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 | */ |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 11 | #define pr_fmt(fmt) "%s: " fmt, __func__ |
| 12 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | #include <linux/kernel.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/errno.h> |
| 16 | #include <linux/time.h> |
| 17 | #include <linux/aio_abi.h> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 18 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | #include <linux/syscalls.h> |
Jens Axboe | b9d128f | 2009-10-29 13:59:26 +0100 | [diff] [blame] | 20 | #include <linux/backing-dev.h> |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 21 | #include <linux/uio.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 29 | #include <linux/percpu.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/slab.h> |
| 31 | #include <linux/timer.h> |
| 32 | #include <linux/aio.h> |
| 33 | #include <linux/highmem.h> |
| 34 | #include <linux/workqueue.h> |
| 35 | #include <linux/security.h> |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 36 | #include <linux/eventfd.h> |
Jeff Moyer | cfb1e33 | 2009-10-02 18:57:36 -0400 | [diff] [blame] | 37 | #include <linux/blkdev.h> |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 38 | #include <linux/compat.h> |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 39 | #include <linux/anon_inodes.h> |
| 40 | #include <linux/migrate.h> |
| 41 | #include <linux/ramfs.h> |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 42 | #include <linux/percpu-refcount.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | |
| 44 | #include <asm/kmap_types.h> |
| 45 | #include <asm/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | |
Al Viro | 68d70d0 | 2013-06-19 15:26:04 +0400 | [diff] [blame] | 47 | #include "internal.h" |
| 48 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 49 | #define AIO_RING_MAGIC 0xa10a10a1 |
| 50 | #define AIO_RING_COMPAT_FEATURES 1 |
| 51 | #define AIO_RING_INCOMPAT_FEATURES 0 |
| 52 | struct aio_ring { |
| 53 | unsigned id; /* kernel internal index number */ |
| 54 | unsigned nr; /* number of io_events */ |
| 55 | unsigned head; |
| 56 | unsigned tail; |
| 57 | |
| 58 | unsigned magic; |
| 59 | unsigned compat_features; |
| 60 | unsigned incompat_features; |
| 61 | unsigned header_length; /* size of aio_ring */ |
| 62 | |
| 63 | |
| 64 | struct io_event io_events[0]; |
| 65 | }; /* 128 bytes + ring size */ |
| 66 | |
| 67 | #define AIO_RING_PAGES 8 |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 68 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 69 | struct kioctx_table { |
| 70 | struct rcu_head rcu; |
| 71 | unsigned nr; |
| 72 | struct kioctx *table[]; |
| 73 | }; |
| 74 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 75 | struct kioctx_cpu { |
| 76 | unsigned reqs_available; |
| 77 | }; |
| 78 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 79 | struct kioctx { |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 80 | struct percpu_ref users; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 81 | atomic_t dead; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 82 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 83 | struct percpu_ref reqs; |
| 84 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 85 | unsigned long user_id; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 86 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 87 | struct __percpu kioctx_cpu *cpu; |
| 88 | |
| 89 | /* |
| 90 | * For percpu reqs_available, number of slots we move to/from global |
| 91 | * counter at a time: |
| 92 | */ |
| 93 | unsigned req_batch; |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 94 | /* |
| 95 | * This is what userspace passed to io_setup(), it's not used for |
| 96 | * anything but counting against the global max_reqs quota. |
| 97 | * |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 98 | * The real limit is nr_events - 1, which will be larger (see |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 99 | * aio_setup_ring()) |
| 100 | */ |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 101 | unsigned max_reqs; |
| 102 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 103 | /* Size of ringbuffer, in units of struct io_event */ |
| 104 | unsigned nr_events; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 105 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 106 | unsigned long mmap_base; |
| 107 | unsigned long mmap_size; |
| 108 | |
| 109 | struct page **ring_pages; |
| 110 | long nr_pages; |
| 111 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 112 | struct work_struct free_work; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 113 | |
| 114 | struct { |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 115 | /* |
| 116 | * This counts the number of available slots in the ringbuffer, |
| 117 | * so we avoid overflowing it: it's decremented (if positive) |
| 118 | * when allocating a kiocb and incremented when the resulting |
| 119 | * io_event is pulled off the ringbuffer. |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 120 | * |
| 121 | * We batch accesses to it with a percpu version. |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 122 | */ |
| 123 | atomic_t reqs_available; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 124 | } ____cacheline_aligned_in_smp; |
| 125 | |
| 126 | struct { |
| 127 | spinlock_t ctx_lock; |
| 128 | struct list_head active_reqs; /* used for cancellation */ |
| 129 | } ____cacheline_aligned_in_smp; |
| 130 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 131 | struct { |
| 132 | struct mutex ring_lock; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 133 | wait_queue_head_t wait; |
| 134 | } ____cacheline_aligned_in_smp; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 135 | |
| 136 | struct { |
| 137 | unsigned tail; |
| 138 | spinlock_t completion_lock; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 139 | } ____cacheline_aligned_in_smp; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 140 | |
| 141 | struct page *internal_pages[AIO_RING_PAGES]; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 142 | struct file *aio_ring_file; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 143 | |
| 144 | unsigned id; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 145 | }; |
| 146 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | /*------ sysctl variables----*/ |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 148 | static DEFINE_SPINLOCK(aio_nr_lock); |
| 149 | unsigned long aio_nr; /* current system wide number of aio requests */ |
| 150 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | /*----end sysctl variables---*/ |
| 152 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 153 | static struct kmem_cache *kiocb_cachep; |
| 154 | static struct kmem_cache *kioctx_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | /* aio_setup |
| 157 | * Creates the slab caches used by the aio routines, panic on |
| 158 | * failure as this is done early during the boot sequence. |
| 159 | */ |
| 160 | static int __init aio_setup(void) |
| 161 | { |
Christoph Lameter | 0a31bd5 | 2007-05-06 14:49:57 -0700 | [diff] [blame] | 162 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
| 163 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 165 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | |
| 167 | return 0; |
| 168 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 169 | __initcall(aio_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 171 | static void put_aio_ring_file(struct kioctx *ctx) |
| 172 | { |
| 173 | struct file *aio_ring_file = ctx->aio_ring_file; |
| 174 | if (aio_ring_file) { |
| 175 | truncate_setsize(aio_ring_file->f_inode, 0); |
| 176 | |
| 177 | /* Prevent further access to the kioctx from migratepages */ |
| 178 | spin_lock(&aio_ring_file->f_inode->i_mapping->private_lock); |
| 179 | aio_ring_file->f_inode->i_mapping->private_data = NULL; |
| 180 | ctx->aio_ring_file = NULL; |
| 181 | spin_unlock(&aio_ring_file->f_inode->i_mapping->private_lock); |
| 182 | |
| 183 | fput(aio_ring_file); |
| 184 | } |
| 185 | } |
| 186 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | static void aio_free_ring(struct kioctx *ctx) |
| 188 | { |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 189 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 191 | for (i = 0; i < ctx->nr_pages; i++) { |
| 192 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, |
| 193 | page_count(ctx->ring_pages[i])); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 194 | put_page(ctx->ring_pages[i]); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 195 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 197 | put_aio_ring_file(ctx); |
| 198 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 199 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) |
| 200 | kfree(ctx->ring_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | } |
| 202 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 203 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) |
| 204 | { |
| 205 | vma->vm_ops = &generic_file_vm_ops; |
| 206 | return 0; |
| 207 | } |
| 208 | |
| 209 | static const struct file_operations aio_ring_fops = { |
| 210 | .mmap = aio_ring_mmap, |
| 211 | }; |
