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; |
41badc15 | 329 | unsigned long size, populate; |
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 | ||
36bc08cc GZ |
350 | for (i = 0; i < nr_pages; i++) { |
351 | struct page *page; | |
352 | page = find_or_create_page(file->f_inode->i_mapping, | |
353 | i, GFP_HIGHUSER | __GFP_ZERO); | |
354 | if (!page) | |
355 | break; | |
356 | pr_debug("pid(%d) page[%d]->count=%d\n", | |
357 | current->pid, i, page_count(page)); | |
358 | SetPageUptodate(page); | |
359 | SetPageDirty(page); | |
360 | unlock_page(page); | |
361 | } | |
362 | ctx->aio_ring_file = file; | |
363 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) | |
364 | / sizeof(struct io_event); | |
1da177e4 | 365 | |
58c85dc2 | 366 | ctx->ring_pages = ctx->internal_pages; |
1da177e4 | 367 | if (nr_pages > AIO_RING_PAGES) { |
58c85dc2 KO |
368 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
369 | GFP_KERNEL); | |
d1b94327 GZ |
370 | if (!ctx->ring_pages) { |
371 | put_aio_ring_file(ctx); | |
1da177e4 | 372 | return -ENOMEM; |
d1b94327 | 373 | } |
1da177e4 LT |
374 | } |
375 | ||
58c85dc2 KO |
376 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
377 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
36bc08cc | 378 | |
41003a7b | 379 | down_write(&mm->mmap_sem); |
36bc08cc GZ |
380 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
381 | PROT_READ | PROT_WRITE, | |
382 | MAP_SHARED | MAP_POPULATE, 0, &populate); | |
58c85dc2 | 383 | if (IS_ERR((void *)ctx->mmap_base)) { |
41003a7b | 384 | up_write(&mm->mmap_sem); |
58c85dc2 | 385 | ctx->mmap_size = 0; |
1da177e4 LT |
386 | aio_free_ring(ctx); |
387 | return -EAGAIN; | |
388 | } | |
389 | ||
58c85dc2 | 390 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
d6c355c7 BL |
391 | |
392 | /* We must do this while still holding mmap_sem for write, as we | |
393 | * need to be protected against userspace attempting to mremap() | |
394 | * or munmap() the ring buffer. | |
395 | */ | |
58c85dc2 KO |
396 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, |
397 | 1, 0, ctx->ring_pages, NULL); | |
d6c355c7 BL |
398 | |
399 | /* Dropping the reference here is safe as the page cache will hold | |
400 | * onto the pages for us. It is also required so that page migration | |
401 | * can unmap the pages and get the right reference count. | |
402 | */ | |
36bc08cc GZ |
403 | for (i = 0; i < ctx->nr_pages; i++) |
404 | put_page(ctx->ring_pages[i]); | |
1da177e4 | 405 | |
d6c355c7 BL |
406 | up_write(&mm->mmap_sem); |
407 | ||
58c85dc2 | 408 | if (unlikely(ctx->nr_pages != nr_pages)) { |
1da177e4 LT |
409 | aio_free_ring(ctx); |
410 | return -EAGAIN; | |
411 | } | |
412 | ||
58c85dc2 KO |
413 | ctx->user_id = ctx->mmap_base; |
414 | ctx->nr_events = nr_events; /* trusted copy */ | |
1da177e4 | 415 | |
58c85dc2 | 416 | ring = kmap_atomic(ctx->ring_pages[0]); |
1da177e4 | 417 | ring->nr = nr_events; /* user copy */ |
db446a08 | 418 | ring->id = ~0U; |
1da177e4 LT |
419 | ring->head = ring->tail = 0; |
420 | ring->magic = AIO_RING_MAGIC; | |
421 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
422 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
423 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 424 | kunmap_atomic(ring); |
58c85dc2 | 425 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 LT |
426 | |
427 | return 0; | |
428 | } | |
429 | ||
1da177e4 LT |
430 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
431 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
432 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
433 | ||
0460fef2 KO |
434 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
435 | { | |
436 | struct kioctx *ctx = req->ki_ctx; | |
437 | unsigned long flags; | |
438 | ||
439 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
440 | ||
441 | if (!req->ki_list.next) | |
442 | list_add(&req->ki_list, &ctx->active_reqs); | |
443 | ||
444 | req->ki_cancel = cancel; | |
445 | ||
446 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
447 | } | |
448 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
449 | ||
bec68faa | 450 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb) |
906b973c | 451 | { |
0460fef2 | 452 | kiocb_cancel_fn *old, *cancel; |
906b973c | 453 | |
0460fef2 KO |
454 | /* |
455 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
456 | * actually has a cancel function, hence the cmpxchg() | |
457 | */ | |
458 | ||
459 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
460 | do { | |
461 | if (!cancel || cancel == KIOCB_CANCELLED) | |
57282d8f | 462 | return -EINVAL; |
906b973c | 463 | |
0460fef2 KO |
464 | old = cancel; |
465 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
466 | } while (cancel != old); | |
906b973c | 467 | |
57282d8f | 468 | return cancel(kiocb); |
906b973c KO |
469 | } |
470 | ||
e34ecee2 | 471 | static void free_ioctx(struct work_struct *work) |
36f55889 | 472 | { |
e34ecee2 | 473 | struct kioctx *ctx = container_of(work, struct kioctx, free_work); |
e1bdd5f2 | 474 | |
e34ecee2 | 475 | pr_debug("freeing %p\n", ctx); |
e1bdd5f2 | 476 | |
e34ecee2 | 477 | aio_free_ring(ctx); |
e1bdd5f2 | 478 | free_percpu(ctx->cpu); |
36f55889 KO |
479 | kmem_cache_free(kioctx_cachep, ctx); |
480 | } | |
481 | ||
e34ecee2 KO |
482 | static void free_ioctx_reqs(struct percpu_ref *ref) |
483 | { | |
484 | struct kioctx *ctx = container_of(ref, struct kioctx, reqs); | |
485 | ||
486 | INIT_WORK(&ctx->free_work, free_ioctx); | |
487 | schedule_work(&ctx->free_work); | |
488 | } | |
489 | ||
36f55889 KO |
490 | /* |
491 | * When this function runs, the kioctx has been removed from the "hash table" | |
