]>
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/migrate.h> |
40 | #include <linux/ramfs.h> | |
723be6e3 | 41 | #include <linux/percpu-refcount.h> |
71ad7490 | 42 | #include <linux/mount.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 | |
71ad7490 BL |
156 | static struct vfsmount *aio_mnt; |
157 | ||
158 | static const struct file_operations aio_ring_fops; | |
159 | static const struct address_space_operations aio_ctx_aops; | |
160 | ||
161 | static struct file *aio_private_file(struct kioctx *ctx, loff_t nr_pages) | |
162 | { | |
163 | struct qstr this = QSTR_INIT("[aio]", 5); | |
164 | struct file *file; | |
165 | struct path path; | |
166 | struct inode *inode = alloc_anon_inode(aio_mnt->mnt_sb); | |
7f62656b DC |
167 | if (IS_ERR(inode)) |
168 | return ERR_CAST(inode); | |
71ad7490 BL |
169 | |
170 | inode->i_mapping->a_ops = &aio_ctx_aops; | |
171 | inode->i_mapping->private_data = ctx; | |
172 | inode->i_size = PAGE_SIZE * nr_pages; | |
173 | ||
174 | path.dentry = d_alloc_pseudo(aio_mnt->mnt_sb, &this); | |
175 | if (!path.dentry) { | |
176 | iput(inode); | |
177 | return ERR_PTR(-ENOMEM); | |
178 | } | |
179 | path.mnt = mntget(aio_mnt); | |
180 | ||
181 | d_instantiate(path.dentry, inode); | |
182 | file = alloc_file(&path, FMODE_READ | FMODE_WRITE, &aio_ring_fops); | |
183 | if (IS_ERR(file)) { | |
184 | path_put(&path); | |
185 | return file; | |
186 | } | |
187 | ||
188 | file->f_flags = O_RDWR; | |
189 | file->private_data = ctx; | |
190 | return file; | |
191 | } | |
192 | ||
193 | static struct dentry *aio_mount(struct file_system_type *fs_type, | |
194 | int flags, const char *dev_name, void *data) | |
195 | { | |
196 | static const struct dentry_operations ops = { | |
197 | .d_dname = simple_dname, | |
198 | }; | |
199 | return mount_pseudo(fs_type, "aio:", NULL, &ops, 0xa10a10a1); | |
200 | } | |
201 | ||
1da177e4 LT |
202 | /* aio_setup |
203 | * Creates the slab caches used by the aio routines, panic on | |
204 | * failure as this is done early during the boot sequence. | |
205 | */ | |
206 | static int __init aio_setup(void) | |
207 | { | |
71ad7490 BL |
208 | static struct file_system_type aio_fs = { |
209 | .name = "aio", | |
210 | .mount = aio_mount, | |
211 | .kill_sb = kill_anon_super, | |
212 | }; | |
213 | aio_mnt = kern_mount(&aio_fs); | |
214 | if (IS_ERR(aio_mnt)) | |
215 | panic("Failed to create aio fs mount."); | |
216 | ||
0a31bd5f CL |
217 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
218 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 219 | |
caf4167a | 220 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
221 | |
222 | return 0; | |
223 | } | |
385773e0 | 224 | __initcall(aio_setup); |
1da177e4 | 225 | |
5e9ae2e5 BL |
226 | static void put_aio_ring_file(struct kioctx *ctx) |
227 | { | |
228 | struct file *aio_ring_file = ctx->aio_ring_file; | |
229 | if (aio_ring_file) { | |
230 | truncate_setsize(aio_ring_file->f_inode, 0); | |
231 | ||
232 | /* Prevent further access to the kioctx from migratepages */ | |
233 | spin_lock(&aio_ring_file->f_inode->i_mapping->private_lock); | |
234 | aio_ring_file->f_inode->i_mapping->private_data = NULL; | |
235 | ctx->aio_ring_file = NULL; | |
236 | spin_unlock(&aio_ring_file->f_inode->i_mapping->private_lock); | |
237 | ||
238 | fput(aio_ring_file); | |
239 | } | |
240 | } | |
241 | ||
1da177e4 LT |
242 | static void aio_free_ring(struct kioctx *ctx) |
243 | { | |
36bc08cc | 244 | int i; |
1da177e4 | 245 | |
36bc08cc GZ |
246 | for (i = 0; i < ctx->nr_pages; i++) { |
247 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, | |
248 | page_count(ctx->ring_pages[i])); | |
58c85dc2 | 249 | put_page(ctx->ring_pages[i]); |
36bc08cc | 250 | } |
1da177e4 | 251 | |
5e9ae2e5 BL |
252 | put_aio_ring_file(ctx); |
253 | ||
ddb8c45b | 254 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) { |
58c85dc2 | 255 | kfree(ctx->ring_pages); |
ddb8c45b SL |
256 | ctx->ring_pages = NULL; |
257 | } | |
36bc08cc GZ |
258 | } |
259 | ||
260 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) | |
261 | { | |
262 | vma->vm_ops = &generic_file_vm_ops; | |
263 | return 0; | |
264 | } | |
265 | ||
266 | static const struct file_operations aio_ring_fops = { | |
267 | .mmap = aio_ring_mmap, | |
268 | }; | |
269 | ||
270 | static int aio_set_page_dirty(struct page *page) | |
271 | { | |
272 | return 0; | |
273 | } | |
274 | ||
0c45355f | 275 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc GZ |
276 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
277 | struct page *old, enum migrate_mode mode) | |
278 | { | |
5e9ae2e5 | 279 | struct kioctx *ctx; |
36bc08cc | 280 | unsigned long flags; |
36bc08cc GZ |
281 | int rc; |
282 | ||
283 | /* Writeback must be complete */ | |
284 | BUG_ON(PageWriteback(old)); | |
285 | put_page(old); | |
286 | ||
287 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode); | |
288 | if (rc != MIGRATEPAGE_SUCCESS) { | |
289 | get_page(old); | |
290 | return rc; | |
291 | } | |
292 | ||
293 | get_page(new); | |
294 | ||
5e9ae2e5 BL |
295 | /* We can potentially race against kioctx teardown here. Use the |
296 | * address_space's private data lock to protect the mapping's | |
297 | * private_data. | |
298 | */ | |
299 | spin_lock(&mapping->private_lock); | |
300 | ctx = mapping->private_data; | |
301 | if (ctx) { | |
302 | pgoff_t idx; | |
303 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
304 | migrate_page_copy(new, old); | |
305 | idx = old->index; | |
306 | if (idx < (pgoff_t)ctx->nr_pages) | |
307 | ctx->ring_pages[idx] = new; | |
308 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
309 | } else | |
310 | rc = -EBUSY; | |
311 | spin_unlock(&mapping->private_lock); | |
36bc08cc GZ |
312 | |
313 | return rc; | |
1da177e4 | 314 | } |
0c45355f | 315 | #endif |
1da177e4 | 316 | |
36bc08cc GZ |
317 | static const struct address_space_operations aio_ctx_aops = { |
318 | .set_page_dirty = aio_set_page_dirty, | |
0c45355f | 319 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc | 320 | .migratepage = aio_migratepage, |
0c45355f | 321 | #endif |
36bc08cc GZ |
322 | }; |
323 | ||
1da177e4 LT |
324 | static int aio_setup_ring(struct kioctx *ctx) |
325 | { | |
326 | struct aio_ring *ring; | |
1da177e4 | 327 | unsigned nr_events = ctx->max_reqs; |
41003a7b | 328 | struct mm_struct *mm = current->mm; |
3dc9acb6 | 329 | unsigned long size, unused; |