| 212 | |
| 213 | static int aio_set_page_dirty(struct page *page) |
| 214 | { |
| 215 | return 0; |
| 216 | } |
| 217 | |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 218 | #if IS_ENABLED(CONFIG_MIGRATION) |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 219 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
| 220 | struct page *old, enum migrate_mode mode) |
| 221 | { |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 222 | struct kioctx *ctx; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 223 | unsigned long flags; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 224 | int rc; |
| 225 | |
| 226 | /* Writeback must be complete */ |
| 227 | BUG_ON(PageWriteback(old)); |
| 228 | put_page(old); |
| 229 | |
| 230 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode); |
| 231 | if (rc != MIGRATEPAGE_SUCCESS) { |
| 232 | get_page(old); |
| 233 | return rc; |
| 234 | } |
| 235 | |
| 236 | get_page(new); |
| 237 | |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 238 | /* We can potentially race against kioctx teardown here. Use the |
| 239 | * address_space's private data lock to protect the mapping's |
| 240 | * private_data. |
| 241 | */ |
| 242 | spin_lock(&mapping->private_lock); |
| 243 | ctx = mapping->private_data; |
| 244 | if (ctx) { |
| 245 | pgoff_t idx; |
| 246 | spin_lock_irqsave(&ctx->completion_lock, flags); |
| 247 | migrate_page_copy(new, old); |
| 248 | idx = old->index; |
| 249 | if (idx < (pgoff_t)ctx->nr_pages) |
| 250 | ctx->ring_pages[idx] = new; |
| 251 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
| 252 | } else |
| 253 | rc = -EBUSY; |
| 254 | spin_unlock(&mapping->private_lock); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 255 | |
| 256 | return rc; |
| 257 | } |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 258 | #endif |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 259 | |
| 260 | static const struct address_space_operations aio_ctx_aops = { |
| 261 | .set_page_dirty = aio_set_page_dirty, |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 262 | #if IS_ENABLED(CONFIG_MIGRATION) |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 263 | .migratepage = aio_migratepage, |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 264 | #endif |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 265 | }; |
| 266 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | static int aio_setup_ring(struct kioctx *ctx) |
| 268 | { |
| 269 | struct aio_ring *ring; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | unsigned nr_events = ctx->max_reqs; |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 271 | struct mm_struct *mm = current->mm; |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 272 | unsigned long size, populate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | int nr_pages; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 274 | int i; |
| 275 | struct file *file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | |
| 277 | /* Compensate for the ring buffer's head/tail overlap entry */ |
| 278 | nr_events += 2; /* 1 is required, 2 for good luck */ |
| 279 | |
| 280 | size = sizeof(struct aio_ring); |
| 281 | size += sizeof(struct io_event) * nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 283 | nr_pages = PFN_UP(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | if (nr_pages < 0) |
| 285 | return -EINVAL; |
| 286 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 287 | file = anon_inode_getfile_private("[aio]", &aio_ring_fops, ctx, O_RDWR); |
| 288 | if (IS_ERR(file)) { |
| 289 | ctx->aio_ring_file = NULL; |
| 290 | return -EAGAIN; |
| 291 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 293 | file->f_inode->i_mapping->a_ops = &aio_ctx_aops; |
| 294 | file->f_inode->i_mapping->private_data = ctx; |
| 295 | file->f_inode->i_size = PAGE_SIZE * (loff_t)nr_pages; |
| 296 | |
| 297 | for (i = 0; i < nr_pages; i++) { |
| 298 | struct page *page; |
| 299 | page = find_or_create_page(file->f_inode->i_mapping, |
| 300 | i, GFP_HIGHUSER | __GFP_ZERO); |
| 301 | if (!page) |
| 302 | break; |
| 303 | pr_debug("pid(%d) page[%d]->count=%d\n", |
| 304 | current->pid, i, page_count(page)); |
| 305 | SetPageUptodate(page); |
| 306 | SetPageDirty(page); |
| 307 | unlock_page(page); |
| 308 | } |
| 309 | ctx->aio_ring_file = file; |
| 310 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) |
| 311 | / sizeof(struct io_event); |
| 312 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 313 | ctx->ring_pages = ctx->internal_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | if (nr_pages > AIO_RING_PAGES) { |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 315 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
| 316 | GFP_KERNEL); |
| 317 | if (!ctx->ring_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | } |
| 320 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 321 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
| 322 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 323 | |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 324 | down_write(&mm->mmap_sem); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 325 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
| 326 | PROT_READ | PROT_WRITE, |
| 327 | MAP_SHARED | MAP_POPULATE, 0, &populate); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 328 | if (IS_ERR((void *)ctx->mmap_base)) { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 329 | up_write(&mm->mmap_sem); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 330 | ctx->mmap_size = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | aio_free_ring(ctx); |
| 332 | return -EAGAIN; |
| 333 | } |
| 334 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 335 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
Benjamin LaHaise | d6c355c | 2013-09-09 11:57:59 -0400 | [diff] [blame] | 336 | |
| 337 | /* We must do this while still holding mmap_sem for write, as we |
| 338 | * need to be protected against userspace attempting to mremap() |
| 339 | * or munmap() the ring buffer. |
| 340 | */ |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 341 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, |
| 342 | 1, 0, ctx->ring_pages, NULL); |
Benjamin LaHaise | d6c355c | 2013-09-09 11:57:59 -0400 | [diff] [blame] | 343 | |
| 344 | /* Dropping the reference here is safe as the page cache will hold |
| 345 | * onto the pages for us. It is also required so that page migration |
| 346 | * can unmap the pages and get the right reference count. |
| 347 | */ |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 348 | for (i = 0; i < ctx->nr_pages; i++) |
| 349 | put_page(ctx->ring_pages[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | |
Benjamin LaHaise | d6c355c | 2013-09-09 11:57:59 -0400 | [diff] [blame] | 351 | up_write(&mm->mmap_sem); |
| 352 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 353 | if (unlikely(ctx->nr_pages != nr_pages)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | aio_free_ring(ctx); |
| 355 | return -EAGAIN; |
| 356 | } |
| 357 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 358 | ctx->user_id = ctx->mmap_base; |
| 359 | ctx->nr_events = nr_events; /* trusted copy */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 360 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 361 | ring = kmap_atomic(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | ring->nr = nr_events; /* user copy */ |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 363 | ring->id = ~0U; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | ring->head = ring->tail = 0; |
| 365 | ring->magic = AIO_RING_MAGIC; |
| 366 | ring->compat_features = AIO_RING_COMPAT_FEATURES; |
| 367 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; |
| 368 | ring->header_length = sizeof(struct aio_ring); |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 369 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 370 | flush_dcache_page(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | |
| 372 | return 0; |
| 373 | } |
| 374 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
| 376 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) |
| 377 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) |
| 378 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 379 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
| 380 | { |
| 381 | struct kioctx *ctx = req->ki_ctx; |
| 382 | unsigned long flags; |
| 383 | |
| 384 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 385 | |
| 386 | if (!req->ki_list.next) |
| 387 | list_add(&req->ki_list, &ctx->active_reqs); |
| 388 | |
| 389 | req->ki_cancel = cancel; |
| 390 | |
| 391 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
| 392 | } |
| 393 | EXPORT_SYMBOL(kiocb_set_cancel_fn); |
| 394 | |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 395 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb) |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 396 | { |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 397 | kiocb_cancel_fn *old, *cancel; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 398 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 399 | /* |
| 400 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it |
| 401 | * actually has a cancel function, hence the cmpxchg() |