492 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
493 | * now it's safe to cancel any that need to be. | |
494 | */ | |
e34ecee2 | 495 | static void free_ioctx_users(struct percpu_ref *ref) |
36f55889 | 496 | { |
e34ecee2 | 497 | struct kioctx *ctx = container_of(ref, struct kioctx, users); |
36f55889 KO |
498 | struct kiocb *req; |
499 | ||
500 | spin_lock_irq(&ctx->ctx_lock); | |
501 | ||
502 | while (!list_empty(&ctx->active_reqs)) { | |
503 | req = list_first_entry(&ctx->active_reqs, | |
504 | struct kiocb, ki_list); | |
505 | ||
506 | list_del_init(&req->ki_list); | |
bec68faa | 507 | kiocb_cancel(ctx, req); |
36f55889 KO |
508 | } |
509 | ||
510 | spin_unlock_irq(&ctx->ctx_lock); | |
511 | ||
e34ecee2 KO |
512 | percpu_ref_kill(&ctx->reqs); |
513 | percpu_ref_put(&ctx->reqs); | |
36f55889 KO |
514 | } |
515 | ||
db446a08 BL |
516 | static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) |
517 | { | |
518 | unsigned i, new_nr; | |
519 | struct kioctx_table *table, *old; | |
520 | struct aio_ring *ring; | |
521 | ||
522 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 523 | rcu_read_lock(); |
77d30b14 | 524 | table = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
525 | |
526 | while (1) { | |
527 | if (table) | |
528 | for (i = 0; i < table->nr; i++) | |
529 | if (!table->table[i]) { | |
530 | ctx->id = i; | |
531 | table->table[i] = ctx; | |
d9b2c871 | 532 | rcu_read_unlock(); |
db446a08 BL |
533 | spin_unlock(&mm->ioctx_lock); |
534 | ||
535 | ring = kmap_atomic(ctx->ring_pages[0]); | |
536 | ring->id = ctx->id; | |
537 | kunmap_atomic(ring); | |
538 | return 0; | |
539 | } | |
540 | ||
541 | new_nr = (table ? table->nr : 1) * 4; | |
542 | ||
d9b2c871 | 543 | rcu_read_unlock(); |
db446a08 BL |
544 | spin_unlock(&mm->ioctx_lock); |
545 | ||
546 | table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * | |
547 | new_nr, GFP_KERNEL); | |
548 | if (!table) | |
549 | return -ENOMEM; | |
550 | ||
551 | table->nr = new_nr; | |
552 | ||
553 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 554 | rcu_read_lock(); |
77d30b14 | 555 | old = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
556 | |
557 | if (!old) { | |
558 | rcu_assign_pointer(mm->ioctx_table, table); | |
559 | } else if (table->nr > old->nr) { | |
560 | memcpy(table->table, old->table, | |
561 | old->nr * sizeof(struct kioctx *)); | |
562 | ||
563 | rcu_assign_pointer(mm->ioctx_table, table); | |
564 | kfree_rcu(old, rcu); | |
565 | } else { | |
566 | kfree(table); | |
567 | table = old; | |
568 | } | |
569 | } | |
570 | } | |
571 | ||
e34ecee2 KO |
572 | static void aio_nr_sub(unsigned nr) |
573 | { | |
574 | spin_lock(&aio_nr_lock); | |
575 | if (WARN_ON(aio_nr - nr > aio_nr)) | |
576 | aio_nr = 0; | |
577 | else | |
578 | aio_nr -= nr; | |
579 | spin_unlock(&aio_nr_lock); | |
580 | } | |
581 | ||
1da177e4 LT |
582 | /* ioctx_alloc |
583 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
584 | */ | |
585 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
586 | { | |
41003a7b | 587 | struct mm_struct *mm = current->mm; |
1da177e4 | 588 | struct kioctx *ctx; |
e23754f8 | 589 | int err = -ENOMEM; |
1da177e4 | 590 | |
e1bdd5f2 KO |
591 | /* |
592 | * We keep track of the number of available ringbuffer slots, to prevent | |
593 | * overflow (reqs_available), and we also use percpu counters for this. | |
594 | * | |
595 | * So since up to half the slots might be on other cpu's percpu counters | |
596 | * and unavailable, double nr_events so userspace sees what they | |
597 | * expected: additionally, we move req_batch slots to/from percpu | |
598 | * counters at a time, so make sure that isn't 0: | |
599 | */ | |
600 | nr_events = max(nr_events, num_possible_cpus() * 4); | |
601 | nr_events *= 2; | |
602 | ||
1da177e4 LT |
603 | /* Prevent overflows */ |
604 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
605 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
606 | pr_debug("ENOMEM: nr_events too high\n"); | |
607 | return ERR_PTR(-EINVAL); | |
608 | } | |
609 | ||
4cd81c3d | 610 | if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) |
1da177e4 LT |
611 | return ERR_PTR(-EAGAIN); |
612 | ||
c3762229 | 613 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
614 | if (!ctx) |
615 | return ERR_PTR(-ENOMEM); | |
616 | ||
1da177e4 | 617 | ctx->max_reqs = nr_events; |
1da177e4 | 618 | |
e34ecee2 KO |
619 | if (percpu_ref_init(&ctx->users, free_ioctx_users)) |
620 | goto err; | |
621 | ||
622 | if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs)) | |
623 | goto err; | |
723be6e3 | 624 | |
1da177e4 | 625 | spin_lock_init(&ctx->ctx_lock); |
0460fef2 | 626 | spin_lock_init(&ctx->completion_lock); |
58c85dc2 | 627 | mutex_init(&ctx->ring_lock); |
1da177e4 LT |
628 | init_waitqueue_head(&ctx->wait); |
629 | ||
630 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 | 631 | |
e1bdd5f2 KO |
632 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
633 | if (!ctx->cpu) | |
e34ecee2 | 634 | goto err; |
1da177e4 | 635 | |
e1bdd5f2 | 636 | if (aio_setup_ring(ctx) < 0) |
e34ecee2 | 637 | goto err; |
e1bdd5f2 | 638 | |
34e83fc6 | 639 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
e1bdd5f2 | 640 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
6878ea72 BL |
641 | if (ctx->req_batch < 1) |
642 | ctx->req_batch = 1; | |
34e83fc6 | 643 | |
1da177e4 | 644 | /* limit the number of system wide aios */ |
9fa1cb39 | 645 | spin_lock(&aio_nr_lock); |
4cd81c3d | 646 | if (aio_nr + nr_events > (aio_max_nr * 2UL) || |
2dd542b7 | 647 | aio_nr + nr_events < aio_nr) { |
9fa1cb39 | 648 | spin_unlock(&aio_nr_lock); |
