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/migrate.h> |
| 40 | #include <linux/ramfs.h> |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 41 | #include <linux/percpu-refcount.h> |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 42 | #include <linux/mount.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 */ |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 55 | unsigned head; /* Written to by userland or under ring_lock |
| 56 | * mutex by aio_read_events_ring(). */ |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 57 | unsigned tail; |
| 58 | |
| 59 | unsigned magic; |
| 60 | unsigned compat_features; |
| 61 | unsigned incompat_features; |
| 62 | unsigned header_length; /* size of aio_ring */ |
| 63 | |
| 64 | |
| 65 | struct io_event io_events[0]; |
| 66 | }; /* 128 bytes + ring size */ |
| 67 | |
| 68 | #define AIO_RING_PAGES 8 |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 69 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 70 | struct kioctx_table { |
| 71 | struct rcu_head rcu; |
| 72 | unsigned nr; |
| 73 | struct kioctx *table[]; |
| 74 | }; |
| 75 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 76 | struct kioctx_cpu { |
| 77 | unsigned reqs_available; |
| 78 | }; |
| 79 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 80 | struct kioctx { |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 81 | struct percpu_ref users; |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 82 | atomic_t dead; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 83 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 84 | struct percpu_ref reqs; |
| 85 | |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 86 | unsigned long user_id; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 87 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 88 | struct __percpu kioctx_cpu *cpu; |
| 89 | |
| 90 | /* |
| 91 | * For percpu reqs_available, number of slots we move to/from global |
| 92 | * counter at a time: |
| 93 | */ |
| 94 | unsigned req_batch; |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 95 | /* |
| 96 | * This is what userspace passed to io_setup(), it's not used for |
| 97 | * anything but counting against the global max_reqs quota. |
| 98 | * |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 99 | * The real limit is nr_events - 1, which will be larger (see |
Kent Overstreet | 3e845ce | 2013-05-07 16:18:51 -0700 | [diff] [blame] | 100 | * aio_setup_ring()) |
| 101 | */ |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 102 | unsigned max_reqs; |
| 103 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 104 | /* Size of ringbuffer, in units of struct io_event */ |
| 105 | unsigned nr_events; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 106 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 107 | unsigned long mmap_base; |
| 108 | unsigned long mmap_size; |
| 109 | |
| 110 | struct page **ring_pages; |
| 111 | long nr_pages; |
| 112 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 113 | struct work_struct free_work; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 114 | |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 115 | /* |
| 116 | * signals when all in-flight requests are done |
| 117 | */ |
| 118 | struct completion *requests_done; |
| 119 | |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 120 | struct { |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 121 | /* |
| 122 | * This counts the number of available slots in the ringbuffer, |
| 123 | * so we avoid overflowing it: it's decremented (if positive) |
| 124 | * when allocating a kiocb and incremented when the resulting |
| 125 | * io_event is pulled off the ringbuffer. |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 126 | * |
| 127 | * We batch accesses to it with a percpu version. |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 128 | */ |
| 129 | atomic_t reqs_available; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 130 | } ____cacheline_aligned_in_smp; |
| 131 | |
| 132 | struct { |
| 133 | spinlock_t ctx_lock; |
| 134 | struct list_head active_reqs; /* used for cancellation */ |
| 135 | } ____cacheline_aligned_in_smp; |
| 136 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 137 | struct { |
| 138 | struct mutex ring_lock; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 139 | wait_queue_head_t wait; |
| 140 | } ____cacheline_aligned_in_smp; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 141 | |
| 142 | struct { |
| 143 | unsigned tail; |
| 144 | spinlock_t completion_lock; |
Kent Overstreet | 4e23bca | 2013-05-07 16:18:56 -0700 | [diff] [blame] | 145 | } ____cacheline_aligned_in_smp; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 146 | |
| 147 | struct page *internal_pages[AIO_RING_PAGES]; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 148 | struct file *aio_ring_file; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 149 | |
| 150 | unsigned id; |
Kent Overstreet | 4e179bc | 2013-05-07 16:18:33 -0700 | [diff] [blame] | 151 | }; |
| 152 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | /*------ sysctl variables----*/ |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 154 | static DEFINE_SPINLOCK(aio_nr_lock); |
| 155 | unsigned long aio_nr; /* current system wide number of aio requests */ |
| 156 | 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] | 157 | /*----end sysctl variables---*/ |
| 158 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 159 | static struct kmem_cache *kiocb_cachep; |
| 160 | static struct kmem_cache *kioctx_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 162 | static struct vfsmount *aio_mnt; |
| 163 | |
| 164 | static const struct file_operations aio_ring_fops; |
| 165 | static const struct address_space_operations aio_ctx_aops; |
| 166 | |
| 167 | static struct file *aio_private_file(struct kioctx *ctx, loff_t nr_pages) |
| 168 | { |
| 169 | struct qstr this = QSTR_INIT("[aio]", 5); |
| 170 | struct file *file; |
| 171 | struct path path; |
| 172 | struct inode *inode = alloc_anon_inode(aio_mnt->mnt_sb); |
Dan Carpenter | 7f62656 | 2013-11-13 10:49:40 +0300 | [diff] [blame] | 173 | if (IS_ERR(inode)) |
| 174 | return ERR_CAST(inode); |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 175 | |
| 176 | inode->i_mapping->a_ops = &aio_ctx_aops; |
| 177 | inode->i_mapping->private_data = ctx; |
| 178 | inode->i_size = PAGE_SIZE * nr_pages; |
| 179 | |
| 180 | path.dentry = d_alloc_pseudo(aio_mnt->mnt_sb, &this); |
| 181 | if (!path.dentry) { |
| 182 | iput(inode); |
| 183 | return ERR_PTR(-ENOMEM); |
| 184 | } |
| 185 | path.mnt = mntget(aio_mnt); |
| 186 | |
| 187 | d_instantiate(path.dentry, inode); |
| 188 | file = alloc_file(&path, FMODE_READ | FMODE_WRITE, &aio_ring_fops); |
| 189 | if (IS_ERR(file)) { |
| 190 | path_put(&path); |
| 191 | return file; |
| 192 | } |
| 193 | |
| 194 | file->f_flags = O_RDWR; |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 195 | return file; |
| 196 | } |
| 197 | |
| 198 | static struct dentry *aio_mount(struct file_system_type *fs_type, |
| 199 | int flags, const char *dev_name, void *data) |
| 200 | { |
| 201 | static const struct dentry_operations ops = { |
| 202 | .d_dname = simple_dname, |
| 203 | }; |
Gu Zheng | 8dc4379 | 2014-07-23 18:03:52 +0800 | [diff] [blame] | 204 | return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC); |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 205 | } |