| 402 | */ |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 403 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 404 | cancel = ACCESS_ONCE(kiocb->ki_cancel); |
| 405 | do { |
| 406 | if (!cancel || cancel == KIOCB_CANCELLED) |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 407 | return -EINVAL; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 408 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 409 | old = cancel; |
| 410 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); |
| 411 | } while (cancel != old); |
| 412 | |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 413 | return cancel(kiocb); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 414 | } |
| 415 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 416 | static void free_ioctx(struct work_struct *work) |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 417 | { |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 418 | struct kioctx *ctx = container_of(work, struct kioctx, free_work); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 419 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 420 | pr_debug("freeing %p\n", ctx); |
| 421 | |
| 422 | aio_free_ring(ctx); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 423 | free_percpu(ctx->cpu); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 424 | kmem_cache_free(kioctx_cachep, ctx); |
| 425 | } |
| 426 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 427 | static void free_ioctx_reqs(struct percpu_ref *ref) |
| 428 | { |
| 429 | struct kioctx *ctx = container_of(ref, struct kioctx, reqs); |
| 430 | |
| 431 | INIT_WORK(&ctx->free_work, free_ioctx); |
| 432 | schedule_work(&ctx->free_work); |
| 433 | } |
| 434 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 435 | /* |
| 436 | * When this function runs, the kioctx has been removed from the "hash table" |
| 437 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - |
| 438 | * now it's safe to cancel any that need to be. |
| 439 | */ |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 440 | static void free_ioctx_users(struct percpu_ref *ref) |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 441 | { |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 442 | struct kioctx *ctx = container_of(ref, struct kioctx, users); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 443 | struct kiocb *req; |
| 444 | |
| 445 | spin_lock_irq(&ctx->ctx_lock); |
| 446 | |
| 447 | while (!list_empty(&ctx->active_reqs)) { |
| 448 | req = list_first_entry(&ctx->active_reqs, |
| 449 | struct kiocb, ki_list); |
| 450 | |
| 451 | list_del_init(&req->ki_list); |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 452 | kiocb_cancel(ctx, req); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 453 | } |
| 454 | |
| 455 | spin_unlock_irq(&ctx->ctx_lock); |
| 456 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 457 | percpu_ref_kill(&ctx->reqs); |
| 458 | percpu_ref_put(&ctx->reqs); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 459 | } |
| 460 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 461 | static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) |
| 462 | { |
| 463 | unsigned i, new_nr; |
| 464 | struct kioctx_table *table, *old; |
| 465 | struct aio_ring *ring; |
| 466 | |
| 467 | spin_lock(&mm->ioctx_lock); |
Artem Savkov | d9b2c87 | 2013-09-09 12:29:35 -0400 | [diff] [blame] | 468 | rcu_read_lock(); |
Benjamin LaHaise | 77d30b1 | 2013-08-30 11:12:50 -0400 | [diff] [blame] | 469 | table = rcu_dereference(mm->ioctx_table); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 470 | |
| 471 | while (1) { |
| 472 | if (table) |
| 473 | for (i = 0; i < table->nr; i++) |
| 474 | if (!table->table[i]) { |
| 475 | ctx->id = i; |
| 476 | table->table[i] = ctx; |
Artem Savkov | d9b2c87 | 2013-09-09 12:29:35 -0400 | [diff] [blame] | 477 | rcu_read_unlock(); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 478 | spin_unlock(&mm->ioctx_lock); |
| 479 | |
| 480 | ring = kmap_atomic(ctx->ring_pages[0]); |
| 481 | ring->id = ctx->id; |
| 482 | kunmap_atomic(ring); |
| 483 | return 0; |
| 484 | } |
| 485 | |
| 486 | new_nr = (table ? table->nr : 1) * 4; |
| 487 | |
Artem Savkov | d9b2c87 | 2013-09-09 12:29:35 -0400 | [diff] [blame] | 488 | rcu_read_unlock(); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 489 | spin_unlock(&mm->ioctx_lock); |
| 490 | |
| 491 | table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * |
| 492 | new_nr, GFP_KERNEL); |
| 493 | if (!table) |
| 494 | return -ENOMEM; |
| 495 | |
| 496 | table->nr = new_nr; |
| 497 | |
| 498 | spin_lock(&mm->ioctx_lock); |
Artem Savkov | d9b2c87 | 2013-09-09 12:29:35 -0400 | [diff] [blame] | 499 | rcu_read_lock(); |
Benjamin LaHaise | 77d30b1 | 2013-08-30 11:12:50 -0400 | [diff] [blame] | 500 | old = rcu_dereference(mm->ioctx_table); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 501 | |
| 502 | if (!old) { |
| 503 | rcu_assign_pointer(mm->ioctx_table, table); |
| 504 | } else if (table->nr > old->nr) { |
| 505 | memcpy(table->table, old->table, |
| 506 | old->nr * sizeof(struct kioctx *)); |
| 507 | |
| 508 | rcu_assign_pointer(mm->ioctx_table, table); |
| 509 | kfree_rcu(old, rcu); |
| 510 | } else { |
| 511 | kfree(table); |
| 512 | table = old; |
| 513 | } |
| 514 | } |
| 515 | } |
| 516 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 517 | static void aio_nr_sub(unsigned nr) |
| 518 | { |
| 519 | spin_lock(&aio_nr_lock); |
| 520 | if (WARN_ON(aio_nr - nr > aio_nr)) |
| 521 | aio_nr = 0; |
| 522 | else |
| 523 | aio_nr -= nr; |
| 524 | spin_unlock(&aio_nr_lock); |
| 525 | } |
| 526 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | /* ioctx_alloc |
| 528 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. |
| 529 | */ |
| 530 | static struct kioctx *ioctx_alloc(unsigned nr_events) |
| 531 | { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 532 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | struct kioctx *ctx; |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 534 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 536 | /* |
| 537 | * We keep track of the number of available ringbuffer slots, to prevent |
| 538 | * overflow (reqs_available), and we also use percpu counters for this. |
| 539 | * |
| 540 | * So since up to half the slots might be on other cpu's percpu counters |
| 541 | * and unavailable, double nr_events so userspace sees what they |
| 542 | * expected: additionally, we move req_batch slots to/from percpu |
| 543 | * counters at a time, so make sure that isn't 0: |
| 544 | */ |
| 545 | nr_events = max(nr_events, num_possible_cpus() * 4); |
| 546 | nr_events *= 2; |
| 547 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | /* Prevent overflows */ |
| 549 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || |
| 550 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { |
| 551 | pr_debug("ENOMEM: nr_events too high\n"); |
| 552 | return ERR_PTR(-EINVAL); |
| 553 | } |
| 554 | |
Benjamin LaHaise | 4cd81c3 | 2013-07-30 12:06:37 -0400 | [diff] [blame] | 555 | if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | return ERR_PTR(-EAGAIN); |
| 557 | |
Robert P. J. Day | c376222 | 2007-02-10 01:45:03 -0800 | [diff] [blame] | 558 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | if (!ctx) |
| 560 | return ERR_PTR(-ENOMEM); |
| 561 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | ctx->max_reqs = nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 564 | if (percpu_ref_init(&ctx->users, free_ioctx_users)) |
| 565 | goto err; |
| 566 | |
| 567 | if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs)) |
| 568 | goto err; |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 569 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | spin_lock_init(&ctx->ctx_lock); |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 571 | spin_lock_init(&ctx->completion_lock); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 572 | mutex_init(&ctx->ring_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | init_waitqueue_head(&ctx->wait); |
| 574 | |
| 575 | INIT_LIST_HEAD(&ctx->active_reqs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 577 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
| 578 | if (!ctx->cpu) |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 579 | goto err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 580 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 581 | if (aio_setup_ring(ctx) < 0) |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 582 | goto err; |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 583 | |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 584 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 585 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
Benjamin LaHaise | 6878ea7 | 2013-07-31 10:34:18 -0400 | [diff] [blame] | 586 | if (ctx->req_batch < 1) |
| 587 | ctx->req_batch = 1; |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 588 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | /* limit the number of system wide aios */ |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 590 | spin_lock(&aio_nr_lock); |