e34ecee2 | 649 | err = -EAGAIN; |
d1b94327 | 650 | goto err_ctx; |
2dd542b7 AV |
651 | } |
652 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 653 | spin_unlock(&aio_nr_lock); |
1da177e4 | 654 | |
723be6e3 KO |
655 | percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ |
656 | ||
da90382c BL |
657 | err = ioctx_add_table(ctx, mm); |
658 | if (err) | |
e34ecee2 | 659 | goto err_cleanup; |
da90382c | 660 | |
caf4167a | 661 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
58c85dc2 | 662 | ctx, ctx->user_id, mm, ctx->nr_events); |
1da177e4 LT |
663 | return ctx; |
664 | ||
e34ecee2 KO |
665 | err_cleanup: |
666 | aio_nr_sub(ctx->max_reqs); | |
d1b94327 GZ |
667 | err_ctx: |
668 | aio_free_ring(ctx); | |
e34ecee2 | 669 | err: |
e1bdd5f2 | 670 | free_percpu(ctx->cpu); |
e34ecee2 | 671 | free_percpu(ctx->reqs.pcpu_count); |
723be6e3 | 672 | free_percpu(ctx->users.pcpu_count); |
1da177e4 | 673 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 674 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 675 | return ERR_PTR(err); |
1da177e4 LT |
676 | } |
677 | ||
36f55889 KO |
678 | /* kill_ioctx |
679 | * Cancels all outstanding aio requests on an aio context. Used | |
680 | * when the processes owning a context have all exited to encourage | |
681 | * the rapid destruction of the kioctx. | |
682 | */ | |
db446a08 | 683 | static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx) |
36f55889 KO |
684 | { |
685 | if (!atomic_xchg(&ctx->dead, 1)) { | |
db446a08 BL |
686 | struct kioctx_table *table; |
687 | ||
688 | spin_lock(&mm->ioctx_lock); | |
d9b2c871 | 689 | rcu_read_lock(); |
77d30b14 | 690 | table = rcu_dereference(mm->ioctx_table); |
db446a08 BL |
691 | |
692 | WARN_ON(ctx != table->table[ctx->id]); | |
693 | table->table[ctx->id] = NULL; | |
d9b2c871 | 694 | rcu_read_unlock(); |
db446a08 BL |
695 | spin_unlock(&mm->ioctx_lock); |
696 | ||
723be6e3 KO |
697 | /* percpu_ref_kill() will do the necessary call_rcu() */ |
698 | wake_up_all(&ctx->wait); | |
dee11c23 | 699 | |
36f55889 | 700 | /* |
4fcc712f KO |
701 | * It'd be more correct to do this in free_ioctx(), after all |
702 | * the outstanding kiocbs have finished - but by then io_destroy | |
703 | * has already returned, so io_setup() could potentially return | |
704 | * -EAGAIN with no ioctxs actually in use (as far as userspace | |
705 | * could tell). | |
36f55889 | 706 | */ |
e34ecee2 | 707 | aio_nr_sub(ctx->max_reqs); |
4fcc712f KO |
708 | |
709 | if (ctx->mmap_size) | |
710 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
711 | ||
723be6e3 | 712 | percpu_ref_kill(&ctx->users); |
36f55889 | 713 | } |
1da177e4 LT |
714 | } |
715 | ||
716 | /* wait_on_sync_kiocb: | |
717 | * Waits on the given sync kiocb to complete. | |
718 | */ | |
57282d8f | 719 | ssize_t wait_on_sync_kiocb(struct kiocb *req) |
1da177e4 | 720 | { |
57282d8f | 721 | while (!req->ki_ctx) { |
1da177e4 | 722 | set_current_state(TASK_UNINTERRUPTIBLE); |
57282d8f | 723 | if (req->ki_ctx) |
1da177e4 | 724 | break; |
41d10da3 | 725 | io_schedule(); |
1da177e4 LT |
726 | } |
727 | __set_current_state(TASK_RUNNING); | |
57282d8f | 728 | return req->ki_user_data; |
1da177e4 | 729 | } |
385773e0 | 730 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 731 | |
36f55889 KO |
732 | /* |
733 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
734 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
735 | * called on the context. | |
736 | * | |
737 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
738 | * them. | |
1da177e4 | 739 | */ |
fc9b52cd | 740 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 741 | { |
db446a08 | 742 | struct kioctx_table *table; |
abf137dd | 743 | struct kioctx *ctx; |
db446a08 BL |
744 | unsigned i = 0; |
745 | ||
746 | while (1) { | |
747 | rcu_read_lock(); | |
748 | table = rcu_dereference(mm->ioctx_table); | |
749 | ||
750 | do { | |
751 | if (!table || i >= table->nr) { | |
752 | rcu_read_unlock(); | |
753 | rcu_assign_pointer(mm->ioctx_table, NULL); | |
754 | if (table) | |
755 | kfree(table); | |
756 | return; | |
757 | } | |
758 | ||
759 | ctx = table->table[i++]; | |
760 | } while (!ctx); | |
761 | ||
762 | rcu_read_unlock(); | |
abf137dd | 763 | |
936af157 AV |
764 | /* |
765 | * We don't need to bother with munmap() here - | |
766 | * exit_mmap(mm) is coming and it'll unmap everything. | |
767 | * Since aio_free_ring() uses non-zero ->mmap_size | |
768 | * as indicator that it needs to unmap the area, | |
769 | * just set it to 0; aio_free_ring() is the only | |
770 | * place that uses ->mmap_size, so it's safe. | |
936af157 | 771 | */ |
58c85dc2 | 772 | ctx->mmap_size = 0; |
36f55889 | 773 | |
db446a08 | 774 | kill_ioctx(mm, ctx); |
1da177e4 LT |
775 | } |
776 | } | |
777 | ||
e1bdd5f2 KO |
778 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
779 | { | |
780 | struct kioctx_cpu *kcpu; | |
781 | ||
782 | preempt_disable(); | |
783 | kcpu = this_cpu_ptr(ctx->cpu); | |
784 | ||
785 | kcpu->reqs_available += nr; | |
786 | while (kcpu->reqs_available >= ctx->req_batch * 2) { | |
787 | kcpu->reqs_available -= ctx->req_batch; | |
788 | atomic_add(ctx->req_batch, &ctx->reqs_available); | |
789 | } | |
790 | ||
791 | preempt_enable(); | |
792 | } | |
793 | ||
794 | static bool get_reqs_available(struct kioctx *ctx) | |
795 | { | |
796 | struct kioctx_cpu *kcpu; | |
797 | bool ret = false; | |
798 | ||
799 | preempt_disable(); | |
800 | kcpu = this_cpu_ptr(ctx->cpu); | |
801 | ||
802 | if (!kcpu->reqs_available) { | |
803 | int old, avail = atomic_read(&ctx->reqs_available); | |
804 | ||
805 | do { | |
806 | if (avail < ctx->req_batch) | |