1da177e4 | 330 | int nr_pages; |
36bc08cc GZ |
331 | int i; |
332 | struct file *file; | |
1da177e4 LT |
333 | |
334 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
335 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
336 | ||
337 | size = sizeof(struct aio_ring); | |
338 | size += sizeof(struct io_event) * nr_events; | |
1da177e4 | 339 | |
36bc08cc | 340 | nr_pages = PFN_UP(size); |
1da177e4 LT |
341 | if (nr_pages < 0) |
342 | return -EINVAL; | |
343 | ||
71ad7490 | 344 | file = aio_private_file(ctx, nr_pages); |
36bc08cc GZ |
345 | if (IS_ERR(file)) { |
346 | ctx->aio_ring_file = NULL; | |
347 | return -EAGAIN; | |
348 | } | |
349 | ||
3dc9acb6 LT |
350 | ctx->aio_ring_file = file; |
351 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) | |
352 | / sizeof(struct io_event); | |
353 | ||
354 | ctx->ring_pages = ctx->internal_pages; | |
355 | if (nr_pages > AIO_RING_PAGES) { | |
356 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), | |
357 | GFP_KERNEL); | |
358 | if (!ctx->ring_pages) { | |
359 | put_aio_ring_file(ctx); | |
360 | return -ENOMEM; | |
361 | } | |
362 | } | |
363 | ||
36bc08cc GZ |
364 | for (i = 0; i < nr_pages; i++) { |
365 | struct page *page; | |
366 | page = find_or_create_page(file->f_inode->i_mapping, | |
367 | i, GFP_HIGHUSER | __GFP_ZERO); | |
368 | if (!page) | |
369 | break; | |
370 | pr_debug("pid(%d) page[%d]->count=%d\n", | |
371 | current->pid, i, page_count(page)); | |
372 | SetPageUptodate(page); | |
373 | SetPageDirty(page); | |
374 | unlock_page(page); | |
3dc9acb6 LT |
375 | |
376 | ctx->ring_pages[i] = page; | |
36bc08cc | 377 | } |
3dc9acb6 | 378 | ctx->nr_pages = i; |
1da177e4 | 379 | |
3dc9acb6 LT |
380 | if (unlikely(i != nr_pages)) { |
381 | aio_free_ring(ctx); | |
382 | return -EAGAIN; | |
1da177e4 LT |
383 | } |
384 | ||
58c85dc2 KO |
385 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
386 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
36bc08cc | 387 | |
41003a7b | 388 | down_write(&mm->mmap_sem); |
36bc08cc GZ |
389 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
390 | PROT_READ | PROT_WRITE, | |
3dc9acb6 LT |
391 | MAP_SHARED, 0, &unused); |
392 | up_write(&mm->mmap_sem); | |
58c85dc2 | 393 | if (IS_ERR((void *)ctx->mmap_base)) { |
58c85dc2 | 394 | ctx->mmap_size = 0; |
1da177e4 LT |
395 | aio_free_ring(ctx); |
396 | return -EAGAIN; | |
397 | } | |
398 | ||
58c85dc2 | 399 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
d6c355c7 | 400 | |
58c85dc2 KO |
401 | ctx->user_id = ctx->mmap_base; |
402 | ctx->nr_events = nr_events; /* trusted copy */ | |
1da177e4 | 403 | |
58c85dc2 | 404 | ring = kmap_atomic(ctx->ring_pages[0]); |
1da177e4 | 405 | ring->nr = nr_events; /* user copy */ |
db446a08 | 406 | ring->id = ~0U; |
1da177e4 LT |
407 | ring->head = ring->tail = 0; |
408 | ring->magic = AIO_RING_MAGIC; | |
409 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
410 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
411 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 412 | kunmap_atomic(ring); |
58c85dc2 | 413 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 LT |
414 | |
415 | return 0; | |
416 | } | |
417 | ||
1da177e4 LT |
418 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
419 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
420 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
421 | ||
0460fef2 KO |
422 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
423 | { | |
424 | struct kioctx *ctx = req->ki_ctx; | |
425 | unsigned long flags; | |
426 | ||
427 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
428 | ||
429 | if (!req->ki_list.next) | |
430 | list_add(&req->ki_list, &ctx->active_reqs); | |
431 | ||
432 | req->ki_cancel = cancel; | |
433 | ||
434 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
435 | } | |
436 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
437 | ||
bec68faa | 438 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb) |
906b973c | 439 | { |
0460fef2 | 440 | kiocb_cancel_fn *old, *cancel; |
906b973c | 441 | |
0460fef2 KO |
442 | /* |
443 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
444 | * actually has a cancel function, hence the cmpxchg() | |
445 | */ | |
446 | ||
447 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
448 | do { | |
449 | if (!cancel || cancel == KIOCB_CANCELLED) | |
57282d8f | 450 | return -EINVAL; |
906b973c | 451 | |
0460fef2 KO |
452 | old = cancel; |
453 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
454 | } while (cancel != old); | |
906b973c | 455 | |
57282d8f | 456 | return cancel(kiocb); |
906b973c KO |
457 | } |
458 | ||
e34ecee2 | 459 | static void free_ioctx(struct work_struct *work) |
36f55889 | 460 | { |
e34ecee2 | 461 | struct kioctx *ctx = container_of(work, struct kioctx, free_work); |
e1bdd5f2 | 462 | |
e34ecee2 | 463 | pr_debug("freeing %p\n", ctx); |
e1bdd5f2 | 464 | |
e34ecee2 | 465 | aio_free_ring(ctx); |
e1bdd5f2 | 466 | free_percpu(ctx->cpu); |
36f55889 KO |
467 | kmem_cache_free(kioctx_cachep, ctx); |
468 | } | |
469 | ||
e34ecee2 KO |
470 | static void free_ioctx_reqs(struct percpu_ref *ref) |
471 | { | |
472 | struct kioctx *ctx = container_of(ref, struct kioctx, reqs); | |
473 | ||
474 | INIT_WORK(&ctx->free_work, free_ioctx); | |
475 | schedule_work(&ctx->free_work); | |
476 | } | |
477 | ||
36f55889 KO |
478 | /* |
479 | * When this function runs, the kioctx has been removed from the "hash table" | |
480 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
481 | * now it's safe to cancel any that need to be. | |
482 | */ | |
e34ecee2 | 483 | static void free_ioctx_users(struct percpu_ref *ref) |
36f55889 | 484 | { |
e34ecee2 | 485 | struct kioctx *ctx = container_of(ref, struct kioctx, users); |
36f55889 KO |
486 | struct kiocb *req; |
487 | ||
488 | spin_lock_irq(&ctx->ctx_lock); | |
489 | ||
490 | while (!list_empty(&ctx->active_reqs)) { | |
491 | req = list_first_entry(&ctx->active_reqs, | |
492 | struct kiocb, ki_list); | |
493 | ||
494 | list_del_init(&req->ki_list); | |
bec68faa | 495 | kiocb_cancel(ctx, req); |
36f55889 KO |
496 | } |
497 | ||
498 | spin_unlock_irq(&ctx->ctx_lock); | |
499 | ||
e34ecee2 KO |
500 | percpu_ref_kill(&ctx->reqs); |
501 | percpu_ref_put(&ctx->reqs); | |
36f55889 KO |
502 | } |
503 | ||
db446a08 BL |
504 | static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) |
505 | { | |
506 | unsigned i, new_nr; | |
507 | struct kioctx_table *table, *old; | |