Benjamin LaHaise | 4cd81c3 | 2013-07-30 12:06:37 -0400 | [diff] [blame] | 591 | if (aio_nr + nr_events > (aio_max_nr * 2UL) || |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 592 | aio_nr + nr_events < aio_nr) { |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 593 | spin_unlock(&aio_nr_lock); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 594 | err = -EAGAIN; |
| 595 | goto err; |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 596 | } |
| 597 | aio_nr += ctx->max_reqs; |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 598 | spin_unlock(&aio_nr_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 600 | percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ |
| 601 | |
Benjamin LaHaise | da90382 | 2013-08-05 13:21:43 -0400 | [diff] [blame] | 602 | err = ioctx_add_table(ctx, mm); |
| 603 | if (err) |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 604 | goto err_cleanup; |
Benjamin LaHaise | da90382 | 2013-08-05 13:21:43 -0400 | [diff] [blame] | 605 | |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 606 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 607 | ctx, ctx->user_id, mm, ctx->nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | return ctx; |
| 609 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 610 | err_cleanup: |
| 611 | aio_nr_sub(ctx->max_reqs); |
| 612 | err: |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 613 | aio_free_ring(ctx); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 614 | free_percpu(ctx->cpu); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 615 | free_percpu(ctx->reqs.pcpu_count); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 616 | free_percpu(ctx->users.pcpu_count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | kmem_cache_free(kioctx_cachep, ctx); |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 618 | pr_debug("error allocating ioctx %d\n", err); |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 619 | return ERR_PTR(err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | } |
| 621 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 622 | /* kill_ioctx |
| 623 | * Cancels all outstanding aio requests on an aio context. Used |
| 624 | * when the processes owning a context have all exited to encourage |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | * the rapid destruction of the kioctx. |
| 626 | */ |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 627 | static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | { |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 629 | if (!atomic_xchg(&ctx->dead, 1)) { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 630 | struct kioctx_table *table; |
| 631 | |
| 632 | spin_lock(&mm->ioctx_lock); |
Artem Savkov | d9b2c87 | 2013-09-09 12:29:35 -0400 | [diff] [blame] | 633 | rcu_read_lock(); |
Benjamin LaHaise | 77d30b1 | 2013-08-30 11:12:50 -0400 | [diff] [blame] | 634 | table = rcu_dereference(mm->ioctx_table); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 635 | |
| 636 | WARN_ON(ctx != table->table[ctx->id]); |
| 637 | table->table[ctx->id] = NULL; |
Artem Savkov | d9b2c87 | 2013-09-09 12:29:35 -0400 | [diff] [blame] | 638 | rcu_read_unlock(); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 639 | spin_unlock(&mm->ioctx_lock); |
| 640 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 641 | /* percpu_ref_kill() will do the necessary call_rcu() */ |
| 642 | wake_up_all(&ctx->wait); |
Al Viro | 06af121 | 2012-03-20 16:26:24 -0400 | [diff] [blame] | 643 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 644 | /* |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 645 | * It'd be more correct to do this in free_ioctx(), after all |
| 646 | * the outstanding kiocbs have finished - but by then io_destroy |
| 647 | * has already returned, so io_setup() could potentially return |
| 648 | * -EAGAIN with no ioctxs actually in use (as far as userspace |
| 649 | * could tell). |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 650 | */ |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 651 | aio_nr_sub(ctx->max_reqs); |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 652 | |
| 653 | if (ctx->mmap_size) |
| 654 | vm_munmap(ctx->mmap_base, ctx->mmap_size); |
| 655 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 656 | percpu_ref_kill(&ctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | } |
| 659 | |
| 660 | /* wait_on_sync_kiocb: |
| 661 | * Waits on the given sync kiocb to complete. |
| 662 | */ |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 663 | ssize_t wait_on_sync_kiocb(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | { |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 665 | while (!req->ki_ctx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | set_current_state(TASK_UNINTERRUPTIBLE); |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 667 | if (req->ki_ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | break; |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 669 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | } |
| 671 | __set_current_state(TASK_RUNNING); |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 672 | return req->ki_user_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 674 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 676 | /* |
| 677 | * exit_aio: called when the last user of mm goes away. At this point, there is |
| 678 | * no way for any new requests to be submited or any of the io_* syscalls to be |
| 679 | * called on the context. |
| 680 | * |
| 681 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on |
| 682 | * them. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 684 | void exit_aio(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 686 | struct kioctx_table *table; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 687 | struct kioctx *ctx; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 688 | unsigned i = 0; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 689 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 690 | while (1) { |
| 691 | rcu_read_lock(); |
| 692 | table = rcu_dereference(mm->ioctx_table); |
| 693 | |
| 694 | do { |
| 695 | if (!table || i >= table->nr) { |
| 696 | rcu_read_unlock(); |
| 697 | rcu_assign_pointer(mm->ioctx_table, NULL); |
| 698 | if (table) |
| 699 | kfree(table); |
| 700 | return; |
| 701 | } |
| 702 | |
| 703 | ctx = table->table[i++]; |
| 704 | } while (!ctx); |
| 705 | |
| 706 | rcu_read_unlock(); |
| 707 | |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 708 | /* |
| 709 | * We don't need to bother with munmap() here - |
| 710 | * exit_mmap(mm) is coming and it'll unmap everything. |
| 711 | * Since aio_free_ring() uses non-zero ->mmap_size |
| 712 | * as indicator that it needs to unmap the area, |
| 713 | * just set it to 0; aio_free_ring() is the only |
| 714 | * place that uses ->mmap_size, so it's safe. |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 715 | */ |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 716 | ctx->mmap_size = 0; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 717 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 718 | kill_ioctx(mm, ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 719 | } |
| 720 | } |
| 721 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 722 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
| 723 | { |
| 724 | struct kioctx_cpu *kcpu; |
| 725 | |
| 726 | preempt_disable(); |
| 727 | kcpu = this_cpu_ptr(ctx->cpu); |
| 728 | |
| 729 | kcpu->reqs_available += nr; |
| 730 | while (kcpu->reqs_available >= ctx->req_batch * 2) { |
| 731 | kcpu->reqs_available -= ctx->req_batch; |
| 732 | atomic_add(ctx->req_batch, &ctx->reqs_available); |
| 733 | } |
| 734 | |
| 735 | preempt_enable(); |
| 736 | } |
| 737 | |
| 738 | static bool get_reqs_available(struct kioctx *ctx) |
| 739 | { |
| 740 | struct kioctx_cpu *kcpu; |
| 741 | bool ret = false; |
| 742 | |
| 743 | preempt_disable(); |
| 744 | kcpu = this_cpu_ptr(ctx->cpu); |
| 745 | |
| 746 | if (!kcpu->reqs_available) { |
| 747 | int old, avail = atomic_read(&ctx->reqs_available); |
| 748 | |
| 749 | do { |
| 750 | if (avail < ctx->req_batch) |
| 751 | goto out; |
| 752 | |
| 753 | old = avail; |
| 754 | avail = atomic_cmpxchg(&ctx->reqs_available, |
| 755 | avail, avail - ctx->req_batch); |
| 756 | } while (avail != old); |
| 757 | |
| 758 | kcpu->reqs_available += ctx->req_batch; |
| 759 | } |
| 760 | |
| 761 | ret = true; |
| 762 | kcpu->reqs_available--; |
| 763 | out: |
| 764 | preempt_enable(); |
| 765 | return ret; |
| 766 | } |
| 767 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | /* aio_get_req |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 769 | * Allocate a slot for an aio request. |
| 770 | * Returns NULL if no requests are free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | */ |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 772 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 773 | { |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 774 | struct kiocb *req; |
| 775 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 776 | if (!get_reqs_available(ctx)) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 777 | return NULL; |
| 778 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 779 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | if (unlikely(!req)) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 781 | goto out_put; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 783 | percpu_ref_get(&ctx->reqs); |