| 206 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | /* aio_setup |
| 208 | * Creates the slab caches used by the aio routines, panic on |
| 209 | * failure as this is done early during the boot sequence. |
| 210 | */ |
| 211 | static int __init aio_setup(void) |
| 212 | { |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 213 | static struct file_system_type aio_fs = { |
| 214 | .name = "aio", |
| 215 | .mount = aio_mount, |
| 216 | .kill_sb = kill_anon_super, |
| 217 | }; |
| 218 | aio_mnt = kern_mount(&aio_fs); |
| 219 | if (IS_ERR(aio_mnt)) |
| 220 | panic("Failed to create aio fs mount."); |
| 221 | |
Christoph Lameter | 0a31bd5 | 2007-05-06 14:49:57 -0700 | [diff] [blame] | 222 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
| 223 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 225 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | |
| 227 | return 0; |
| 228 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 229 | __initcall(aio_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 231 | static void put_aio_ring_file(struct kioctx *ctx) |
| 232 | { |
| 233 | struct file *aio_ring_file = ctx->aio_ring_file; |
| 234 | if (aio_ring_file) { |
| 235 | truncate_setsize(aio_ring_file->f_inode, 0); |
| 236 | |
| 237 | /* Prevent further access to the kioctx from migratepages */ |
| 238 | spin_lock(&aio_ring_file->f_inode->i_mapping->private_lock); |
| 239 | aio_ring_file->f_inode->i_mapping->private_data = NULL; |
| 240 | ctx->aio_ring_file = NULL; |
| 241 | spin_unlock(&aio_ring_file->f_inode->i_mapping->private_lock); |
| 242 | |
| 243 | fput(aio_ring_file); |
| 244 | } |
| 245 | } |
| 246 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | static void aio_free_ring(struct kioctx *ctx) |
| 248 | { |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 249 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 250 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 251 | /* Disconnect the kiotx from the ring file. This prevents future |
| 252 | * accesses to the kioctx from page migration. |
| 253 | */ |
| 254 | put_aio_ring_file(ctx); |
| 255 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 256 | for (i = 0; i < ctx->nr_pages; i++) { |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 257 | struct page *page; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 258 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, |
| 259 | page_count(ctx->ring_pages[i])); |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 260 | page = ctx->ring_pages[i]; |
| 261 | if (!page) |
| 262 | continue; |
| 263 | ctx->ring_pages[i] = NULL; |
| 264 | put_page(page); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 265 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | |
Sasha Levin | ddb8c45 | 2013-11-19 17:33:03 -0500 | [diff] [blame] | 267 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) { |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 268 | kfree(ctx->ring_pages); |
Sasha Levin | ddb8c45 | 2013-11-19 17:33:03 -0500 | [diff] [blame] | 269 | ctx->ring_pages = NULL; |
| 270 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | } |
| 272 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 273 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) |
| 274 | { |
| 275 | vma->vm_ops = &generic_file_vm_ops; |
| 276 | return 0; |
| 277 | } |
| 278 | |
| 279 | static const struct file_operations aio_ring_fops = { |
| 280 | .mmap = aio_ring_mmap, |
| 281 | }; |
| 282 | |
| 283 | static int aio_set_page_dirty(struct page *page) |
| 284 | { |
| 285 | return 0; |
| 286 | } |
| 287 | |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 288 | #if IS_ENABLED(CONFIG_MIGRATION) |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 289 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
| 290 | struct page *old, enum migrate_mode mode) |
| 291 | { |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 292 | struct kioctx *ctx; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 293 | unsigned long flags; |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 294 | pgoff_t idx; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 295 | int rc; |
| 296 | |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 297 | rc = 0; |
| 298 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 299 | /* mapping->private_lock here protects against the kioctx teardown. */ |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 300 | spin_lock(&mapping->private_lock); |
| 301 | ctx = mapping->private_data; |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 302 | if (!ctx) { |
| 303 | rc = -EINVAL; |
| 304 | goto out; |
| 305 | } |
| 306 | |
| 307 | /* The ring_lock mutex. The prevents aio_read_events() from writing |
| 308 | * to the ring's head, and prevents page migration from mucking in |
| 309 | * a partially initialized kiotx. |
| 310 | */ |
| 311 | if (!mutex_trylock(&ctx->ring_lock)) { |
| 312 | rc = -EAGAIN; |
| 313 | goto out; |
| 314 | } |
| 315 | |
| 316 | idx = old->index; |
| 317 | if (idx < (pgoff_t)ctx->nr_pages) { |
| 318 | /* Make sure the old page hasn't already been changed */ |
| 319 | if (ctx->ring_pages[idx] != old) |
| 320 | rc = -EAGAIN; |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 321 | } else |
| 322 | rc = -EINVAL; |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 323 | |
| 324 | if (rc != 0) |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 325 | goto out_unlock; |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 326 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 327 | /* Writeback must be complete */ |
| 328 | BUG_ON(PageWriteback(old)); |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 329 | get_page(new); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 330 | |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 331 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 332 | if (rc != MIGRATEPAGE_SUCCESS) { |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 333 | put_page(new); |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 334 | goto out_unlock; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 335 | } |
| 336 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 337 | /* Take completion_lock to prevent other writes to the ring buffer |
| 338 | * while the old page is copied to the new. This prevents new |
| 339 | * events from being lost. |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 340 | */ |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 341 | spin_lock_irqsave(&ctx->completion_lock, flags); |
| 342 | migrate_page_copy(new, old); |
| 343 | BUG_ON(ctx->ring_pages[idx] != old); |
| 344 | ctx->ring_pages[idx] = new; |
| 345 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
| 346 | |
| 347 | /* The old page is no longer accessible. */ |
| 348 | put_page(old); |
| 349 | |
| 350 | out_unlock: |
| 351 | mutex_unlock(&ctx->ring_lock); |
| 352 | out: |
Benjamin LaHaise | 5e9ae2e | 2013-09-26 20:34:51 -0400 | [diff] [blame] | 353 | spin_unlock(&mapping->private_lock); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 354 | return rc; |
| 355 | } |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 356 | #endif |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 357 | |
| 358 | static const struct address_space_operations aio_ctx_aops = { |