807 | goto out; | |
808 | ||
809 | old = avail; | |
810 | avail = atomic_cmpxchg(&ctx->reqs_available, | |
811 | avail, avail - ctx->req_batch); | |
812 | } while (avail != old); | |
813 | ||
814 | kcpu->reqs_available += ctx->req_batch; | |
815 | } | |
816 | ||
817 | ret = true; | |
818 | kcpu->reqs_available--; | |
819 | out: | |
820 | preempt_enable(); | |
821 | return ret; | |
822 | } | |
823 | ||
1da177e4 | 824 | /* aio_get_req |
57282d8f KO |
825 | * Allocate a slot for an aio request. |
826 | * Returns NULL if no requests are free. | |
1da177e4 | 827 | */ |
a1c8eae7 | 828 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
1da177e4 | 829 | { |
a1c8eae7 KO |
830 | struct kiocb *req; |
831 | ||
e1bdd5f2 | 832 | if (!get_reqs_available(ctx)) |
a1c8eae7 KO |
833 | return NULL; |
834 | ||
0460fef2 | 835 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
1da177e4 | 836 | if (unlikely(!req)) |
a1c8eae7 | 837 | goto out_put; |
1da177e4 | 838 | |
e34ecee2 KO |
839 | percpu_ref_get(&ctx->reqs); |
840 | ||
1da177e4 | 841 | req->ki_ctx = ctx; |
080d676d | 842 | return req; |
a1c8eae7 | 843 | out_put: |
e1bdd5f2 | 844 | put_reqs_available(ctx, 1); |
a1c8eae7 | 845 | return NULL; |
1da177e4 LT |
846 | } |
847 | ||
11599eba | 848 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 849 | { |
1d98ebfc KO |
850 | if (req->ki_filp) |
851 | fput(req->ki_filp); | |
13389010 DL |
852 | if (req->ki_eventfd != NULL) |
853 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 | 854 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
855 | } |
856 | ||
d5470b59 | 857 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 858 | { |
db446a08 | 859 | struct aio_ring __user *ring = (void __user *)ctx_id; |
abf137dd | 860 | struct mm_struct *mm = current->mm; |
65c24491 | 861 | struct kioctx *ctx, *ret = NULL; |
db446a08 BL |
862 | struct kioctx_table *table; |
863 | unsigned id; | |
864 | ||
865 | if (get_user(id, &ring->id)) | |
866 | return NULL; | |
1da177e4 | 867 | |
abf137dd | 868 | rcu_read_lock(); |
db446a08 | 869 | table = rcu_dereference(mm->ioctx_table); |
abf137dd | 870 | |
db446a08 BL |
871 | if (!table || id >= table->nr) |
872 | goto out; | |
1da177e4 | 873 | |
db446a08 | 874 | ctx = table->table[id]; |
f30d704f | 875 | if (ctx && ctx->user_id == ctx_id) { |
db446a08 BL |
876 | percpu_ref_get(&ctx->users); |
877 | ret = ctx; | |
878 | } | |
879 | out: | |
abf137dd | 880 | rcu_read_unlock(); |
65c24491 | 881 | return ret; |
1da177e4 LT |
882 | } |
883 | ||
1da177e4 LT |
884 | /* aio_complete |
885 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 886 | */ |
2d68449e | 887 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
888 | { |
889 | struct kioctx *ctx = iocb->ki_ctx; | |
1da177e4 | 890 | struct aio_ring *ring; |
21b40200 | 891 | struct io_event *ev_page, *event; |
1da177e4 | 892 | unsigned long flags; |
21b40200 | 893 | unsigned tail, pos; |
1da177e4 | 894 | |
20dcae32 ZB |
895 | /* |
896 | * Special case handling for sync iocbs: | |
897 | * - events go directly into the iocb for fast handling | |
898 | * - the sync task with the iocb in its stack holds the single iocb | |
899 | * ref, no other paths have a way to get another ref | |
900 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
901 | */ |
902 | if (is_sync_kiocb(iocb)) { | |
1da177e4 | 903 | iocb->ki_user_data = res; |
57282d8f KO |
904 | smp_wmb(); |
905 | iocb->ki_ctx = ERR_PTR(-EXDEV); | |
1da177e4 | 906 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 907 | return; |
1da177e4 LT |
908 | } |
909 | ||
0460fef2 KO |
910 | if (iocb->ki_list.next) { |
911 | unsigned long flags; | |
912 | ||
913 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
914 | list_del(&iocb->ki_list); | |
915 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
916 | } | |
11599eba | 917 | |
0460fef2 KO |
918 | /* |
919 | * Add a completion event to the ring buffer. Must be done holding | |
4b30f07e | 920 | * ctx->completion_lock to prevent other code from messing with the tail |
0460fef2 KO |
921 | * pointer since we might be called from irq context. |
922 | */ | |
923 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
924 | ||
58c85dc2 | 925 | tail = ctx->tail; |
21b40200 KO |
926 | pos = tail + AIO_EVENTS_OFFSET; |
927 | ||
58c85dc2 | 928 | if (++tail >= ctx->nr_events) |
4bf69b2a | 929 | tail = 0; |
1da177e4 | 930 | |
58c85dc2 | 931 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
932 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
933 | ||
1da177e4 LT |
934 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
935 | event->data = iocb->ki_user_data; | |
936 | event->res = res; | |
937 | event->res2 = res2; | |
938 | ||
21b40200 | 939 | kunmap_atomic(ev_page); |
58c85dc2 | 940 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
941 | |
942 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
caf4167a KO |
943 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
944 | res, res2); | |
1da177e4 LT |
945 | |
946 | /* after flagging the request as done, we | |
947 | * must never even look at it again | |
948 | */ | |
949 | smp_wmb(); /* make event visible before updating tail */ | |
950 | ||
58c85dc2 | 951 | ctx->tail = tail; |
1da177e4 | 952 | |
58c85dc2 | 953 | ring = kmap_atomic(ctx->ring_pages[0]); |
21b40200 | 954 | ring->tail = tail; |
e8e3c3d6 | 955 | kunmap_atomic(ring); |
58c85dc2 | 956 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 | 957 | |
0460fef2 KO |
958 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
959 | ||
21b40200 | 960 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
8d1c98b0 DL |
961 | |
962 | /* | |