508 | struct aio_ring *ring; | |
509 | ||
510 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 511 | rcu_read_lock(); |
77d30b14 | 512 | table = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
513 | |
514 | while (1) { | |
515 | if (table) | |
516 | for (i = 0; i < table->nr; i++) | |
517 | if (!table->table[i]) { | |
518 | ctx->id = i; | |
519 | table->table[i] = ctx; | |
d9b2c871 | 520 | rcu_read_unlock(); |
db446a08 BL |
521 | spin_unlock(&mm->ioctx_lock); |
522 | ||
523 | ring = kmap_atomic(ctx->ring_pages[0]); | |
524 | ring->id = ctx->id; | |
525 | kunmap_atomic(ring); | |
526 | return 0; | |
527 | } | |
528 | ||
529 | new_nr = (table ? table->nr : 1) * 4; | |
530 | ||
d9b2c871 | 531 | rcu_read_unlock(); |
db446a08 BL |
532 | spin_unlock(&mm->ioctx_lock); |
533 | ||
534 | table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * | |
535 | new_nr, GFP_KERNEL); | |
536 | if (!table) | |
537 | return -ENOMEM; | |
538 | ||
539 | table->nr = new_nr; | |
540 | ||
541 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 542 | rcu_read_lock(); |
77d30b14 | 543 | old = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
544 | |
545 | if (!old) { | |
546 | rcu_assign_pointer(mm->ioctx_table, table); | |
547 | } else if (table->nr > old->nr) { | |
548 | memcpy(table->table, old->table, | |
549 | old->nr * sizeof(struct kioctx *)); | |
550 | ||
551 | rcu_assign_pointer(mm->ioctx_table, table); | |
552 | kfree_rcu(old, rcu); | |
553 | } else { | |
554 | kfree(table); | |
555 | table = old; | |
556 | } | |
557 | } | |
558 | } | |
559 | ||
e34ecee2 KO |
560 | static void aio_nr_sub(unsigned nr) |
561 | { | |
562 | spin_lock(&aio_nr_lock); | |
563 | if (WARN_ON(aio_nr - nr > aio_nr)) | |
564 | aio_nr = 0; | |
565 | else | |
566 | aio_nr -= nr; | |
567 | spin_unlock(&aio_nr_lock); | |
568 | } | |
569 | ||
1da177e4 LT |
570 | /* ioctx_alloc |
571 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
572 | */ | |
573 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
574 | { | |
41003a7b | 575 | struct mm_struct *mm = current->mm; |
1da177e4 | 576 | struct kioctx *ctx; |
e23754f8 | 577 | int err = -ENOMEM; |
1da177e4 | 578 | |
e1bdd5f2 KO |
579 | /* |
580 | * We keep track of the number of available ringbuffer slots, to prevent | |
581 | * overflow (reqs_available), and we also use percpu counters for this. | |
582 | * | |
583 | * So since up to half the slots might be on other cpu's percpu counters | |
584 | * and unavailable, double nr_events so userspace sees what they | |
585 | * expected: additionally, we move req_batch slots to/from percpu | |
586 | * counters at a time, so make sure that isn't 0: | |
587 | */ | |
588 | nr_events = max(nr_events, num_possible_cpus() * 4); | |
589 | nr_events *= 2; | |
590 | ||
1da177e4 LT |
591 | /* Prevent overflows */ |
592 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
593 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
594 | pr_debug("ENOMEM: nr_events too high\n"); | |
595 | return ERR_PTR(-EINVAL); | |
596 | } | |
597 | ||
4cd81c3d | 598 | if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) |
1da177e4 LT |
599 | return ERR_PTR(-EAGAIN); |
600 | ||
c3762229 | 601 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
602 | if (!ctx) |
603 | return ERR_PTR(-ENOMEM); | |
604 | ||
1da177e4 | 605 | ctx->max_reqs = nr_events; |
1da177e4 | 606 | |
e34ecee2 KO |
607 | if (percpu_ref_init(&ctx->users, free_ioctx_users)) |
608 | goto err; | |
609 | ||
610 | if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs)) | |
611 | goto err; | |
723be6e3 | 612 | |
1da177e4 | 613 | spin_lock_init(&ctx->ctx_lock); |
0460fef2 | 614 | spin_lock_init(&ctx->completion_lock); |
58c85dc2 | 615 | mutex_init(&ctx->ring_lock); |
1da177e4 LT |
616 | init_waitqueue_head(&ctx->wait); |
617 | ||
618 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 | 619 | |
e1bdd5f2 KO |
620 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
621 | if (!ctx->cpu) | |
e34ecee2 | 622 | goto err; |
1da177e4 | 623 | |
e1bdd5f2 | 624 | if (aio_setup_ring(ctx) < 0) |
e34ecee2 | 625 | goto err; |
e1bdd5f2 | 626 | |
34e83fc6 | 627 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
e1bdd5f2 | 628 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
6878ea72 BL |
629 | if (ctx->req_batch < 1) |
630 | ctx->req_batch = 1; | |
34e83fc6 | 631 | |
1da177e4 | 632 | /* limit the number of system wide aios */ |
9fa1cb39 | 633 | spin_lock(&aio_nr_lock); |
4cd81c3d | 634 | if (aio_nr + nr_events > (aio_max_nr * 2UL) || |
2dd542b7 | 635 | aio_nr + nr_events < aio_nr) { |
9fa1cb39 | 636 | spin_unlock(&aio_nr_lock); |
e34ecee2 | 637 | err = -EAGAIN; |
d1b94327 | 638 | goto err_ctx; |
2dd542b7 AV |
639 | } |
640 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 641 | spin_unlock(&aio_nr_lock); |
1da177e4 | 642 | |
723be6e3 KO |
643 | percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ |
644 | ||
da90382c BL |
645 | err = ioctx_add_table(ctx, mm); |
646 | if (err) | |
e34ecee2 | 647 | goto err_cleanup; |
da90382c | 648 | |
caf4167a | 649 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
58c85dc2 | 650 | ctx, ctx->user_id, mm, ctx->nr_events); |
1da177e4 LT |
651 | return ctx; |
652 | ||
e34ecee2 KO |
653 | err_cleanup: |
654 | aio_nr_sub(ctx->max_reqs); | |
d1b94327 GZ |
655 | err_ctx: |
656 | aio_free_ring(ctx); | |
e34ecee2 | 657 | err: |
e1bdd5f2 | 658 | free_percpu(ctx->cpu); |
e34ecee2 | 659 | free_percpu(ctx->reqs.pcpu_count); |
723be6e3 | 660 | free_percpu(ctx->users.pcpu_count); |
1da177e4 | 661 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 662 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 663 | return ERR_PTR(err); |
1da177e4 LT |
664 | } |
665 | ||
36f55889 KO |
666 | /* kill_ioctx |
667 | * Cancels all outstanding aio requests on an aio context. Used | |
668 | * when the processes owning a context have all exited to encourage | |
669 | * the rapid destruction of the kioctx. | |
670 | */ | |
db446a08 | 671 | static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx) |
36f55889 KO |
672 | { |
673 | if (!atomic_xchg(&ctx->dead, 1)) { | |
db446a08 BL |
674 | struct kioctx_table *table; |
675 | ||
676 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 677 | rcu_read_lock(); |
77d30b14 | 678 | table = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
679 | |
680 | WARN_ON(ctx != table->table[ctx->id]); | |