| 359 | .set_page_dirty = aio_set_page_dirty, |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 360 | #if IS_ENABLED(CONFIG_MIGRATION) |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 361 | .migratepage = aio_migratepage, |
Benjamin LaHaise | 0c45355 | 2013-07-17 09:34:24 -0400 | [diff] [blame] | 362 | #endif |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 363 | }; |
| 364 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | static int aio_setup_ring(struct kioctx *ctx) |
| 366 | { |
| 367 | struct aio_ring *ring; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | unsigned nr_events = ctx->max_reqs; |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 369 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 3dc9acb | 2013-12-20 05:11:12 +0900 | [diff] [blame] | 370 | unsigned long size, unused; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | int nr_pages; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 372 | int i; |
| 373 | struct file *file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | |
| 375 | /* Compensate for the ring buffer's head/tail overlap entry */ |
| 376 | nr_events += 2; /* 1 is required, 2 for good luck */ |
| 377 | |
| 378 | size = sizeof(struct aio_ring); |
| 379 | size += sizeof(struct io_event) * nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 380 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 381 | nr_pages = PFN_UP(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | if (nr_pages < 0) |
| 383 | return -EINVAL; |
| 384 | |
Benjamin LaHaise | 71ad749 | 2013-09-17 10:18:25 -0400 | [diff] [blame] | 385 | file = aio_private_file(ctx, nr_pages); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 386 | if (IS_ERR(file)) { |
| 387 | ctx->aio_ring_file = NULL; |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 388 | return -ENOMEM; |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 389 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 391 | ctx->aio_ring_file = file; |
| 392 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) |
| 393 | / sizeof(struct io_event); |
| 394 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 395 | ctx->ring_pages = ctx->internal_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | if (nr_pages > AIO_RING_PAGES) { |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 397 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
| 398 | GFP_KERNEL); |
Gu Zheng | d1b9432 | 2013-12-04 18:19:06 +0800 | [diff] [blame] | 399 | if (!ctx->ring_pages) { |
| 400 | put_aio_ring_file(ctx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | return -ENOMEM; |
Gu Zheng | d1b9432 | 2013-12-04 18:19:06 +0800 | [diff] [blame] | 402 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | } |
| 404 | |
Linus Torvalds | 3dc9acb | 2013-12-20 05:11:12 +0900 | [diff] [blame] | 405 | for (i = 0; i < nr_pages; i++) { |
| 406 | struct page *page; |
| 407 | page = find_or_create_page(file->f_inode->i_mapping, |
| 408 | i, GFP_HIGHUSER | __GFP_ZERO); |
| 409 | if (!page) |
| 410 | break; |
| 411 | pr_debug("pid(%d) page[%d]->count=%d\n", |
| 412 | current->pid, i, page_count(page)); |
| 413 | SetPageUptodate(page); |
| 414 | SetPageDirty(page); |
| 415 | unlock_page(page); |
| 416 | |
| 417 | ctx->ring_pages[i] = page; |
| 418 | } |
| 419 | ctx->nr_pages = i; |
| 420 | |
| 421 | if (unlikely(i != nr_pages)) { |
| 422 | aio_free_ring(ctx); |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 423 | return -ENOMEM; |
Linus Torvalds | 3dc9acb | 2013-12-20 05:11:12 +0900 | [diff] [blame] | 424 | } |
| 425 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 426 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
| 427 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 428 | |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 429 | down_write(&mm->mmap_sem); |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 430 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
| 431 | PROT_READ | PROT_WRITE, |
Linus Torvalds | 3dc9acb | 2013-12-20 05:11:12 +0900 | [diff] [blame] | 432 | MAP_SHARED, 0, &unused); |
| 433 | up_write(&mm->mmap_sem); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 434 | if (IS_ERR((void *)ctx->mmap_base)) { |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 435 | ctx->mmap_size = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | aio_free_ring(ctx); |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 437 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | } |
| 439 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 440 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
Benjamin LaHaise | d6c355c | 2013-09-09 11:57:59 -0400 | [diff] [blame] | 441 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 442 | ctx->user_id = ctx->mmap_base; |
| 443 | ctx->nr_events = nr_events; /* trusted copy */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 445 | ring = kmap_atomic(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | ring->nr = nr_events; /* user copy */ |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 447 | ring->id = ~0U; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 448 | ring->head = ring->tail = 0; |
| 449 | ring->magic = AIO_RING_MAGIC; |
| 450 | ring->compat_features = AIO_RING_COMPAT_FEATURES; |
| 451 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; |
| 452 | ring->header_length = sizeof(struct aio_ring); |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 453 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 454 | flush_dcache_page(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 455 | |
| 456 | return 0; |
| 457 | } |
| 458 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
| 460 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) |
| 461 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) |
| 462 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 463 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
| 464 | { |
| 465 | struct kioctx *ctx = req->ki_ctx; |
| 466 | unsigned long flags; |
| 467 | |
| 468 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 469 | |
| 470 | if (!req->ki_list.next) |
| 471 | list_add(&req->ki_list, &ctx->active_reqs); |
| 472 | |
| 473 | req->ki_cancel = cancel; |
| 474 | |
| 475 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
| 476 | } |
| 477 | EXPORT_SYMBOL(kiocb_set_cancel_fn); |
| 478 | |
Fabian Frederick | d52a8f9 | 2014-04-22 07:26:58 +0200 | [diff] [blame] | 479 | static int kiocb_cancel(struct kiocb *kiocb) |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 480 | { |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 481 | kiocb_cancel_fn *old, *cancel; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 482 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 483 | /* |
| 484 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it |
| 485 | * actually has a cancel function, hence the cmpxchg() |
| 486 | */ |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 487 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 488 | cancel = ACCESS_ONCE(kiocb->ki_cancel); |
| 489 | do { |
| 490 | if (!cancel || cancel == KIOCB_CANCELLED) |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 491 | return -EINVAL; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 492 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 493 | old = cancel; |
| 494 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); |
| 495 | } while (cancel != old); |
| 496 | |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 497 | return cancel(kiocb); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 498 | } |