963 | * Check if the user asked us to deliver the result through an | |
964 | * eventfd. The eventfd_signal() function is safe to be called | |
965 | * from IRQ context. | |
966 | */ | |
87c3a86e | 967 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
968 | eventfd_signal(iocb->ki_eventfd, 1); |
969 | ||
1da177e4 | 970 | /* everything turned out well, dispose of the aiocb. */ |
57282d8f | 971 | kiocb_free(iocb); |
1da177e4 | 972 | |
6cb2a210 QB |
973 | /* |
974 | * We have to order our ring_info tail store above and test | |
975 | * of the wait list below outside the wait lock. This is | |
976 | * like in wake_up_bit() where clearing a bit has to be | |
977 | * ordered with the unlocked test. | |
978 | */ | |
979 | smp_mb(); | |
980 | ||
1da177e4 LT |
981 | if (waitqueue_active(&ctx->wait)) |
982 | wake_up(&ctx->wait); | |
983 | ||
e34ecee2 | 984 | percpu_ref_put(&ctx->reqs); |
1da177e4 | 985 | } |
385773e0 | 986 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 987 | |
a31ad380 KO |
988 | /* aio_read_events |
989 | * Pull an event off of the ioctx's event ring. Returns the number of | |
990 | * events fetched | |
1da177e4 | 991 | */ |
a31ad380 KO |
992 | static long aio_read_events_ring(struct kioctx *ctx, |
993 | struct io_event __user *event, long nr) | |
1da177e4 | 994 | { |
1da177e4 | 995 | struct aio_ring *ring; |
5ffac122 | 996 | unsigned head, tail, pos; |
a31ad380 KO |
997 | long ret = 0; |
998 | int copy_ret; | |
999 | ||
58c85dc2 | 1000 | mutex_lock(&ctx->ring_lock); |
1da177e4 | 1001 | |
58c85dc2 | 1002 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 1003 | head = ring->head; |
5ffac122 | 1004 | tail = ring->tail; |
a31ad380 KO |
1005 | kunmap_atomic(ring); |
1006 | ||
5ffac122 | 1007 | pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); |
1da177e4 | 1008 | |
5ffac122 | 1009 | if (head == tail) |
1da177e4 LT |
1010 | goto out; |
1011 | ||
a31ad380 KO |
1012 | while (ret < nr) { |
1013 | long avail; | |
1014 | struct io_event *ev; | |
1015 | struct page *page; | |
1016 | ||
5ffac122 KO |
1017 | avail = (head <= tail ? tail : ctx->nr_events) - head; |
1018 | if (head == tail) | |
a31ad380 KO |
1019 | break; |
1020 | ||
1021 | avail = min(avail, nr - ret); | |
1022 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
1023 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
1024 | ||
1025 | pos = head + AIO_EVENTS_OFFSET; | |
58c85dc2 | 1026 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
a31ad380 KO |
1027 | pos %= AIO_EVENTS_PER_PAGE; |
1028 | ||
1029 | ev = kmap(page); | |
1030 | copy_ret = copy_to_user(event + ret, ev + pos, | |
1031 | sizeof(*ev) * avail); | |
1032 | kunmap(page); | |
1033 | ||
1034 | if (unlikely(copy_ret)) { | |
1035 | ret = -EFAULT; | |
1036 | goto out; | |
1037 | } | |
1038 | ||
1039 | ret += avail; | |
1040 | head += avail; | |
58c85dc2 | 1041 | head %= ctx->nr_events; |
1da177e4 | 1042 | } |
1da177e4 | 1043 | |
58c85dc2 | 1044 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 1045 | ring->head = head; |
91d80a84 | 1046 | kunmap_atomic(ring); |
58c85dc2 | 1047 | flush_dcache_page(ctx->ring_pages[0]); |
a31ad380 | 1048 | |
5ffac122 | 1049 | pr_debug("%li h%u t%u\n", ret, head, tail); |
3e845ce0 | 1050 | |
e1bdd5f2 | 1051 | put_reqs_available(ctx, ret); |
a31ad380 | 1052 | out: |
58c85dc2 | 1053 | mutex_unlock(&ctx->ring_lock); |
a31ad380 | 1054 | |
1da177e4 LT |
1055 | return ret; |
1056 | } | |
1057 | ||
a31ad380 KO |
1058 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
1059 | struct io_event __user *event, long *i) | |
1da177e4 | 1060 | { |
a31ad380 | 1061 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 1062 | |
a31ad380 KO |
1063 | if (ret > 0) |
1064 | *i += ret; | |
1da177e4 | 1065 | |
a31ad380 KO |
1066 | if (unlikely(atomic_read(&ctx->dead))) |
1067 | ret = -EINVAL; | |
1da177e4 | 1068 | |
a31ad380 KO |
1069 | if (!*i) |
1070 | *i = ret; | |
1da177e4 | 1071 | |
a31ad380 | 1072 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
1073 | } |
1074 | ||
a31ad380 | 1075 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
1076 | struct io_event __user *event, |
1077 | struct timespec __user *timeout) | |
1078 | { | |
a31ad380 KO |
1079 | ktime_t until = { .tv64 = KTIME_MAX }; |
1080 | long ret = 0; | |
1da177e4 | 1081 | |
1da177e4 LT |
1082 | if (timeout) { |
1083 | struct timespec ts; | |
a31ad380 | 1084 | |
1da177e4 | 1085 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 1086 | return -EFAULT; |
1da177e4 | 1087 | |
a31ad380 | 1088 | until = timespec_to_ktime(ts); |
1da177e4 LT |
1089 | } |
1090 | ||
a31ad380 KO |
1091 | /* |
1092 | * Note that aio_read_events() is being called as the conditional - i.e. | |
1093 | * we're calling it after prepare_to_wait() has set task state to | |
1094 | * TASK_INTERRUPTIBLE. | |
1095 | * | |
1096 | * But aio_read_events() can block, and if it blocks it's going to flip | |
1097 | * the task state back to TASK_RUNNING. | |
1098 | * | |
1099 | * This should be ok, provided it doesn't flip the state back to | |
1100 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
1101 | * will only happen if the mutex_lock() call blocks, and we then find | |
1102 | * the ringbuffer empty. So in practice we should be ok, but it's | |
1103 | * something to be aware of when touching this code. | |
1104 | */ | |
1105 | wait_event_interruptible_hrtimeout(ctx->wait, | |
1106 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
1da177e4 | 1107 | |
a31ad380 KO |
1108 | if (!ret && signal_pending(current)) |
1109 | ret = -EINTR; | |
1da177e4 | 1110 | |
a31ad380 | 1111 | return ret; |
1da177e4 LT |
1112 | } |
1113 | ||
1da177e4 LT |
