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> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 16 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 37 | #include <linux/compat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | |
| 39 | #include <asm/kmap_types.h> |
| 40 | #include <asm/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | |
| 42 | #if DEBUG > 1 |
| 43 | #define dprintk printk |
| 44 | #else |
| 45 | #define dprintk(x...) do { ; } while (0) |
| 46 | #endif |
| 47 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame^] | 48 | #define AIO_RING_MAGIC 0xa10a10a1 |
| 49 | #define AIO_RING_COMPAT_FEATURES 1 |
| 50 | #define AIO_RING_INCOMPAT_FEATURES 0 |
| 51 | struct aio_ring { |
| 52 | unsigned id; /* kernel internal index number */ |
| 53 | unsigned nr; /* number of io_events */ |
| 54 | unsigned head; |
| 55 | unsigned tail; |
| 56 | |
| 57 | unsigned magic; |
| 58 | unsigned compat_features; |
| 59 | unsigned incompat_features; |
| 60 | unsigned header_length; /* size of aio_ring */ |
| 61 | |
| 62 | |
| 63 | struct io_event io_events[0]; |
| 64 | }; /* 128 bytes + ring size */ |
| 65 | |
| 66 | #define AIO_RING_PAGES 8 |
| 67 | struct aio_ring_info { |
| 68 | unsigned long mmap_base; |
| 69 | unsigned long mmap_size; |
| 70 | |
| 71 | struct page **ring_pages; |
| 72 | spinlock_t ring_lock; |
| 73 | long nr_pages; |
| 74 | |
| 75 | unsigned nr, tail; |
| 76 | |
| 77 | struct page *internal_pages[AIO_RING_PAGES]; |
| 78 | }; |
| 79 | |
| 80 | static inline unsigned aio_ring_avail(struct aio_ring_info *info, |
| 81 | struct aio_ring *ring) |
| 82 | { |
| 83 | return (ring->head + info->nr - 1 - ring->tail) % info->nr; |
| 84 | } |
| 85 | |
| 86 | struct kioctx { |
| 87 | atomic_t users; |
| 88 | int dead; |
| 89 | |
| 90 | /* This needs improving */ |
| 91 | unsigned long user_id; |
| 92 | struct hlist_node list; |
| 93 | |
| 94 | wait_queue_head_t wait; |
| 95 | |
| 96 | spinlock_t ctx_lock; |
| 97 | |
| 98 | int reqs_active; |
| 99 | struct list_head active_reqs; /* used for cancellation */ |
| 100 | |
| 101 | /* sys_io_setup currently limits this to an unsigned int */ |
| 102 | unsigned max_reqs; |
| 103 | |
| 104 | struct aio_ring_info ring_info; |
| 105 | |
| 106 | struct rcu_head rcu_head; |
| 107 | }; |
| 108 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | /*------ sysctl variables----*/ |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 110 | static DEFINE_SPINLOCK(aio_nr_lock); |
| 111 | unsigned long aio_nr; /* current system wide number of aio requests */ |
| 112 | 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] | 113 | /*----end sysctl variables---*/ |
| 114 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 115 | static struct kmem_cache *kiocb_cachep; |
| 116 | static struct kmem_cache *kioctx_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | /* aio_setup |
| 119 | * Creates the slab caches used by the aio routines, panic on |
| 120 | * failure as this is done early during the boot sequence. |
| 121 | */ |
| 122 | static int __init aio_setup(void) |
| 123 | { |
Christoph Lameter | 0a31bd5 | 2007-05-06 14:49:57 -0700 | [diff] [blame] | 124 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
| 125 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); |
| 128 | |
| 129 | return 0; |
| 130 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 131 | __initcall(aio_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | |
| 133 | static void aio_free_ring(struct kioctx *ctx) |
| 134 | { |
| 135 | struct aio_ring_info *info = &ctx->ring_info; |
| 136 | long i; |
| 137 | |
| 138 | for (i=0; i<info->nr_pages; i++) |
| 139 | put_page(info->ring_pages[i]); |
| 140 | |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 141 | if (info->mmap_size) { |
Al Viro | bfce281 | 2012-04-20 21:57:04 -0400 | [diff] [blame] | 142 | vm_munmap(info->mmap_base, info->mmap_size); |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 143 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | |
| 145 | if (info->ring_pages && info->ring_pages != info->internal_pages) |
| 146 | kfree(info->ring_pages); |
| 147 | info->ring_pages = NULL; |
| 148 | info->nr = 0; |
| 149 | } |
| 150 | |
| 151 | static int aio_setup_ring(struct kioctx *ctx) |
| 152 | { |
| 153 | struct aio_ring *ring; |
| 154 | struct aio_ring_info *info = &ctx->ring_info; |
| 155 | unsigned nr_events = ctx->max_reqs; |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 156 | struct mm_struct *mm = current->mm; |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 157 | unsigned long size, populate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | int nr_pages; |
| 159 | |
| 160 | /* Compensate for the ring buffer's head/tail overlap entry */ |
| 161 | nr_events += 2; /* 1 is required, 2 for good luck */ |
| 162 | |
| 163 | size = sizeof(struct aio_ring); |
| 164 | size += sizeof(struct io_event) * nr_events; |
| 165 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; |
| 166 | |
| 167 | if (nr_pages < 0) |
| 168 | return -EINVAL; |
| 169 | |
| 170 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); |
| 171 | |
| 172 | info->nr = 0; |
| 173 | info->ring_pages = info->internal_pages; |
| 174 | if (nr_pages > AIO_RING_PAGES) { |
Oliver Neukum | 11b0b5a | 2006-03-25 03:08:13 -0800 | [diff] [blame] | 175 | info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | if (!info->ring_pages) |
| 177 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | } |
| 179 | |
| 180 | info->mmap_size = nr_pages * PAGE_SIZE; |
| 181 | dprintk("attempting mmap of %lu bytes\n", info->mmap_size); |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 182 | down_write(&mm->mmap_sem); |
Al Viro | e3fc629 | 2012-05-30 20:08:42 -0400 | [diff] [blame] | 183 | info->mmap_base = do_mmap_pgoff(NULL, 0, info->mmap_size, |
| 184 | PROT_READ|PROT_WRITE, |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 185 | MAP_ANONYMOUS|MAP_PRIVATE, 0, |
| 186 | &populate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | if (IS_ERR((void *)info->mmap_base)) { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 188 | up_write(&mm->mmap_sem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | info->mmap_size = 0; |
| 190 | aio_free_ring(ctx); |
| 191 | return -EAGAIN; |
| 192 | } |
| 193 | |
| 194 | dprintk("mmap address: 0x%08lx\n", info->mmap_base); |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 195 | info->nr_pages = get_user_pages(current, mm, info->mmap_base, nr_pages, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | 1, 0, info->ring_pages, NULL); |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 197 | up_write(&mm->mmap_sem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | |
| 199 | if (unlikely(info->nr_pages != nr_pages)) { |
| 200 | aio_free_ring(ctx); |
| 201 | return -EAGAIN; |
| 202 | } |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 203 | if (populate) |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 204 | mm_populate(info->mmap_base, populate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | |
| 206 | ctx->user_id = info->mmap_base; |
| 207 | |
| 208 | info->nr = nr_events; /* trusted copy */ |
| 209 | |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 210 | ring = kmap_atomic(info->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | ring->nr = nr_events; /* user copy */ |
| 212 | ring->id = ctx->user_id; |
| 213 | ring->head = ring->tail = 0; |
| 214 | ring->magic = AIO_RING_MAGIC; |
| 215 | ring->compat_features = AIO_RING_COMPAT_FEATURES; |
| 216 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; |
| 217 | ring->header_length = sizeof(struct aio_ring); |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 218 | kunmap_atomic(ring); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | |
| 224 | /* aio_ring_event: returns a pointer to the event at the given index from |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 225 | * kmap_atomic(). Release the pointer with put_aio_ring_event(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | */ |
| 227 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
| 228 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) |
| 229 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) |
| 230 | |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 231 | #define aio_ring_event(info, nr) ({ \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ |
| 233 | struct io_event *__event; \ |
| 234 | __event = kmap_atomic( \ |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 235 | (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE]); \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | __event += pos % AIO_EVENTS_PER_PAGE; \ |
