Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/init.h> |
| 13 | #include <linux/errno.h> |
| 14 | #include <linux/time.h> |
| 15 | #include <linux/aio_abi.h> |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/syscalls.h> |
Jens Axboe | b9d128f | 2009-10-29 13:59:26 +0100 | [diff] [blame] | 18 | #include <linux/backing-dev.h> |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 19 | #include <linux/uio.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | |
| 21 | #define DEBUG 0 |
| 22 | |
| 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> |
Michael S. Tsirkin | 3d2d827 | 2009-09-21 17:03:51 -0700 | [diff] [blame] | 28 | #include <linux/mmu_context.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/slab.h> |
| 30 | #include <linux/timer.h> |
| 31 | #include <linux/aio.h> |
| 32 | #include <linux/highmem.h> |
| 33 | #include <linux/workqueue.h> |
| 34 | #include <linux/security.h> |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 35 | #include <linux/eventfd.h> |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 36 | #include <linux/blkdev.h> |
| 37 | #include <linux/mempool.h> |
| 38 | #include <linux/hash.h> |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 39 | #include <linux/compat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | |
| 41 | #include <asm/kmap_types.h> |
| 42 | #include <asm/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | |
| 44 | #if DEBUG > 1 |
| 45 | #define dprintk printk |
| 46 | #else |
| 47 | #define dprintk(x...) do { ; } while (0) |
| 48 | #endif |
| 49 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | /*------ sysctl variables----*/ |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 51 | static DEFINE_SPINLOCK(aio_nr_lock); |
| 52 | unsigned long aio_nr; /* current system wide number of aio requests */ |
| 53 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | /*----end sysctl variables---*/ |
| 55 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 56 | static struct kmem_cache *kiocb_cachep; |
| 57 | static struct kmem_cache *kioctx_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | |
| 59 | static struct workqueue_struct *aio_wq; |
| 60 | |
| 61 | /* Used for rare fput completion. */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 62 | static void aio_fput_routine(struct work_struct *); |
| 63 | static DECLARE_WORK(fput_work, aio_fput_routine); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | |
| 65 | static DEFINE_SPINLOCK(fput_lock); |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 66 | static LIST_HEAD(fput_head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 68 | #define AIO_BATCH_HASH_BITS 3 /* allocated on-stack, so don't go crazy */ |
| 69 | #define AIO_BATCH_HASH_SIZE (1 << AIO_BATCH_HASH_BITS) |
| 70 | struct aio_batch_entry { |
| 71 | struct hlist_node list; |
| 72 | struct address_space *mapping; |
| 73 | }; |
| 74 | mempool_t *abe_pool; |
| 75 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 76 | static void aio_kick_handler(struct work_struct *); |
Sébastien Dugu | c016e22 | 2005-06-28 20:44:59 -0700 | [diff] [blame] | 77 | static void aio_queue_work(struct kioctx *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | |
| 79 | /* aio_setup |
| 80 | * Creates the slab caches used by the aio routines, panic on |
| 81 | * failure as this is done early during the boot sequence. |
| 82 | */ |
| 83 | static int __init aio_setup(void) |
| 84 | { |
Christoph Lameter | 0a31bd5 | 2007-05-06 14:49:57 -0700 | [diff] [blame] | 85 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
| 86 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | |
| 88 | aio_wq = create_workqueue("aio"); |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 89 | abe_pool = mempool_create_kmalloc_pool(1, sizeof(struct aio_batch_entry)); |
| 90 | BUG_ON(!abe_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | |
| 92 | pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); |
| 93 | |
| 94 | return 0; |
| 95 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 96 | __initcall(aio_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | |
| 98 | static void aio_free_ring(struct kioctx *ctx) |
| 99 | { |
| 100 | struct aio_ring_info *info = &ctx->ring_info; |
| 101 | long i; |
| 102 | |
| 103 | for (i=0; i<info->nr_pages; i++) |
| 104 | put_page(info->ring_pages[i]); |
| 105 | |
| 106 | if (info->mmap_size) { |
| 107 | down_write(&ctx->mm->mmap_sem); |
| 108 | do_munmap(ctx->mm, info->mmap_base, info->mmap_size); |
| 109 | up_write(&ctx->mm->mmap_sem); |
| 110 | } |
| 111 | |
| 112 | if (info->ring_pages && info->ring_pages != info->internal_pages) |
| 113 | kfree(info->ring_pages); |
| 114 | info->ring_pages = NULL; |
| 115 | info->nr = 0; |
| 116 | } |
| 117 | |
| 118 | static int aio_setup_ring(struct kioctx *ctx) |
| 119 | { |
| 120 | struct aio_ring *ring; |
| 121 | struct aio_ring_info *info = &ctx->ring_info; |
| 122 | unsigned nr_events = ctx->max_reqs; |
| 123 | unsigned long size; |
| 124 | int nr_pages; |
| 125 | |
| 126 | /* Compensate for the ring buffer's head/tail overlap entry */ |
| 127 | nr_events += 2; /* 1 is required, 2 for good luck */ |
| 128 | |
| 129 | size = sizeof(struct aio_ring); |
| 130 | size += sizeof(struct io_event) * nr_events; |
| 131 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; |
| 132 | |
| 133 | if (nr_pages < 0) |
| 134 | return -EINVAL; |
| 135 | |
| 136 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); |
| 137 | |
| 138 | info->nr = 0; |
| 139 | info->ring_pages = info->internal_pages; |
| 140 | if (nr_pages > AIO_RING_PAGES) { |
Oliver Neukum | 11b0b5a | 2006-03-25 03:08:13 -0800 | [diff] [blame] | 141 | info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | if (!info->ring_pages) |
| 143 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | } |
| 145 | |
| 146 | info->mmap_size = nr_pages * PAGE_SIZE; |
| 147 | dprintk("attempting mmap of %lu bytes\n", info->mmap_size); |
| 148 | down_write(&ctx->mm->mmap_sem); |
| 149 | info->mmap_base = do_mmap(NULL, 0, info->mmap_size, |
Robert P. J. Day | e10a443 | 2007-02-10 01:42:56 -0800 | [diff] [blame] | 150 | PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | 0); |
| 152 | if (IS_ERR((void *)info->mmap_base)) { |
| 153 | up_write(&ctx->mm->mmap_sem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | info->mmap_size = 0; |
| 155 | aio_free_ring(ctx); |
| 156 | return -EAGAIN; |
| 157 | } |
| 158 | |
| 159 | dprintk("mmap address: 0x%08lx\n", info->mmap_base); |
| 160 | info->nr_pages = get_user_pages(current, ctx->mm, |
| 161 | info->mmap_base, nr_pages, |
| 162 | 1, 0, info->ring_pages, NULL); |
| 163 | up_write(&ctx->mm->mmap_sem); |
| 164 | |
| 165 | if (unlikely(info->nr_pages != nr_pages)) { |
| 166 | aio_free_ring(ctx); |
| 167 | return -EAGAIN; |
| 168 | } |
| 169 | |
| 170 | ctx->user_id = info->mmap_base; |
| 171 | |
| 172 | info->nr = nr_events; /* trusted copy */ |
| 173 | |
| 174 | ring = kmap_atomic(info->ring_pages[0], KM_USER0); |
| 175 | ring->nr = nr_events; /* user copy */ |
| 176 | ring->id = ctx->user_id; |
| 177 | ring->head = ring->tail = 0; |
| 178 | ring->magic = AIO_RING_MAGIC; |
| 179 | ring->compat_features = AIO_RING_COMPAT_FEATURES; |
| 180 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; |
| 181 | ring->header_length = sizeof(struct aio_ring); |
| 182 | kunmap_atomic(ring, KM_USER0); |
| 183 | |
| 184 | return 0; |
| 185 | } |
| 186 | |
| 187 | |
| 188 | /* aio_ring_event: returns a pointer to the event at the given index from |
| 189 | * kmap_atomic(, km). Release the pointer with put_aio_ring_event(); |
| 190 | */ |
| 191 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
| 192 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) |
| 193 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) |
| 194 | |
| 195 | #define aio_ring_event(info, nr, km) ({ \ |
| 196 | unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ |
| 197 | struct io_event *__event; \ |
| 198 | __event = kmap_atomic( \ |
| 199 | (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE], km); \ |
| 200 | __event += pos % AIO_EVENTS_PER_PAGE; \ |
| 201 | __event; \ |
| 202 | }) |
| 203 | |
| 204 | #define put_aio_ring_event(event, km) do { \ |
| 205 | struct io_event *__event = (event); \ |
| 206 | (void)__event; \ |
| 207 | kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \ |
| 208 | } while(0) |
| 209 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 210 | static void ctx_rcu_free(struct rcu_head *head) |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 211 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 212 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 213 | unsigned nr_events = ctx->max_reqs; |
| 214 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 215 | kmem_cache_free(kioctx_cachep, ctx); |
| 216 | |
| 217 | if (nr_events) { |
| 218 | spin_lock(&aio_nr_lock); |
| 219 | BUG_ON(aio_nr - nr_events > aio_nr); |
| 220 | aio_nr -= nr_events; |
| 221 | spin_unlock(&aio_nr_lock); |
| 222 | } |
| 223 | } |
| 224 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 225 | /* __put_ioctx |
| 226 | * Called when the last user of an aio context has gone away, |
| 227 | * and the struct needs to be freed. |
| 228 | */ |
| 229 | static void __put_ioctx(struct kioctx *ctx) |
| 230 | { |
| 231 | BUG_ON(ctx->reqs_active); |
| 232 | |
| 233 | cancel_delayed_work(&ctx->wq); |
| 234 | cancel_work_sync(&ctx->wq.work); |
| 235 | aio_free_ring(ctx); |
| 236 | mmdrop(ctx->mm); |
| 237 | ctx->mm = NULL; |