| 499 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 500 | static void free_ioctx(struct work_struct *work) |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 501 | { |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 502 | struct kioctx *ctx = container_of(work, struct kioctx, free_work); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 503 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 504 | pr_debug("freeing %p\n", ctx); |
| 505 | |
| 506 | aio_free_ring(ctx); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 507 | free_percpu(ctx->cpu); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 508 | kmem_cache_free(kioctx_cachep, ctx); |
| 509 | } |
| 510 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 511 | static void free_ioctx_reqs(struct percpu_ref *ref) |
| 512 | { |
| 513 | struct kioctx *ctx = container_of(ref, struct kioctx, reqs); |
| 514 | |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 515 | /* At this point we know that there are no any in-flight requests */ |
| 516 | if (ctx->requests_done) |
| 517 | complete(ctx->requests_done); |
| 518 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 519 | INIT_WORK(&ctx->free_work, free_ioctx); |
| 520 | schedule_work(&ctx->free_work); |
| 521 | } |
| 522 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 523 | /* |
| 524 | * When this function runs, the kioctx has been removed from the "hash table" |
| 525 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - |
| 526 | * now it's safe to cancel any that need to be. |
| 527 | */ |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 528 | static void free_ioctx_users(struct percpu_ref *ref) |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 529 | { |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 530 | struct kioctx *ctx = container_of(ref, struct kioctx, users); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 531 | struct kiocb *req; |
| 532 | |
| 533 | spin_lock_irq(&ctx->ctx_lock); |
| 534 | |
| 535 | while (!list_empty(&ctx->active_reqs)) { |
| 536 | req = list_first_entry(&ctx->active_reqs, |
| 537 | struct kiocb, ki_list); |
| 538 | |
| 539 | list_del_init(&req->ki_list); |
Fabian Frederick | d52a8f9 | 2014-04-22 07:26:58 +0200 | [diff] [blame] | 540 | kiocb_cancel(req); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 541 | } |
| 542 | |
| 543 | spin_unlock_irq(&ctx->ctx_lock); |
| 544 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 545 | percpu_ref_kill(&ctx->reqs); |
| 546 | percpu_ref_put(&ctx->reqs); |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 547 | } |
| 548 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 549 | static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) |
| 550 | { |
| 551 | unsigned i, new_nr; |
| 552 | struct kioctx_table *table, *old; |
| 553 | struct aio_ring *ring; |
| 554 | |
| 555 | spin_lock(&mm->ioctx_lock); |
Oleg Nesterov | 855ef0d | 2014-04-30 16:16:36 +0200 | [diff] [blame] | 556 | table = rcu_dereference_raw(mm->ioctx_table); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 557 | |
| 558 | while (1) { |
| 559 | if (table) |
| 560 | for (i = 0; i < table->nr; i++) |
| 561 | if (!table->table[i]) { |
| 562 | ctx->id = i; |
| 563 | table->table[i] = ctx; |
| 564 | spin_unlock(&mm->ioctx_lock); |
| 565 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 566 | /* While kioctx setup is in progress, |
| 567 | * we are protected from page migration |
| 568 | * changes ring_pages by ->ring_lock. |
| 569 | */ |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 570 | ring = kmap_atomic(ctx->ring_pages[0]); |
| 571 | ring->id = ctx->id; |
| 572 | kunmap_atomic(ring); |
| 573 | return 0; |
| 574 | } |
| 575 | |
| 576 | new_nr = (table ? table->nr : 1) * 4; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 577 | spin_unlock(&mm->ioctx_lock); |
| 578 | |
| 579 | table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * |
| 580 | new_nr, GFP_KERNEL); |
| 581 | if (!table) |
| 582 | return -ENOMEM; |
| 583 | |
| 584 | table->nr = new_nr; |
| 585 | |
| 586 | spin_lock(&mm->ioctx_lock); |
Oleg Nesterov | 855ef0d | 2014-04-30 16:16:36 +0200 | [diff] [blame] | 587 | old = rcu_dereference_raw(mm->ioctx_table); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 588 | |
| 589 | if (!old) { |
| 590 | rcu_assign_pointer(mm->ioctx_table, table); |
| 591 | } else if (table->nr > old->nr) { |
| 592 | memcpy(table->table, old->table, |
| 593 | old->nr * sizeof(struct kioctx *)); |
| 594 | |
| 595 | rcu_assign_pointer(mm->ioctx_table, table); |
| 596 | kfree_rcu(old, rcu); |
| 597 | } else { |
| 598 | kfree(table); |
| 599 | table = old; |
| 600 | } |
| 601 | } |
| 602 | } |
| 603 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 604 | static void aio_nr_sub(unsigned nr) |
| 605 | { |
| 606 | spin_lock(&aio_nr_lock); |
| 607 | if (WARN_ON(aio_nr - nr > aio_nr)) |
| 608 | aio_nr = 0; |
| 609 | else |
| 610 | aio_nr -= nr; |
| 611 | spin_unlock(&aio_nr_lock); |
| 612 | } |
| 613 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | /* ioctx_alloc |
| 615 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. |
| 616 | */ |
| 617 | static struct kioctx *ioctx_alloc(unsigned nr_events) |
| 618 | { |
Zach Brown | 41003a7 | 2013-05-07 16:18:25 -0700 | [diff] [blame] | 619 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | struct kioctx *ctx; |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 621 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 623 | /* |
| 624 | * We keep track of the number of available ringbuffer slots, to prevent |
| 625 | * overflow (reqs_available), and we also use percpu counters for this. |
| 626 | * |
| 627 | * So since up to half the slots might be on other cpu's percpu counters |
| 628 | * and unavailable, double nr_events so userspace sees what they |
| 629 | * expected: additionally, we move req_batch slots to/from percpu |
| 630 | * counters at a time, so make sure that isn't 0: |
| 631 | */ |
| 632 | nr_events = max(nr_events, num_possible_cpus() * 4); |
| 633 | nr_events *= 2; |
| 634 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | /* Prevent overflows */ |
| 636 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || |
| 637 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { |
| 638 | pr_debug("ENOMEM: nr_events too high\n"); |
| 639 | return ERR_PTR(-EINVAL); |
| 640 | } |
| 641 | |
Benjamin LaHaise | 4cd81c3 | 2013-07-30 12:06:37 -0400 | [diff] [blame] | 642 | if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | return ERR_PTR(-EAGAIN); |
| 644 | |
Robert P. J. Day | c376222 | 2007-02-10 01:45:03 -0800 | [diff] [blame] | 645 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | if (!ctx) |
| 647 | return ERR_PTR(-ENOMEM); |
| 648 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | ctx->max_reqs = nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 651 | spin_lock_init(&ctx->ctx_lock); |
| 652 | spin_lock_init(&ctx->completion_lock); |
| 653 | mutex_init(&ctx->ring_lock); |
| 654 | /* Protect against page migration throughout kiotx setup by keeping |
| 655 | * the ring_lock mutex held until setup is complete. */ |
| 656 | mutex_lock(&ctx->ring_lock); |
| 657 | init_waitqueue_head(&ctx->wait); |
| 658 | |
| 659 | INIT_LIST_HEAD(&ctx->active_reqs); |
| 660 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 661 | if (percpu_ref_init(&ctx->users, free_ioctx_users)) |