| 237 | __event; \ |
| 238 | }) |
| 239 | |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 240 | #define put_aio_ring_event(event) do { \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | struct io_event *__event = (event); \ |
| 242 | (void)__event; \ |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 243 | kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK)); \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | } while(0) |
| 245 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 246 | static void ctx_rcu_free(struct rcu_head *head) |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 247 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 248 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 249 | kmem_cache_free(kioctx_cachep, ctx); |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 250 | } |
| 251 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 252 | /* __put_ioctx |
| 253 | * Called when the last user of an aio context has gone away, |
| 254 | * and the struct needs to be freed. |
| 255 | */ |
| 256 | static void __put_ioctx(struct kioctx *ctx) |
| 257 | { |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 258 | unsigned nr_events = ctx->max_reqs; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 259 | BUG_ON(ctx->reqs_active); |
| 260 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 261 | aio_free_ring(ctx); |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 262 | if (nr_events) { |
| 263 | spin_lock(&aio_nr_lock); |
| 264 | BUG_ON(aio_nr - nr_events > aio_nr); |
| 265 | aio_nr -= nr_events; |
| 266 | spin_unlock(&aio_nr_lock); |
| 267 | } |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 268 | pr_debug("__put_ioctx: freeing %p\n", ctx); |
| 269 | call_rcu(&ctx->rcu_head, ctx_rcu_free); |
| 270 | } |
| 271 | |
Nick Piggin | 3bd9a5d | 2011-02-25 14:44:26 -0800 | [diff] [blame] | 272 | static inline int try_get_ioctx(struct kioctx *kioctx) |
| 273 | { |
| 274 | return atomic_inc_not_zero(&kioctx->users); |
| 275 | } |
| 276 | |
| 277 | static inline void put_ioctx(struct kioctx *kioctx) |
| 278 | { |
| 279 | BUG_ON(atomic_read(&kioctx->users) <= 0); |
| 280 | if (unlikely(atomic_dec_and_test(&kioctx->users))) |
| 281 | __put_ioctx(kioctx); |
| 282 | } |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 283 | |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 284 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, |
| 285 | struct io_event *res) |
| 286 | { |
| 287 | int (*cancel)(struct kiocb *, struct io_event *); |
| 288 | int ret = -EINVAL; |
| 289 | |
| 290 | cancel = kiocb->ki_cancel; |
| 291 | kiocbSetCancelled(kiocb); |
| 292 | if (cancel) { |
| 293 | kiocb->ki_users++; |
| 294 | spin_unlock_irq(&ctx->ctx_lock); |
| 295 | |
| 296 | memset(res, 0, sizeof(*res)); |
| 297 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; |
| 298 | res->data = kiocb->ki_user_data; |
| 299 | ret = cancel(kiocb, res); |
| 300 | |
| 301 | spin_lock_irq(&ctx->ctx_lock); |
| 302 | } |
| 303 | |
| 304 | return ret; |
| 305 | } |
| 306 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | /* ioctx_alloc |
| 308 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. |
| 309 | */ |
| 310 | static struct kioctx *ioctx_alloc(unsigned nr_events) |
| 311 | { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 312 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 313 | struct kioctx *ctx; |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 314 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | |
| 316 | /* Prevent overflows */ |
| 317 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || |
| 318 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { |
| 319 | pr_debug("ENOMEM: nr_events too high\n"); |
| 320 | return ERR_PTR(-EINVAL); |
| 321 | } |
| 322 | |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 323 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | return ERR_PTR(-EAGAIN); |
| 325 | |
Robert P. J. Day | c376222 | 2007-02-10 01:45:03 -0800 | [diff] [blame] | 326 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | if (!ctx) |
| 328 | return ERR_PTR(-ENOMEM); |
| 329 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | ctx->max_reqs = nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | |
Al Viro | 86b62a2 | 2012-03-07 05:16:35 +0000 | [diff] [blame] | 332 | atomic_set(&ctx->users, 2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | spin_lock_init(&ctx->ctx_lock); |
| 334 | spin_lock_init(&ctx->ring_info.ring_lock); |
| 335 | init_waitqueue_head(&ctx->wait); |
| 336 | |
| 337 | INIT_LIST_HEAD(&ctx->active_reqs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 338 | |
| 339 | if (aio_setup_ring(ctx) < 0) |
| 340 | goto out_freectx; |
| 341 | |
| 342 | /* limit the number of system wide aios */ |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 343 | spin_lock(&aio_nr_lock); |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 344 | if (aio_nr + nr_events > aio_max_nr || |
| 345 | aio_nr + nr_events < aio_nr) { |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 346 | spin_unlock(&aio_nr_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 347 | goto out_cleanup; |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 348 | } |
| 349 | aio_nr += ctx->max_reqs; |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 350 | spin_unlock(&aio_nr_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | |
Jeff Moyer | 39fa003 | 2008-04-29 01:03:48 -0700 | [diff] [blame] | 352 | /* now link into global list. */ |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 353 | spin_lock(&mm->ioctx_lock); |
| 354 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); |
| 355 | spin_unlock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | |
| 357 | dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 358 | ctx, ctx->user_id, mm, ctx->ring_info.nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | return ctx; |
| 360 | |
| 361 | out_cleanup: |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 362 | err = -EAGAIN; |
| 363 | aio_free_ring(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | out_freectx: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | kmem_cache_free(kioctx_cachep, ctx); |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 366 | dprintk("aio: error allocating ioctx %d\n", err); |
| 367 | return ERR_PTR(err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | } |
| 369 | |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 370 | /* kill_ctx |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | * Cancels all outstanding aio requests on an aio context. Used |
| 372 | * when the processes owning a context have all exited to encourage |
| 373 | * the rapid destruction of the kioctx. |
| 374 | */ |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 375 | static void kill_ctx(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | { |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 377 | struct task_struct *tsk = current; |
| 378 | DECLARE_WAITQUEUE(wait, tsk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | struct io_event res; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 380 | struct kiocb *req; |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 381 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | spin_lock_irq(&ctx->ctx_lock); |
| 383 | ctx->dead = 1; |
| 384 | while (!list_empty(&ctx->active_reqs)) { |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 385 | req = list_first_entry(&ctx->active_reqs, |
| 386 | struct kiocb, ki_list); |
| 387 | |
| 388 | list_del_init(&req->ki_list); |
| 389 | kiocb_cancel(ctx, req, &res); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | if (!ctx->reqs_active) |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 393 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | |
| 395 | add_wait_queue(&ctx->wait, &wait); |
| 396 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
| 397 | while (ctx->reqs_active) { |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 398 | spin_unlock_irq(&ctx->ctx_lock); |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 399 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 401 | spin_lock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | } |
| 403 | __set_task_state(tsk, TASK_RUNNING); |
| 404 | remove_wait_queue(&ctx->wait, &wait); |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 405 | |
| 406 | out: |
| 407 | spin_unlock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | } |
| 409 | |
| 410 | /* wait_on_sync_kiocb: |