| 238 | pr_debug("__put_ioctx: freeing %p\n", ctx); |
| 239 | call_rcu(&ctx->rcu_head, ctx_rcu_free); |
| 240 | } |
| 241 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 242 | #define get_ioctx(kioctx) do { \ |
| 243 | BUG_ON(atomic_read(&(kioctx)->users) <= 0); \ |
| 244 | atomic_inc(&(kioctx)->users); \ |
| 245 | } while (0) |
| 246 | #define put_ioctx(kioctx) do { \ |
| 247 | BUG_ON(atomic_read(&(kioctx)->users) <= 0); \ |
| 248 | if (unlikely(atomic_dec_and_test(&(kioctx)->users))) \ |
| 249 | __put_ioctx(kioctx); \ |
| 250 | } while (0) |
| 251 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 252 | /* ioctx_alloc |
| 253 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. |
| 254 | */ |
| 255 | static struct kioctx *ioctx_alloc(unsigned nr_events) |
| 256 | { |
| 257 | struct mm_struct *mm; |
| 258 | struct kioctx *ctx; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 259 | int did_sync = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | |
| 261 | /* Prevent overflows */ |
| 262 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || |
| 263 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { |
| 264 | pr_debug("ENOMEM: nr_events too high\n"); |
| 265 | return ERR_PTR(-EINVAL); |
| 266 | } |
| 267 | |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 268 | if ((unsigned long)nr_events > aio_max_nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | return ERR_PTR(-EAGAIN); |
| 270 | |
Robert P. J. Day | c376222 | 2007-02-10 01:45:03 -0800 | [diff] [blame] | 271 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | if (!ctx) |
| 273 | return ERR_PTR(-ENOMEM); |
| 274 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | ctx->max_reqs = nr_events; |
| 276 | mm = ctx->mm = current->mm; |
| 277 | atomic_inc(&mm->mm_count); |
| 278 | |
| 279 | atomic_set(&ctx->users, 1); |
| 280 | spin_lock_init(&ctx->ctx_lock); |
| 281 | spin_lock_init(&ctx->ring_info.ring_lock); |
| 282 | init_waitqueue_head(&ctx->wait); |
| 283 | |
| 284 | INIT_LIST_HEAD(&ctx->active_reqs); |
| 285 | INIT_LIST_HEAD(&ctx->run_list); |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 286 | INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | |
| 288 | if (aio_setup_ring(ctx) < 0) |
| 289 | goto out_freectx; |
| 290 | |
| 291 | /* limit the number of system wide aios */ |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 292 | do { |
| 293 | spin_lock_bh(&aio_nr_lock); |
| 294 | if (aio_nr + nr_events > aio_max_nr || |
| 295 | aio_nr + nr_events < aio_nr) |
| 296 | ctx->max_reqs = 0; |
| 297 | else |
| 298 | aio_nr += ctx->max_reqs; |
| 299 | spin_unlock_bh(&aio_nr_lock); |
| 300 | if (ctx->max_reqs || did_sync) |
| 301 | break; |
| 302 | |
| 303 | /* wait for rcu callbacks to have completed before giving up */ |
| 304 | synchronize_rcu(); |
| 305 | did_sync = 1; |
| 306 | ctx->max_reqs = nr_events; |
| 307 | } while (1); |
| 308 | |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 309 | if (ctx->max_reqs == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | goto out_cleanup; |
| 311 | |
Jeff Moyer | 39fa003 | 2008-04-29 01:03:48 -0700 | [diff] [blame] | 312 | /* now link into global list. */ |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 313 | spin_lock(&mm->ioctx_lock); |
| 314 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); |
| 315 | spin_unlock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
| 317 | dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
| 318 | ctx, ctx->user_id, current->mm, ctx->ring_info.nr); |
| 319 | return ctx; |
| 320 | |
| 321 | out_cleanup: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | __put_ioctx(ctx); |
| 323 | return ERR_PTR(-EAGAIN); |
| 324 | |
| 325 | out_freectx: |
| 326 | mmdrop(mm); |
| 327 | kmem_cache_free(kioctx_cachep, ctx); |
| 328 | ctx = ERR_PTR(-ENOMEM); |
| 329 | |
| 330 | dprintk("aio: error allocating ioctx %p\n", ctx); |
| 331 | return ctx; |
| 332 | } |
| 333 | |
| 334 | /* aio_cancel_all |
| 335 | * Cancels all outstanding aio requests on an aio context. Used |
| 336 | * when the processes owning a context have all exited to encourage |
| 337 | * the rapid destruction of the kioctx. |
| 338 | */ |
| 339 | static void aio_cancel_all(struct kioctx *ctx) |
| 340 | { |
| 341 | int (*cancel)(struct kiocb *, struct io_event *); |
| 342 | struct io_event res; |
| 343 | spin_lock_irq(&ctx->ctx_lock); |
| 344 | ctx->dead = 1; |
| 345 | while (!list_empty(&ctx->active_reqs)) { |
| 346 | struct list_head *pos = ctx->active_reqs.next; |
| 347 | struct kiocb *iocb = list_kiocb(pos); |
| 348 | list_del_init(&iocb->ki_list); |
| 349 | cancel = iocb->ki_cancel; |
| 350 | kiocbSetCancelled(iocb); |
| 351 | if (cancel) { |
| 352 | iocb->ki_users++; |
| 353 | spin_unlock_irq(&ctx->ctx_lock); |
| 354 | cancel(iocb, &res); |
| 355 | spin_lock_irq(&ctx->ctx_lock); |
| 356 | } |
| 357 | } |
| 358 | spin_unlock_irq(&ctx->ctx_lock); |
| 359 | } |
| 360 | |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 361 | static void wait_for_all_aios(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | { |
| 363 | struct task_struct *tsk = current; |
| 364 | DECLARE_WAITQUEUE(wait, tsk); |
| 365 | |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 366 | spin_lock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | if (!ctx->reqs_active) |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 368 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | |
| 370 | add_wait_queue(&ctx->wait, &wait); |
| 371 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
| 372 | while (ctx->reqs_active) { |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 373 | spin_unlock_irq(&ctx->ctx_lock); |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 374 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 376 | spin_lock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | } |
| 378 | __set_task_state(tsk, TASK_RUNNING); |
| 379 | remove_wait_queue(&ctx->wait, &wait); |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 380 | |
| 381 | out: |
| 382 | spin_unlock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | } |
| 384 | |
| 385 | /* wait_on_sync_kiocb: |
| 386 | * Waits on the given sync kiocb to complete. |
| 387 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 388 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | { |
| 390 | while (iocb->ki_users) { |
| 391 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 392 | if (!iocb->ki_users) |
| 393 | break; |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 394 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | } |
| 396 | __set_current_state(TASK_RUNNING); |
| 397 | return iocb->ki_user_data; |
| 398 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 399 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | |
| 401 | /* exit_aio: called when the last user of mm goes away. At this point, |
| 402 | * there is no way for any new requests to be submited or any of the |
| 403 | * io_* syscalls to be called on the context. However, there may be |
| 404 | * outstanding requests which hold references to the context; as they |
| 405 | * go away, they will call put_ioctx and release any pinned memory |
| 406 | * associated with the request (held via struct page * references). |
| 407 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 408 | void exit_aio(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 410 | struct kioctx *ctx; |
| 411 | |
| 412 | while (!hlist_empty(&mm->ioctx_list)) { |
| 413 | ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); |
| 414 | hlist_del_rcu(&ctx->list); |
| 415 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | aio_cancel_all(ctx); |
| 417 | |
| 418 | wait_for_all_aios(ctx); |
| 419 | /* |
Andrew Morton | a9df62c | 2007-05-09 02:33:54 -0700 | [diff] [blame] | 420 | * Ensure we don't leave the ctx on the aio_wq |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | */ |
Oleg Nesterov | 28e53bd | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 422 | cancel_work_sync(&ctx->wq.work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | |
| 424 | if (1 != atomic_read(&ctx->users)) |
| 425 | printk(KERN_DEBUG |
| 426 | "exit_aio:ioctx still alive: %d %d %d\n", |
| 427 | atomic_read(&ctx->users), ctx->dead, |
| 428 | ctx->reqs_active); |
| 429 | put_ioctx(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | } |
| 431 | } |
| 432 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | /* aio_get_req |
| 434 | * Allocate a slot for an aio request. Increments the users count |
| 435 | * of the kioctx so that the kioctx stays around until all requests are |
| 436 | * complete. Returns NULL if no requests are free. |
| 437 | * |
| 438 | * Returns with kiocb->users set to 2. The io submit code path holds |
| 439 | * an extra reference while submitting the i/o. |
| 440 | * This prevents races between the aio code path referencing the |
| 441 | * req (after submitting it) and aio_complete() freeing the req. |
| 442 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 443 | static struct kiocb *__aio_get_req(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | { |
| 445 | struct kiocb *req = NULL; |
| 446 | struct aio_ring *ring; |
| 447 | int okay = 0; |
| 448 | |
| 449 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); |
| 450 | if (unlikely(!req)) |
| 451 | return NULL; |
| 452 | |
Zach Brown | 4faa528 | 2005-10-17 16:43:33 -0700 | [diff] [blame] | 453 | req->ki_flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | req->ki_users = 2; |
| 455 | req->ki_key = 0; |
| 456 | req->ki_ctx = ctx; |
| 457 | req->ki_cancel = NULL; |