1114 | /* sys_io_setup: |
1115 | * Create an aio_context capable of receiving at least nr_events. | |
1116 | * ctxp must not point to an aio_context that already exists, and | |
1117 | * must be initialized to 0 prior to the call. On successful | |
1118 | * creation of the aio_context, *ctxp is filled in with the resulting | |
1119 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
1120 | * if the specified nr_events exceeds internal limits. May fail | |
1121 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
1122 | * of available events. May fail with -ENOMEM if insufficient kernel | |
1123 | * resources are available. May fail with -EFAULT if an invalid | |
1124 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
1125 | * implemented. | |
1126 | */ | |
002c8976 | 1127 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
1128 | { |
1129 | struct kioctx *ioctx = NULL; | |
1130 | unsigned long ctx; | |
1131 | long ret; | |
1132 | ||
1133 | ret = get_user(ctx, ctxp); | |
1134 | if (unlikely(ret)) | |
1135 | goto out; | |
1136 | ||
1137 | ret = -EINVAL; | |
d55b5fda ZB |
1138 | if (unlikely(ctx || nr_events == 0)) { |
1139 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
1140 | ctx, nr_events); | |
1da177e4 LT |
1141 | goto out; |
1142 | } | |
1143 | ||
1144 | ioctx = ioctx_alloc(nr_events); | |
1145 | ret = PTR_ERR(ioctx); | |
1146 | if (!IS_ERR(ioctx)) { | |
1147 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a | 1148 | if (ret) |
db446a08 | 1149 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1150 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1151 | } |
1152 | ||
1153 | out: | |
1154 | return ret; | |
1155 | } | |
1156 | ||
1157 | /* sys_io_destroy: | |
1158 | * Destroy the aio_context specified. May cancel any outstanding | |
1159 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 1160 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
1161 | * is invalid. |
1162 | */ | |
002c8976 | 1163 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
1164 | { |
1165 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
1166 | if (likely(NULL != ioctx)) { | |
db446a08 | 1167 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1168 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1169 | return 0; |
1170 | } | |
1171 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
1172 | return -EINVAL; | |
1173 | } | |
1174 | ||
41ef4eb8 KO |
1175 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
1176 | unsigned long, loff_t); | |
1177 | ||
8bc92afc KO |
1178 | static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb, |
1179 | int rw, char __user *buf, | |
1180 | unsigned long *nr_segs, | |
1181 | struct iovec **iovec, | |
1182 | bool compat) | |
eed4e51f BP |
1183 | { |
1184 | ssize_t ret; | |
1185 | ||
8bc92afc | 1186 | *nr_segs = kiocb->ki_nbytes; |
41ef4eb8 | 1187 | |
9d85cba7 JM |
1188 | #ifdef CONFIG_COMPAT |
1189 | if (compat) | |
41ef4eb8 | 1190 | ret = compat_rw_copy_check_uvector(rw, |
8bc92afc KO |
1191 | (struct compat_iovec __user *)buf, |
1192 | *nr_segs, 1, *iovec, iovec); | |
9d85cba7 JM |
1193 | else |
1194 | #endif | |
41ef4eb8 | 1195 | ret = rw_copy_check_uvector(rw, |
8bc92afc KO |
1196 | (struct iovec __user *)buf, |
1197 | *nr_segs, 1, *iovec, iovec); | |
eed4e51f | 1198 | if (ret < 0) |
41ef4eb8 | 1199 | return ret; |
a70b52ec | 1200 | |
41ef4eb8 | 1201 | /* ki_nbytes now reflect bytes instead of segs */ |
eed4e51f | 1202 | kiocb->ki_nbytes = ret; |
41ef4eb8 | 1203 | return 0; |
eed4e51f BP |
1204 | } |
1205 | ||
8bc92afc KO |
1206 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb, |
1207 | int rw, char __user *buf, | |
1208 | unsigned long *nr_segs, | |
1209 | struct iovec *iovec) | |
eed4e51f | 1210 | { |
8bc92afc | 1211 | if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes))) |
41ef4eb8 | 1212 | return -EFAULT; |
a70b52ec | 1213 | |
8bc92afc KO |
1214 | iovec->iov_base = buf; |
1215 | iovec->iov_len = kiocb->ki_nbytes; | |
1216 | *nr_segs = 1; | |
eed4e51f BP |
1217 | return 0; |
1218 | } | |
1219 | ||
1da177e4 LT |
1220 | /* |
1221 | * aio_setup_iocb: | |
1222 | * Performs the initial checks and aio retry method | |
1223 | * setup for the kiocb at the time of io submission. | |
1224 | */ | |
8bc92afc KO |
1225 | static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode, |
1226 | char __user *buf, bool compat) | |
1da177e4 | 1227 | { |
41ef4eb8 KO |
1228 | struct file *file = req->ki_filp; |
1229 | ssize_t ret; | |
8bc92afc | 1230 | unsigned long nr_segs; |
41ef4eb8 KO |
1231 | int rw; |
1232 | fmode_t mode; | |
1233 | aio_rw_op *rw_op; | |
8bc92afc | 1234 | struct iovec inline_vec, *iovec = &inline_vec; |
1da177e4 | 1235 | |
8bc92afc | 1236 | switch (opcode) { |
1da177e4 | 1237 | case IOCB_CMD_PREAD: |
eed4e51f | 1238 | case IOCB_CMD_PREADV: |
41ef4eb8 KO |
1239 | mode = FMODE_READ; |
1240 | rw = READ; | |
1241 | rw_op = file->f_op->aio_read; | |
1242 | goto rw_common; | |
1243 | ||
1244 | case IOCB_CMD_PWRITE: | |
eed4e51f | 1245 | case IOCB_CMD_PWRITEV: |
41ef4eb8 KO |
1246 | mode = FMODE_WRITE; |
1247 | rw = WRITE; | |
1248 | rw_op = file->f_op->aio_write; | |
1249 | goto rw_common; | |
1250 | rw_common: | |
1251 | if (unlikely(!(file->f_mode & mode))) | |
1252 | return -EBADF; | |
1253 | ||
1254 | if (!rw_op) | |
1255 | return -EINVAL; | |
1256 | ||
8bc92afc KO |
1257 | ret = (opcode == IOCB_CMD_PREADV || |
1258 | opcode == IOCB_CMD_PWRITEV) | |
1259 | ? aio_setup_vectored_rw(req, rw, buf, &nr_segs, | |
1260 | &iovec, compat) | |
1261 | : aio_setup_single_vector(req, rw, buf, &nr_segs, | |
1262 | iovec); | |
eed4e51f | 1263 | if (ret) |
41ef4eb8 KO |