| 411 | * Waits on the given sync kiocb to complete. |
| 412 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 413 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | { |
| 415 | while (iocb->ki_users) { |
| 416 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 417 | if (!iocb->ki_users) |
| 418 | break; |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 419 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | } |
| 421 | __set_current_state(TASK_RUNNING); |
| 422 | return iocb->ki_user_data; |
| 423 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 424 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | |
| 426 | /* exit_aio: called when the last user of mm goes away. At this point, |
| 427 | * there is no way for any new requests to be submited or any of the |
| 428 | * io_* syscalls to be called on the context. However, there may be |
| 429 | * outstanding requests which hold references to the context; as they |
| 430 | * go away, they will call put_ioctx and release any pinned memory |
| 431 | * associated with the request (held via struct page * references). |
| 432 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 433 | void exit_aio(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 435 | struct kioctx *ctx; |
| 436 | |
| 437 | while (!hlist_empty(&mm->ioctx_list)) { |
| 438 | ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); |
| 439 | hlist_del_rcu(&ctx->list); |
| 440 | |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 441 | kill_ctx(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | |
| 443 | if (1 != atomic_read(&ctx->users)) |
| 444 | printk(KERN_DEBUG |
| 445 | "exit_aio:ioctx still alive: %d %d %d\n", |
| 446 | atomic_read(&ctx->users), ctx->dead, |
| 447 | ctx->reqs_active); |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 448 | /* |
| 449 | * We don't need to bother with munmap() here - |
| 450 | * exit_mmap(mm) is coming and it'll unmap everything. |
| 451 | * Since aio_free_ring() uses non-zero ->mmap_size |
| 452 | * as indicator that it needs to unmap the area, |
| 453 | * just set it to 0; aio_free_ring() is the only |
| 454 | * place that uses ->mmap_size, so it's safe. |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 455 | */ |
| 456 | ctx->ring_info.mmap_size = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | put_ioctx(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | } |
| 459 | } |
| 460 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | /* aio_get_req |
| 462 | * Allocate a slot for an aio request. Increments the users count |
| 463 | * of the kioctx so that the kioctx stays around until all requests are |
| 464 | * complete. Returns NULL if no requests are free. |
| 465 | * |
| 466 | * Returns with kiocb->users set to 2. The io submit code path holds |
| 467 | * an extra reference while submitting the i/o. |
| 468 | * This prevents races between the aio code path referencing the |
| 469 | * req (after submitting it) and aio_complete() freeing the req. |
| 470 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 471 | static struct kiocb *__aio_get_req(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | { |
| 473 | struct kiocb *req = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | |
| 475 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); |
| 476 | if (unlikely(!req)) |
| 477 | return NULL; |
| 478 | |
Zach Brown | 4faa528 | 2005-10-17 16:43:33 -0700 | [diff] [blame] | 479 | req->ki_flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 480 | req->ki_users = 2; |
| 481 | req->ki_key = 0; |
| 482 | req->ki_ctx = ctx; |
| 483 | req->ki_cancel = NULL; |
| 484 | req->ki_retry = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | req->ki_dtor = NULL; |
| 486 | req->private = NULL; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 487 | req->ki_iovec = NULL; |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 488 | req->ki_eventfd = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | return req; |
| 491 | } |
| 492 | |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 493 | /* |
| 494 | * struct kiocb's are allocated in batches to reduce the number of |
| 495 | * times the ctx lock is acquired and released. |
| 496 | */ |
| 497 | #define KIOCB_BATCH_SIZE 32L |
| 498 | struct kiocb_batch { |
| 499 | struct list_head head; |
| 500 | long count; /* number of requests left to allocate */ |
| 501 | }; |
| 502 | |
| 503 | static void kiocb_batch_init(struct kiocb_batch *batch, long total) |
| 504 | { |
| 505 | INIT_LIST_HEAD(&batch->head); |
| 506 | batch->count = total; |
| 507 | } |
| 508 | |
Gleb Natapov | 69e4747 | 2012-01-08 17:07:28 +0200 | [diff] [blame] | 509 | static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch) |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 510 | { |
| 511 | struct kiocb *req, *n; |
| 512 | |
Gleb Natapov | 69e4747 | 2012-01-08 17:07:28 +0200 | [diff] [blame] | 513 | if (list_empty(&batch->head)) |
| 514 | return; |
| 515 | |
| 516 | spin_lock_irq(&ctx->ctx_lock); |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 517 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
| 518 | list_del(&req->ki_batch); |
Gleb Natapov | 69e4747 | 2012-01-08 17:07:28 +0200 | [diff] [blame] | 519 | list_del(&req->ki_list); |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 520 | kmem_cache_free(kiocb_cachep, req); |
Gleb Natapov | 69e4747 | 2012-01-08 17:07:28 +0200 | [diff] [blame] | 521 | ctx->reqs_active--; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 522 | } |
Jeff Moyer | 880641b | 2012-03-05 14:59:12 -0800 | [diff] [blame] | 523 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
| 524 | wake_up_all(&ctx->wait); |
Gleb Natapov | 69e4747 | 2012-01-08 17:07:28 +0200 | [diff] [blame] | 525 | spin_unlock_irq(&ctx->ctx_lock); |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 526 | } |
| 527 | |
| 528 | /* |
| 529 | * Allocate a batch of kiocbs. This avoids taking and dropping the |
| 530 | * context lock a lot during setup. |
| 531 | */ |
| 532 | static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch) |
| 533 | { |
| 534 | unsigned short allocated, to_alloc; |
| 535 | long avail; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 536 | struct kiocb *req, *n; |
| 537 | struct aio_ring *ring; |
| 538 | |
| 539 | to_alloc = min(batch->count, KIOCB_BATCH_SIZE); |
| 540 | for (allocated = 0; allocated < to_alloc; allocated++) { |
| 541 | req = __aio_get_req(ctx); |
| 542 | if (!req) |
| 543 | /* allocation failed, go with what we've got */ |
| 544 | break; |
| 545 | list_add(&req->ki_batch, &batch->head); |
| 546 | } |
| 547 | |
| 548 | if (allocated == 0) |
| 549 | goto out; |
| 550 | |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 551 | spin_lock_irq(&ctx->ctx_lock); |
| 552 | ring = kmap_atomic(ctx->ring_info.ring_pages[0]); |
| 553 | |
| 554 | avail = aio_ring_avail(&ctx->ring_info, ring) - ctx->reqs_active; |
| 555 | BUG_ON(avail < 0); |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 556 | if (avail < allocated) { |
| 557 | /* Trim back the number of requests. */ |
| 558 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
| 559 | list_del(&req->ki_batch); |
| 560 | kmem_cache_free(kiocb_cachep, req); |
| 561 | if (--allocated <= avail) |
| 562 | break; |
| 563 | } |
| 564 | } |
| 565 | |
| 566 | batch->count -= allocated; |
| 567 | list_for_each_entry(req, &batch->head, ki_batch) { |
| 568 | list_add(&req->ki_list, &ctx->active_reqs); |
| 569 | ctx->reqs_active++; |
| 570 | } |
| 571 | |
| 572 | kunmap_atomic(ring); |
| 573 | spin_unlock_irq(&ctx->ctx_lock); |
| 574 | |
| 575 | out: |
| 576 | return allocated; |
| 577 | } |
| 578 | |
| 579 | static inline struct kiocb *aio_get_req(struct kioctx *ctx, |
| 580 | struct kiocb_batch *batch) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 581 | { |
| 582 | struct kiocb *req; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 583 | |
| 584 | if (list_empty(&batch->head)) |
| 585 | if (kiocb_batch_refill(ctx, batch) == 0) |
| 586 | return NULL; |
| 587 | req = list_first_entry(&batch->head, struct kiocb, ki_batch); |
| 588 | list_del(&req->ki_batch); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | return req; |
| 590 | } |
| 591 | |
| 592 | static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) |
| 593 | { |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 594 | assert_spin_locked(&ctx->ctx_lock); |
| 595 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 596 | if (req->ki_eventfd != NULL) |
| 597 | eventfd_ctx_put(req->ki_eventfd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | if (req->ki_dtor) |