681 | table->table[ctx->id] = NULL; | |
d9b2c871 | 682 | rcu_read_unlock(); |
db446a08 BL |
683 | spin_unlock(&mm->ioctx_lock); |
684 | ||
723be6e3 KO |
685 | /* percpu_ref_kill() will do the necessary call_rcu() */ |
686 | wake_up_all(&ctx->wait); | |
dee11c23 | 687 | |
36f55889 | 688 | /* |
4fcc712f KO |
689 | * It'd be more correct to do this in free_ioctx(), after all |
690 | * the outstanding kiocbs have finished - but by then io_destroy | |
691 | * has already returned, so io_setup() could potentially return | |
692 | * -EAGAIN with no ioctxs actually in use (as far as userspace | |
693 | * could tell). | |
36f55889 | 694 | */ |
e34ecee2 | 695 | aio_nr_sub(ctx->max_reqs); |
4fcc712f KO |
696 | |
697 | if (ctx->mmap_size) | |
698 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
699 | ||
723be6e3 | 700 | percpu_ref_kill(&ctx->users); |
36f55889 | 701 | } |
1da177e4 LT |
702 | } |
703 | ||
704 | /* wait_on_sync_kiocb: | |
705 | * Waits on the given sync kiocb to complete. | |
706 | */ | |
57282d8f | 707 | ssize_t wait_on_sync_kiocb(struct kiocb *req) |
1da177e4 | 708 | { |
57282d8f | 709 | while (!req->ki_ctx) { |
1da177e4 | 710 | set_current_state(TASK_UNINTERRUPTIBLE); |
57282d8f | 711 | if (req->ki_ctx) |
1da177e4 | 712 | break; |
41d10da3 | 713 | io_schedule(); |
1da177e4 LT |
714 | } |
715 | __set_current_state(TASK_RUNNING); | |
57282d8f | 716 | return req->ki_user_data; |
1da177e4 | 717 | } |
385773e0 | 718 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 719 | |
36f55889 KO |
720 | /* |
721 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
722 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
723 | * called on the context. | |
724 | * | |
725 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
726 | * them. | |
1da177e4 | 727 | */ |
fc9b52cd | 728 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 729 | { |
db446a08 | 730 | struct kioctx_table *table; |
abf137dd | 731 | struct kioctx *ctx; |
db446a08 BL |
732 | unsigned i = 0; |
733 | ||
734 | while (1) { | |
735 | rcu_read_lock(); | |
736 | table = rcu_dereference(mm->ioctx_table); | |
737 | ||
738 | do { | |
739 | if (!table || i >= table->nr) { | |
740 | rcu_read_unlock(); | |
741 | rcu_assign_pointer(mm->ioctx_table, NULL); | |
742 | if (table) | |
743 | kfree(table); | |
744 | return; | |
745 | } | |
746 | ||
747 | ctx = table->table[i++]; | |
748 | } while (!ctx); | |
749 | ||
750 | rcu_read_unlock(); | |
abf137dd | 751 | |
936af157 AV |
752 | /* |
753 | * We don't need to bother with munmap() here - | |
754 | * exit_mmap(mm) is coming and it'll unmap everything. | |
755 | * Since aio_free_ring() uses non-zero ->mmap_size | |
756 | * as indicator that it needs to unmap the area, | |
757 | * just set it to 0; aio_free_ring() is the only | |
758 | * place that uses ->mmap_size, so it's safe. | |
936af157 | 759 | */ |
58c85dc2 | 760 | ctx->mmap_size = 0; |
36f55889 | 761 | |
db446a08 | 762 | kill_ioctx(mm, ctx); |
1da177e4 LT |
763 | } |
764 | } | |
765 | ||
e1bdd5f2 KO |
766 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
767 | { | |
768 | struct kioctx_cpu *kcpu; | |
769 | ||
770 | preempt_disable(); | |
771 | kcpu = this_cpu_ptr(ctx->cpu); | |
772 | ||
773 | kcpu->reqs_available += nr; | |
774 | while (kcpu->reqs_available >= ctx->req_batch * 2) { | |
775 | kcpu->reqs_available -= ctx->req_batch; | |
776 | atomic_add(ctx->req_batch, &ctx->reqs_available); | |
777 | } | |
778 | ||
779 | preempt_enable(); | |
780 | } | |
781 | ||
782 | static bool get_reqs_available(struct kioctx *ctx) | |
783 | { | |
784 | struct kioctx_cpu *kcpu; | |
785 | bool ret = false; | |
786 | ||
787 | preempt_disable(); | |
788 | kcpu = this_cpu_ptr(ctx->cpu); | |
789 | ||
790 | if (!kcpu->reqs_available) { | |
791 | int old, avail = atomic_read(&ctx->reqs_available); | |
792 | ||
793 | do { | |
794 | if (avail < ctx->req_batch) | |
795 | goto out; | |
796 | ||
797 | old = avail; | |
798 | avail = atomic_cmpxchg(&ctx->reqs_available, | |
799 | avail, avail - ctx->req_batch); | |
800 | } while (avail != old); | |
801 | ||
802 | kcpu->reqs_available += ctx->req_batch; | |
803 | } | |
804 | ||
805 | ret = true; | |
806 | kcpu->reqs_available--; | |
807 | out: | |
808 | preempt_enable(); | |
809 | return ret; | |
810 | } | |
811 | ||
1da177e4 | 812 | /* aio_get_req |
57282d8f KO |
813 | * Allocate a slot for an aio request. |
814 | * Returns NULL if no requests are free. | |
1da177e4 | 815 | */ |
a1c8eae7 | 816 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
1da177e4 | 817 | { |
a1c8eae7 KO |
818 | struct kiocb *req; |
819 | ||
e1bdd5f2 | 820 | if (!get_reqs_available(ctx)) |
a1c8eae7 KO |
821 | return NULL; |
822 | ||
0460fef2 | 823 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
1da177e4 | 824 | if (unlikely(!req)) |
a1c8eae7 | 825 | goto out_put; |
1da177e4 | 826 | |
e34ecee2 KO |
827 | percpu_ref_get(&ctx->reqs); |
828 | ||
1da177e4 | 829 | req->ki_ctx = ctx; |
080d676d | 830 | return req; |
a1c8eae7 | 831 | out_put: |
e1bdd5f2 | 832 | put_reqs_available(ctx, 1); |
a1c8eae7 | 833 | return NULL; |
1da177e4 LT |
834 | } |
835 | ||
11599eba | 836 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 837 | { |
1d98ebfc KO |
838 | if (req->ki_filp) |
839 | fput(req->ki_filp); | |
13389010 DL |
840 | if (req->ki_eventfd != NULL) |
841 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 | 842 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
843 | } |
844 | ||
d5470b59 | 845 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 846 | { |
db446a08 | 847 | struct aio_ring __user *ring = (void __user *)ctx_id; |
abf137dd | 848 | struct mm_struct *mm = current->mm; |
65c24491 | 849 | struct kioctx *ctx, *ret = NULL; |
db446a08 BL |
850 | struct kioctx_table *table; |
851 | unsigned id; | |
852 | ||
853 | if (get_user(id, &ring->id)) | |
854 | return NULL; | |
1da177e4 | 855 | |
abf137dd | 856 | rcu_read_lock(); |
db446a08 | 857 | table = rcu_dereference(mm->ioctx_table); |
abf137dd | 858 | |
db446a08 BL |
859 | if (!table || id >= table->nr) |
860 | goto out; | |
1da177e4 | 861 | |
db446a08 | 862 | ctx = table->table[id]; |
f30d704f | 863 | if (ctx && ctx->user_id == ctx_id) { |
db446a08 BL |
864 | percpu_ref_get(&ctx->users); |
865 | ret = ctx; | |
866 | } | |
867 | out: | |
abf137dd | 868 | rcu_read_unlock(); |
65c24491 | 869 | return ret; |
1da177e4 LT |
870 | } |
871 | ||
1da177e4 LT |
872 | /* aio_complete |