1264 | return ret; |
1265 | ||
1266 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); | |
8bc92afc KO |
1267 | if (ret < 0) { |
1268 | if (iovec != &inline_vec) | |
1269 | kfree(iovec); | |
41ef4eb8 | 1270 | return ret; |
8bc92afc | 1271 | } |
41ef4eb8 KO |
1272 | |
1273 | req->ki_nbytes = ret; | |
41ef4eb8 | 1274 | |
73a7075e KO |
1275 | /* XXX: move/kill - rw_verify_area()? */ |
1276 | /* This matches the pread()/pwrite() logic */ | |
1277 | if (req->ki_pos < 0) { | |
1278 | ret = -EINVAL; | |
1279 | break; | |
1280 | } | |
1281 | ||
1282 | if (rw == WRITE) | |
1283 | file_start_write(file); | |
1284 | ||
8bc92afc | 1285 | ret = rw_op(req, iovec, nr_segs, req->ki_pos); |
73a7075e KO |
1286 | |
1287 | if (rw == WRITE) | |
1288 | file_end_write(file); | |
1da177e4 | 1289 | break; |
41ef4eb8 | 1290 | |
1da177e4 | 1291 | case IOCB_CMD_FDSYNC: |
41ef4eb8 KO |
1292 | if (!file->f_op->aio_fsync) |
1293 | return -EINVAL; | |
1294 | ||
1295 | ret = file->f_op->aio_fsync(req, 1); | |
1da177e4 | 1296 | break; |
41ef4eb8 | 1297 | |
1da177e4 | 1298 | case IOCB_CMD_FSYNC: |
41ef4eb8 KO |
1299 | if (!file->f_op->aio_fsync) |
1300 | return -EINVAL; | |
1301 | ||
1302 | ret = file->f_op->aio_fsync(req, 0); | |
1da177e4 | 1303 | break; |
41ef4eb8 | 1304 | |
1da177e4 | 1305 | default: |
caf4167a | 1306 | pr_debug("EINVAL: no operation provided\n"); |
41ef4eb8 | 1307 | return -EINVAL; |
1da177e4 LT |
1308 | } |
1309 | ||
8bc92afc KO |
1310 | if (iovec != &inline_vec) |
1311 | kfree(iovec); | |
1312 | ||
41ef4eb8 KO |
1313 | if (ret != -EIOCBQUEUED) { |
1314 | /* | |
1315 | * There's no easy way to restart the syscall since other AIO's | |
1316 | * may be already running. Just fail this IO with EINTR. | |
1317 | */ | |
1318 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1319 | ret == -ERESTARTNOHAND || | |
1320 | ret == -ERESTART_RESTARTBLOCK)) | |
1321 | ret = -EINTR; | |
1322 | aio_complete(req, ret, 0); | |
1323 | } | |
1da177e4 LT |
1324 | |
1325 | return 0; | |
1326 | } | |
1327 | ||
d5470b59 | 1328 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
a1c8eae7 | 1329 | struct iocb *iocb, bool compat) |
1da177e4 LT |
1330 | { |
1331 | struct kiocb *req; | |
1da177e4 LT |
1332 | ssize_t ret; |
1333 | ||
1334 | /* enforce forwards compatibility on users */ | |
9c3060be | 1335 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1336 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1337 | return -EINVAL; |
1338 | } | |
1339 | ||
1340 | /* prevent overflows */ | |
1341 | if (unlikely( | |
1342 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1343 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1344 | ((ssize_t)iocb->aio_nbytes < 0) | |
1345 | )) { | |
1346 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1347 | return -EINVAL; | |
1348 | } | |
1349 | ||
41ef4eb8 | 1350 | req = aio_get_req(ctx); |
1d98ebfc | 1351 | if (unlikely(!req)) |
1da177e4 | 1352 | return -EAGAIN; |
1d98ebfc KO |
1353 | |
1354 | req->ki_filp = fget(iocb->aio_fildes); | |
1355 | if (unlikely(!req->ki_filp)) { | |
1356 | ret = -EBADF; | |
1357 | goto out_put_req; | |
1da177e4 | 1358 | } |
1d98ebfc | 1359 | |
9c3060be DL |
1360 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1361 | /* | |
1362 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1363 | * instance of the file* now. The file descriptor must be | |
1364 | * an eventfd() fd, and will be signaled for each completed | |
1365 | * event using the eventfd_signal() function. | |
1366 | */ | |
13389010 | 1367 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1368 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1369 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1370 | req->ki_eventfd = NULL; |
9c3060be DL |
1371 | goto out_put_req; |
1372 | } | |
1373 | } | |
1da177e4 | 1374 | |
8a660890 | 1375 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
1da177e4 | 1376 | if (unlikely(ret)) { |
caf4167a | 1377 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1378 | goto out_put_req; |
1379 | } | |
1380 | ||
1381 | req->ki_obj.user = user_iocb; | |
1382 | req->ki_user_data = iocb->aio_data; | |
1383 | req->ki_pos = iocb->aio_offset; | |
73a7075e | 1384 | req->ki_nbytes = iocb->aio_nbytes; |
1da177e4 | 1385 | |
8bc92afc KO |
1386 | ret = aio_run_iocb(req, iocb->aio_lio_opcode, |
1387 | (char __user *)(unsigned long)iocb->aio_buf, | |
1388 | compat); | |
41003a7b | 1389 | if (ret) |
7137c6bd | 1390 | goto out_put_req; |
41003a7b | 1391 | |
1da177e4 | 1392 | return 0; |
1da177e4 | 1393 | out_put_req: |
e1bdd5f2 | 1394 | put_reqs_available(ctx, 1); |
e34ecee2 | 1395 | percpu_ref_put(&ctx->reqs); |
57282d8f | 1396 | kiocb_free(req); |
1da177e4 LT |
1397 | return ret; |
1398 | } | |
1399 | ||
9d85cba7 JM |
1400 | long do_io_submit(aio_context_t ctx_id, long nr, |
1401 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1402 | { |
1403 | struct kioctx *ctx; | |
1404 | long ret = 0; | |
080d676d | 1405 | int i = 0; |
9f5b9425 | 1406 | struct blk_plug plug; |
1da177e4 LT |
1407 | |
1408 | if (unlikely(nr < 0)) | |
1409 | return -EINVAL; | |
1410 | ||
75e1c70f JM |
1411 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1412 | nr = LONG_MAX/sizeof(*iocbpp); | |
1413 | ||
1da177e4 LT |
1414 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1415 | return -EFAULT; | |
1416 | ||
1417 | ctx = lookup_ioctx(ctx_id); | |
1418 | if (unlikely(!ctx)) { | |
caf4167a | 1419 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1420 | return -EINVAL; |
1421 | } | |
1422 | ||
9f5b9425 SL |
1423 | blk_start_plug(&plug); |
1424 | ||
1da177e4 LT |
1425 | /* |
1426 | * AKPM: should this return a partial result if some of the IOs were | |