| 599 | req->ki_dtor(req); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 600 | if (req->ki_iovec != &req->ki_inline_vec) |
| 601 | kfree(req->ki_iovec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | kmem_cache_free(kiocb_cachep, req); |
| 603 | ctx->reqs_active--; |
| 604 | |
| 605 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
Roland Dreier | e91f90b | 2011-03-22 16:35:10 -0700 | [diff] [blame] | 606 | wake_up_all(&ctx->wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | } |
| 608 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 609 | /* __aio_put_req |
| 610 | * Returns true if this put was the last user of the request. |
| 611 | */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 612 | static void __aio_put_req(struct kioctx *ctx, struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | { |
Al Viro | 516e0cc | 2008-07-26 00:39:17 -0400 | [diff] [blame] | 614 | dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n", |
| 615 | req, atomic_long_read(&req->ki_filp->f_count)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 617 | assert_spin_locked(&ctx->ctx_lock); |
| 618 | |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 619 | req->ki_users--; |
Eric Sesterhenn | 93e06b4 | 2006-11-30 05:29:23 +0100 | [diff] [blame] | 620 | BUG_ON(req->ki_users < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 621 | if (likely(req->ki_users)) |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 622 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | list_del(&req->ki_list); /* remove from active_reqs */ |
| 624 | req->ki_cancel = NULL; |
| 625 | req->ki_retry = NULL; |
| 626 | |
Al Viro | 3ffa3c0 | 2012-06-24 10:00:10 +0400 | [diff] [blame] | 627 | fput(req->ki_filp); |
| 628 | req->ki_filp = NULL; |
| 629 | really_put_req(ctx, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | } |
| 631 | |
| 632 | /* aio_put_req |
| 633 | * Returns true if this put was the last user of the kiocb, |
| 634 | * false if the request is still in use. |
| 635 | */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 636 | void aio_put_req(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | { |
| 638 | struct kioctx *ctx = req->ki_ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | spin_lock_irq(&ctx->ctx_lock); |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 640 | __aio_put_req(ctx, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | spin_unlock_irq(&ctx->ctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 643 | EXPORT_SYMBOL(aio_put_req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 645 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | { |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 647 | struct mm_struct *mm = current->mm; |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 648 | struct kioctx *ctx, *ret = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 650 | rcu_read_lock(); |
| 651 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 652 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
Nick Piggin | 3bd9a5d | 2011-02-25 14:44:26 -0800 | [diff] [blame] | 653 | /* |
| 654 | * RCU protects us against accessing freed memory but |
| 655 | * we have to be careful not to get a reference when the |
| 656 | * reference count already dropped to 0 (ctx->dead test |
| 657 | * is unreliable because of races). |
| 658 | */ |
| 659 | if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){ |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 660 | ret = ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | break; |
| 662 | } |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 663 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 665 | rcu_read_unlock(); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 666 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 667 | } |
| 668 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | /* aio_complete |
| 670 | * Called when the io request on the given iocb is complete. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 672 | void aio_complete(struct kiocb *iocb, long res, long res2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | { |
| 674 | struct kioctx *ctx = iocb->ki_ctx; |
| 675 | struct aio_ring_info *info; |
| 676 | struct aio_ring *ring; |
| 677 | struct io_event *event; |
| 678 | unsigned long flags; |
| 679 | unsigned long tail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 681 | /* |
| 682 | * Special case handling for sync iocbs: |
| 683 | * - events go directly into the iocb for fast handling |
| 684 | * - the sync task with the iocb in its stack holds the single iocb |
| 685 | * ref, no other paths have a way to get another ref |
| 686 | * - the sync task helpfully left a reference to itself in the iocb |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | */ |
| 688 | if (is_sync_kiocb(iocb)) { |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 689 | BUG_ON(iocb->ki_users != 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | iocb->ki_user_data = res; |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 691 | iocb->ki_users = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | wake_up_process(iocb->ki_obj.tsk); |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 693 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | } |
| 695 | |
| 696 | info = &ctx->ring_info; |
| 697 | |
| 698 | /* add a completion event to the ring buffer. |
| 699 | * must be done holding ctx->ctx_lock to prevent |
| 700 | * other code from messing with the tail |
| 701 | * pointer since we might be called from irq |
| 702 | * context. |
| 703 | */ |
| 704 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 705 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | /* |
| 707 | * cancelled requests don't get events, userland was given one |
| 708 | * when the event got cancelled. |
| 709 | */ |
| 710 | if (kiocbIsCancelled(iocb)) |
| 711 | goto put_rq; |
| 712 | |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 713 | ring = kmap_atomic(info->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | |
| 715 | tail = info->tail; |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 716 | event = aio_ring_event(info, tail); |
Ken Chen | 4bf69b2 | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 717 | if (++tail >= info->nr) |
| 718 | tail = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 719 | |
| 720 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
| 721 | event->data = iocb->ki_user_data; |
| 722 | event->res = res; |
| 723 | event->res2 = res2; |
| 724 | |
| 725 | dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n", |
| 726 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
| 727 | res, res2); |
| 728 | |
| 729 | /* after flagging the request as done, we |
| 730 | * must never even look at it again |
| 731 | */ |
| 732 | smp_wmb(); /* make event visible before updating tail */ |
| 733 | |
| 734 | info->tail = tail; |
| 735 | ring->tail = tail; |
| 736 | |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 737 | put_aio_ring_event(event); |
| 738 | kunmap_atomic(ring); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | |
| 740 | pr_debug("added to ring %p at [%lu]\n", iocb, tail); |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 741 | |
| 742 | /* |
| 743 | * Check if the user asked us to deliver the result through an |
| 744 | * eventfd. The eventfd_signal() function is safe to be called |
| 745 | * from IRQ context. |
| 746 | */ |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 747 | if (iocb->ki_eventfd != NULL) |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 748 | eventfd_signal(iocb->ki_eventfd, 1); |
| 749 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | put_rq: |
| 751 | /* everything turned out well, dispose of the aiocb. */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 752 | __aio_put_req(ctx, iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | |
Quentin Barnes | 6cb2a21 | 2008-03-19 17:00:39 -0700 | [diff] [blame] | 754 | /* |
| 755 | * We have to order our ring_info tail store above and test |
| 756 | * of the wait list below outside the wait lock. This is |
| 757 | * like in wake_up_bit() where clearing a bit has to be |
| 758 | * ordered with the unlocked test. |
| 759 | */ |
| 760 | smp_mb(); |
| 761 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | if (waitqueue_active(&ctx->wait)) |
| 763 | wake_up(&ctx->wait); |
| 764 | |
Ken Chen | dee11c2 | 2007-02-03 01:13:45 -0800 | [diff] [blame] | 765 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 767 | EXPORT_SYMBOL(aio_complete); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | |
| 769 | /* aio_read_evt |
| 770 | * Pull an event off of the ioctx's event ring. Returns the number of |
| 771 | * events fetched (0 or 1 ;-) |
| 772 | * FIXME: make this use cmpxchg. |
| 773 | * TODO: make the ringbuffer user mmap()able (requires FIXME). |
| 774 | */ |
| 775 | static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) |
| 776 | { |
| 777 | struct aio_ring_info *info = &ioctx->ring_info; |
| 778 | struct aio_ring *ring; |
| 779 | unsigned long head; |
| 780 | int ret = 0; |
| 781 | |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 782 | ring = kmap_atomic(info->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | dprintk("in aio_read_evt h%lu t%lu m%lu\n", |
| 784 | (unsigned long)ring->head, (unsigned long)ring->tail, |
| 785 | (unsigned long)ring->nr); |
| 786 | |
| 787 | if (ring->head == ring->tail) |
| 788 | goto out; |
| 789 | |
| 790 | spin_lock(&info->ring_lock); |
| 791 | |
| 792 | head = ring->head % info->nr; |
| 793 | if (head != ring->tail) { |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 794 | struct io_event *evp = aio_ring_event(info, head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | *ent = *evp; |
| 796 | head = (head + 1) % info->nr; |
| 797 | smp_mb(); /* finish reading the event before updatng the head */ |
| 798 | ring->head = head; |
| 799 | ret = 1; |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 800 | put_aio_ring_event(evp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | } |
| 802 | spin_unlock(&info->ring_lock); |
| 803 | |
| 804 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 805 | dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret, |
| 806 | (unsigned long)ring->head, (unsigned long)ring->tail); |
Zhao Hongjiang | 91d80a8 | 2013-04-26 11:03:53 +0800 | [diff] [blame] | 807 | kunmap_atomic(ring); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | return ret; |
| 809 | } |
| 810 | |
| 811 | struct aio_timeout { |
| 812 | struct timer_list timer; |
| 813 | int timed_out; |
| 814 | struct task_struct *p; |
| 815 | }; |
| 816 | |
| 817 | static void timeout_func(unsigned long data) |
| 818 | { |
| 819 | struct aio_timeout *to = (struct aio_timeout *)data; |
| 820 | |
| 821 | to->timed_out = 1; |
| 822 | wake_up_process(to->p); |
| 823 | } |
| 824 | |
| 825 | static inline void init_timeout(struct aio_timeout *to) |
| 826 | { |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 827 | setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | to->timed_out = 0; |
| 829 | to->p = current; |
| 830 | } |
| 831 | |
| 832 | static inline void set_timeout(long start_jiffies, struct aio_timeout *to, |
| 833 | const struct timespec *ts) |
| 834 | { |
| 835 | to->timer.expires = start_jiffies + timespec_to_jiffies(ts); |
| 836 | if (time_after(to->timer.expires, jiffies)) |
| 837 | add_timer(&to->timer); |
| 838 | else |
| 839 | to->timed_out = 1; |
| 840 | } |
| 841 | |
| 842 | static inline void clear_timeout(struct aio_timeout *to) |
| 843 | { |
| 844 | del_singleshot_timer_sync(&to->timer); |
| 845 | } |
| 846 | |
| 847 | static int read_events(struct kioctx *ctx, |
| 848 | long min_nr, long nr, |
| 849 | struct io_event __user *event, |
| 850 | struct timespec __user *timeout) |
| 851 | { |
| 852 | long start_jiffies = jiffies; |
| 853 | struct task_struct *tsk = current; |
| 854 | DECLARE_WAITQUEUE(wait, tsk); |
| 855 | int ret; |
| 856 | int i = 0; |
| 857 | struct io_event ent; |
| 858 | struct aio_timeout to; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 859 | |
| 860 | /* needed to zero any padding within an entry (there shouldn't be |
| 861 | * any, but C is fun! |
| 862 | */ |
| 863 | memset(&ent, 0, sizeof(ent)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 864 | ret = 0; |
| 865 | while (likely(i < nr)) { |
| 866 | ret = aio_read_evt(ctx, &ent); |
| 867 | if (unlikely(ret <= 0)) |
| 868 | break; |
| 869 | |
| 870 | dprintk("read event: %Lx %Lx %Lx %Lx\n", |
| 871 | ent.data, ent.obj, ent.res, ent.res2); |
| 872 | |
| 873 | /* Could we split the check in two? */ |
| 874 | ret = -EFAULT; |
| 875 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { |
| 876 | dprintk("aio: lost an event due to EFAULT.\n"); |
| 877 | break; |
| 878 | } |
| 879 | ret = 0; |
| 880 | |
| 881 | /* Good, event copied to userland, update counts. */ |
| 882 | event ++; |
| 883 | i ++; |
| 884 | } |
| 885 | |
| 886 | if (min_nr <= i) |
| 887 | return i; |
| 888 | if (ret) |
| 889 | return ret; |
| 890 | |
| 891 | /* End fast path */ |
| 892 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | init_timeout(&to); |
| 894 | if (timeout) { |
| 895 | struct timespec ts; |
| 896 | ret = -EFAULT; |
| 897 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
| 898 | goto out; |
| 899 | |
| 900 | set_timeout(start_jiffies, &to, &ts); |
| 901 | } |
| 902 | |
| 903 | while (likely(i < nr)) { |
| 904 | add_wait_queue_exclusive(&ctx->wait, &wait); |
| 905 | do { |
| 906 | set_task_state(tsk, TASK_INTERRUPTIBLE); |
| 907 | ret = aio_read_evt(ctx, &ent); |
| 908 | if (ret) |
| 909 | break; |
| 910 | if (min_nr <= i) |
| 911 | break; |
Jeff Moyer | e92adcb | 2008-04-28 02:12:04 -0700 | [diff] [blame] | 912 | if (unlikely(ctx->dead)) { |
| 913 | ret = -EINVAL; |
| 914 | break; |
| 915 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | if (to.timed_out) /* Only check after read evt */ |
| 917 | break; |
Jeff Moyer | e00ba3d | 2007-12-04 23:45:02 -0800 | [diff] [blame] | 918 | /* Try to only show up in io wait if there are ops |
| 919 | * in flight */ |
| 920 | if (ctx->reqs_active) |
| 921 | io_schedule(); |
| 922 | else |
| 923 | schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | if (signal_pending(tsk)) { |
| 925 | ret = -EINTR; |
| 926 | break; |
| 927 | } |
| 928 | /*ret = aio_read_evt(ctx, &ent);*/ |
| 929 | } while (1) ; |
| 930 | |
| 931 | set_task_state(tsk, TASK_RUNNING); |
| 932 | remove_wait_queue(&ctx->wait, &wait); |
| 933 | |
| 934 | if (unlikely(ret <= 0)) |
| 935 | break; |
| 936 | |
| 937 | ret = -EFAULT; |
| 938 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { |
| 939 | dprintk("aio: lost an event due to EFAULT.\n"); |
| 940 | break; |
| 941 | } |
| 942 | |
| 943 | /* Good, event copied to userland, update counts. */ |
| 944 | event ++; |
| 945 | i ++; |
| 946 | } |
| 947 | |
| 948 | if (timeout) |
| 949 | clear_timeout(&to); |
| 950 | out: |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 951 | destroy_timer_on_stack(&to.timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | return i ? i : ret; |
| 953 | } |
| 954 | |
| 955 | /* Take an ioctx and remove it from the list of ioctx's. Protects |
| 956 | * against races with itself via ->dead. |
| 957 | */ |
| 958 | static void io_destroy(struct kioctx *ioctx) |
| 959 | { |
| 960 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | int was_dead; |
| 962 | |
| 963 | /* delete the entry from the list is someone else hasn't already */ |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 964 | spin_lock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 965 | was_dead = ioctx->dead; |
| 966 | ioctx->dead = 1; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 967 | hlist_del_rcu(&ioctx->list); |
| 968 | spin_unlock(&mm->ioctx_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
| 970 | dprintk("aio_release(%p)\n", ioctx); |
| 971 | if (likely(!was_dead)) |
| 972 | put_ioctx(ioctx); /* twice for the list */ |
| 973 | |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 974 | kill_ctx(ioctx); |
Jeff Moyer | e92adcb | 2008-04-28 02:12:04 -0700 | [diff] [blame] | 975 | |
| 976 | /* |
| 977 | * Wake up any waiters. The setting of ctx->dead must be seen |
| 978 | * by other CPUs at this point. Right now, we rely on the |
| 979 | * locking done by the above calls to ensure this consistency. |
| 980 | */ |
Roland Dreier | e91f90b | 2011-03-22 16:35:10 -0700 | [diff] [blame] | 981 | wake_up_all(&ioctx->wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | } |
| 983 | |
| 984 | /* sys_io_setup: |
| 985 | * Create an aio_context capable of receiving at least nr_events. |
| 986 | * ctxp must not point to an aio_context that already exists, and |
| 987 | * must be initialized to 0 prior to the call. On successful |
| 988 | * creation of the aio_context, *ctxp is filled in with the resulting |
| 989 | * handle. May fail with -EINVAL if *ctxp is not initialized, |
| 990 | * if the specified nr_events exceeds internal limits. May fail |
| 991 | * with -EAGAIN if the specified nr_events exceeds the user's limit |
| 992 | * of available events. May fail with -ENOMEM if insufficient kernel |