873 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 874 | */ |
2d68449e | 875 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
876 | { |
877 | struct kioctx *ctx = iocb->ki_ctx; | |
1da177e4 | 878 | struct aio_ring *ring; |
21b40200 | 879 | struct io_event *ev_page, *event; |
1da177e4 | 880 | unsigned long flags; |
21b40200 | 881 | unsigned tail, pos; |
1da177e4 | 882 | |
20dcae32 ZB |
883 | /* |
884 | * Special case handling for sync iocbs: | |
885 | * - events go directly into the iocb for fast handling | |
886 | * - the sync task with the iocb in its stack holds the single iocb | |
887 | * ref, no other paths have a way to get another ref | |
888 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
889 | */ |
890 | if (is_sync_kiocb(iocb)) { | |
1da177e4 | 891 | iocb->ki_user_data = res; |
57282d8f KO |
892 | smp_wmb(); |
893 | iocb->ki_ctx = ERR_PTR(-EXDEV); | |
1da177e4 | 894 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 895 | return; |
1da177e4 LT |
896 | } |
897 | ||
0460fef2 KO |
898 | if (iocb->ki_list.next) { |
899 | unsigned long flags; | |
900 | ||
901 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
902 | list_del(&iocb->ki_list); | |
903 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
904 | } | |
11599eba | 905 | |
0460fef2 KO |
906 | /* |
907 | * Add a completion event to the ring buffer. Must be done holding | |
4b30f07e | 908 | * ctx->completion_lock to prevent other code from messing with the tail |
0460fef2 KO |
909 | * pointer since we might be called from irq context. |
910 | */ | |
911 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
912 | ||
58c85dc2 | 913 | tail = ctx->tail; |
21b40200 KO |
914 | pos = tail + AIO_EVENTS_OFFSET; |
915 | ||
58c85dc2 | 916 | if (++tail >= ctx->nr_events) |
4bf69b2a | 917 | tail = 0; |
1da177e4 | 918 | |
58c85dc2 | 919 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
920 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
921 | ||
1da177e4 LT |
922 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
923 | event->data = iocb->ki_user_data; | |
924 | event->res = res; | |
925 | event->res2 = res2; | |
926 | ||
21b40200 | 927 | kunmap_atomic(ev_page); |
58c85dc2 | 928 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
929 | |
930 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
caf4167a KO |
931 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
932 | res, res2); | |
1da177e4 LT |
933 | |
934 | /* after flagging the request as done, we | |
935 | * must never even look at it again | |
936 | */ | |
937 | smp_wmb(); /* make event visible before updating tail */ | |
938 | ||
58c85dc2 | 939 | ctx->tail = tail; |
1da177e4 | 940 | |
58c85dc2 | 941 | ring = kmap_atomic(ctx->ring_pages[0]); |
21b40200 | 942 | ring->tail = tail; |
e8e3c3d6 | 943 | kunmap_atomic(ring); |
58c85dc2 | 944 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 | 945 | |
0460fef2 KO |
946 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
947 | ||
21b40200 | 948 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
8d1c98b0 DL |
949 | |
950 | /* | |
951 | * Check if the user asked us to deliver the result through an | |
952 | * eventfd. The eventfd_signal() function is safe to be called | |
953 | * from IRQ context. | |
954 | */ | |
87c3a86e | 955 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
956 | eventfd_signal(iocb->ki_eventfd, 1); |
957 | ||
1da177e4 | 958 | /* everything turned out well, dispose of the aiocb. */ |
57282d8f | 959 | kiocb_free(iocb); |
1da177e4 | 960 | |
6cb2a210 QB |
961 | /* |
962 | * We have to order our ring_info tail store above and test | |
963 | * of the wait list below outside the wait lock. This is | |
964 | * like in wake_up_bit() where clearing a bit has to be | |
965 | * ordered with the unlocked test. | |
966 | */ | |
967 | smp_mb(); | |
968 | ||
1da177e4 LT |
969 | if (waitqueue_active(&ctx->wait)) |
970 | wake_up(&ctx->wait); | |
971 | ||
e34ecee2 | 972 | percpu_ref_put(&ctx->reqs); |
1da177e4 | 973 | } |
385773e0 | 974 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 975 | |
a31ad380 KO |
976 | /* aio_read_events |
977 | * Pull an event off of the ioctx's event ring. Returns the number of | |
978 | * events fetched | |
1da177e4 | 979 | */ |
a31ad380 KO |
980 | static long aio_read_events_ring(struct kioctx *ctx, |
981 | struct io_event __user *event, long nr) | |
1da177e4 | 982 | { |
1da177e4 | 983 | struct aio_ring *ring; |
5ffac122 | 984 | unsigned head, tail, pos; |
a31ad380 KO |
985 | long ret = 0; |
986 | int copy_ret; | |
987 | ||
58c85dc2 | 988 | mutex_lock(&ctx->ring_lock); |
1da177e4 | 989 | |
58c85dc2 | 990 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 991 | head = ring->head; |
5ffac122 | 992 | tail = ring->tail; |
a31ad380 KO |
993 | kunmap_atomic(ring); |
994 | ||
5ffac122 | 995 | pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); |
1da177e4 | 996 | |
5ffac122 | 997 | if (head == tail) |
1da177e4 LT |
998 | goto out; |
999 | ||
a31ad380 KO |
1000 | while (ret < nr) { |
1001 | long avail; | |
1002 | struct io_event *ev; | |
1003 | struct page *page; | |
1004 | ||
5ffac122 KO |
1005 | avail = (head <= tail ? tail : ctx->nr_events) - head; |
1006 | if (head == tail) | |
a31ad380 KO |
1007 | break; |
1008 | ||
1009 | avail = min(avail, nr - ret); | |
1010 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
1011 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
1012 | ||
1013 | pos = head + AIO_EVENTS_OFFSET; | |
58c85dc2 | 1014 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
a31ad380 KO |
1015 | pos %= AIO_EVENTS_PER_PAGE; |
1016 | ||
1017 | ev = kmap(page); | |
1018 | copy_ret = copy_to_user(event + ret, ev + pos, | |
1019 | sizeof(*ev) * avail); | |
1020 | kunmap(page); | |
1021 | ||
1022 | if (unlikely(copy_ret)) { | |
1023 | ret = -EFAULT; | |
1024 | goto out; | |
1025 | } | |
1026 | ||
1027 | ret += avail; | |
1028 | head += avail; | |
58c85dc2 | 1029 | head %= ctx->nr_events; |
1da177e4 | 1030 | } |
1da177e4 | 1031 | |
58c85dc2 | 1032 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 1033 | ring->head = head; |
91d80a84 | 1034 | kunmap_atomic(ring); |
58c85dc2 | 1035 | flush_dcache_page(ctx->ring_pages[0]); |
a31ad380 | 1036 | |
5ffac122 | 1037 | pr_debug("%li h%u t%u\n", ret, head, tail); |
3e845ce0 | 1038 | |
e1bdd5f2 | 1039 | put_reqs_available(ctx, ret); |
a31ad380 | 1040 | out: |