| 458 | req->ki_retry = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | req->ki_dtor = NULL; |
| 460 | req->private = NULL; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 461 | req->ki_iovec = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | INIT_LIST_HEAD(&req->ki_run_list); |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 463 | req->ki_eventfd = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | |
| 465 | /* Check if the completion queue has enough free space to |
| 466 | * accept an event from this io. |
| 467 | */ |
| 468 | spin_lock_irq(&ctx->ctx_lock); |
| 469 | ring = kmap_atomic(ctx->ring_info.ring_pages[0], KM_USER0); |
| 470 | if (ctx->reqs_active < aio_ring_avail(&ctx->ring_info, ring)) { |
| 471 | list_add(&req->ki_list, &ctx->active_reqs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | ctx->reqs_active++; |
| 473 | okay = 1; |
| 474 | } |
| 475 | kunmap_atomic(ring, KM_USER0); |
| 476 | spin_unlock_irq(&ctx->ctx_lock); |
| 477 | |
| 478 | if (!okay) { |
| 479 | kmem_cache_free(kiocb_cachep, req); |
| 480 | req = NULL; |
| 481 | } |
| 482 | |
| 483 | return req; |
| 484 | } |
| 485 | |
| 486 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
| 487 | { |
| 488 | struct kiocb *req; |
| 489 | /* Handle a potential starvation case -- should be exceedingly rare as |
| 490 | * requests will be stuck on fput_head only if the aio_fput_routine is |
| 491 | * delayed and the requests were the last user of the struct file. |
| 492 | */ |
| 493 | req = __aio_get_req(ctx); |
| 494 | if (unlikely(NULL == req)) { |
| 495 | aio_fput_routine(NULL); |
| 496 | req = __aio_get_req(ctx); |
| 497 | } |
| 498 | return req; |
| 499 | } |
| 500 | |
| 501 | static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) |
| 502 | { |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 503 | assert_spin_locked(&ctx->ctx_lock); |
| 504 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 505 | if (req->ki_eventfd != NULL) |
| 506 | eventfd_ctx_put(req->ki_eventfd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | if (req->ki_dtor) |
| 508 | req->ki_dtor(req); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 509 | if (req->ki_iovec != &req->ki_inline_vec) |
| 510 | kfree(req->ki_iovec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | kmem_cache_free(kiocb_cachep, req); |
| 512 | ctx->reqs_active--; |
| 513 | |
| 514 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
| 515 | wake_up(&ctx->wait); |
| 516 | } |
| 517 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 518 | static void aio_fput_routine(struct work_struct *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | { |
| 520 | spin_lock_irq(&fput_lock); |
| 521 | while (likely(!list_empty(&fput_head))) { |
| 522 | struct kiocb *req = list_kiocb(fput_head.next); |
| 523 | struct kioctx *ctx = req->ki_ctx; |
| 524 | |
| 525 | list_del(&req->ki_list); |
| 526 | spin_unlock_irq(&fput_lock); |
| 527 | |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 528 | /* Complete the fput(s) */ |
| 529 | if (req->ki_filp != NULL) |
| 530 | __fput(req->ki_filp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | |
| 532 | /* Link the iocb into the context's free list */ |
| 533 | spin_lock_irq(&ctx->ctx_lock); |
| 534 | really_put_req(ctx, req); |
| 535 | spin_unlock_irq(&ctx->ctx_lock); |
| 536 | |
| 537 | put_ioctx(ctx); |
| 538 | spin_lock_irq(&fput_lock); |
| 539 | } |
| 540 | spin_unlock_irq(&fput_lock); |
| 541 | } |
| 542 | |
| 543 | /* __aio_put_req |
| 544 | * Returns true if this put was the last user of the request. |
| 545 | */ |
| 546 | static int __aio_put_req(struct kioctx *ctx, struct kiocb *req) |
| 547 | { |
Al Viro | 516e0cc | 2008-07-26 00:39:17 -0400 | [diff] [blame] | 548 | dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n", |
| 549 | req, atomic_long_read(&req->ki_filp->f_count)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 551 | assert_spin_locked(&ctx->ctx_lock); |
| 552 | |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 553 | req->ki_users--; |
Eric Sesterhenn | 93e06b4 | 2006-11-30 05:29:23 +0100 | [diff] [blame] | 554 | BUG_ON(req->ki_users < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | if (likely(req->ki_users)) |
| 556 | return 0; |
| 557 | list_del(&req->ki_list); /* remove from active_reqs */ |
| 558 | req->ki_cancel = NULL; |
| 559 | req->ki_retry = NULL; |
| 560 | |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 561 | /* |
| 562 | * Try to optimize the aio and eventfd file* puts, by avoiding to |
| 563 | * schedule work in case it is not __fput() time. In normal cases, |
| 564 | * we would not be holding the last reference to the file*, so |
| 565 | * this function will be executed w/out any aio kthread wakeup. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | */ |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 567 | if (unlikely(atomic_long_dec_and_test(&req->ki_filp->f_count))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | get_ioctx(ctx); |
| 569 | spin_lock(&fput_lock); |
| 570 | list_add(&req->ki_list, &fput_head); |
| 571 | spin_unlock(&fput_lock); |
| 572 | queue_work(aio_wq, &fput_work); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 573 | } else { |
| 574 | req->ki_filp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | really_put_req(ctx, req); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 576 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | return 1; |
| 578 | } |
| 579 | |
| 580 | /* aio_put_req |
| 581 | * Returns true if this put was the last user of the kiocb, |
| 582 | * false if the request is still in use. |
| 583 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 584 | int aio_put_req(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | { |
| 586 | struct kioctx *ctx = req->ki_ctx; |
| 587 | int ret; |
| 588 | spin_lock_irq(&ctx->ctx_lock); |
| 589 | ret = __aio_put_req(ctx, req); |
| 590 | spin_unlock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | return ret; |
| 592 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 593 | EXPORT_SYMBOL(aio_put_req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 594 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 595 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 597 | struct mm_struct *mm = current->mm; |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 598 | struct kioctx *ctx, *ret = NULL; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 599 | struct hlist_node *n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 601 | rcu_read_lock(); |
| 602 | |
| 603 | hlist_for_each_entry_rcu(ctx, n, &mm->ioctx_list, list) { |
| 604 | if (ctx->user_id == ctx_id && !ctx->dead) { |
| 605 | get_ioctx(ctx); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 606 | ret = ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | break; |
| 608 | } |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 609 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 611 | rcu_read_unlock(); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 612 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | } |
| 614 | |
| 615 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | * Queue up a kiocb to be retried. Assumes that the kiocb |
| 617 | * has already been marked as kicked, and places it on |
| 618 | * the retry run list for the corresponding ioctx, if it |
| 619 | * isn't already queued. Returns 1 if it actually queued |
| 620 | * the kiocb (to tell the caller to activate the work |
| 621 | * queue to process it), or 0, if it found that it was |
| 622 | * already queued. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | */ |
| 624 | static inline int __queue_kicked_iocb(struct kiocb *iocb) |
| 625 | { |
| 626 | struct kioctx *ctx = iocb->ki_ctx; |
| 627 | |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 628 | assert_spin_locked(&ctx->ctx_lock); |
| 629 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | if (list_empty(&iocb->ki_run_list)) { |
| 631 | list_add_tail(&iocb->ki_run_list, |
| 632 | &ctx->run_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | return 1; |
| 634 | } |
| 635 | return 0; |
| 636 | } |
| 637 | |
| 638 | /* aio_run_iocb |
| 639 | * This is the core aio execution routine. It is |
| 640 | * invoked both for initial i/o submission and |
| 641 | * subsequent retries via the aio_kick_handler. |
| 642 | * Expects to be invoked with iocb->ki_ctx->lock |
Andreas Mohr | d6e05ed | 2006-06-26 18:35:02 +0200 | [diff] [blame] | 643 | * already held. The lock is released and reacquired |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | * as needed during processing. |
| 645 | * |
| 646 | * Calls the iocb retry method (already setup for the |
| 647 | * iocb on initial submission) for operation specific |
| 648 | * handling, but takes care of most of common retry |
| 649 | * execution details for a given iocb. The retry method |
| 650 | * needs to be non-blocking as far as possible, to avoid |
| 651 | * holding up other iocbs waiting to be serviced by the |
| 652 | * retry kernel thread. |
| 653 | * |
| 654 | * The trickier parts in this code have to do with |
| 655 | * ensuring that only one retry instance is in progress |
| 656 | * for a given iocb at any time. Providing that guarantee |
| 657 | * simplifies the coding of individual aio operations as |
| 658 | * it avoids various potential races. |
| 659 | */ |
| 660 | static ssize_t aio_run_iocb(struct kiocb *iocb) |
| 661 | { |
| 662 | struct kioctx *ctx = iocb->ki_ctx; |
| 663 | ssize_t (*retry)(struct kiocb *); |