| 784 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | req->ki_ctx = ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | return req; |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 787 | out_put: |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 788 | put_reqs_available(ctx, 1); |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 789 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | } |
| 791 | |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 792 | static void kiocb_free(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | { |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 794 | if (req->ki_filp) |
| 795 | fput(req->ki_filp); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 796 | if (req->ki_eventfd != NULL) |
| 797 | eventfd_ctx_put(req->ki_eventfd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | kmem_cache_free(kiocb_cachep, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | } |
| 800 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 801 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 803 | struct aio_ring __user *ring = (void __user *)ctx_id; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 804 | struct mm_struct *mm = current->mm; |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 805 | struct kioctx *ctx, *ret = NULL; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 806 | struct kioctx_table *table; |
| 807 | unsigned id; |
| 808 | |
| 809 | if (get_user(id, &ring->id)) |
| 810 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 811 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 812 | rcu_read_lock(); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 813 | table = rcu_dereference(mm->ioctx_table); |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 814 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 815 | if (!table || id >= table->nr) |
| 816 | goto out; |
| 817 | |
| 818 | ctx = table->table[id]; |
Benjamin LaHaise | f30d704 | 2013-08-07 18:23:48 -0400 | [diff] [blame] | 819 | if (ctx && ctx->user_id == ctx_id) { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 820 | percpu_ref_get(&ctx->users); |
| 821 | ret = ctx; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 822 | } |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 823 | out: |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 824 | rcu_read_unlock(); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 825 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | } |
| 827 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | /* aio_complete |
| 829 | * Called when the io request on the given iocb is complete. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 831 | void aio_complete(struct kiocb *iocb, long res, long res2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 832 | { |
| 833 | struct kioctx *ctx = iocb->ki_ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | struct aio_ring *ring; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 835 | struct io_event *ev_page, *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | unsigned long flags; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 837 | unsigned tail, pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 839 | /* |
| 840 | * Special case handling for sync iocbs: |
| 841 | * - events go directly into the iocb for fast handling |
| 842 | * - the sync task with the iocb in its stack holds the single iocb |
| 843 | * ref, no other paths have a way to get another ref |
| 844 | * - the sync task helpfully left a reference to itself in the iocb |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 845 | */ |
| 846 | if (is_sync_kiocb(iocb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 847 | iocb->ki_user_data = res; |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 848 | smp_wmb(); |
| 849 | iocb->ki_ctx = ERR_PTR(-EXDEV); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | wake_up_process(iocb->ki_obj.tsk); |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 851 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | } |
| 853 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 854 | if (iocb->ki_list.next) { |
| 855 | unsigned long flags; |
| 856 | |
| 857 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 858 | list_del(&iocb->ki_list); |
| 859 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
| 860 | } |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 861 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 862 | /* |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 863 | * Add a completion event to the ring buffer. Must be done holding |
Tang Chen | 4b30f07 | 2013-07-03 15:09:16 -0700 | [diff] [blame] | 864 | * ctx->completion_lock to prevent other code from messing with the tail |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 865 | * pointer since we might be called from irq context. |
| 866 | */ |
| 867 | spin_lock_irqsave(&ctx->completion_lock, flags); |
| 868 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 869 | tail = ctx->tail; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 870 | pos = tail + AIO_EVENTS_OFFSET; |
| 871 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 872 | if (++tail >= ctx->nr_events) |
Ken Chen | 4bf69b2 | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 873 | tail = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 874 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 875 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 876 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
| 877 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 878 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
| 879 | event->data = iocb->ki_user_data; |
| 880 | event->res = res; |
| 881 | event->res2 = res2; |
| 882 | |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 883 | kunmap_atomic(ev_page); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 884 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 885 | |
| 886 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 887 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
| 888 | res, res2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | |
| 890 | /* after flagging the request as done, we |
| 891 | * must never even look at it again |
| 892 | */ |
| 893 | smp_wmb(); /* make event visible before updating tail */ |
| 894 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 895 | ctx->tail = tail; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 896 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 897 | ring = kmap_atomic(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 898 | ring->tail = tail; |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 899 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 900 | flush_dcache_page(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 902 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
| 903 | |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 904 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 905 | |
| 906 | /* |
| 907 | * Check if the user asked us to deliver the result through an |
| 908 | * eventfd. The eventfd_signal() function is safe to be called |
| 909 | * from IRQ context. |
| 910 | */ |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 911 | if (iocb->ki_eventfd != NULL) |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 912 | eventfd_signal(iocb->ki_eventfd, 1); |
| 913 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 914 | /* everything turned out well, dispose of the aiocb. */ |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 915 | kiocb_free(iocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | |
Quentin Barnes | 6cb2a21 | 2008-03-19 17:00:39 -0700 | [diff] [blame] | 917 | /* |
| 918 | * We have to order our ring_info tail store above and test |
| 919 | * of the wait list below outside the wait lock. This is |
| 920 | * like in wake_up_bit() where clearing a bit has to be |
| 921 | * ordered with the unlocked test. |
| 922 | */ |
| 923 | smp_mb(); |
| 924 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | if (waitqueue_active(&ctx->wait)) |
| 926 | wake_up(&ctx->wait); |
| 927 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 928 | percpu_ref_put(&ctx->reqs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 930 | EXPORT_SYMBOL(aio_complete); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 931 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 932 | /* aio_read_events |
| 933 | * Pull an event off of the ioctx's event ring. Returns the number of |
| 934 | * events fetched |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | */ |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 936 | static long aio_read_events_ring(struct kioctx *ctx, |
| 937 | struct io_event __user *event, long nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | struct aio_ring *ring; |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 940 | unsigned head, tail, pos; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 941 | long ret = 0; |
| 942 | int copy_ret; |