58c85dc2 | 1041 | mutex_unlock(&ctx->ring_lock); |
a31ad380 | 1042 | |
1da177e4 LT |
1043 | return ret; |
1044 | } | |
1045 | ||
a31ad380 KO |
1046 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
1047 | struct io_event __user *event, long *i) | |
1da177e4 | 1048 | { |
a31ad380 | 1049 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 1050 | |
a31ad380 KO |
1051 | if (ret > 0) |
1052 | *i += ret; | |
1da177e4 | 1053 | |
a31ad380 KO |
1054 | if (unlikely(atomic_read(&ctx->dead))) |
1055 | ret = -EINVAL; | |
1da177e4 | 1056 | |
a31ad380 KO |
1057 | if (!*i) |
1058 | *i = ret; | |
1da177e4 | 1059 | |
a31ad380 | 1060 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
1061 | } |
1062 | ||
a31ad380 | 1063 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
1064 | struct io_event __user *event, |
1065 | struct timespec __user *timeout) | |
1066 | { | |
a31ad380 KO |
1067 | ktime_t until = { .tv64 = KTIME_MAX }; |
1068 | long ret = 0; | |
1da177e4 | 1069 | |
1da177e4 LT |
1070 | if (timeout) { |
1071 | struct timespec ts; | |
a31ad380 | 1072 | |
1da177e4 | 1073 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 1074 | return -EFAULT; |
1da177e4 | 1075 | |
a31ad380 | 1076 | until = timespec_to_ktime(ts); |
1da177e4 LT |
1077 | } |
1078 | ||
a31ad380 KO |
1079 | /* |
1080 | * Note that aio_read_events() is being called as the conditional - i.e. | |
1081 | * we're calling it after prepare_to_wait() has set task state to | |
1082 | * TASK_INTERRUPTIBLE. | |
1083 | * | |
1084 | * But aio_read_events() can block, and if it blocks it's going to flip | |
1085 | * the task state back to TASK_RUNNING. | |
1086 | * | |
1087 | * This should be ok, provided it doesn't flip the state back to | |
1088 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
1089 | * will only happen if the mutex_lock() call blocks, and we then find | |
1090 | * the ringbuffer empty. So in practice we should be ok, but it's | |
1091 | * something to be aware of when touching this code. | |
1092 | */ | |
1093 | wait_event_interruptible_hrtimeout(ctx->wait, | |
1094 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
1da177e4 | 1095 | |
a31ad380 KO |
1096 | if (!ret && signal_pending(current)) |
1097 | ret = -EINTR; | |
1da177e4 | 1098 | |
a31ad380 | 1099 | return ret; |
1da177e4 LT |
1100 | } |
1101 | ||
1da177e4 LT |
1102 | /* sys_io_setup: |
1103 | * Create an aio_context capable of receiving at least nr_events. | |
1104 | * ctxp must not point to an aio_context that already exists, and | |
1105 | * must be initialized to 0 prior to the call. On successful | |
1106 | * creation of the aio_context, *ctxp is filled in with the resulting | |
1107 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
1108 | * if the specified nr_events exceeds internal limits. May fail | |
1109 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
1110 | * of available events. May fail with -ENOMEM if insufficient kernel | |
1111 | * resources are available. May fail with -EFAULT if an invalid | |
1112 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
1113 | * implemented. | |
1114 | */ | |
002c8976 | 1115 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
1116 | { |
1117 | struct kioctx *ioctx = NULL; | |
1118 | unsigned long ctx; | |
1119 | long ret; | |
1120 | ||
1121 | ret = get_user(ctx, ctxp); | |
1122 | if (unlikely(ret)) | |
1123 | goto out; | |
1124 | ||
1125 | ret = -EINVAL; | |
d55b5fda ZB |
1126 | if (unlikely(ctx || nr_events == 0)) { |
1127 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
1128 | ctx, nr_events); | |
1da177e4 LT |
1129 | goto out; |
1130 | } | |
1131 | ||
1132 | ioctx = ioctx_alloc(nr_events); | |
1133 | ret = PTR_ERR(ioctx); | |
1134 | if (!IS_ERR(ioctx)) { | |
1135 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a | 1136 | if (ret) |
db446a08 | 1137 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1138 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1139 | } |
1140 | ||
1141 | out: | |
1142 | return ret; | |
1143 | } | |
1144 | ||
1145 | /* sys_io_destroy: | |
1146 | * Destroy the aio_context specified. May cancel any outstanding | |
1147 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 1148 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
1149 | * is invalid. |
1150 | */ | |
002c8976 | 1151 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
1152 | { |
1153 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
1154 | if (likely(NULL != ioctx)) { | |
db446a08 | 1155 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1156 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1157 | return 0; |
1158 | } | |
1159 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
1160 | return -EINVAL; | |
1161 | } | |
1162 | ||
41ef4eb8 KO |
1163 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
1164 | unsigned long, loff_t); | |
1165 | ||
8bc92afc KO |
1166 | static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb, |
1167 | int rw, char __user *buf, | |
1168 | unsigned long *nr_segs, | |
1169 | struct iovec **iovec, | |
1170 | bool compat) | |
eed4e51f BP |
1171 | { |
1172 | ssize_t ret; | |
1173 | ||
8bc92afc | 1174 | *nr_segs = kiocb->ki_nbytes; |
41ef4eb8 | 1175 | |
9d85cba7 JM |
1176 | #ifdef CONFIG_COMPAT |
1177 | if (compat) | |
41ef4eb8 | 1178 | ret = compat_rw_copy_check_uvector(rw, |
8bc92afc KO |
1179 | (struct compat_iovec __user *)buf, |
1180 | *nr_segs, 1, *iovec, iovec); | |
9d85cba7 JM |
1181 | else |
1182 | #endif | |
41ef4eb8 | 1183 | ret = rw_copy_check_uvector(rw, |
8bc92afc KO |
1184 | (struct iovec __user *)buf, |
1185 | *nr_segs, 1, *iovec, iovec); | |
eed4e51f | 1186 | if (ret < 0) |
41ef4eb8 | 1187 | return ret; |
a70b52ec | 1188 | |
41ef4eb8 | 1189 | /* ki_nbytes now reflect bytes instead of segs */ |
eed4e51f | 1190 | kiocb->ki_nbytes = ret; |
41ef4eb8 | 1191 | return 0; |
eed4e51f BP |
1192 | } |
1193 | ||
8bc92afc KO |
1194 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb, |
1195 | int rw, char __user *buf, | |
1196 | unsigned long *nr_segs, | |
1197 | struct iovec *iovec) | |
eed4e51f | 1198 | { |
8bc92afc | 1199 | if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes))) |
41ef4eb8 | 1200 | return -EFAULT; |
a70b52ec | 1201 | |
8bc92afc KO |
1202 | iovec->iov_base = buf; |
1203 | iovec->iov_len = kiocb->ki_nbytes; | |
1204 | *nr_segs = 1; | |
eed4e51f BP |
1205 | return 0; |
1206 | } | |
1207 | ||
1da177e4 LT |
1208 | /* |
1209 | * aio_setup_iocb: | |