| 993 | * resources are available. May fail with -EFAULT if an invalid |
| 994 | * pointer is passed for ctxp. Will fail with -ENOSYS if not |
| 995 | * implemented. |
| 996 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 997 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | { |
| 999 | struct kioctx *ioctx = NULL; |
| 1000 | unsigned long ctx; |
| 1001 | long ret; |
| 1002 | |
| 1003 | ret = get_user(ctx, ctxp); |
| 1004 | if (unlikely(ret)) |
| 1005 | goto out; |
| 1006 | |
| 1007 | ret = -EINVAL; |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 1008 | if (unlikely(ctx || nr_events == 0)) { |
| 1009 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", |
| 1010 | ctx, nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | goto out; |
| 1012 | } |
| 1013 | |
| 1014 | ioctx = ioctx_alloc(nr_events); |
| 1015 | ret = PTR_ERR(ioctx); |
| 1016 | if (!IS_ERR(ioctx)) { |
| 1017 | ret = put_user(ioctx->user_id, ctxp); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 1018 | if (ret) |
| 1019 | io_destroy(ioctx); |
| 1020 | put_ioctx(ioctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1021 | } |
| 1022 | |
| 1023 | out: |
| 1024 | return ret; |
| 1025 | } |
| 1026 | |
| 1027 | /* sys_io_destroy: |
| 1028 | * Destroy the aio_context specified. May cancel any outstanding |
| 1029 | * AIOs and block on completion. Will fail with -ENOSYS if not |
Satoru Takeuchi | 642b512 | 2010-08-05 11:23:11 -0700 | [diff] [blame] | 1030 | * implemented. May fail with -EINVAL if the context pointed to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | * is invalid. |
| 1032 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1033 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | { |
| 1035 | struct kioctx *ioctx = lookup_ioctx(ctx); |
| 1036 | if (likely(NULL != ioctx)) { |
| 1037 | io_destroy(ioctx); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 1038 | put_ioctx(ioctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1039 | return 0; |
| 1040 | } |
| 1041 | pr_debug("EINVAL: io_destroy: invalid context id\n"); |
| 1042 | return -EINVAL; |
| 1043 | } |
| 1044 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1045 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
| 1046 | { |
| 1047 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
| 1048 | |
| 1049 | BUG_ON(ret <= 0); |
| 1050 | |
| 1051 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { |
| 1052 | ssize_t this = min((ssize_t)iov->iov_len, ret); |
| 1053 | iov->iov_base += this; |
| 1054 | iov->iov_len -= this; |
| 1055 | iocb->ki_left -= this; |
| 1056 | ret -= this; |
| 1057 | if (iov->iov_len == 0) { |
| 1058 | iocb->ki_cur_seg++; |
| 1059 | iov++; |
| 1060 | } |
| 1061 | } |
| 1062 | |
| 1063 | /* the caller should not have done more io than what fit in |
| 1064 | * the remaining iovecs */ |
| 1065 | BUG_ON(ret > 0 && iocb->ki_left == 0); |
| 1066 | } |
| 1067 | |
| 1068 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | { |
| 1070 | struct file *file = iocb->ki_filp; |
| 1071 | struct address_space *mapping = file->f_mapping; |
| 1072 | struct inode *inode = mapping->host; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1073 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, |
| 1074 | unsigned long, loff_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1075 | ssize_t ret = 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1076 | unsigned short opcode; |
| 1077 | |
| 1078 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || |
| 1079 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { |
| 1080 | rw_op = file->f_op->aio_read; |
| 1081 | opcode = IOCB_CMD_PREADV; |
| 1082 | } else { |
| 1083 | rw_op = file->f_op->aio_write; |
| 1084 | opcode = IOCB_CMD_PWRITEV; |
| 1085 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | |
Rusty Russell | c2ec668 | 2008-02-08 04:20:15 -0800 | [diff] [blame] | 1087 | /* This matches the pread()/pwrite() logic */ |
| 1088 | if (iocb->ki_pos < 0) |
| 1089 | return -EINVAL; |
| 1090 | |
Al Viro | 8d71db4 | 2013-03-19 21:01:03 -0400 | [diff] [blame] | 1091 | if (opcode == IOCB_CMD_PWRITEV) |
| 1092 | file_start_write(file); |
Zach Brown | 897f15f | 2005-09-30 11:58:55 -0700 | [diff] [blame] | 1093 | do { |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1094 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
| 1095 | iocb->ki_nr_segs - iocb->ki_cur_seg, |
| 1096 | iocb->ki_pos); |
| 1097 | if (ret > 0) |
| 1098 | aio_advance_iovec(iocb, ret); |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1099 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1100 | /* retry all partial writes. retry partial reads as long as its a |
| 1101 | * regular file. */ |
Zach Brown | 353fb07 | 2005-09-30 11:58:56 -0700 | [diff] [blame] | 1102 | } while (ret > 0 && iocb->ki_left > 0 && |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1103 | (opcode == IOCB_CMD_PWRITEV || |
| 1104 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); |
Al Viro | 8d71db4 | 2013-03-19 21:01:03 -0400 | [diff] [blame] | 1105 | if (opcode == IOCB_CMD_PWRITEV) |
| 1106 | file_end_write(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | |
| 1108 | /* This means we must have transferred all that we could */ |
| 1109 | /* No need to retry anymore */ |
| 1110 | if ((ret == 0) || (iocb->ki_left == 0)) |
| 1111 | ret = iocb->ki_nbytes - iocb->ki_left; |
| 1112 | |
Rusty Russell | 7adfa2f | 2008-02-08 04:20:14 -0800 | [diff] [blame] | 1113 | /* If we managed to write some out we return that, rather than |
| 1114 | * the eventual error. */ |
| 1115 | if (opcode == IOCB_CMD_PWRITEV |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 1116 | && ret < 0 && ret != -EIOCBQUEUED |
Rusty Russell | 7adfa2f | 2008-02-08 04:20:14 -0800 | [diff] [blame] | 1117 | && iocb->ki_nbytes - iocb->ki_left) |
| 1118 | ret = iocb->ki_nbytes - iocb->ki_left; |
| 1119 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | return ret; |
| 1121 | } |
| 1122 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | static ssize_t aio_fdsync(struct kiocb *iocb) |
| 1124 | { |
| 1125 | struct file *file = iocb->ki_filp; |
| 1126 | ssize_t ret = -EINVAL; |
| 1127 | |
| 1128 | if (file->f_op->aio_fsync) |
| 1129 | ret = file->f_op->aio_fsync(iocb, 1); |
| 1130 | return ret; |
| 1131 | } |
| 1132 | |
| 1133 | static ssize_t aio_fsync(struct kiocb *iocb) |
| 1134 | { |
| 1135 | struct file *file = iocb->ki_filp; |
| 1136 | ssize_t ret = -EINVAL; |
| 1137 | |
| 1138 | if (file->f_op->aio_fsync) |
| 1139 | ret = file->f_op->aio_fsync(iocb, 0); |
| 1140 | return ret; |
| 1141 | } |
| 1142 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1143 | 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] | 1144 | { |
| 1145 | ssize_t ret; |
| 1146 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1147 | #ifdef CONFIG_COMPAT |
| 1148 | if (compat) |
| 1149 | ret = compat_rw_copy_check_uvector(type, |
| 1150 | (struct compat_iovec __user *)kiocb->ki_buf, |
| 1151 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, |
Christopher Yeoh | ac34ebb | 2012-05-31 16:26:42 -0700 | [diff] [blame] | 1152 | &kiocb->ki_iovec); |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1153 | else |
| 1154 | #endif |
| 1155 | ret = rw_copy_check_uvector(type, |
| 1156 | (struct iovec __user *)kiocb->ki_buf, |
| 1157 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, |
Christopher Yeoh | ac34ebb | 2012-05-31 16:26:42 -0700 | [diff] [blame] | 1158 | &kiocb->ki_iovec); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1159 | if (ret < 0) |
| 1160 | goto out; |
| 1161 | |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1162 | ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret); |
| 1163 | if (ret < 0) |
| 1164 | goto out; |
| 1165 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1166 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
| 1167 | kiocb->ki_cur_seg = 0; |
| 1168 | /* ki_nbytes/left now reflect bytes instead of segs */ |
| 1169 | kiocb->ki_nbytes = ret; |
| 1170 | kiocb->ki_left = ret; |
| 1171 | |
| 1172 | ret = 0; |
| 1173 | out: |
| 1174 | return ret; |
| 1175 | } |
| 1176 | |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1177 | static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1178 | { |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1179 | int bytes; |
| 1180 | |
| 1181 | bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left); |
| 1182 | if (bytes < 0) |
| 1183 | return bytes; |
| 1184 | |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1185 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