| 664 | ssize_t ret; |
| 665 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | if (!(retry = iocb->ki_retry)) { |
| 667 | printk("aio_run_iocb: iocb->ki_retry = NULL\n"); |
| 668 | return 0; |
| 669 | } |
| 670 | |
| 671 | /* |
| 672 | * We don't want the next retry iteration for this |
| 673 | * operation to start until this one has returned and |
| 674 | * updated the iocb state. However, wait_queue functions |
| 675 | * can trigger a kick_iocb from interrupt context in the |
| 676 | * meantime, indicating that data is available for the next |
| 677 | * iteration. We want to remember that and enable the |
| 678 | * next retry iteration _after_ we are through with |
| 679 | * this one. |
| 680 | * |
| 681 | * So, in order to be able to register a "kick", but |
| 682 | * prevent it from being queued now, we clear the kick |
| 683 | * flag, but make the kick code *think* that the iocb is |
| 684 | * still on the run list until we are actually done. |
| 685 | * When we are done with this iteration, we check if |
| 686 | * the iocb was kicked in the meantime and if so, queue |
| 687 | * it up afresh. |
| 688 | */ |
| 689 | |
| 690 | kiocbClearKicked(iocb); |
| 691 | |
| 692 | /* |
| 693 | * This is so that aio_complete knows it doesn't need to |
| 694 | * pull the iocb off the run list (We can't just call |
| 695 | * INIT_LIST_HEAD because we don't want a kick_iocb to |
| 696 | * queue this on the run list yet) |
| 697 | */ |
| 698 | iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL; |
| 699 | spin_unlock_irq(&ctx->ctx_lock); |
| 700 | |
| 701 | /* Quit retrying if the i/o has been cancelled */ |
| 702 | if (kiocbIsCancelled(iocb)) { |
| 703 | ret = -EINTR; |
| 704 | aio_complete(iocb, ret, 0); |
| 705 | /* must not access the iocb after this */ |
| 706 | goto out; |
| 707 | } |
| 708 | |
| 709 | /* |
| 710 | * Now we are all set to call the retry method in async |
Alexey Dobriyan | 6212e3a | 2007-10-18 03:04:56 -0700 | [diff] [blame] | 711 | * context. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | ret = retry(iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | |
Shaohua Li | fac046a | 2009-12-15 16:47:47 -0800 | [diff] [blame] | 715 | if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) |
Zach Brown | 897f15f | 2005-09-30 11:58:55 -0700 | [diff] [blame] | 716 | aio_complete(iocb, ret, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | out: |
| 718 | spin_lock_irq(&ctx->ctx_lock); |
| 719 | |
| 720 | if (-EIOCBRETRY == ret) { |
| 721 | /* |
| 722 | * OK, now that we are done with this iteration |
| 723 | * and know that there is more left to go, |
| 724 | * this is where we let go so that a subsequent |
| 725 | * "kick" can start the next iteration |
| 726 | */ |
| 727 | |
| 728 | /* will make __queue_kicked_iocb succeed from here on */ |
| 729 | INIT_LIST_HEAD(&iocb->ki_run_list); |
| 730 | /* we must queue the next iteration ourselves, if it |
| 731 | * has already been kicked */ |
| 732 | if (kiocbIsKicked(iocb)) { |
| 733 | __queue_kicked_iocb(iocb); |
Sébastien Dugu | c016e22 | 2005-06-28 20:44:59 -0700 | [diff] [blame] | 734 | |
| 735 | /* |
| 736 | * __queue_kicked_iocb will always return 1 here, because |
| 737 | * iocb->ki_run_list is empty at this point so it should |
| 738 | * be safe to unconditionally queue the context into the |
| 739 | * work queue. |
| 740 | */ |
| 741 | aio_queue_work(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | } |
| 743 | } |
| 744 | return ret; |
| 745 | } |
| 746 | |
| 747 | /* |
| 748 | * __aio_run_iocbs: |
| 749 | * Process all pending retries queued on the ioctx |
| 750 | * run list. |
| 751 | * Assumes it is operating within the aio issuer's mm |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 752 | * context. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | */ |
| 754 | static int __aio_run_iocbs(struct kioctx *ctx) |
| 755 | { |
| 756 | struct kiocb *iocb; |
Oleg Nesterov | 626ab0e | 2006-06-23 02:05:55 -0700 | [diff] [blame] | 757 | struct list_head run_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 759 | assert_spin_locked(&ctx->ctx_lock); |
| 760 | |
Oleg Nesterov | 626ab0e | 2006-06-23 02:05:55 -0700 | [diff] [blame] | 761 | list_replace_init(&ctx->run_list, &run_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | while (!list_empty(&run_list)) { |
| 763 | iocb = list_entry(run_list.next, struct kiocb, |
| 764 | ki_run_list); |
| 765 | list_del(&iocb->ki_run_list); |
| 766 | /* |
| 767 | * Hold an extra reference while retrying i/o. |
| 768 | */ |
| 769 | iocb->ki_users++; /* grab extra reference */ |
| 770 | aio_run_iocb(iocb); |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 771 | __aio_put_req(ctx, iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 773 | if (!list_empty(&ctx->run_list)) |
| 774 | return 1; |
| 775 | return 0; |
| 776 | } |
| 777 | |
| 778 | static void aio_queue_work(struct kioctx * ctx) |
| 779 | { |
| 780 | unsigned long timeout; |
| 781 | /* |
| 782 | * if someone is waiting, get the work started right |
| 783 | * away, otherwise, use a longer delay |
| 784 | */ |
| 785 | smp_mb(); |
| 786 | if (waitqueue_active(&ctx->wait)) |
| 787 | timeout = 1; |
| 788 | else |
| 789 | timeout = HZ/10; |
| 790 | queue_delayed_work(aio_wq, &ctx->wq, timeout); |
| 791 | } |
| 792 | |
| 793 | |
| 794 | /* |
| 795 | * aio_run_iocbs: |
| 796 | * Process all pending retries queued on the ioctx |
| 797 | * run list. |
| 798 | * Assumes it is operating within the aio issuer's mm |
| 799 | * context. |
| 800 | */ |
| 801 | static inline void aio_run_iocbs(struct kioctx *ctx) |
| 802 | { |
| 803 | int requeue; |
| 804 | |
| 805 | spin_lock_irq(&ctx->ctx_lock); |
| 806 | |
| 807 | requeue = __aio_run_iocbs(ctx); |
| 808 | spin_unlock_irq(&ctx->ctx_lock); |
| 809 | if (requeue) |
| 810 | aio_queue_work(ctx); |
| 811 | } |
| 812 | |
| 813 | /* |
| 814 | * just like aio_run_iocbs, but keeps running them until |
| 815 | * the list stays empty |
| 816 | */ |
| 817 | static inline void aio_run_all_iocbs(struct kioctx *ctx) |
| 818 | { |
| 819 | spin_lock_irq(&ctx->ctx_lock); |
| 820 | while (__aio_run_iocbs(ctx)) |
| 821 | ; |
| 822 | spin_unlock_irq(&ctx->ctx_lock); |
| 823 | } |
| 824 | |
| 825 | /* |
| 826 | * aio_kick_handler: |
| 827 | * Work queue handler triggered to process pending |
| 828 | * retries on an ioctx. Takes on the aio issuer's |
| 829 | * mm context before running the iocbs, so that |
| 830 | * copy_xxx_user operates on the issuer's address |
| 831 | * space. |
| 832 | * Run on aiod's context. |
| 833 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 834 | static void aio_kick_handler(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | { |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 836 | struct kioctx *ctx = container_of(work, struct kioctx, wq.work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | mm_segment_t oldfs = get_fs(); |
Zach Brown | 1ebb110 | 2006-12-29 16:47:02 -0800 | [diff] [blame] | 838 | struct mm_struct *mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | int requeue; |
| 840 | |
| 841 | set_fs(USER_DS); |
| 842 | use_mm(ctx->mm); |
| 843 | spin_lock_irq(&ctx->ctx_lock); |
| 844 | requeue =__aio_run_iocbs(ctx); |
Zach Brown | 1ebb110 | 2006-12-29 16:47:02 -0800 | [diff] [blame] | 845 | mm = ctx->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | spin_unlock_irq(&ctx->ctx_lock); |
Zach Brown | 1ebb110 | 2006-12-29 16:47:02 -0800 | [diff] [blame] | 847 | unuse_mm(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | set_fs(oldfs); |
| 849 | /* |
| 850 | * we're in a worker thread already, don't use queue_delayed_work, |
| 851 | */ |
| 852 | if (requeue) |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 853 | queue_delayed_work(aio_wq, &ctx->wq, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | } |
| 855 | |
| 856 | |
| 857 | /* |
| 858 | * Called by kick_iocb to queue the kiocb for retry |
| 859 | * and if required activate the aio work queue to process |
| 860 | * it |
| 861 | */ |
Zach Brown | 998765e | 2005-09-30 11:58:54 -0700 | [diff] [blame] | 862 | static void try_queue_kicked_iocb(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | { |
| 864 | struct kioctx *ctx = iocb->ki_ctx; |
| 865 | unsigned long flags; |
| 866 | int run = 0; |
| 867 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 868 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
Zach Brown | 998765e | 2005-09-30 11:58:54 -0700 | [diff] [blame] | 869 | /* set this inside the lock so that we can't race with aio_run_iocb() |
| 870 | * testing it and putting the iocb on the run list under the lock */ |
| 871 | if (!kiocbTryKick(iocb)) |
| 872 | run = __queue_kicked_iocb(iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 873 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
Ken Chen | 644d3a0 | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 874 | if (run) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | aio_queue_work(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | } |
| 877 | |
| 878 | /* |
| 879 | * kick_iocb: |
| 880 | * Called typically from a wait queue callback context |
Shaohua Li | fac046a | 2009-12-15 16:47:47 -0800 | [diff] [blame] | 881 | * to trigger a retry of the iocb. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | * The retry is usually executed by aio workqueue |
| 883 | * threads (See aio_kick_handler). |
| 884 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 885 | void kick_iocb(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | { |
| 887 | /* sync iocbs are easy: they can only ever be executing from a |
| 888 | * single context. */ |
| 889 | if (is_sync_kiocb(iocb)) { |
| 890 | kiocbSetKicked(iocb); |
| 891 | wake_up_process(iocb->ki_obj.tsk); |
| 892 | return; |
| 893 | } |
| 894 | |
Zach Brown | 998765e | 2005-09-30 11:58:54 -0700 | [diff] [blame] | 895 | try_queue_kicked_iocb(iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | } |
| 897 | EXPORT_SYMBOL(kick_iocb); |
| 898 | |
| 899 | /* aio_complete |
| 900 | * Called when the io request on the given iocb is complete. |
| 901 | * Returns true if this is the last user of the request. The |
| 902 | * only other user of the request can be the cancellation code. |
| 903 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 904 | int aio_complete(struct kiocb *iocb, long res, long res2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | { |
| 906 | struct kioctx *ctx = iocb->ki_ctx; |
| 907 | struct aio_ring_info *info; |
| 908 | struct aio_ring *ring; |
| 909 | struct io_event *event; |
| 910 | unsigned long flags; |
| 911 | unsigned long tail; |
| 912 | int ret; |
| 913 | |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 914 | /* |
| 915 | * Special case handling for sync iocbs: |
| 916 | * - events go directly into the iocb for fast handling |
| 917 | * - the sync task with the iocb in its stack holds the single iocb |
| 918 | * ref, no other paths have a way to get another ref |
| 919 | * - the sync task helpfully left a reference to itself in the iocb |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 920 | */ |
| 921 | if (is_sync_kiocb(iocb)) { |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 922 | BUG_ON(iocb->ki_users != 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | iocb->ki_user_data = res; |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 924 | iocb->ki_users = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | wake_up_process(iocb->ki_obj.tsk); |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 926 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 927 | } |
| 928 | |
| 929 | info = &ctx->ring_info; |
| 930 | |
| 931 | /* add a completion event to the ring buffer. |
| 932 | * must be done holding ctx->ctx_lock to prevent |
| 933 | * other code from messing with the tail |
| 934 | * pointer since we might be called from irq |
| 935 | * context. |
| 936 | */ |
| 937 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 938 | |
| 939 | if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list)) |
| 940 | list_del_init(&iocb->ki_run_list); |
| 941 | |
| 942 | /* |
| 943 | * cancelled requests don't get events, userland was given one |
| 944 | * when the event got cancelled. |
| 945 | */ |
| 946 | if (kiocbIsCancelled(iocb)) |
| 947 | goto put_rq; |
| 948 | |
| 949 | ring = kmap_atomic(info->ring_pages[0], KM_IRQ1); |
| 950 | |
| 951 | tail = info->tail; |
| 952 | event = aio_ring_event(info, tail, KM_IRQ0); |
Ken Chen | 4bf69b2 | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 953 | if (++tail >= info->nr) |
| 954 | tail = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | |
| 956 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
| 957 | event->data = iocb->ki_user_data; |
| 958 | event->res = res; |
| 959 | event->res2 = res2; |
| 960 | |
| 961 | dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n", |
| 962 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
| 963 | res, res2); |
| 964 | |
| 965 | /* after flagging the request as done, we |
| 966 | * must never even look at it again |
| 967 | */ |
| 968 | smp_wmb(); /* make event visible before updating tail */ |
| 969 | |
| 970 | info->tail = tail; |
| 971 | ring->tail = tail; |
| 972 | |
| 973 | put_aio_ring_event(event, KM_IRQ0); |
| 974 | kunmap_atomic(ring, KM_IRQ1); |
| 975 | |
| 976 | pr_debug("added to ring %p at [%lu]\n", iocb, tail); |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 977 | |
| 978 | /* |
| 979 | * Check if the user asked us to deliver the result through an |
| 980 | * eventfd. The eventfd_signal() function is safe to be called |
| 981 | * from IRQ context. |
| 982 | */ |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 983 | if (iocb->ki_eventfd != NULL) |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 984 | eventfd_signal(iocb->ki_eventfd, 1); |
| 985 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 986 | put_rq: |
| 987 | /* everything turned out well, dispose of the aiocb. */ |
| 988 | ret = __aio_put_req(ctx, iocb); |
| 989 | |
Quentin Barnes | 6cb2a21 | 2008-03-19 17:00:39 -0700 | [diff] [blame] | 990 | /* |
| 991 | * We have to order our ring_info tail store above and test |
| 992 | * of the wait list below outside the wait lock. This is |
| 993 | * like in wake_up_bit() where clearing a bit has to be |
| 994 | * ordered with the unlocked test. |
| 995 | */ |
| 996 | smp_mb(); |
| 997 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | if (waitqueue_active(&ctx->wait)) |
| 999 | wake_up(&ctx->wait); |
| 1000 | |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 1001 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | return ret; |
| 1003 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 1004 | EXPORT_SYMBOL(aio_complete); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | |
| 1006 | /* aio_read_evt |
| 1007 | * Pull an event off of the ioctx's event ring. Returns the number of |
| 1008 | * events fetched (0 or 1 ;-) |
| 1009 | * FIXME: make this use cmpxchg. |
| 1010 | * TODO: make the ringbuffer user mmap()able (requires FIXME). |
| 1011 | */ |
| 1012 | static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) |
| 1013 | { |
| 1014 | struct aio_ring_info *info = &ioctx->ring_info; |
| 1015 | struct aio_ring *ring; |
| 1016 | unsigned long head; |
| 1017 | int ret = 0; |
| 1018 | |
| 1019 | ring = kmap_atomic(info->ring_pages[0], KM_USER0); |
| 1020 | dprintk("in aio_read_evt h%lu t%lu m%lu\n", |
| 1021 | (unsigned long)ring->head, (unsigned long)ring->tail, |
| 1022 | (unsigned long)ring->nr); |
| 1023 | |
| 1024 | if (ring->head == ring->tail) |
| 1025 | goto out; |
| 1026 | |
| 1027 | spin_lock(&info->ring_lock); |
| 1028 | |
| 1029 | head = ring->head % info->nr; |
| 1030 | if (head != ring->tail) { |
| 1031 | struct io_event *evp = aio_ring_event(info, head, KM_USER1); |
| 1032 | *ent = *evp; |
| 1033 | head = (head + 1) % info->nr; |
| 1034 | smp_mb(); /* finish reading the event before updatng the head */ |
| 1035 | ring->head = head; |
| 1036 | ret = 1; |
| 1037 | put_aio_ring_event(evp, KM_USER1); |
| 1038 | } |
| 1039 | spin_unlock(&info->ring_lock); |
| 1040 | |
| 1041 | out: |
| 1042 | kunmap_atomic(ring, KM_USER0); |
| 1043 | dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret, |
| 1044 | (unsigned long)ring->head, (unsigned long)ring->tail); |
| 1045 | return ret; |
| 1046 | } |
| 1047 | |
| 1048 | struct aio_timeout { |
| 1049 | struct timer_list timer; |
| 1050 | int timed_out; |
| 1051 | struct task_struct *p; |
| 1052 | }; |
| 1053 | |
| 1054 | static void timeout_func(unsigned long data) |
| 1055 | { |
| 1056 | struct aio_timeout *to = (struct aio_timeout *)data; |
| 1057 | |
| 1058 | to->timed_out = 1; |
| 1059 | wake_up_process(to->p); |
| 1060 | } |
| 1061 | |
| 1062 | static inline void init_timeout(struct aio_timeout *to) |
| 1063 | { |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 1064 | setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | to->timed_out = 0; |
| 1066 | to->p = current; |
| 1067 | } |
| 1068 | |
| 1069 | static inline void set_timeout(long start_jiffies, struct aio_timeout *to, |
| 1070 | const struct timespec *ts) |
| 1071 | { |
| 1072 | to->timer.expires = start_jiffies + timespec_to_jiffies(ts); |
| 1073 | if (time_after(to->timer.expires, jiffies)) |
| 1074 | add_timer(&to->timer); |
| 1075 | else |
| 1076 | to->timed_out = 1; |
| 1077 | } |
| 1078 | |
| 1079 | static inline void clear_timeout(struct aio_timeout *to) |
| 1080 | { |
| 1081 | del_singleshot_timer_sync(&to->timer); |
| 1082 | } |
| 1083 | |
| 1084 | static int read_events(struct kioctx *ctx, |
| 1085 | long min_nr, long nr, |
| 1086 | struct io_event __user *event, |
| 1087 | struct timespec __user *timeout) |
| 1088 | { |
| 1089 | long start_jiffies = jiffies; |
| 1090 | struct task_struct *tsk = current; |
| 1091 | DECLARE_WAITQUEUE(wait, tsk); |
| 1092 | int ret; |
| 1093 | int i = 0; |
| 1094 | struct io_event ent; |
| 1095 | struct aio_timeout to; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | int retry = 0; |
| 1097 | |
| 1098 | /* needed to zero any padding within an entry (there shouldn't be |
| 1099 | * any, but C is fun! |
| 1100 | */ |
| 1101 | memset(&ent, 0, sizeof(ent)); |
| 1102 | retry: |
| 1103 | ret = 0; |
| 1104 | while (likely(i < nr)) { |
| 1105 | ret = aio_read_evt(ctx, &ent); |
| 1106 | if (unlikely(ret <= 0)) |
| 1107 | break; |
| 1108 | |
| 1109 | dprintk("read event: %Lx %Lx %Lx %Lx\n", |
| 1110 | ent.data, ent.obj, ent.res, ent.res2); |
| 1111 | |
| 1112 | /* Could we split the check in two? */ |
| 1113 | ret = -EFAULT; |
| 1114 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { |
| 1115 | dprintk("aio: lost an event due to EFAULT.\n"); |
| 1116 | break; |
| 1117 | } |
| 1118 | ret = 0; |
| 1119 | |
| 1120 | /* Good, event copied to userland, update counts. */ |
| 1121 | event ++; |
| 1122 | i ++; |
| 1123 | } |
| 1124 | |
| 1125 | if (min_nr <= i) |
| 1126 | return i; |
| 1127 | if (ret) |
| 1128 | return ret; |
| 1129 | |
| 1130 | /* End fast path */ |
| 1131 | |
| 1132 | /* racey check, but it gets redone */ |