| 943 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 944 | mutex_lock(&ctx->ring_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 945 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 946 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 947 | head = ring->head; |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 948 | tail = ring->tail; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 949 | kunmap_atomic(ring); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 951 | pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 952 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 953 | if (head == tail) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | goto out; |
| 955 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 956 | while (ret < nr) { |
| 957 | long avail; |
| 958 | struct io_event *ev; |
| 959 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 960 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 961 | avail = (head <= tail ? tail : ctx->nr_events) - head; |
| 962 | if (head == tail) |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 963 | break; |
| 964 | |
| 965 | avail = min(avail, nr - ret); |
| 966 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - |
| 967 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); |
| 968 | |
| 969 | pos = head + AIO_EVENTS_OFFSET; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 970 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 971 | pos %= AIO_EVENTS_PER_PAGE; |
| 972 | |
| 973 | ev = kmap(page); |
| 974 | copy_ret = copy_to_user(event + ret, ev + pos, |
| 975 | sizeof(*ev) * avail); |
| 976 | kunmap(page); |
| 977 | |
| 978 | if (unlikely(copy_ret)) { |
| 979 | ret = -EFAULT; |
| 980 | goto out; |
| 981 | } |
| 982 | |
| 983 | ret += avail; |
| 984 | head += avail; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 985 | head %= ctx->nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 986 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 988 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 989 | ring->head = head; |
Zhao Hongjiang | 91d80a8 | 2013-04-26 11:03:53 +0800 | [diff] [blame] | 990 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 991 | flush_dcache_page(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 992 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 993 | pr_debug("%li h%u t%u\n", ret, head, tail); |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 994 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 995 | put_reqs_available(ctx, ret); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 996 | out: |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 997 | mutex_unlock(&ctx->ring_lock); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 998 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | return ret; |
| 1000 | } |
| 1001 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1002 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
| 1003 | struct io_event __user *event, long *i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1004 | { |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1005 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1006 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1007 | if (ret > 0) |
| 1008 | *i += ret; |
| 1009 | |
| 1010 | if (unlikely(atomic_read(&ctx->dead))) |
| 1011 | ret = -EINVAL; |
| 1012 | |
| 1013 | if (!*i) |
| 1014 | *i = ret; |
| 1015 | |
| 1016 | return ret < 0 || *i >= min_nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | } |
| 1018 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1019 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | struct io_event __user *event, |
| 1021 | struct timespec __user *timeout) |
| 1022 | { |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1023 | ktime_t until = { .tv64 = KTIME_MAX }; |
| 1024 | long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1025 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | if (timeout) { |
| 1027 | struct timespec ts; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1028 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1030 | return -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1032 | until = timespec_to_ktime(ts); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1033 | } |
| 1034 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1035 | /* |
| 1036 | * Note that aio_read_events() is being called as the conditional - i.e. |
| 1037 | * we're calling it after prepare_to_wait() has set task state to |
| 1038 | * TASK_INTERRUPTIBLE. |
| 1039 | * |
| 1040 | * But aio_read_events() can block, and if it blocks it's going to flip |
| 1041 | * the task state back to TASK_RUNNING. |
| 1042 | * |
| 1043 | * This should be ok, provided it doesn't flip the state back to |
| 1044 | * TASK_RUNNING and return 0 too much - that causes us to spin. That |
| 1045 | * will only happen if the mutex_lock() call blocks, and we then find |
| 1046 | * the ringbuffer empty. So in practice we should be ok, but it's |
| 1047 | * something to be aware of when touching this code. |
| 1048 | */ |
| 1049 | wait_event_interruptible_hrtimeout(ctx->wait, |
| 1050 | aio_read_events(ctx, min_nr, nr, event, &ret), until); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1052 | if (!ret && signal_pending(current)) |
| 1053 | ret = -EINTR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1055 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | } |
| 1057 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | /* sys_io_setup: |
| 1059 | * Create an aio_context capable of receiving at least nr_events. |
| 1060 | * ctxp must not point to an aio_context that already exists, and |
| 1061 | * must be initialized to 0 prior to the call. On successful |
| 1062 | * creation of the aio_context, *ctxp is filled in with the resulting |
| 1063 | * handle. May fail with -EINVAL if *ctxp is not initialized, |
| 1064 | * if the specified nr_events exceeds internal limits. May fail |
| 1065 | * with -EAGAIN if the specified nr_events exceeds the user's limit |
| 1066 | * of available events. May fail with -ENOMEM if insufficient kernel |
| 1067 | * resources are available. May fail with -EFAULT if an invalid |
| 1068 | * pointer is passed for ctxp. Will fail with -ENOSYS if not |
| 1069 | * implemented. |
| 1070 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1071 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | { |
| 1073 | struct kioctx *ioctx = NULL; |
| 1074 | unsigned long ctx; |
| 1075 | long ret; |
| 1076 | |
| 1077 | ret = get_user(ctx, ctxp); |
| 1078 | if (unlikely(ret)) |
| 1079 | goto out; |
| 1080 | |
| 1081 | ret = -EINVAL; |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 1082 | if (unlikely(ctx || nr_events == 0)) { |
| 1083 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", |
| 1084 | ctx, nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | goto out; |
| 1086 | } |
| 1087 | |
| 1088 | ioctx = ioctx_alloc(nr_events); |
| 1089 | ret = PTR_ERR(ioctx); |
| 1090 | if (!IS_ERR(ioctx)) { |
| 1091 | ret = put_user(ioctx->user_id, ctxp); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 1092 | if (ret) |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 1093 | kill_ioctx(current->mm, ioctx); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1094 | percpu_ref_put(&ioctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | } |
| 1096 | |
| 1097 | out: |
| 1098 | return ret; |
| 1099 | } |
| 1100 | |
| 1101 | /* sys_io_destroy: |
| 1102 | * Destroy the aio_context specified. May cancel any outstanding |
| 1103 | * AIOs and block on completion. Will fail with -ENOSYS if not |
Satoru Takeuchi | 642b512 | 2010-08-05 11:23:11 -0700 | [diff] [blame] | 1104 | * implemented. May fail with -EINVAL if the context pointed to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | * is invalid. |
| 1106 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1107 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | { |
| 1109 | struct kioctx *ioctx = lookup_ioctx(ctx); |
| 1110 | if (likely(NULL != ioctx)) { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 1111 | kill_ioctx(current->mm, ioctx); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1112 | percpu_ref_put(&ioctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | return 0; |
| 1114 | } |
| 1115 | pr_debug("EINVAL: io_destroy: invalid context id\n"); |
| 1116 | return -EINVAL; |
| 1117 | } |
| 1118 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1119 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
| 1120 | unsigned long, loff_t); |
| 1121 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1122 | static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb, |
| 1123 | int rw, char __user *buf, |
| 1124 | unsigned long *nr_segs, |
| 1125 | struct iovec **iovec, |
| 1126 | bool compat) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1127 | { |
| 1128 | ssize_t ret; |
| 1129 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1130 | *nr_segs = kiocb->ki_nbytes; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1131 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1132 | #ifdef CONFIG_COMPAT |
| 1133 | if (compat) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1134 | ret = compat_rw_copy_check_uvector(rw, |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1135 | (struct compat_iovec __user *)buf, |
| 1136 | *nr_segs, 1, *iovec, iovec); |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1137 | else |