| 662 | goto err; |
| 663 | |
| 664 | if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs)) |
| 665 | goto err; |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 666 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 667 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
| 668 | if (!ctx->cpu) |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 669 | goto err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 671 | err = aio_setup_ring(ctx); |
| 672 | if (err < 0) |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 673 | goto err; |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 674 | |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 675 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 676 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
Benjamin LaHaise | 6878ea7 | 2013-07-31 10:34:18 -0400 | [diff] [blame] | 677 | if (ctx->req_batch < 1) |
| 678 | ctx->req_batch = 1; |
Kent Overstreet | 34e83fc | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 679 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | /* limit the number of system wide aios */ |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 681 | spin_lock(&aio_nr_lock); |
Benjamin LaHaise | 4cd81c3 | 2013-07-30 12:06:37 -0400 | [diff] [blame] | 682 | if (aio_nr + nr_events > (aio_max_nr * 2UL) || |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 683 | aio_nr + nr_events < aio_nr) { |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 684 | spin_unlock(&aio_nr_lock); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 685 | err = -EAGAIN; |
Gu Zheng | d1b9432 | 2013-12-04 18:19:06 +0800 | [diff] [blame] | 686 | goto err_ctx; |
Al Viro | 2dd542b | 2012-03-10 23:10:35 -0500 | [diff] [blame] | 687 | } |
| 688 | aio_nr += ctx->max_reqs; |
Al Viro | 9fa1cb3 | 2012-03-10 23:14:05 -0500 | [diff] [blame] | 689 | spin_unlock(&aio_nr_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | |
Benjamin LaHaise | 1881686 | 2013-12-21 15:49:28 -0500 | [diff] [blame] | 691 | percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ |
| 692 | percpu_ref_get(&ctx->reqs); /* free_ioctx_users() will drop this */ |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 693 | |
Benjamin LaHaise | da90382 | 2013-08-05 13:21:43 -0400 | [diff] [blame] | 694 | err = ioctx_add_table(ctx, mm); |
| 695 | if (err) |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 696 | goto err_cleanup; |
Benjamin LaHaise | da90382 | 2013-08-05 13:21:43 -0400 | [diff] [blame] | 697 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 698 | /* Release the ring_lock mutex now that all setup is complete. */ |
| 699 | mutex_unlock(&ctx->ring_lock); |
| 700 | |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 701 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 702 | ctx, ctx->user_id, mm, ctx->nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | return ctx; |
| 704 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 705 | err_cleanup: |
| 706 | aio_nr_sub(ctx->max_reqs); |
Gu Zheng | d1b9432 | 2013-12-04 18:19:06 +0800 | [diff] [blame] | 707 | err_ctx: |
| 708 | aio_free_ring(ctx); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 709 | err: |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 710 | mutex_unlock(&ctx->ring_lock); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 711 | free_percpu(ctx->cpu); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 712 | free_percpu(ctx->reqs.pcpu_count); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 713 | free_percpu(ctx->users.pcpu_count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | kmem_cache_free(kioctx_cachep, ctx); |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 715 | pr_debug("error allocating ioctx %d\n", err); |
Al Viro | e23754f | 2012-03-06 14:33:22 -0500 | [diff] [blame] | 716 | return ERR_PTR(err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | } |
| 718 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 719 | /* kill_ioctx |
| 720 | * Cancels all outstanding aio requests on an aio context. Used |
| 721 | * when the processes owning a context have all exited to encourage |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | * the rapid destruction of the kioctx. |
| 723 | */ |
Benjamin LaHaise | fb2d448 | 2014-04-29 12:45:17 -0400 | [diff] [blame] | 724 | static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx, |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 725 | struct completion *requests_done) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | { |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 727 | struct kioctx_table *table; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 728 | |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 729 | if (atomic_xchg(&ctx->dead, 1)) |
| 730 | return -EINVAL; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 731 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 732 | |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 733 | spin_lock(&mm->ioctx_lock); |
Oleg Nesterov | 855ef0d | 2014-04-30 16:16:36 +0200 | [diff] [blame] | 734 | table = rcu_dereference_raw(mm->ioctx_table); |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 735 | WARN_ON(ctx != table->table[ctx->id]); |
| 736 | table->table[ctx->id] = NULL; |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 737 | spin_unlock(&mm->ioctx_lock); |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 738 | |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 739 | /* percpu_ref_kill() will do the necessary call_rcu() */ |
| 740 | wake_up_all(&ctx->wait); |
Kent Overstreet | 4fcc712 | 2013-06-12 14:04:59 -0700 | [diff] [blame] | 741 | |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 742 | /* |
| 743 | * It'd be more correct to do this in free_ioctx(), after all |
| 744 | * the outstanding kiocbs have finished - but by then io_destroy |
| 745 | * has already returned, so io_setup() could potentially return |
| 746 | * -EAGAIN with no ioctxs actually in use (as far as userspace |
| 747 | * could tell). |
| 748 | */ |
| 749 | aio_nr_sub(ctx->max_reqs); |
Benjamin LaHaise | fb2d448 | 2014-04-29 12:45:17 -0400 | [diff] [blame] | 750 | |
Benjamin LaHaise | fa88b6f | 2014-04-29 12:55:48 -0400 | [diff] [blame] | 751 | if (ctx->mmap_size) |
| 752 | vm_munmap(ctx->mmap_base, ctx->mmap_size); |
| 753 | |
| 754 | ctx->requests_done = requests_done; |
| 755 | percpu_ref_kill(&ctx->users); |
| 756 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | } |
| 758 | |
| 759 | /* wait_on_sync_kiocb: |
| 760 | * Waits on the given sync kiocb to complete. |
| 761 | */ |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 762 | ssize_t wait_on_sync_kiocb(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | { |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 764 | while (!req->ki_ctx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | set_current_state(TASK_UNINTERRUPTIBLE); |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 766 | if (req->ki_ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | break; |
Jeff Moyer | 41d10da | 2007-10-16 23:27:20 -0700 | [diff] [blame] | 768 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | } |
| 770 | __set_current_state(TASK_RUNNING); |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 771 | return req->ki_user_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 773 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | |
Kent Overstreet | 36f5588 | 2013-05-07 16:18:41 -0700 | [diff] [blame] | 775 | /* |
| 776 | * exit_aio: called when the last user of mm goes away. At this point, there is |
| 777 | * no way for any new requests to be submited or any of the io_* syscalls to be |
| 778 | * called on the context. |
| 779 | * |
| 780 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on |
| 781 | * them. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 783 | void exit_aio(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | { |
Oleg Nesterov | 4b70ac5 | 2014-04-30 19:02:48 +0200 | [diff] [blame] | 785 | struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table); |
| 786 | int i; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 787 | |
Oleg Nesterov | 4b70ac5 | 2014-04-30 19:02:48 +0200 | [diff] [blame] | 788 | if (!table) |
| 789 | return; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 790 | |
Oleg Nesterov | 4b70ac5 | 2014-04-30 19:02:48 +0200 | [diff] [blame] | 791 | for (i = 0; i < table->nr; ++i) { |
| 792 | struct kioctx *ctx = table->table[i]; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 793 | |
Oleg Nesterov | 4b70ac5 | 2014-04-30 19:02:48 +0200 | [diff] [blame] | 794 | if (!ctx) |
| 795 | continue; |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 796 | /* |
Oleg Nesterov | 4b70ac5 | 2014-04-30 19:02:48 +0200 | [diff] [blame] | 797 | * We don't need to bother with munmap() here - exit_mmap(mm) |
| 798 | * is coming and it'll unmap everything. And we simply can't, |
| 799 | * this is not necessarily our ->mm. |
| 800 | * Since kill_ioctx() uses non-zero ->mmap_size as indicator |
| 801 | * that it needs to unmap the area, just set it to 0. |
Al Viro | 936af15 | 2012-04-20 21:49:41 -0400 | [diff] [blame] | 802 | */ |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 803 | ctx->mmap_size = 0; |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 804 | kill_ioctx(mm, ctx, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 805 | } |
Oleg Nesterov | 4b70ac5 | 2014-04-30 19:02:48 +0200 | [diff] [blame] | 806 | |
| 807 | RCU_INIT_POINTER(mm->ioctx_table, NULL); |
| 808 | kfree(table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | } |
| 810 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 811 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
| 812 | { |
| 813 | struct kioctx_cpu *kcpu; |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 814 | unsigned long flags; |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 815 | |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 816 | local_irq_save(flags); |
Benjamin LaHaise | be6fb45 | 2014-07-22 09:56:56 -0400 | [diff] [blame] | 817 | kcpu = this_cpu_ptr(ctx->cpu); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 818 | kcpu->reqs_available += nr; |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 819 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 820 | while (kcpu->reqs_available >= ctx->req_batch * 2) { |
| 821 | kcpu->reqs_available -= ctx->req_batch; |
| 822 | atomic_add(ctx->req_batch, &ctx->reqs_available); |
| 823 | } |
| 824 | |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 825 | local_irq_restore(flags); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 826 | } |
| 827 | |
| 828 | static bool get_reqs_available(struct kioctx *ctx) |
| 829 | { |
| 830 | struct kioctx_cpu *kcpu; |
| 831 | bool ret = false; |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 832 | unsigned long flags; |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 833 | |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 834 | local_irq_save(flags); |
Benjamin LaHaise | be6fb45 | 2014-07-22 09:56:56 -0400 | [diff] [blame] | 835 | kcpu = this_cpu_ptr(ctx->cpu); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 836 | if (!kcpu->reqs_available) { |
| 837 | int old, avail = atomic_read(&ctx->reqs_available); |
| 838 | |
| 839 | do { |
| 840 | if (avail < ctx->req_batch) |
| 841 | goto out; |
| 842 | |
| 843 | old = avail; |
| 844 | avail = atomic_cmpxchg(&ctx->reqs_available, |
| 845 | avail, avail - ctx->req_batch); |
| 846 | } while (avail != old); |
| 847 | |
| 848 | kcpu->reqs_available += ctx->req_batch; |
| 849 | } |
| 850 | |
| 851 | ret = true; |
| 852 | kcpu->reqs_available--; |
| 853 | out: |
Benjamin LaHaise | 263782c | 2014-07-14 12:49:26 -0400 | [diff] [blame] | 854 | local_irq_restore(flags); |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 855 | return ret; |
| 856 | } |
| 857 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | /* aio_get_req |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 859 | * Allocate a slot for an aio request. |
| 860 | * Returns NULL if no requests are free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | */ |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 862 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | { |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 864 | struct kiocb *req; |
| 865 | |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 866 | if (!get_reqs_available(ctx)) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 867 | return NULL; |
| 868 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 869 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 870 | if (unlikely(!req)) |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 871 | goto out_put; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 872 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 873 | percpu_ref_get(&ctx->reqs); |
| 874 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | req->ki_ctx = ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | return req; |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 877 | out_put: |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 878 | put_reqs_available(ctx, 1); |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 879 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 880 | } |
| 881 | |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 882 | static void kiocb_free(struct kiocb *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 883 | { |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 884 | if (req->ki_filp) |
| 885 | fput(req->ki_filp); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 886 | if (req->ki_eventfd != NULL) |
| 887 | eventfd_ctx_put(req->ki_eventfd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | kmem_cache_free(kiocb_cachep, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | } |
| 890 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 891 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 892 | { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 893 | struct aio_ring __user *ring = (void __user *)ctx_id; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 894 | struct mm_struct *mm = current->mm; |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 895 | struct kioctx *ctx, *ret = NULL; |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 896 | struct kioctx_table *table; |
| 897 | unsigned id; |
| 898 | |
| 899 | if (get_user(id, &ring->id)) |
| 900 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 902 | rcu_read_lock(); |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 903 | table = rcu_dereference(mm->ioctx_table); |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 904 | |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 905 | if (!table || id >= table->nr) |
| 906 | goto out; |
| 907 | |