| 1133 | if (!retry && unlikely(!list_empty(&ctx->run_list))) { |
| 1134 | retry = 1; |
| 1135 | aio_run_all_iocbs(ctx); |
| 1136 | goto retry; |
| 1137 | } |
| 1138 | |
| 1139 | init_timeout(&to); |
| 1140 | if (timeout) { |
| 1141 | struct timespec ts; |
| 1142 | ret = -EFAULT; |
| 1143 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
| 1144 | goto out; |
| 1145 | |
| 1146 | set_timeout(start_jiffies, &to, &ts); |
| 1147 | } |
| 1148 | |
| 1149 | while (likely(i < nr)) { |
| 1150 | add_wait_queue_exclusive(&ctx->wait, &wait); |
| 1151 | do { |
| 1152 | set_task_state(tsk, TASK_INTERRUPTIBLE); |
| 1153 | ret = aio_read_evt(ctx, &ent); |
| 1154 | if (ret) |
| 1155 | break; |
| 1156 | if (min_nr <= i) |
| 1157 | break; |
Jeff Moyer | e92adcb | 2008-04-28 02:12:04 -0700 | [diff] [blame] | 1158 | if (unlikely(ctx->dead)) { |
| 1159 | ret = -EINVAL; |
| 1160 | break; |
| 1161 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | if (to.timed_out) /* Only check after read evt */ |
| 1163 | break; |
Jeff Moyer | e00ba3d | 2007-12-04 23:45:02 -0800 | [diff] [blame] | 1164 | /* Try to only show up in io wait if there are ops |
| 1165 | * in flight */ |
| 1166 | if (ctx->reqs_active) |
| 1167 | io_schedule(); |
| 1168 | else |
| 1169 | schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | if (signal_pending(tsk)) { |
| 1171 | ret = -EINTR; |
| 1172 | break; |
| 1173 | } |
| 1174 | /*ret = aio_read_evt(ctx, &ent);*/ |
| 1175 | } while (1) ; |
| 1176 | |
| 1177 | set_task_state(tsk, TASK_RUNNING); |
| 1178 | remove_wait_queue(&ctx->wait, &wait); |
| 1179 | |
| 1180 | if (unlikely(ret <= 0)) |
| 1181 | break; |
| 1182 | |
| 1183 | ret = -EFAULT; |
| 1184 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { |
| 1185 | dprintk("aio: lost an event due to EFAULT.\n"); |
| 1186 | break; |
| 1187 | } |
| 1188 | |
| 1189 | /* Good, event copied to userland, update counts. */ |
| 1190 | event ++; |
| 1191 | i ++; |
| 1192 | } |
| 1193 | |
| 1194 | if (timeout) |
| 1195 | clear_timeout(&to); |
| 1196 | out: |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 1197 | destroy_timer_on_stack(&to.timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | return i ? i : ret; |
| 1199 | } |
| 1200 | |
| 1201 | /* Take an ioctx and remove it from the list of ioctx's. Protects |
| 1202 | * against races with itself via ->dead. |
| 1203 | */ |
| 1204 | static void io_destroy(struct kioctx *ioctx) |
| 1205 | { |
| 1206 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | int was_dead; |
| 1208 | |
| 1209 | /* delete the entry from the list is someone else hasn't already */ |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 1210 | spin_lock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | was_dead = ioctx->dead; |
| 1212 | ioctx->dead = 1; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 1213 | hlist_del_rcu(&ioctx->list); |
| 1214 | spin_unlock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | |
| 1216 | dprintk("aio_release(%p)\n", ioctx); |
| 1217 | if (likely(!was_dead)) |
| 1218 | put_ioctx(ioctx); /* twice for the list */ |
| 1219 | |
| 1220 | aio_cancel_all(ioctx); |
| 1221 | wait_for_all_aios(ioctx); |
Jeff Moyer | e92adcb | 2008-04-28 02:12:04 -0700 | [diff] [blame] | 1222 | |
| 1223 | /* |
| 1224 | * Wake up any waiters. The setting of ctx->dead must be seen |
| 1225 | * by other CPUs at this point. Right now, we rely on the |
| 1226 | * locking done by the above calls to ensure this consistency. |
| 1227 | */ |
| 1228 | wake_up(&ioctx->wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | put_ioctx(ioctx); /* once for the lookup */ |
| 1230 | } |
| 1231 | |
| 1232 | /* sys_io_setup: |
| 1233 | * Create an aio_context capable of receiving at least nr_events. |
| 1234 | * ctxp must not point to an aio_context that already exists, and |
| 1235 | * must be initialized to 0 prior to the call. On successful |
| 1236 | * creation of the aio_context, *ctxp is filled in with the resulting |
| 1237 | * handle. May fail with -EINVAL if *ctxp is not initialized, |
| 1238 | * if the specified nr_events exceeds internal limits. May fail |
| 1239 | * with -EAGAIN if the specified nr_events exceeds the user's limit |
| 1240 | * of available events. May fail with -ENOMEM if insufficient kernel |
| 1241 | * resources are available. May fail with -EFAULT if an invalid |
| 1242 | * pointer is passed for ctxp. Will fail with -ENOSYS if not |
| 1243 | * implemented. |
| 1244 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1245 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1246 | { |
| 1247 | struct kioctx *ioctx = NULL; |
| 1248 | unsigned long ctx; |
| 1249 | long ret; |
| 1250 | |
| 1251 | ret = get_user(ctx, ctxp); |
| 1252 | if (unlikely(ret)) |
| 1253 | goto out; |
| 1254 | |
| 1255 | ret = -EINVAL; |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 1256 | if (unlikely(ctx || nr_events == 0)) { |
| 1257 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", |
| 1258 | ctx, nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | goto out; |
| 1260 | } |
| 1261 | |
| 1262 | ioctx = ioctx_alloc(nr_events); |
| 1263 | ret = PTR_ERR(ioctx); |
| 1264 | if (!IS_ERR(ioctx)) { |
| 1265 | ret = put_user(ioctx->user_id, ctxp); |
| 1266 | if (!ret) |
| 1267 | return 0; |
| 1268 | |
| 1269 | get_ioctx(ioctx); /* io_destroy() expects us to hold a ref */ |
| 1270 | io_destroy(ioctx); |
| 1271 | } |
| 1272 | |
| 1273 | out: |
| 1274 | return ret; |
| 1275 | } |
| 1276 | |
| 1277 | /* sys_io_destroy: |
| 1278 | * Destroy the aio_context specified. May cancel any outstanding |
| 1279 | * AIOs and block on completion. Will fail with -ENOSYS if not |
| 1280 | * implemented. May fail with -EFAULT if the context pointed to |
| 1281 | * is invalid. |
| 1282 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1283 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1284 | { |
| 1285 | struct kioctx *ioctx = lookup_ioctx(ctx); |
| 1286 | if (likely(NULL != ioctx)) { |
| 1287 | io_destroy(ioctx); |
| 1288 | return 0; |
| 1289 | } |
| 1290 | pr_debug("EINVAL: io_destroy: invalid context id\n"); |
| 1291 | return -EINVAL; |
| 1292 | } |
| 1293 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1294 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
| 1295 | { |
| 1296 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
| 1297 | |
| 1298 | BUG_ON(ret <= 0); |
| 1299 | |
| 1300 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { |
| 1301 | ssize_t this = min((ssize_t)iov->iov_len, ret); |
| 1302 | iov->iov_base += this; |
| 1303 | iov->iov_len -= this; |
| 1304 | iocb->ki_left -= this; |
| 1305 | ret -= this; |
| 1306 | if (iov->iov_len == 0) { |
| 1307 | iocb->ki_cur_seg++; |
| 1308 | iov++; |
| 1309 | } |
| 1310 | } |
| 1311 | |
| 1312 | /* the caller should not have done more io than what fit in |
| 1313 | * the remaining iovecs */ |
| 1314 | BUG_ON(ret > 0 && iocb->ki_left == 0); |
| 1315 | } |
| 1316 | |
| 1317 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1318 | { |
| 1319 | struct file *file = iocb->ki_filp; |
| 1320 | struct address_space *mapping = file->f_mapping; |
| 1321 | struct inode *inode = mapping->host; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1322 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, |
| 1323 | unsigned long, loff_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1324 | ssize_t ret = 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1325 | unsigned short opcode; |
| 1326 | |
| 1327 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || |
| 1328 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { |
| 1329 | rw_op = file->f_op->aio_read; |
| 1330 | opcode = IOCB_CMD_PREADV; |
| 1331 | } else { |
| 1332 | rw_op = file->f_op->aio_write; |
| 1333 | opcode = IOCB_CMD_PWRITEV; |
| 1334 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | |
Rusty Russell | c2ec668 | 2008-02-08 04:20:15 -0800 | [diff] [blame] | 1336 | /* This matches the pread()/pwrite() logic */ |
| 1337 | if (iocb->ki_pos < 0) |
| 1338 | return -EINVAL; |
| 1339 | |
Zach Brown | 897f15f | 2005-09-30 11:58:55 -0700 | [diff] [blame] | 1340 | do { |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1341 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
| 1342 | iocb->ki_nr_segs - iocb->ki_cur_seg, |
| 1343 | iocb->ki_pos); |
| 1344 | if (ret > 0) |
| 1345 | aio_advance_iovec(iocb, ret); |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1346 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1347 | /* retry all partial writes. retry partial reads as long as its a |
| 1348 | * regular file. */ |
Zach Brown | 353fb07 | 2005-09-30 11:58:56 -0700 | [diff] [blame] | 1349 | } while (ret > 0 && iocb->ki_left > 0 && |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1350 | (opcode == IOCB_CMD_PWRITEV || |
| 1351 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | |
| 1353 | /* This means we must have transferred all that we could */ |
| 1354 | /* No need to retry anymore */ |
| 1355 | if ((ret == 0) || (iocb->ki_left == 0)) |
| 1356 | ret = iocb->ki_nbytes - iocb->ki_left; |
| 1357 | |
Rusty Russell | 7adfa2f | 2008-02-08 04:20:14 -0800 | [diff] [blame] | 1358 | /* If we managed to write some out we return that, rather than |
| 1359 | * the eventual error. */ |
| 1360 | if (opcode == IOCB_CMD_PWRITEV |
| 1361 | && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY |
| 1362 | && iocb->ki_nbytes - iocb->ki_left) |
| 1363 | ret = iocb->ki_nbytes - iocb->ki_left; |
| 1364 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | return ret; |