1210 | * Performs the initial checks and aio retry method | |
1211 | * setup for the kiocb at the time of io submission. | |
1212 | */ | |
8bc92afc KO |
1213 | static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode, |
1214 | char __user *buf, bool compat) | |
1da177e4 | 1215 | { |
41ef4eb8 KO |
1216 | struct file *file = req->ki_filp; |
1217 | ssize_t ret; | |
8bc92afc | 1218 | unsigned long nr_segs; |
41ef4eb8 KO |
1219 | int rw; |
1220 | fmode_t mode; | |
1221 | aio_rw_op *rw_op; | |
8bc92afc | 1222 | struct iovec inline_vec, *iovec = &inline_vec; |
1da177e4 | 1223 | |
8bc92afc | 1224 | switch (opcode) { |
1da177e4 | 1225 | case IOCB_CMD_PREAD: |
eed4e51f | 1226 | case IOCB_CMD_PREADV: |
41ef4eb8 KO |
1227 | mode = FMODE_READ; |
1228 | rw = READ; | |
1229 | rw_op = file->f_op->aio_read; | |
1230 | goto rw_common; | |
1231 | ||
1232 | case IOCB_CMD_PWRITE: | |
eed4e51f | 1233 | case IOCB_CMD_PWRITEV: |
41ef4eb8 KO |
1234 | mode = FMODE_WRITE; |
1235 | rw = WRITE; | |
1236 | rw_op = file->f_op->aio_write; | |
1237 | goto rw_common; | |
1238 | rw_common: | |
1239 | if (unlikely(!(file->f_mode & mode))) | |
1240 | return -EBADF; | |
1241 | ||
1242 | if (!rw_op) | |
1243 | return -EINVAL; | |
1244 | ||
8bc92afc KO |
1245 | ret = (opcode == IOCB_CMD_PREADV || |
1246 | opcode == IOCB_CMD_PWRITEV) | |
1247 | ? aio_setup_vectored_rw(req, rw, buf, &nr_segs, | |
1248 | &iovec, compat) | |
1249 | : aio_setup_single_vector(req, rw, buf, &nr_segs, | |
1250 | iovec); | |
eed4e51f | 1251 | if (ret) |
41ef4eb8 KO |
1252 | return ret; |
1253 | ||
1254 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); | |
8bc92afc KO |
1255 | if (ret < 0) { |
1256 | if (iovec != &inline_vec) | |
1257 | kfree(iovec); | |
41ef4eb8 | 1258 | return ret; |
8bc92afc | 1259 | } |
41ef4eb8 KO |
1260 | |
1261 | req->ki_nbytes = ret; | |
41ef4eb8 | 1262 | |
73a7075e KO |
1263 | /* XXX: move/kill - rw_verify_area()? */ |
1264 | /* This matches the pread()/pwrite() logic */ | |
1265 | if (req->ki_pos < 0) { | |
1266 | ret = -EINVAL; | |
1267 | break; | |
1268 | } | |
1269 | ||
1270 | if (rw == WRITE) | |
1271 | file_start_write(file); | |
1272 | ||
8bc92afc | 1273 | ret = rw_op(req, iovec, nr_segs, req->ki_pos); |
73a7075e KO |
1274 | |
1275 | if (rw == WRITE) | |
1276 | file_end_write(file); | |
1da177e4 | 1277 | break; |
41ef4eb8 | 1278 | |
1da177e4 | 1279 | case IOCB_CMD_FDSYNC: |
41ef4eb8 KO |
1280 | if (!file->f_op->aio_fsync) |
1281 | return -EINVAL; | |
1282 | ||
1283 | ret = file->f_op->aio_fsync(req, 1); | |
1da177e4 | 1284 | break; |
41ef4eb8 | 1285 | |
1da177e4 | 1286 | case IOCB_CMD_FSYNC: |
41ef4eb8 KO |
1287 | if (!file->f_op->aio_fsync) |
1288 | return -EINVAL; | |
1289 | ||
1290 | ret = file->f_op->aio_fsync(req, 0); | |
1da177e4 | 1291 | break; |
41ef4eb8 | 1292 | |
1da177e4 | 1293 | default: |
caf4167a | 1294 | pr_debug("EINVAL: no operation provided\n"); |
41ef4eb8 | 1295 | return -EINVAL; |
1da177e4 LT |
1296 | } |
1297 | ||
8bc92afc KO |
1298 | if (iovec != &inline_vec) |
1299 | kfree(iovec); | |
1300 | ||
41ef4eb8 KO |
1301 | if (ret != -EIOCBQUEUED) { |
1302 | /* | |
1303 | * There's no easy way to restart the syscall since other AIO's | |
1304 | * may be already running. Just fail this IO with EINTR. | |
1305 | */ | |
1306 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1307 | ret == -ERESTARTNOHAND || | |
1308 | ret == -ERESTART_RESTARTBLOCK)) | |
1309 | ret = -EINTR; | |
1310 | aio_complete(req, ret, 0); | |
1311 | } | |
1da177e4 LT |
1312 | |
1313 | return 0; | |
1314 | } | |
1315 | ||
d5470b59 | 1316 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
a1c8eae7 | 1317 | struct iocb *iocb, bool compat) |
1da177e4 LT |
1318 | { |
1319 | struct kiocb *req; | |
1da177e4 LT |
1320 | ssize_t ret; |
1321 | ||
1322 | /* enforce forwards compatibility on users */ | |
9c3060be | 1323 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1324 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1325 | return -EINVAL; |
1326 | } | |
1327 | ||
1328 | /* prevent overflows */ | |
1329 | if (unlikely( | |
1330 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1331 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1332 | ((ssize_t)iocb->aio_nbytes < 0) | |
1333 | )) { | |
1334 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1335 | return -EINVAL; | |
1336 | } | |
1337 | ||
41ef4eb8 | 1338 | req = aio_get_req(ctx); |
1d98ebfc | 1339 | if (unlikely(!req)) |
1da177e4 | 1340 | return -EAGAIN; |
1d98ebfc KO |
1341 | |
1342 | req->ki_filp = fget(iocb->aio_fildes); | |
1343 | if (unlikely(!req->ki_filp)) { | |
1344 | ret = -EBADF; | |
1345 | goto out_put_req; | |
1da177e4 | 1346 | } |
1d98ebfc | 1347 | |
9c3060be DL |
1348 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1349 | /* | |
1350 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1351 | * instance of the file* now. The file descriptor must be | |
1352 | * an eventfd() fd, and will be signaled for each completed | |
1353 | * event using the eventfd_signal() function. | |
1354 | */ | |
13389010 | 1355 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1356 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1357 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1358 | req->ki_eventfd = NULL; |
9c3060be DL |
1359 | goto out_put_req; |
1360 | } | |
1361 | } | |
1da177e4 | 1362 | |
8a660890 | 1363 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
1da177e4 | 1364 | if (unlikely(ret)) { |
caf4167a | 1365 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1366 | goto out_put_req; |
1367 | } | |
1368 | ||
1369 | req->ki_obj.user = user_iocb; | |
1370 | req->ki_user_data = iocb->aio_data; | |
1371 | req->ki_pos = iocb->aio_offset; | |
73a7075e | 1372 | req->ki_nbytes = iocb->aio_nbytes; |
1da177e4 | 1373 | |
8bc92afc KO |
1374 | ret = aio_run_iocb(req, iocb->aio_lio_opcode, |
1375 | (char __user *)(unsigned long)iocb->aio_buf, | |
1376 | compat); | |
41003a7b | 1377 | if (ret) |
7137c6bd | 1378 | goto out_put_req; |
41003a7b | 1379 | |
1da177e4 | 1380 | return 0; |
1da177e4 | 1381 | out_put_req: |
e1bdd5f2 | 1382 | put_reqs_available(ctx, 1); |
e34ecee2 | 1383 | percpu_ref_put(&ctx->reqs); |
57282d8f | 1384 | kiocb_free(req); |
1da177e4 LT |
1385 | return ret; |
1386 | } | |
1387 | ||
9d85cba7 JM |
1388 | long do_io_submit(aio_context_t ctx_id, long nr, |
1389 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1390 | { |