| 1138 | #endif |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1139 | ret = rw_copy_check_uvector(rw, |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1140 | (struct iovec __user *)buf, |
| 1141 | *nr_segs, 1, *iovec, iovec); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1142 | if (ret < 0) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1143 | return ret; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1144 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1145 | /* ki_nbytes now reflect bytes instead of segs */ |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1146 | kiocb->ki_nbytes = ret; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1147 | return 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1148 | } |
| 1149 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1150 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb, |
| 1151 | int rw, char __user *buf, |
| 1152 | unsigned long *nr_segs, |
| 1153 | struct iovec *iovec) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1154 | { |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1155 | if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes))) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1156 | return -EFAULT; |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1157 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1158 | iovec->iov_base = buf; |
| 1159 | iovec->iov_len = kiocb->ki_nbytes; |
| 1160 | *nr_segs = 1; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1161 | return 0; |
| 1162 | } |
| 1163 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1164 | /* |
| 1165 | * aio_setup_iocb: |
| 1166 | * Performs the initial checks and aio retry method |
| 1167 | * setup for the kiocb at the time of io submission. |
| 1168 | */ |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1169 | static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode, |
| 1170 | char __user *buf, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | { |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1172 | struct file *file = req->ki_filp; |
| 1173 | ssize_t ret; |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1174 | unsigned long nr_segs; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1175 | int rw; |
| 1176 | fmode_t mode; |
| 1177 | aio_rw_op *rw_op; |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1178 | struct iovec inline_vec, *iovec = &inline_vec; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1179 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1180 | switch (opcode) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1181 | case IOCB_CMD_PREAD: |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1182 | case IOCB_CMD_PREADV: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1183 | mode = FMODE_READ; |
| 1184 | rw = READ; |
| 1185 | rw_op = file->f_op->aio_read; |
| 1186 | goto rw_common; |
| 1187 | |
| 1188 | case IOCB_CMD_PWRITE: |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1189 | case IOCB_CMD_PWRITEV: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1190 | mode = FMODE_WRITE; |
| 1191 | rw = WRITE; |
| 1192 | rw_op = file->f_op->aio_write; |
| 1193 | goto rw_common; |
| 1194 | rw_common: |
| 1195 | if (unlikely(!(file->f_mode & mode))) |
| 1196 | return -EBADF; |
| 1197 | |
| 1198 | if (!rw_op) |
| 1199 | return -EINVAL; |
| 1200 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1201 | ret = (opcode == IOCB_CMD_PREADV || |
| 1202 | opcode == IOCB_CMD_PWRITEV) |
| 1203 | ? aio_setup_vectored_rw(req, rw, buf, &nr_segs, |
| 1204 | &iovec, compat) |
| 1205 | : aio_setup_single_vector(req, rw, buf, &nr_segs, |
| 1206 | iovec); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1207 | if (ret) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1208 | return ret; |
| 1209 | |
| 1210 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1211 | if (ret < 0) { |
| 1212 | if (iovec != &inline_vec) |
| 1213 | kfree(iovec); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1214 | return ret; |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1215 | } |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1216 | |
| 1217 | req->ki_nbytes = ret; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1218 | |
Kent Overstreet | 73a7075 | 2013-05-09 15:03:42 -0700 | [diff] [blame] | 1219 | /* XXX: move/kill - rw_verify_area()? */ |
| 1220 | /* This matches the pread()/pwrite() logic */ |
| 1221 | if (req->ki_pos < 0) { |
| 1222 | ret = -EINVAL; |
| 1223 | break; |
| 1224 | } |
| 1225 | |
| 1226 | if (rw == WRITE) |
| 1227 | file_start_write(file); |
| 1228 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1229 | ret = rw_op(req, iovec, nr_segs, req->ki_pos); |
Kent Overstreet | 73a7075 | 2013-05-09 15:03:42 -0700 | [diff] [blame] | 1230 | |
| 1231 | if (rw == WRITE) |
| 1232 | file_end_write(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1234 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1235 | case IOCB_CMD_FDSYNC: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1236 | if (!file->f_op->aio_fsync) |
| 1237 | return -EINVAL; |
| 1238 | |
| 1239 | ret = file->f_op->aio_fsync(req, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1240 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1241 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | case IOCB_CMD_FSYNC: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1243 | if (!file->f_op->aio_fsync) |
| 1244 | return -EINVAL; |
| 1245 | |
| 1246 | ret = file->f_op->aio_fsync(req, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1247 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1248 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | default: |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1250 | pr_debug("EINVAL: no operation provided\n"); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1251 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | } |
| 1253 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1254 | if (iovec != &inline_vec) |
| 1255 | kfree(iovec); |
| 1256 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1257 | if (ret != -EIOCBQUEUED) { |
| 1258 | /* |
| 1259 | * There's no easy way to restart the syscall since other AIO's |
| 1260 | * may be already running. Just fail this IO with EINTR. |
| 1261 | */ |
| 1262 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || |
| 1263 | ret == -ERESTARTNOHAND || |
| 1264 | ret == -ERESTART_RESTARTBLOCK)) |
| 1265 | ret = -EINTR; |
| 1266 | aio_complete(req, ret, 0); |
| 1267 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | |
| 1269 | return 0; |
| 1270 | } |
| 1271 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 1272 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 1273 | struct iocb *iocb, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | { |
| 1275 | struct kiocb *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | ssize_t ret; |
| 1277 | |
| 1278 | /* enforce forwards compatibility on users */ |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1279 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1280 | pr_debug("EINVAL: reserve field set\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1281 | return -EINVAL; |
| 1282 | } |
| 1283 | |
| 1284 | /* prevent overflows */ |
| 1285 | if (unlikely( |
| 1286 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || |
| 1287 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || |
| 1288 | ((ssize_t)iocb->aio_nbytes < 0) |
| 1289 | )) { |
| 1290 | pr_debug("EINVAL: io_submit: overflow check\n"); |
| 1291 | return -EINVAL; |
| 1292 | } |
| 1293 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1294 | req = aio_get_req(ctx); |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1295 | if (unlikely(!req)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | return -EAGAIN; |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1297 | |
| 1298 | req->ki_filp = fget(iocb->aio_fildes); |
| 1299 | if (unlikely(!req->ki_filp)) { |
| 1300 | ret = -EBADF; |
| 1301 | goto out_put_req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1302 | } |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1303 | |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1304 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
| 1305 | /* |
| 1306 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an |
| 1307 | * instance of the file* now. The file descriptor must be |
| 1308 | * an eventfd() fd, and will be signaled for each completed |
| 1309 | * event using the eventfd_signal() function. |
| 1310 | */ |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 1311 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
Hirofumi Nakagawa | 801678c | 2008-04-29 01:03:09 -0700 | [diff] [blame] | 1312 | if (IS_ERR(req->ki_eventfd)) { |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1313 | ret = PTR_ERR(req->ki_eventfd); |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 1314 | req->ki_eventfd = NULL; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1315 | goto out_put_req; |
| 1316 | } |
| 1317 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1318 | |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1319 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | if (unlikely(ret)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1321 | pr_debug("EFAULT: aio_key\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1322 | goto out_put_req; |