1427 | * successfully submitted? | |
1428 | */ | |
1429 | for (i=0; i<nr; i++) { | |
1430 | struct iocb __user *user_iocb; | |
1431 | struct iocb tmp; | |
1432 | ||
1433 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1434 | ret = -EFAULT; | |
1435 | break; | |
1436 | } | |
1437 | ||
1438 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1439 | ret = -EFAULT; | |
1440 | break; | |
1441 | } | |
1442 | ||
a1c8eae7 | 1443 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
1da177e4 LT |
1444 | if (ret) |
1445 | break; | |
1446 | } | |
9f5b9425 | 1447 | blk_finish_plug(&plug); |
1da177e4 | 1448 | |
723be6e3 | 1449 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1450 | return i ? i : ret; |
1451 | } | |
1452 | ||
9d85cba7 JM |
1453 | /* sys_io_submit: |
1454 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1455 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1456 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1457 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1458 | * is invalid for the file descriptor in the iocb. May fail with | |
1459 | * -EFAULT if any of the data structures point to invalid data. May | |
1460 | * fail with -EBADF if the file descriptor specified in the first | |
1461 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1462 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1463 | * fail with -ENOSYS if not implemented. | |
1464 | */ | |
1465 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1466 | struct iocb __user * __user *, iocbpp) | |
1467 | { | |
1468 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1469 | } | |
1470 | ||
1da177e4 LT |
1471 | /* lookup_kiocb |
1472 | * Finds a given iocb for cancellation. | |
1da177e4 | 1473 | */ |
25ee7e38 AB |
1474 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1475 | u32 key) | |
1da177e4 LT |
1476 | { |
1477 | struct list_head *pos; | |
d00689af ZB |
1478 | |
1479 | assert_spin_locked(&ctx->ctx_lock); | |
1480 | ||
8a660890 KO |
1481 | if (key != KIOCB_KEY) |
1482 | return NULL; | |
1483 | ||
1da177e4 LT |
1484 | /* TODO: use a hash or array, this sucks. */ |
1485 | list_for_each(pos, &ctx->active_reqs) { | |
1486 | struct kiocb *kiocb = list_kiocb(pos); | |
8a660890 | 1487 | if (kiocb->ki_obj.user == iocb) |
1da177e4 LT |
1488 | return kiocb; |
1489 | } | |
1490 | return NULL; | |
1491 | } | |
1492 | ||
1493 | /* sys_io_cancel: | |
1494 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1495 | * the operation is successfully cancelled, the resulting event is | |
1496 | * copied into the memory pointed to by result without being placed | |
1497 | * into the completion queue and 0 is returned. May fail with | |
1498 | * -EFAULT if any of the data structures pointed to are invalid. | |
1499 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1500 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1501 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1502 | */ | |
002c8976 HC |
1503 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1504 | struct io_event __user *, result) | |
1da177e4 | 1505 | { |
1da177e4 LT |
1506 | struct kioctx *ctx; |
1507 | struct kiocb *kiocb; | |
1508 | u32 key; | |
1509 | int ret; | |
1510 | ||
1511 | ret = get_user(key, &iocb->aio_key); | |
1512 | if (unlikely(ret)) | |
1513 | return -EFAULT; | |
1514 | ||
1515 | ctx = lookup_ioctx(ctx_id); | |
1516 | if (unlikely(!ctx)) | |
1517 | return -EINVAL; | |
1518 | ||
1519 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1520 | |
1da177e4 | 1521 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c | 1522 | if (kiocb) |
bec68faa | 1523 | ret = kiocb_cancel(ctx, kiocb); |
906b973c KO |
1524 | else |
1525 | ret = -EINVAL; | |
1526 | ||
1da177e4 LT |
1527 | spin_unlock_irq(&ctx->ctx_lock); |
1528 | ||
906b973c | 1529 | if (!ret) { |
bec68faa KO |
1530 | /* |
1531 | * The result argument is no longer used - the io_event is | |
1532 | * always delivered via the ring buffer. -EINPROGRESS indicates | |
1533 | * cancellation is progress: | |
906b973c | 1534 | */ |
bec68faa | 1535 | ret = -EINPROGRESS; |
906b973c | 1536 | } |
1da177e4 | 1537 | |
723be6e3 | 1538 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1539 | |
1540 | return ret; | |
1541 | } | |
1542 | ||
1543 | /* io_getevents: | |
1544 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1545 | * the completion queue for the aio_context specified by ctx_id. If |
1546 | * it succeeds, the number of read events is returned. May fail with | |
1547 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1548 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1549 | * if any of the memory specified is invalid. May return 0 or | |
1550 | * < min_nr if the timeout specified by timeout has elapsed | |
1551 | * before sufficient events are available, where timeout == NULL | |
1552 | * specifies an infinite timeout. Note that the timeout pointed to by | |
6900807c | 1553 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
1da177e4 | 1554 | */ |
002c8976 HC |
1555 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1556 | long, min_nr, | |
1557 | long, nr, | |
1558 | struct io_event __user *, events, | |
1559 | struct timespec __user *, timeout) | |
1da177e4 LT |
1560 | { |
1561 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1562 | long ret = -EINVAL; | |
1563 | ||
1564 | if (likely(ioctx)) { | |
2e410255 | 1565 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 | 1566 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
723be6e3 | 1567 | percpu_ref_put(&ioctx->users); |
1da177e4 | 1568 | } |
1da177e4 LT |
1569 | return ret; |
1570 | } |