FMODE_READ; |
1206 | rw = READ; | |
1207 | rw_op = file->f_op->aio_read; | |
1208 | goto rw_common; | |
1209 | ||
1210 | case IOCB_CMD_PWRITE: | |
eed4e51f | 1211 | case IOCB_CMD_PWRITEV: |
41ef4eb8 KO |
1212 | mode = FMODE_WRITE; |
1213 | rw = WRITE; | |
1214 | rw_op = file->f_op->aio_write; | |
1215 | goto rw_common; | |
1216 | rw_common: | |
1217 | if (unlikely(!(file->f_mode & mode))) | |
1218 | return -EBADF; | |
1219 | ||
1220 | if (!rw_op) | |
1221 | return -EINVAL; | |
1222 | ||
1223 | ret = (req->ki_opcode == IOCB_CMD_PREADV || | |
1224 | req->ki_opcode == IOCB_CMD_PWRITEV) | |
1225 | ? aio_setup_vectored_rw(rw, req, compat) | |
1226 | : aio_setup_single_vector(rw, req); | |
eed4e51f | 1227 | if (ret) |
41ef4eb8 KO |
1228 | return ret; |
1229 | ||
1230 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); | |
1231 | if (ret < 0) | |
1232 | return ret; | |
1233 | ||
1234 | req->ki_nbytes = ret; | |
1235 | req->ki_left = ret; | |
1236 | ||
1237 | ret = aio_rw_vect_retry(req, rw, rw_op); | |
1da177e4 | 1238 | break; |
41ef4eb8 | 1239 | |
1da177e4 | 1240 | case IOCB_CMD_FDSYNC: |
41ef4eb8 KO |
1241 | if (!file->f_op->aio_fsync) |
1242 | return -EINVAL; | |
1243 | ||
1244 | ret = file->f_op->aio_fsync(req, 1); | |
1da177e4 | 1245 | break; |
41ef4eb8 | 1246 | |
1da177e4 | 1247 | case IOCB_CMD_FSYNC: |
41ef4eb8 KO |
1248 | if (!file->f_op->aio_fsync) |
1249 | return -EINVAL; | |
1250 | ||
1251 | ret = file->f_op->aio_fsync(req, 0); | |
1da177e4 | 1252 | break; |
41ef4eb8 | 1253 | |
1da177e4 | 1254 | default: |
caf4167a | 1255 | pr_debug("EINVAL: no operation provided\n"); |
41ef4eb8 | 1256 | return -EINVAL; |
1da177e4 LT |
1257 | } |
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; | |
1330 | ||
1331 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; | |
1332 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; | |
1333 | req->ki_opcode = iocb->aio_lio_opcode; | |
1da177e4 | 1334 | |
41ef4eb8 | 1335 | ret = aio_run_iocb(req, compat); |
41003a7b | 1336 | if (ret) |
7137c6bd | 1337 | goto out_put_req; |
41003a7b | 1338 | |
1da177e4 LT |
1339 | aio_put_req(req); /* drop extra ref to req */ |
1340 | return 0; | |
1da177e4 | 1341 | out_put_req: |
e1bdd5f2 | 1342 | put_reqs_available(ctx, 1); |
1da177e4 LT |
1343 | aio_put_req(req); /* drop extra ref to req */ |
1344 | aio_put_req(req); /* drop i/o ref to req */ | |
1345 | return ret; | |
1346 | } | |
1347 | ||
9d85cba7 JM |
1348 | long do_io_submit(aio_context_t ctx_id, long nr, |
1349 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1350 | { |
1351 | struct kioctx *ctx; | |
1352 | long ret = 0; | |
080d676d | 1353 | int i = 0; |
9f5b9425 | 1354 | struct blk_plug plug; |
1da177e4 LT |
1355 | |
1356 | if (unlikely(nr < 0)) | |
1357 | return -EINVAL; | |
1358 | ||
75e1c70f JM |
1359 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1360 | nr = LONG_MAX/sizeof(*iocbpp); | |
1361 | ||
1da177e4 LT |
1362 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1363 | return -EFAULT; | |
1364 | ||
1365 | ctx = lookup_ioctx(ctx_id); | |
1366 | if (unlikely(!ctx)) { | |
caf4167a | 1367 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1368 | return -EINVAL; |
1369 | } | |
1370 | ||
9f5b9425 SL |
1371 | blk_start_plug(&plug); |
1372 | ||
1da177e4 LT |
1373 | /* |
1374 | * AKPM: should this return a partial result if some of the IOs were | |
1375 | * successfully submitted? | |
1376 | */ | |
1377 | for (i=0; i<nr; i++) { | |
1378 | struct iocb __user *user_iocb; | |
1379 | struct iocb tmp; | |
1380 | ||
1381 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1382 | ret = -EFAULT; | |
1383 | break; | |
1384 | } | |
1385 | ||
1386 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1387 | ret = -EFAULT; | |
1388 | break; | |
1389 | } | |
1390 | ||
a1c8eae7 | 1391 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
1da177e4 LT |
1392 | if (ret) |
1393 | break; | |
1394 | } | |
9f5b9425 | 1395 | blk_finish_plug(&plug); |
1da177e4 LT |
1396 | |
1397 | put_ioctx(ctx); | |
1398 | return i ? i : ret; | |
1399 | } | |
1400 | ||
9d85cba7 JM |
1401 | /* sys_io_submit: |
1402 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1403 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1404 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1405 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1406 | * is invalid for the file descriptor in the iocb. May fail with | |