| 1366 | } |
| 1367 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | static ssize_t aio_fdsync(struct kiocb *iocb) |
| 1369 | { |
| 1370 | struct file *file = iocb->ki_filp; |
| 1371 | ssize_t ret = -EINVAL; |
| 1372 | |
| 1373 | if (file->f_op->aio_fsync) |
| 1374 | ret = file->f_op->aio_fsync(iocb, 1); |
| 1375 | return ret; |
| 1376 | } |
| 1377 | |
| 1378 | static ssize_t aio_fsync(struct kiocb *iocb) |
| 1379 | { |
| 1380 | struct file *file = iocb->ki_filp; |
| 1381 | ssize_t ret = -EINVAL; |
| 1382 | |
| 1383 | if (file->f_op->aio_fsync) |
| 1384 | ret = file->f_op->aio_fsync(iocb, 0); |
| 1385 | return ret; |
| 1386 | } |
| 1387 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1388 | static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1389 | { |
| 1390 | ssize_t ret; |
| 1391 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1392 | #ifdef CONFIG_COMPAT |
| 1393 | if (compat) |
| 1394 | ret = compat_rw_copy_check_uvector(type, |
| 1395 | (struct compat_iovec __user *)kiocb->ki_buf, |
| 1396 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, |
| 1397 | &kiocb->ki_iovec); |
| 1398 | else |
| 1399 | #endif |
| 1400 | ret = rw_copy_check_uvector(type, |
| 1401 | (struct iovec __user *)kiocb->ki_buf, |
| 1402 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, |
| 1403 | &kiocb->ki_iovec); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1404 | if (ret < 0) |
| 1405 | goto out; |
| 1406 | |
| 1407 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
| 1408 | kiocb->ki_cur_seg = 0; |
| 1409 | /* ki_nbytes/left now reflect bytes instead of segs */ |
| 1410 | kiocb->ki_nbytes = ret; |
| 1411 | kiocb->ki_left = ret; |
| 1412 | |
| 1413 | ret = 0; |
| 1414 | out: |
| 1415 | return ret; |
| 1416 | } |
| 1417 | |
| 1418 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb) |
| 1419 | { |
| 1420 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
| 1421 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; |
| 1422 | kiocb->ki_iovec->iov_len = kiocb->ki_left; |
| 1423 | kiocb->ki_nr_segs = 1; |
| 1424 | kiocb->ki_cur_seg = 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1425 | return 0; |
| 1426 | } |
| 1427 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1428 | /* |
| 1429 | * aio_setup_iocb: |
| 1430 | * Performs the initial checks and aio retry method |
| 1431 | * setup for the kiocb at the time of io submission. |
| 1432 | */ |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1433 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1434 | { |
| 1435 | struct file *file = kiocb->ki_filp; |
| 1436 | ssize_t ret = 0; |
| 1437 | |
| 1438 | switch (kiocb->ki_opcode) { |
| 1439 | case IOCB_CMD_PREAD: |
| 1440 | ret = -EBADF; |
| 1441 | if (unlikely(!(file->f_mode & FMODE_READ))) |
| 1442 | break; |
| 1443 | ret = -EFAULT; |
| 1444 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, |
| 1445 | kiocb->ki_left))) |
| 1446 | break; |
Kostik Belousov | 8766ce4 | 2005-10-23 12:57:13 -0700 | [diff] [blame] | 1447 | ret = security_file_permission(file, MAY_READ); |
| 1448 | if (unlikely(ret)) |
| 1449 | break; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1450 | ret = aio_setup_single_vector(kiocb); |
| 1451 | if (ret) |
| 1452 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | ret = -EINVAL; |
| 1454 | if (file->f_op->aio_read) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1455 | kiocb->ki_retry = aio_rw_vect_retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1456 | break; |
| 1457 | case IOCB_CMD_PWRITE: |
| 1458 | ret = -EBADF; |
| 1459 | if (unlikely(!(file->f_mode & FMODE_WRITE))) |
| 1460 | break; |
| 1461 | ret = -EFAULT; |
| 1462 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, |
| 1463 | kiocb->ki_left))) |
| 1464 | break; |
Kostik Belousov | 8766ce4 | 2005-10-23 12:57:13 -0700 | [diff] [blame] | 1465 | ret = security_file_permission(file, MAY_WRITE); |
| 1466 | if (unlikely(ret)) |
| 1467 | break; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1468 | ret = aio_setup_single_vector(kiocb); |
| 1469 | if (ret) |
| 1470 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | ret = -EINVAL; |
| 1472 | if (file->f_op->aio_write) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1473 | kiocb->ki_retry = aio_rw_vect_retry; |
| 1474 | break; |
| 1475 | case IOCB_CMD_PREADV: |
| 1476 | ret = -EBADF; |
| 1477 | if (unlikely(!(file->f_mode & FMODE_READ))) |
| 1478 | break; |
| 1479 | ret = security_file_permission(file, MAY_READ); |
| 1480 | if (unlikely(ret)) |
| 1481 | break; |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1482 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1483 | if (ret) |
| 1484 | break; |
| 1485 | ret = -EINVAL; |
| 1486 | if (file->f_op->aio_read) |
| 1487 | kiocb->ki_retry = aio_rw_vect_retry; |
| 1488 | break; |
| 1489 | case IOCB_CMD_PWRITEV: |
| 1490 | ret = -EBADF; |
| 1491 | if (unlikely(!(file->f_mode & FMODE_WRITE))) |
| 1492 | break; |
| 1493 | ret = security_file_permission(file, MAY_WRITE); |
| 1494 | if (unlikely(ret)) |
| 1495 | break; |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1496 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1497 | if (ret) |
| 1498 | break; |
| 1499 | ret = -EINVAL; |
| 1500 | if (file->f_op->aio_write) |
| 1501 | kiocb->ki_retry = aio_rw_vect_retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | break; |
| 1503 | case IOCB_CMD_FDSYNC: |
| 1504 | ret = -EINVAL; |
| 1505 | if (file->f_op->aio_fsync) |
| 1506 | kiocb->ki_retry = aio_fdsync; |
| 1507 | break; |
| 1508 | case IOCB_CMD_FSYNC: |
| 1509 | ret = -EINVAL; |
| 1510 | if (file->f_op->aio_fsync) |
| 1511 | kiocb->ki_retry = aio_fsync; |
| 1512 | break; |
| 1513 | default: |
| 1514 | dprintk("EINVAL: io_submit: no operation provided\n"); |
| 1515 | ret = -EINVAL; |
| 1516 | } |
| 1517 | |
| 1518 | if (!kiocb->ki_retry) |
| 1519 | return ret; |
| 1520 | |
| 1521 | return 0; |
| 1522 | } |
| 1523 | |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 1524 | static void aio_batch_add(struct address_space *mapping, |
| 1525 | struct hlist_head *batch_hash) |
| 1526 | { |
| 1527 | struct aio_batch_entry *abe; |
| 1528 | struct hlist_node *pos; |
| 1529 | unsigned bucket; |
| 1530 | |
| 1531 | bucket = hash_ptr(mapping, AIO_BATCH_HASH_BITS); |
| 1532 | hlist_for_each_entry(abe, pos, &batch_hash[bucket], list) { |
| 1533 | if (abe->mapping == mapping) |
| 1534 | return; |
| 1535 | } |
| 1536 | |
| 1537 | abe = mempool_alloc(abe_pool, GFP_KERNEL); |
| 1538 | BUG_ON(!igrab(mapping->host)); |
| 1539 | abe->mapping = mapping; |
| 1540 | hlist_add_head(&abe->list, &batch_hash[bucket]); |
| 1541 | return; |
| 1542 | } |
| 1543 | |
| 1544 | static void aio_batch_free(struct hlist_head *batch_hash) |
| 1545 | { |
| 1546 | struct aio_batch_entry *abe; |
| 1547 | struct hlist_node *pos, *n; |
| 1548 | int i; |
| 1549 | |
| 1550 | for (i = 0; i < AIO_BATCH_HASH_SIZE; i++) { |
| 1551 | hlist_for_each_entry_safe(abe, pos, n, &batch_hash[i], list) { |
| 1552 | blk_run_address_space(abe->mapping); |
| 1553 | iput(abe->mapping->host); |
| 1554 | hlist_del(&abe->list); |
| 1555 | mempool_free(abe, abe_pool); |
| 1556 | } |
| 1557 | } |
| 1558 | } |
| 1559 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 1560 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1561 | struct iocb *iocb, struct hlist_head *batch_hash, |
| 1562 | bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1563 | { |
| 1564 | struct kiocb *req; |
| 1565 | struct file *file; |
| 1566 | ssize_t ret; |
| 1567 | |
| 1568 | /* enforce forwards compatibility on users */ |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1569 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | pr_debug("EINVAL: io_submit: reserve field set\n"); |
| 1571 | return -EINVAL; |
| 1572 | } |
| 1573 | |
| 1574 | /* prevent overflows */ |
| 1575 | if (unlikely( |
| 1576 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || |
| 1577 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || |
| 1578 | ((ssize_t)iocb->aio_nbytes < 0) |
| 1579 | )) { |
| 1580 | pr_debug("EINVAL: io_submit: overflow check\n"); |
| 1581 | return -EINVAL; |
| 1582 | } |
| 1583 | |
| 1584 | file = fget(iocb->aio_fildes); |
| 1585 | if (unlikely(!file)) |
| 1586 | return -EBADF; |
| 1587 | |
| 1588 | req = aio_get_req(ctx); /* returns with 2 references to req */ |
| 1589 | if (unlikely(!req)) { |
| 1590 | fput(file); |
| 1591 | return -EAGAIN; |
| 1592 | } |
Yan Zheng | 87e2831 | 2007-10-08 12:16:20 -0700 | [diff] [blame] | 1593 | req->ki_filp = file; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1594 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
| 1595 | /* |
| 1596 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an |
| 1597 | * instance of the file* now. The file descriptor must be |
| 1598 | * an eventfd() fd, and will be signaled for each completed |
| 1599 | * event using the eventfd_signal() function. |
| 1600 | */ |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 1601 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
Hirofumi Nakagawa | 801678c | 2008-04-29 01:03:09 -0700 | [diff] [blame] | 1602 | if (IS_ERR(req->ki_eventfd)) { |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1603 | ret = PTR_ERR(req->ki_eventfd); |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 1604 | req->ki_eventfd = NULL; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1605 | goto out_put_req; |
| 1606 | } |
| 1607 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | |
Ken Chen | 212079c | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 1609 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | if (unlikely(ret)) { |
| 1611 | dprintk("EFAULT: aio_key\n"); |
| 1612 | goto out_put_req; |
| 1613 | } |
| 1614 | |
| 1615 | req->ki_obj.user = user_iocb; |
| 1616 | req->ki_user_data = iocb->aio_data; |
| 1617 | req->ki_pos = iocb->aio_offset; |
| 1618 | |
| 1619 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; |
| 1620 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; |
| 1621 | req->ki_opcode = iocb->aio_lio_opcode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1622 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1623 | ret = aio_setup_iocb(req, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1624 | |
| 1625 | if (ret) |
| 1626 | goto out_put_req; |
| 1627 | |
| 1628 | spin_lock_irq(&ctx->ctx_lock); |
Benjamin LaHaise | ac0b1bc | 2005-09-09 13:02:09 -0700 | [diff] [blame] | 1629 | aio_run_iocb(req); |
Benjamin LaHaise | ac0b1bc | 2005-09-09 13:02:09 -0700 | [diff] [blame] | 1630 | if (!list_empty(&ctx->run_list)) { |
Ken Chen | 954d3e9 | 2005-05-01 08:59:16 -0700 | [diff] [blame] | 1631 | /* drain the run list */ |
| 1632 | while (__aio_run_iocbs(ctx)) |
| 1633 | ; |
| 1634 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1635 | spin_unlock_irq(&ctx->ctx_lock); |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 1636 | if (req->ki_opcode == IOCB_CMD_PREAD || |
| 1637 | req->ki_opcode == IOCB_CMD_PREADV || |
| 1638 | req->ki_opcode == IOCB_CMD_PWRITE || |
| 1639 | req->ki_opcode == IOCB_CMD_PWRITEV) |
| 1640 | aio_batch_add(file->f_mapping, batch_hash); |
| 1641 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | aio_put_req(req); /* drop extra ref to req */ |
| 1643 | return 0; |
| 1644 | |
| 1645 | out_put_req: |
| 1646 | aio_put_req(req); /* drop extra ref to req */ |
| 1647 | aio_put_req(req); /* drop i/o ref to req */ |
| 1648 | return ret; |
| 1649 | } |
| 1650 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1651 | long do_io_submit(aio_context_t ctx_id, long nr, |
| 1652 | struct iocb __user *__user *iocbpp, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | { |
| 1654 | struct kioctx *ctx; |
| 1655 | long ret = 0; |
| 1656 | int i; |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 1657 | struct hlist_head batch_hash[AIO_BATCH_HASH_SIZE] = { { 0, }, }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1658 | |
| 1659 | if (unlikely(nr < 0)) |
| 1660 | return -EINVAL; |
| 1661 | |
| 1662 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
| 1663 | return -EFAULT; |
| 1664 | |
| 1665 | ctx = lookup_ioctx(ctx_id); |
| 1666 | if (unlikely(!ctx)) { |
| 1667 | pr_debug("EINVAL: io_submit: invalid context id\n"); |
| 1668 | return -EINVAL; |
| 1669 | } |
| 1670 | |
| 1671 | /* |
| 1672 | * AKPM: should this return a partial result if some of the IOs were |
| 1673 | * successfully submitted? |
| 1674 | */ |
| 1675 | for (i=0; i<nr; i++) { |
| 1676 | struct iocb __user *user_iocb; |
| 1677 | struct iocb tmp; |
| 1678 | |
| 1679 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { |
| 1680 | ret = -EFAULT; |
| 1681 | break; |
| 1682 | } |
| 1683 | |
| 1684 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { |
| 1685 | ret = -EFAULT; |
| 1686 | break; |
| 1687 | } |
| 1688 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1689 | ret = io_submit_one(ctx, user_iocb, &tmp, batch_hash, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1690 | if (ret) |
| 1691 | break; |
| 1692 | } |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 1693 | aio_batch_free(batch_hash); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1694 | |
| 1695 | put_ioctx(ctx); |
| 1696 | return i ? i : ret; |
| 1697 | } |
| 1698 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame^] | 1699 | /* sys_io_submit: |
| 1700 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns |
| 1701 | * the number of iocbs queued. May return -EINVAL if the aio_context |
| 1702 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at |
| 1703 | * *iocbpp[0] is not properly initialized, if the operation specified |
| 1704 | * is invalid for the file descriptor in the iocb. May fail with |
| 1705 | * -EFAULT if any of the data structures point to invalid data. May |
| 1706 | * fail with -EBADF if the file descriptor specified in the first |
| 1707 | * iocb is invalid. May fail with -EAGAIN if insufficient resources |
| 1708 | * are available to queue any iocbs. Will return 0 if nr is 0. Will |
| 1709 | * fail with -ENOSYS if not implemented. |
| 1710 | */ |
| 1711 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, |
| 1712 | struct iocb __user * __user *, iocbpp) |
| 1713 | { |
| 1714 | return do_io_submit(ctx_id, nr, iocbpp, 0); |
| 1715 | } |
| 1716 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1717 | /* lookup_kiocb |
| 1718 | * Finds a given iocb for cancellation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1719 | */ |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 1720 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
| 1721 | u32 key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | { |
| 1723 | struct list_head *pos; |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 1724 | |
| 1725 | assert_spin_locked(&ctx->ctx_lock); |
| 1726 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1727 | /* TODO: use a hash or array, this sucks. */ |
| 1728 | list_for_each(pos, &ctx->active_reqs) { |
| 1729 | struct kiocb *kiocb = list_kiocb(pos); |
| 1730 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) |
| 1731 | return kiocb; |
| 1732 | } |
| 1733 | return NULL; |
| 1734 | } |
| 1735 | |
| 1736 | /* sys_io_cancel: |
| 1737 | * Attempts to cancel an iocb previously passed to io_submit. If |
| 1738 | * the operation is successfully cancelled, the resulting event is |
| 1739 | * copied into the memory pointed to by result without being placed |
| 1740 | * into the completion queue and 0 is returned. May fail with |
| 1741 | * -EFAULT if any of the data structures pointed to are invalid. |
| 1742 | * May fail with -EINVAL if aio_context specified by ctx_id is |
| 1743 | * invalid. May fail with -EAGAIN if the iocb specified was not |
| 1744 | * cancelled. Will fail with -ENOSYS if not implemented. |
| 1745 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1746 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
| 1747 | struct io_event __user *, result) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | { |
| 1749 | int (*cancel)(struct kiocb *iocb, struct io_event *res); |
| 1750 | struct kioctx *ctx; |
| 1751 | struct kiocb *kiocb; |
| 1752 | u32 key; |
| 1753 | int ret; |
| 1754 | |
| 1755 | ret = get_user(key, &iocb->aio_key); |
| 1756 | if (unlikely(ret)) |
| 1757 | return -EFAULT; |
| 1758 | |
| 1759 | ctx = lookup_ioctx(ctx_id); |
| 1760 | if (unlikely(!ctx)) |
| 1761 | return -EINVAL; |
| 1762 | |
| 1763 | spin_lock_irq(&ctx->ctx_lock); |
| 1764 | ret = -EAGAIN; |
| 1765 | kiocb = lookup_kiocb(ctx, iocb, key); |
| 1766 | if (kiocb && kiocb->ki_cancel) { |
| 1767 | cancel = kiocb->ki_cancel; |
| 1768 | kiocb->ki_users ++; |
| 1769 | kiocbSetCancelled(kiocb); |
| 1770 | } else |
| 1771 | cancel = NULL; |
| 1772 | spin_unlock_irq(&ctx->ctx_lock); |
| 1773 | |
| 1774 | if (NULL != cancel) { |
| 1775 | struct io_event tmp; |
| 1776 | pr_debug("calling cancel\n"); |
| 1777 | memset(&tmp, 0, sizeof(tmp)); |
| 1778 | tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user; |
| 1779 | tmp.data = kiocb->ki_user_data; |
| 1780 | ret = cancel(kiocb, &tmp); |
| 1781 | if (!ret) { |
| 1782 | /* Cancellation succeeded -- copy the result |
| 1783 | * into the user's buffer. |
| 1784 | */ |
| 1785 | if (copy_to_user(result, &tmp, sizeof(tmp))) |
| 1786 | ret = -EFAULT; |
| 1787 | } |
| 1788 | } else |
Wendy Cheng | 8f58202 | 2005-09-09 13:02:08 -0700 | [diff] [blame] | 1789 | ret = -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | |
| 1791 | put_ioctx(ctx); |
| 1792 | |
| 1793 | return ret; |
| 1794 | } |
| 1795 | |
| 1796 | /* io_getevents: |
| 1797 | * Attempts to read at least min_nr events and up to nr events from |
| 1798 | * the completion queue for the aio_context specified by ctx_id. May |
| 1799 | * fail with -EINVAL if ctx_id is invalid, if min_nr is out of range, |
| 1800 | * if nr is out of range, if when is out of range. May fail with |
| 1801 | * -EFAULT if any of the memory specified to is invalid. May return |
| 1802 | * 0 or < min_nr if no events are available and the timeout specified |
| 1803 | * by when has elapsed, where when == NULL specifies an infinite |
| 1804 | * timeout. Note that the timeout pointed to by when is relative and |
| 1805 | * will be updated if not NULL and the operation blocks. Will fail |
| 1806 | * with -ENOSYS if not implemented. |
| 1807 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1808 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
| 1809 | long, min_nr, |
| 1810 | long, nr, |
| 1811 | struct io_event __user *, events, |
| 1812 | struct timespec __user *, timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | { |
| 1814 | struct kioctx *ioctx = lookup_ioctx(ctx_id); |
| 1815 | long ret = -EINVAL; |
| 1816 | |
| 1817 | if (likely(ioctx)) { |
| 1818 | if (likely(min_nr <= nr && min_nr >= 0 && nr >= 0)) |
| 1819 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
| 1820 | put_ioctx(ioctx); |
| 1821 | } |
| 1822 | |
Roland McGrath | 598af05 | 2008-04-10 15:38:45 -0700 | [diff] [blame] | 1823 | asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1824 | return ret; |
| 1825 | } |