| 1323 | } |
| 1324 | |
| 1325 | req->ki_obj.user = user_iocb; |
| 1326 | req->ki_user_data = iocb->aio_data; |
| 1327 | req->ki_pos = iocb->aio_offset; |
Kent Overstreet | 73a7075 | 2013-05-09 15:03:42 -0700 | [diff] [blame] | 1328 | req->ki_nbytes = iocb->aio_nbytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1330 | ret = aio_run_iocb(req, iocb->aio_lio_opcode, |
| 1331 | (char __user *)(unsigned long)iocb->aio_buf, |
| 1332 | compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | if (ret) |
| 1334 | goto out_put_req; |
| 1335 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1336 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | out_put_req: |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 1338 | put_reqs_available(ctx, 1); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame^] | 1339 | percpu_ref_put(&ctx->reqs); |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 1340 | kiocb_free(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | return ret; |
| 1342 | } |
| 1343 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1344 | long do_io_submit(aio_context_t ctx_id, long nr, |
| 1345 | struct iocb __user *__user *iocbpp, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | { |
| 1347 | struct kioctx *ctx; |
| 1348 | long ret = 0; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1349 | int i = 0; |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1350 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | |
| 1352 | if (unlikely(nr < 0)) |
| 1353 | return -EINVAL; |
| 1354 | |
Jeff Moyer | 75e1c70 | 2010-09-10 14:16:00 -0700 | [diff] [blame] | 1355 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
| 1356 | nr = LONG_MAX/sizeof(*iocbpp); |
| 1357 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1358 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
| 1359 | return -EFAULT; |
| 1360 | |
| 1361 | ctx = lookup_ioctx(ctx_id); |
| 1362 | if (unlikely(!ctx)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1363 | pr_debug("EINVAL: invalid context id\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | return -EINVAL; |
| 1365 | } |
| 1366 | |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1367 | blk_start_plug(&plug); |
| 1368 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | /* |
| 1370 | * AKPM: should this return a partial result if some of the IOs were |
| 1371 | * successfully submitted? |
| 1372 | */ |
| 1373 | for (i=0; i<nr; i++) { |
| 1374 | struct iocb __user *user_iocb; |
| 1375 | struct iocb tmp; |
| 1376 | |
| 1377 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { |
| 1378 | ret = -EFAULT; |
| 1379 | break; |
| 1380 | } |
| 1381 | |
| 1382 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { |
| 1383 | ret = -EFAULT; |
| 1384 | break; |
| 1385 | } |
| 1386 | |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 1387 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | if (ret) |
| 1389 | break; |
| 1390 | } |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1391 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1393 | percpu_ref_put(&ctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | return i ? i : ret; |
| 1395 | } |
| 1396 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1397 | /* sys_io_submit: |
| 1398 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns |
| 1399 | * the number of iocbs queued. May return -EINVAL if the aio_context |
| 1400 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at |
| 1401 | * *iocbpp[0] is not properly initialized, if the operation specified |
| 1402 | * is invalid for the file descriptor in the iocb. May fail with |
| 1403 | * -EFAULT if any of the data structures point to invalid data. May |
| 1404 | * fail with -EBADF if the file descriptor specified in the first |
| 1405 | * iocb is invalid. May fail with -EAGAIN if insufficient resources |
| 1406 | * are available to queue any iocbs. Will return 0 if nr is 0. Will |
| 1407 | * fail with -ENOSYS if not implemented. |
| 1408 | */ |
| 1409 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, |
| 1410 | struct iocb __user * __user *, iocbpp) |
| 1411 | { |
| 1412 | return do_io_submit(ctx_id, nr, iocbpp, 0); |
| 1413 | } |
| 1414 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1415 | /* lookup_kiocb |
| 1416 | * Finds a given iocb for cancellation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1417 | */ |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 1418 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
| 1419 | u32 key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1420 | { |
| 1421 | struct list_head *pos; |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 1422 | |
| 1423 | assert_spin_locked(&ctx->ctx_lock); |
| 1424 | |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1425 | if (key != KIOCB_KEY) |
| 1426 | return NULL; |
| 1427 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1428 | /* TODO: use a hash or array, this sucks. */ |
| 1429 | list_for_each(pos, &ctx->active_reqs) { |
| 1430 | struct kiocb *kiocb = list_kiocb(pos); |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1431 | if (kiocb->ki_obj.user == iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1432 | return kiocb; |
| 1433 | } |
| 1434 | return NULL; |
| 1435 | } |
| 1436 | |
| 1437 | /* sys_io_cancel: |
| 1438 | * Attempts to cancel an iocb previously passed to io_submit. If |
| 1439 | * the operation is successfully cancelled, the resulting event is |
| 1440 | * copied into the memory pointed to by result without being placed |
| 1441 | * into the completion queue and 0 is returned. May fail with |
| 1442 | * -EFAULT if any of the data structures pointed to are invalid. |
| 1443 | * May fail with -EINVAL if aio_context specified by ctx_id is |
| 1444 | * invalid. May fail with -EAGAIN if the iocb specified was not |
| 1445 | * cancelled. Will fail with -ENOSYS if not implemented. |
| 1446 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1447 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
| 1448 | struct io_event __user *, result) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1449 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | struct kioctx *ctx; |
| 1451 | struct kiocb *kiocb; |
| 1452 | u32 key; |
| 1453 | int ret; |
| 1454 | |
| 1455 | ret = get_user(key, &iocb->aio_key); |
| 1456 | if (unlikely(ret)) |
| 1457 | return -EFAULT; |
| 1458 | |
| 1459 | ctx = lookup_ioctx(ctx_id); |
| 1460 | if (unlikely(!ctx)) |
| 1461 | return -EINVAL; |
| 1462 | |
| 1463 | spin_lock_irq(&ctx->ctx_lock); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1464 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | kiocb = lookup_kiocb(ctx, iocb, key); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1466 | if (kiocb) |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 1467 | ret = kiocb_cancel(ctx, kiocb); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1468 | else |
| 1469 | ret = -EINVAL; |
| 1470 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | spin_unlock_irq(&ctx->ctx_lock); |
| 1472 | |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1473 | if (!ret) { |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 1474 | /* |
| 1475 | * The result argument is no longer used - the io_event is |
| 1476 | * always delivered via the ring buffer. -EINPROGRESS indicates |
| 1477 | * cancellation is progress: |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1478 | */ |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 1479 | ret = -EINPROGRESS; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1480 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1481 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1482 | percpu_ref_put(&ctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1483 | |
| 1484 | return ret; |
| 1485 | } |
| 1486 | |
| 1487 | /* io_getevents: |
| 1488 | * 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] | 1489 | * the completion queue for the aio_context specified by ctx_id. If |
| 1490 | * it succeeds, the number of read events is returned. May fail with |
| 1491 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is |
| 1492 | * out of range, if timeout is out of range. May fail with -EFAULT |
| 1493 | * if any of the memory specified is invalid. May return 0 or |
| 1494 | * < min_nr if the timeout specified by timeout has elapsed |
| 1495 | * before sufficient events are available, where timeout == NULL |
| 1496 | * specifies an infinite timeout. Note that the timeout pointed to by |
Jeff Moyer | 6900807 | 2013-05-24 15:55:24 -0700 | [diff] [blame] | 1497 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1499 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
| 1500 | long, min_nr, |
| 1501 | long, nr, |
| 1502 | struct io_event __user *, events, |
| 1503 | struct timespec __user *, timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | { |
| 1505 | struct kioctx *ioctx = lookup_ioctx(ctx_id); |
| 1506 | long ret = -EINVAL; |
| 1507 | |
| 1508 | if (likely(ioctx)) { |
Namhyung Kim | 2e41025 | 2011-01-12 17:01:08 -0800 | [diff] [blame] | 1509 | if (likely(min_nr <= nr && min_nr >= 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1511 | percpu_ref_put(&ioctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1513 | return ret; |
| 1514 | } |