| 908 | ctx = table->table[id]; |
Benjamin LaHaise | f30d704 | 2013-08-07 18:23:48 -0400 | [diff] [blame] | 909 | if (ctx && ctx->user_id == ctx_id) { |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 910 | percpu_ref_get(&ctx->users); |
| 911 | ret = ctx; |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 912 | } |
Benjamin LaHaise | db446a0 | 2013-07-30 12:54:40 -0400 | [diff] [blame] | 913 | out: |
Jens Axboe | abf137d | 2008-12-09 08:11:22 +0100 | [diff] [blame] | 914 | rcu_read_unlock(); |
Jeff Moyer | 65c2449 | 2009-03-18 17:04:21 -0700 | [diff] [blame] | 915 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | } |
| 917 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | /* aio_complete |
| 919 | * Called when the io request on the given iocb is complete. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 920 | */ |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 921 | void aio_complete(struct kiocb *iocb, long res, long res2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | { |
| 923 | struct kioctx *ctx = iocb->ki_ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | struct aio_ring *ring; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 925 | struct io_event *ev_page, *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 926 | unsigned long flags; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 927 | unsigned tail, pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | |
Zach Brown | 20dcae3 | 2005-11-13 16:07:33 -0800 | [diff] [blame] | 929 | /* |
| 930 | * Special case handling for sync iocbs: |
| 931 | * - events go directly into the iocb for fast handling |
| 932 | * - the sync task with the iocb in its stack holds the single iocb |
| 933 | * ref, no other paths have a way to get another ref |
| 934 | * - the sync task helpfully left a reference to itself in the iocb |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | */ |
| 936 | if (is_sync_kiocb(iocb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 937 | iocb->ki_user_data = res; |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 938 | smp_wmb(); |
| 939 | iocb->ki_ctx = ERR_PTR(-EXDEV); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 940 | wake_up_process(iocb->ki_obj.tsk); |
Kent Overstreet | 2d68449 | 2013-05-07 16:18:29 -0700 | [diff] [blame] | 941 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | } |
| 943 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 944 | if (iocb->ki_list.next) { |
| 945 | unsigned long flags; |
| 946 | |
| 947 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
| 948 | list_del(&iocb->ki_list); |
| 949 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
| 950 | } |
Kent Overstreet | 11599eb | 2013-05-07 16:18:39 -0700 | [diff] [blame] | 951 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | /* |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 953 | * Add a completion event to the ring buffer. Must be done holding |
Tang Chen | 4b30f07 | 2013-07-03 15:09:16 -0700 | [diff] [blame] | 954 | * ctx->completion_lock to prevent other code from messing with the tail |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 955 | * pointer since we might be called from irq context. |
| 956 | */ |
| 957 | spin_lock_irqsave(&ctx->completion_lock, flags); |
| 958 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 959 | tail = ctx->tail; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 960 | pos = tail + AIO_EVENTS_OFFSET; |
| 961 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 962 | if (++tail >= ctx->nr_events) |
Ken Chen | 4bf69b2 | 2005-05-01 08:59:15 -0700 | [diff] [blame] | 963 | tail = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 965 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 966 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
| 967 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 968 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
| 969 | event->data = iocb->ki_user_data; |
| 970 | event->res = res; |
| 971 | event->res2 = res2; |
| 972 | |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 973 | kunmap_atomic(ev_page); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 974 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 975 | |
| 976 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 977 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
| 978 | res, res2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | |
| 980 | /* after flagging the request as done, we |
| 981 | * must never even look at it again |
| 982 | */ |
| 983 | smp_wmb(); /* make event visible before updating tail */ |
| 984 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 985 | ctx->tail = tail; |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 986 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 987 | ring = kmap_atomic(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 988 | ring->tail = tail; |
Cong Wang | e8e3c3d | 2011-11-25 23:14:27 +0800 | [diff] [blame] | 989 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 990 | flush_dcache_page(ctx->ring_pages[0]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 991 | |
Kent Overstreet | 0460fef | 2013-05-07 16:18:49 -0700 | [diff] [blame] | 992 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
| 993 | |
Kent Overstreet | 21b4020 | 2013-05-07 16:18:47 -0700 | [diff] [blame] | 994 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 995 | |
| 996 | /* |
| 997 | * Check if the user asked us to deliver the result through an |
| 998 | * eventfd. The eventfd_signal() function is safe to be called |
| 999 | * from IRQ context. |
| 1000 | */ |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 1001 | if (iocb->ki_eventfd != NULL) |
Davide Libenzi | 8d1c98b | 2008-04-10 21:29:19 -0700 | [diff] [blame] | 1002 | eventfd_signal(iocb->ki_eventfd, 1); |
| 1003 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1004 | /* everything turned out well, dispose of the aiocb. */ |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 1005 | kiocb_free(iocb); |
Benjamin LaHaise | f8567a3 | 2014-06-24 13:12:55 -0400 | [diff] [blame] | 1006 | put_reqs_available(ctx, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | |
Quentin Barnes | 6cb2a21 | 2008-03-19 17:00:39 -0700 | [diff] [blame] | 1008 | /* |
| 1009 | * We have to order our ring_info tail store above and test |
| 1010 | * of the wait list below outside the wait lock. This is |
| 1011 | * like in wake_up_bit() where clearing a bit has to be |
| 1012 | * ordered with the unlocked test. |
| 1013 | */ |
| 1014 | smp_mb(); |
| 1015 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1016 | if (waitqueue_active(&ctx->wait)) |
| 1017 | wake_up(&ctx->wait); |
| 1018 | |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 1019 | percpu_ref_put(&ctx->reqs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | } |
H Hartley Sweeten | 385773e | 2009-09-22 16:43:53 -0700 | [diff] [blame] | 1021 | EXPORT_SYMBOL(aio_complete); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | |
Gu Zheng | 2be4e7d | 2014-07-23 18:03:53 +0800 | [diff] [blame^] | 1023 | /* aio_read_events_ring |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1024 | * Pull an event off of the ioctx's event ring. Returns the number of |
| 1025 | * events fetched |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | */ |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1027 | static long aio_read_events_ring(struct kioctx *ctx, |