| 1186 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1187 | kiocb->ki_iovec->iov_len = bytes; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1188 | kiocb->ki_nr_segs = 1; |
| 1189 | kiocb->ki_cur_seg = 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1190 | return 0; |
| 1191 | } |
| 1192 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | /* |
| 1194 | * aio_setup_iocb: |
| 1195 | * Performs the initial checks and aio retry method |
| 1196 | * setup for the kiocb at the time of io submission. |
| 1197 | */ |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1198 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1199 | { |
| 1200 | struct file *file = kiocb->ki_filp; |
| 1201 | ssize_t ret = 0; |
| 1202 | |
| 1203 | switch (kiocb->ki_opcode) { |
| 1204 | case IOCB_CMD_PREAD: |
| 1205 | ret = -EBADF; |
| 1206 | if (unlikely(!(file->f_mode & FMODE_READ))) |
| 1207 | break; |
| 1208 | ret = -EFAULT; |
| 1209 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, |
| 1210 | kiocb->ki_left))) |
| 1211 | break; |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1212 | ret = aio_setup_single_vector(READ, file, kiocb); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1213 | if (ret) |
| 1214 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | ret = -EINVAL; |
| 1216 | if (file->f_op->aio_read) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1217 | kiocb->ki_retry = aio_rw_vect_retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | break; |
| 1219 | case IOCB_CMD_PWRITE: |
| 1220 | ret = -EBADF; |
| 1221 | if (unlikely(!(file->f_mode & FMODE_WRITE))) |
| 1222 | break; |
| 1223 | ret = -EFAULT; |
| 1224 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, |
| 1225 | kiocb->ki_left))) |
| 1226 | break; |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1227 | ret = aio_setup_single_vector(WRITE, file, kiocb); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1228 | if (ret) |
| 1229 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1230 | ret = -EINVAL; |
| 1231 | if (file->f_op->aio_write) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1232 | kiocb->ki_retry = aio_rw_vect_retry; |
| 1233 | break; |
| 1234 | case IOCB_CMD_PREADV: |
| 1235 | ret = -EBADF; |
| 1236 | if (unlikely(!(file->f_mode & FMODE_READ))) |
| 1237 | break; |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1238 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1239 | if (ret) |
| 1240 | break; |
| 1241 | ret = -EINVAL; |
| 1242 | if (file->f_op->aio_read) |
| 1243 | kiocb->ki_retry = aio_rw_vect_retry; |
| 1244 | break; |
| 1245 | case IOCB_CMD_PWRITEV: |
| 1246 | ret = -EBADF; |
| 1247 | if (unlikely(!(file->f_mode & FMODE_WRITE))) |
| 1248 | break; |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1249 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1250 | if (ret) |
| 1251 | break; |
| 1252 | ret = -EINVAL; |
| 1253 | if (file->f_op->aio_write) |
| 1254 | kiocb->ki_retry = aio_rw_vect_retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | break; |
| 1256 | case IOCB_CMD_FDSYNC: |
| 1257 | ret = -EINVAL; |
| 1258 | if (file->f_op->aio_fsync) |
| 1259 | kiocb->ki_retry = aio_fdsync; |
| 1260 | break; |
| 1261 | case IOCB_CMD_FSYNC: |
| 1262 | ret = -EINVAL; |
| 1263 | if (file->f_op->aio_fsync) |
| 1264 | kiocb->ki_retry = aio_fsync; |
| 1265 | break; |
| 1266 | default: |
| 1267 | dprintk("EINVAL: io_submit: no operation provided\n"); |
| 1268 | ret = -EINVAL; |
| 1269 | } |
| 1270 | |
| 1271 | if (!kiocb->ki_retry) |
| 1272 | return ret; |
| 1273 | |
| 1274 | return 0; |
| 1275 | } |
| 1276 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 1277 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1278 | struct iocb *iocb, struct kiocb_batch *batch, |
| 1279 | bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | { |
| 1281 | struct kiocb *req; |
| 1282 | struct file *file; |
| 1283 | ssize_t ret; |
| 1284 | |
| 1285 | /* enforce forwards compatibility on users */ |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1286 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | pr_debug("EINVAL: io_submit: reserve field set\n"); |
| 1288 | return -EINVAL; |
| 1289 | } |
| 1290 | |
| 1291 | /* prevent overflows */ |
| 1292 | if (unlikely( |
| 1293 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || |
| 1294 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || |
| 1295 | ((ssize_t)iocb->aio_nbytes < 0) |
| 1296 | )) { |
| 1297 | pr_debug("EINVAL: io_submit: overflow check\n"); |
| 1298 | return -EINVAL; |
| 1299 | } |
| 1300 | |
| 1301 | file = fget(iocb->aio_fildes); |
| 1302 | if (unlikely(!file)) |
| 1303 | return -EBADF; |
| 1304 | |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1305 | req = aio_get_req(ctx, batch); /* returns with 2 references to req */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1306 | if (unlikely(!req)) { |
| 1307 | fput(file); |
| 1308 | return -EAGAIN; |
| 1309 | } |
Yan Zheng | 87e2831 | 2007-10-08 12:16:20 -0700 | [diff] [blame] | 1310 | req->ki_filp = file; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1311 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
| 1312 | /* |
| 1313 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an |
| 1314 | * instance of the file* now. The file descriptor must be |
| 1315 | * an eventfd() fd, and will be signaled for each completed |
| 1316 | * event using the eventfd_signal() function. |
| 1317 | */ |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 1318 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
Hirofumi Nakagawa | 801678c | 2008-04-29 01:03:09 -0700 | [diff] [blame] | 1319 | if (IS_ERR(req->ki_eventfd)) { |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1320 | ret = PTR_ERR(req->ki_eventfd); |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 1321 | req->ki_eventfd = NULL; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1322 | goto out_put_req; |
| 1323 | } |
| 1324 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | |
Ken Chen | 212079c | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 1326 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | if (unlikely(ret)) { |
| 1328 | dprintk("EFAULT: aio_key\n"); |
| 1329 | goto out_put_req; |
| 1330 | } |
| 1331 | |
| 1332 | req->ki_obj.user = user_iocb; |
| 1333 | req->ki_user_data = iocb->aio_data; |
| 1334 | req->ki_pos = iocb->aio_offset; |
| 1335 | |
| 1336 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; |
| 1337 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; |
| 1338 | req->ki_opcode = iocb->aio_lio_opcode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1339 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1340 | ret = aio_setup_iocb(req, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | |
| 1342 | if (ret) |
| 1343 | goto out_put_req; |
| 1344 | |
| 1345 | spin_lock_irq(&ctx->ctx_lock); |
Jan Kara | 7137c6bd | 2011-02-25 14:44:27 -0800 | [diff] [blame] | 1346 | /* |
| 1347 | * We could have raced with io_destroy() and are currently holding a |
| 1348 | * reference to ctx which should be destroyed. We cannot submit IO |
| 1349 | * since ctx gets freed as soon as io_submit() puts its reference. The |
| 1350 | * check here is reliable: io_destroy() sets ctx->dead before waiting |
| 1351 | * for outstanding IO and the barrier between these two is realized by |
| 1352 | * unlock of mm->ioctx_lock and lock of ctx->ctx_lock. Analogously we |
| 1353 | * increment ctx->reqs_active before checking for ctx->dead and the |
| 1354 | * barrier is realized by unlock and lock of ctx->ctx_lock. Thus if we |
| 1355 | * don't see ctx->dead set here, io_destroy() waits for our IO to |
| 1356 | * finish. |
| 1357 | */ |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 1358 | if (ctx->dead) |
Jan Kara | 7137c6bd | 2011-02-25 14:44:27 -0800 | [diff] [blame] | 1359 | ret = -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1360 | spin_unlock_irq(&ctx->ctx_lock); |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 1361 | if (ret) |
| 1362 | goto out_put_req; |
| 1363 | |
| 1364 | if (unlikely(kiocbIsCancelled(req))) |
| 1365 | ret = -EINTR; |
| 1366 | else |
| 1367 | ret = req->ki_retry(req); |
| 1368 | |
| 1369 | if (ret != -EIOCBQUEUED) { |
| 1370 | /* |
| 1371 | * There's no easy way to restart the syscall since other AIO's |
| 1372 | * may be already running. Just fail this IO with EINTR. |
| 1373 | */ |
| 1374 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || |
| 1375 | ret == -ERESTARTNOHAND || |
| 1376 | ret == -ERESTART_RESTARTBLOCK)) |
| 1377 | ret = -EINTR; |
| 1378 | aio_complete(req, ret, 0); |