1391 | struct kioctx *ctx; | |
1392 | long ret = 0; | |
080d676d | 1393 | int i = 0; |
9f5b9425 | 1394 | struct blk_plug plug; |
1da177e4 LT |
1395 | |
1396 | if (unlikely(nr < 0)) | |
1397 | return -EINVAL; | |
1398 | ||
75e1c70f JM |
1399 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1400 | nr = LONG_MAX/sizeof(*iocbpp); | |
1401 | ||
1da177e4 LT |
1402 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1403 | return -EFAULT; | |
1404 | ||
1405 | ctx = lookup_ioctx(ctx_id); | |
1406 | if (unlikely(!ctx)) { | |
caf4167a | 1407 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1408 | return -EINVAL; |
1409 | } | |
1410 | ||
9f5b9425 SL |
1411 | blk_start_plug(&plug); |
1412 | ||
1da177e4 LT |
1413 | /* |
1414 | * AKPM: should this return a partial result if some of the IOs were | |
1415 | * successfully submitted? | |
1416 | */ | |
1417 | for (i=0; i<nr; i++) { | |
1418 | struct iocb __user *user_iocb; | |
1419 | struct iocb tmp; | |
1420 | ||
1421 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1422 | ret = -EFAULT; | |
1423 | break; | |
1424 | } | |
1425 | ||
1426 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1427 | ret = -EFAULT; | |
1428 | break; | |
1429 | } | |
1430 | ||
a1c8eae7 | 1431 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
1da177e4 LT |
1432 | if (ret) |
1433 | break; | |
1434 | } | |
9f5b9425 | 1435 | blk_finish_plug(&plug); |
1da177e4 | 1436 | |
723be6e3 | 1437 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1438 | return i ? i : ret; |
1439 | } | |
1440 | ||
9d85cba7 JM |
1441 | /* sys_io_submit: |
1442 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1443 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1444 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1445 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1446 | * is invalid for the file descriptor in the iocb. May fail with | |
1447 | * -EFAULT if any of the data structures point to invalid data. May | |
1448 | * fail with -EBADF if the file descriptor specified in the first | |
1449 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1450 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1451 | * fail with -ENOSYS if not implemented. | |
1452 | */ | |
1453 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1454 | struct iocb __user * __user *, iocbpp) | |
1455 | { | |
1456 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1457 | } | |
1458 | ||
1da177e4 LT |
1459 | /* lookup_kiocb |
1460 | * Finds a given iocb for cancellation. | |
1da177e4 | 1461 | */ |
25ee7e38 AB |
1462 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1463 | u32 key) | |
1da177e4 LT |
1464 | { |
1465 | struct list_head *pos; | |
d00689af ZB |
1466 | |
1467 | assert_spin_locked(&ctx->ctx_lock); | |
1468 | ||
8a660890 KO |
1469 | if (key != KIOCB_KEY) |
1470 | return NULL; | |
1471 | ||
1da177e4 LT |
1472 | /* TODO: use a hash or array, this sucks. */ |
1473 | list_for_each(pos, &ctx->active_reqs) { | |
1474 | struct kiocb *kiocb = list_kiocb(pos); | |
8a660890 | 1475 | if (kiocb->ki_obj.user == iocb) |
1da177e4 LT |
1476 | return kiocb; |
1477 | } | |
1478 | return NULL; | |
1479 | } | |
1480 | ||
1481 | /* sys_io_cancel: | |
1482 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1483 | * the operation is successfully cancelled, the resulting event is | |
1484 | * copied into the memory pointed to by result without being placed | |
1485 | * into the completion queue and 0 is returned. May fail with | |
1486 | * -EFAULT if any of the data structures pointed to are invalid. | |
1487 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1488 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1489 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1490 | */ | |
002c8976 HC |
1491 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1492 | struct io_event __user *, result) | |
1da177e4 | 1493 | { |
1da177e4 LT |
1494 | struct kioctx *ctx; |
1495 | struct kiocb *kiocb; | |
1496 | u32 key; | |
1497 | int ret; | |
1498 | ||
1499 | ret = get_user(key, &iocb->aio_key); | |
1500 | if (unlikely(ret)) | |
1501 | return -EFAULT; | |
1502 | ||
1503 | ctx = lookup_ioctx(ctx_id); | |
1504 | if (unlikely(!ctx)) | |
1505 | return -EINVAL; | |
1506 | ||
1507 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1508 | |
1da177e4 | 1509 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c | 1510 | if (kiocb) |
bec68faa | 1511 | ret = kiocb_cancel(ctx, kiocb); |
906b973c KO |
1512 | else |
1513 | ret = -EINVAL; | |
1514 | ||
1da177e4 LT |
1515 | spin_unlock_irq(&ctx->ctx_lock); |
1516 | ||
906b973c | 1517 | if (!ret) { |
bec68faa KO |
1518 | /* |
1519 | * The result argument is no longer used - the io_event is | |
1520 | * always delivered via the ring buffer. -EINPROGRESS indicates | |
1521 | * cancellation is progress: | |
906b973c | 1522 | */ |
bec68faa | 1523 | ret = -EINPROGRESS; |
906b973c | 1524 | } |
1da177e4 | 1525 | |
723be6e3 | 1526 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1527 | |
1528 | return ret; | |
1529 | } | |
1530 | ||
1531 | /* io_getevents: | |
1532 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1533 | * the completion queue for the aio_context specified by ctx_id. If |
1534 | * it succeeds, the number of read events is returned. May fail with | |
1535 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1536 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1537 | * if any of the memory specified is invalid. May return 0 or | |
1538 | * < min_nr if the timeout specified by timeout has elapsed | |
1539 | * before sufficient events are available, where timeout == NULL | |
1540 | * specifies an infinite timeout. Note that the timeout pointed to by | |
6900807c | 1541 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
1da177e4 | 1542 | */ |
002c8976 HC |
1543 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1544 | long, min_nr, | |
1545 | long, nr, | |
1546 | struct io_event __user *, events, | |
1547 | struct timespec __user *, timeout) | |
1da177e4 LT |
1548 | { |
1549 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1550 | long ret = -EINVAL; | |
1551 | ||
1552 | if (likely(ioctx)) { | |
2e410255 | 1553 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 | 1554 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
723be6e3 | 1555 | percpu_ref_put(&ioctx->users); |
1da177e4 | 1556 | } |
1da177e4 LT |
1557 | return ret; |
1558 | } |