1407 | * -EFAULT if any of the data structures point to invalid data. May | |
1408 | * fail with -EBADF if the file descriptor specified in the first | |
1409 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1410 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1411 | * fail with -ENOSYS if not implemented. | |
1412 | */ | |
1413 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1414 | struct iocb __user * __user *, iocbpp) | |
1415 | { | |
1416 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1417 | } | |
1418 | ||
1da177e4 LT |
1419 | /* lookup_kiocb |
1420 | * Finds a given iocb for cancellation. | |
1da177e4 | 1421 | */ |
25ee7e38 AB |
1422 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1423 | u32 key) | |
1da177e4 LT |
1424 | { |
1425 | struct list_head *pos; | |
d00689af ZB |
1426 | |
1427 | assert_spin_locked(&ctx->ctx_lock); | |
1428 | ||
8a660890 KO |
1429 | if (key != KIOCB_KEY) |
1430 | return NULL; | |
1431 | ||
1da177e4 LT |
1432 | /* TODO: use a hash or array, this sucks. */ |
1433 | list_for_each(pos, &ctx->active_reqs) { | |
1434 | struct kiocb *kiocb = list_kiocb(pos); | |
8a660890 | 1435 | if (kiocb->ki_obj.user == iocb) |
1da177e4 LT |
1436 | return kiocb; |
1437 | } | |
1438 | return NULL; | |
1439 | } | |
1440 | ||
1441 | /* sys_io_cancel: | |
1442 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1443 | * the operation is successfully cancelled, the resulting event is | |
1444 | * copied into the memory pointed to by result without being placed | |
1445 | * into the completion queue and 0 is returned. May fail with | |
1446 | * -EFAULT if any of the data structures pointed to are invalid. | |
1447 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1448 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1449 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1450 | */ | |
002c8976 HC |
1451 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1452 | struct io_event __user *, result) | |
1da177e4 | 1453 | { |
906b973c | 1454 | struct io_event res; |
1da177e4 LT |
1455 | struct kioctx *ctx; |
1456 | struct kiocb *kiocb; | |
1457 | u32 key; | |
1458 | int ret; | |
1459 | ||
1460 | ret = get_user(key, &iocb->aio_key); | |
1461 | if (unlikely(ret)) | |
1462 | return -EFAULT; | |
1463 | ||
1464 | ctx = lookup_ioctx(ctx_id); | |
1465 | if (unlikely(!ctx)) | |
1466 | return -EINVAL; | |
1467 | ||
1468 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1469 | |
1da177e4 | 1470 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c KO |
1471 | if (kiocb) |
1472 | ret = kiocb_cancel(ctx, kiocb, &res); | |
1473 | else | |
1474 | ret = -EINVAL; | |
1475 | ||
1da177e4 LT |
1476 | spin_unlock_irq(&ctx->ctx_lock); |
1477 | ||
906b973c KO |
1478 | if (!ret) { |
1479 | /* Cancellation succeeded -- copy the result | |
1480 | * into the user's buffer. | |
1481 | */ | |
1482 | if (copy_to_user(result, &res, sizeof(res))) | |
1483 | ret = -EFAULT; | |
1484 | } | |
1da177e4 LT |
1485 | |
1486 | put_ioctx(ctx); | |
1487 | ||
1488 | return ret; | |
1489 | } | |
1490 | ||
1491 | /* io_getevents: | |
1492 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1493 | * the completion queue for the aio_context specified by ctx_id. If |
1494 | * it succeeds, the number of read events is returned. May fail with | |
1495 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1496 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1497 | * if any of the memory specified is invalid. May return 0 or | |
1498 | * < min_nr if the timeout specified by timeout has elapsed | |
1499 | * before sufficient events are available, where timeout == NULL | |
1500 | * specifies an infinite timeout. Note that the timeout pointed to by | |
6900807c | 1501 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
1da177e4 | 1502 | */ |
002c8976 HC |
1503 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1504 | long, min_nr, | |
1505 | long, nr, | |
1506 | struct io_event __user *, events, | |
1507 | struct timespec __user *, timeout) | |
1da177e4 LT |
1508 | { |
1509 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1510 | long ret = -EINVAL; | |
1511 | ||
1512 | if (likely(ioctx)) { | |
2e410255 | 1513 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 LT |
1514 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
1515 | put_ioctx(ioctx); | |
1516 | } | |
1da177e4 LT |
1517 | return ret; |
1518 | } |