| 1379 | } |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 1380 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1381 | aio_put_req(req); /* drop extra ref to req */ |
| 1382 | return 0; |
| 1383 | |
| 1384 | out_put_req: |
| 1385 | aio_put_req(req); /* drop extra ref to req */ |
| 1386 | aio_put_req(req); /* drop i/o ref to req */ |
| 1387 | return ret; |
| 1388 | } |
| 1389 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1390 | long do_io_submit(aio_context_t ctx_id, long nr, |
| 1391 | struct iocb __user *__user *iocbpp, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | { |
| 1393 | struct kioctx *ctx; |
| 1394 | long ret = 0; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1395 | int i = 0; |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1396 | struct blk_plug plug; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1397 | struct kiocb_batch batch; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1398 | |
| 1399 | if (unlikely(nr < 0)) |
| 1400 | return -EINVAL; |
| 1401 | |
Jeff Moyer | 75e1c70 | 2010-09-10 14:16:00 -0700 | [diff] [blame] | 1402 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
| 1403 | nr = LONG_MAX/sizeof(*iocbpp); |
| 1404 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1405 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
| 1406 | return -EFAULT; |
| 1407 | |
| 1408 | ctx = lookup_ioctx(ctx_id); |
| 1409 | if (unlikely(!ctx)) { |
| 1410 | pr_debug("EINVAL: io_submit: invalid context id\n"); |
| 1411 | return -EINVAL; |
| 1412 | } |
| 1413 | |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1414 | kiocb_batch_init(&batch, nr); |
| 1415 | |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1416 | blk_start_plug(&plug); |
| 1417 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | /* |
| 1419 | * AKPM: should this return a partial result if some of the IOs were |
| 1420 | * successfully submitted? |
| 1421 | */ |
| 1422 | for (i=0; i<nr; i++) { |
| 1423 | struct iocb __user *user_iocb; |
| 1424 | struct iocb tmp; |
| 1425 | |
| 1426 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { |
| 1427 | ret = -EFAULT; |
| 1428 | break; |
| 1429 | } |
| 1430 | |
| 1431 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { |
| 1432 | ret = -EFAULT; |
| 1433 | break; |
| 1434 | } |
| 1435 | |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1436 | ret = io_submit_one(ctx, user_iocb, &tmp, &batch, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | if (ret) |
| 1438 | break; |
| 1439 | } |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1440 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1441 | |
Gleb Natapov | 69e4747 | 2012-01-08 17:07:28 +0200 | [diff] [blame] | 1442 | kiocb_batch_free(ctx, &batch); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | put_ioctx(ctx); |
| 1444 | return i ? i : ret; |
| 1445 | } |
| 1446 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1447 | /* sys_io_submit: |
| 1448 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns |
| 1449 | * the number of iocbs queued. May return -EINVAL if the aio_context |
| 1450 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at |
| 1451 | * *iocbpp[0] is not properly initialized, if the operation specified |
| 1452 | * is invalid for the file descriptor in the iocb. May fail with |
| 1453 | * -EFAULT if any of the data structures point to invalid data. May |
| 1454 | * fail with -EBADF if the file descriptor specified in the first |
| 1455 | * iocb is invalid. May fail with -EAGAIN if insufficient resources |
| 1456 | * are available to queue any iocbs. Will return 0 if nr is 0. Will |
| 1457 | * fail with -ENOSYS if not implemented. |
| 1458 | */ |
| 1459 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, |
| 1460 | struct iocb __user * __user *, iocbpp) |
| 1461 | { |
| 1462 | return do_io_submit(ctx_id, nr, iocbpp, 0); |
| 1463 | } |
| 1464 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | /* lookup_kiocb |
| 1466 | * Finds a given iocb for cancellation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | */ |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 1468 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
| 1469 | u32 key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | { |
| 1471 | struct list_head *pos; |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 1472 | |
| 1473 | assert_spin_locked(&ctx->ctx_lock); |
| 1474 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | /* TODO: use a hash or array, this sucks. */ |
| 1476 | list_for_each(pos, &ctx->active_reqs) { |
| 1477 | struct kiocb *kiocb = list_kiocb(pos); |
| 1478 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) |
| 1479 | return kiocb; |
| 1480 | } |
| 1481 | return NULL; |
| 1482 | } |
| 1483 | |
| 1484 | /* sys_io_cancel: |
| 1485 | * Attempts to cancel an iocb previously passed to io_submit. If |
| 1486 | * the operation is successfully cancelled, the resulting event is |
| 1487 | * copied into the memory pointed to by result without being placed |
| 1488 | * into the completion queue and 0 is returned. May fail with |
| 1489 | * -EFAULT if any of the data structures pointed to are invalid. |
| 1490 | * May fail with -EINVAL if aio_context specified by ctx_id is |
| 1491 | * invalid. May fail with -EAGAIN if the iocb specified was not |
| 1492 | * cancelled. Will fail with -ENOSYS if not implemented. |
| 1493 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1494 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
| 1495 | struct io_event __user *, result) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1496 | { |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1497 | struct io_event res; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | struct kioctx *ctx; |
| 1499 | struct kiocb *kiocb; |
| 1500 | u32 key; |
| 1501 | int ret; |
| 1502 | |
| 1503 | ret = get_user(key, &iocb->aio_key); |
| 1504 | if (unlikely(ret)) |
| 1505 | return -EFAULT; |
| 1506 | |
| 1507 | ctx = lookup_ioctx(ctx_id); |
| 1508 | if (unlikely(!ctx)) |
| 1509 | return -EINVAL; |
| 1510 | |
| 1511 | spin_lock_irq(&ctx->ctx_lock); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1512 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1513 | kiocb = lookup_kiocb(ctx, iocb, key); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1514 | if (kiocb) |
| 1515 | ret = kiocb_cancel(ctx, kiocb, &res); |
| 1516 | else |
| 1517 | ret = -EINVAL; |
| 1518 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | spin_unlock_irq(&ctx->ctx_lock); |
| 1520 | |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1521 | if (!ret) { |
| 1522 | /* Cancellation succeeded -- copy the result |
| 1523 | * into the user's buffer. |
| 1524 | */ |
| 1525 | if (copy_to_user(result, &res, sizeof(res))) |
| 1526 | ret = -EFAULT; |
| 1527 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | |
| 1529 | put_ioctx(ctx); |
| 1530 | |
| 1531 | return ret; |
| 1532 | } |
| 1533 | |
| 1534 | /* io_getevents: |
| 1535 | * Attempts to read at least min_nr events and up to nr events from |
Satoru Takeuchi | 642b512 | 2010-08-05 11:23:11 -0700 | [diff] [blame] | 1536 | * the completion queue for the aio_context specified by ctx_id. If |
| 1537 | * it succeeds, the number of read events is returned. May fail with |
| 1538 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is |
| 1539 | * out of range, if timeout is out of range. May fail with -EFAULT |
| 1540 | * if any of the memory specified is invalid. May return 0 or |
| 1541 | * < min_nr if the timeout specified by timeout has elapsed |
| 1542 | * before sufficient events are available, where timeout == NULL |
| 1543 | * specifies an infinite timeout. Note that the timeout pointed to by |
| 1544 | * timeout is relative and will be updated if not NULL and the |
| 1545 | * operation blocks. Will fail with -ENOSYS if not implemented. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1547 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
| 1548 | long, min_nr, |
| 1549 | long, nr, |
| 1550 | struct io_event __user *, events, |
| 1551 | struct timespec __user *, timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1552 | { |
| 1553 | struct kioctx *ioctx = lookup_ioctx(ctx_id); |
| 1554 | long ret = -EINVAL; |
| 1555 | |
| 1556 | if (likely(ioctx)) { |
Namhyung Kim | 2e41025 | 2011-01-12 17:01:08 -0800 | [diff] [blame] | 1557 | if (likely(min_nr <= nr && min_nr >= 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1558 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
| 1559 | put_ioctx(ioctx); |
| 1560 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | return ret; |
| 1562 | } |