| 1028 | struct io_event __user *event, long nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | struct aio_ring *ring; |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 1031 | unsigned head, tail, pos; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1032 | long ret = 0; |
| 1033 | int copy_ret; |
| 1034 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1035 | mutex_lock(&ctx->ring_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | |
Benjamin LaHaise | fa8a53c | 2014-03-28 10:14:45 -0400 | [diff] [blame] | 1037 | /* Access to ->ring_pages here is protected by ctx->ring_lock. */ |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1038 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1039 | head = ring->head; |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 1040 | tail = ring->tail; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1041 | kunmap_atomic(ring); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1042 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 1043 | 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] | 1044 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 1045 | if (head == tail) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | goto out; |
| 1047 | |
Benjamin LaHaise | edfbbf3 | 2014-06-24 13:32:51 -0400 | [diff] [blame] | 1048 | head %= ctx->nr_events; |
| 1049 | tail %= ctx->nr_events; |
| 1050 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1051 | while (ret < nr) { |
| 1052 | long avail; |
| 1053 | struct io_event *ev; |
| 1054 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 1056 | avail = (head <= tail ? tail : ctx->nr_events) - head; |
| 1057 | if (head == tail) |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1058 | break; |
| 1059 | |
| 1060 | avail = min(avail, nr - ret); |
| 1061 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - |
| 1062 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); |
| 1063 | |
| 1064 | pos = head + AIO_EVENTS_OFFSET; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1065 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1066 | pos %= AIO_EVENTS_PER_PAGE; |
| 1067 | |
| 1068 | ev = kmap(page); |
| 1069 | copy_ret = copy_to_user(event + ret, ev + pos, |
| 1070 | sizeof(*ev) * avail); |
| 1071 | kunmap(page); |
| 1072 | |
| 1073 | if (unlikely(copy_ret)) { |
| 1074 | ret = -EFAULT; |
| 1075 | goto out; |
| 1076 | } |
| 1077 | |
| 1078 | ret += avail; |
| 1079 | head += avail; |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1080 | head %= ctx->nr_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1083 | ring = kmap_atomic(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1084 | ring->head = head; |
Zhao Hongjiang | 91d80a8 | 2013-04-26 11:03:53 +0800 | [diff] [blame] | 1085 | kunmap_atomic(ring); |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1086 | flush_dcache_page(ctx->ring_pages[0]); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1087 | |
Kent Overstreet | 5ffac12 | 2013-05-09 15:36:07 -0700 | [diff] [blame] | 1088 | pr_debug("%li h%u t%u\n", ret, head, tail); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1089 | out: |
Kent Overstreet | 58c85dc | 2013-05-07 16:18:55 -0700 | [diff] [blame] | 1090 | mutex_unlock(&ctx->ring_lock); |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1091 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | return ret; |
| 1093 | } |
| 1094 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1095 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
| 1096 | struct io_event __user *event, long *i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | { |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1098 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1099 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1100 | if (ret > 0) |
| 1101 | *i += ret; |
| 1102 | |
| 1103 | if (unlikely(atomic_read(&ctx->dead))) |
| 1104 | ret = -EINVAL; |
| 1105 | |
| 1106 | if (!*i) |
| 1107 | *i = ret; |
| 1108 | |
| 1109 | return ret < 0 || *i >= min_nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | } |
| 1111 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1112 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | struct io_event __user *event, |
| 1114 | struct timespec __user *timeout) |
| 1115 | { |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1116 | ktime_t until = { .tv64 = KTIME_MAX }; |
| 1117 | long ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | if (timeout) { |
| 1120 | struct timespec ts; |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1121 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1122 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1123 | return -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1125 | until = timespec_to_ktime(ts); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | } |
| 1127 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1128 | /* |
| 1129 | * Note that aio_read_events() is being called as the conditional - i.e. |
| 1130 | * we're calling it after prepare_to_wait() has set task state to |
| 1131 | * TASK_INTERRUPTIBLE. |
| 1132 | * |
| 1133 | * But aio_read_events() can block, and if it blocks it's going to flip |
| 1134 | * the task state back to TASK_RUNNING. |
| 1135 | * |
| 1136 | * This should be ok, provided it doesn't flip the state back to |
| 1137 | * TASK_RUNNING and return 0 too much - that causes us to spin. That |
| 1138 | * will only happen if the mutex_lock() call blocks, and we then find |
| 1139 | * the ringbuffer empty. So in practice we should be ok, but it's |
| 1140 | * something to be aware of when touching this code. |
| 1141 | */ |
| 1142 | wait_event_interruptible_hrtimeout(ctx->wait, |
| 1143 | aio_read_events(ctx, min_nr, nr, event, &ret), until); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1144 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1145 | if (!ret && signal_pending(current)) |
| 1146 | ret = -EINTR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1147 | |
Kent Overstreet | a31ad38 | 2013-05-07 16:18:45 -0700 | [diff] [blame] | 1148 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | } |
| 1150 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1151 | /* sys_io_setup: |
| 1152 | * Create an aio_context capable of receiving at least nr_events. |
| 1153 | * ctxp must not point to an aio_context that already exists, and |
| 1154 | * must be initialized to 0 prior to the call. On successful |
| 1155 | * creation of the aio_context, *ctxp is filled in with the resulting |
| 1156 | * handle. May fail with -EINVAL if *ctxp is not initialized, |
| 1157 | * if the specified nr_events exceeds internal limits. May fail |
| 1158 | * with -EAGAIN if the specified nr_events exceeds the user's limit |
| 1159 | * of available events. May fail with -ENOMEM if insufficient kernel |
| 1160 | * resources are available. May fail with -EFAULT if an invalid |
| 1161 | * pointer is passed for ctxp. Will fail with -ENOSYS if not |
| 1162 | * implemented. |
| 1163 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1164 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1165 | { |
| 1166 | struct kioctx *ioctx = NULL; |
| 1167 | unsigned long ctx; |
| 1168 | long ret; |
| 1169 | |
| 1170 | ret = get_user(ctx, ctxp); |
| 1171 | if (unlikely(ret)) |
| 1172 | goto out; |
| 1173 | |
| 1174 | ret = -EINVAL; |
Zach Brown | d55b5fd | 2005-11-07 00:59:31 -0800 | [diff] [blame] | 1175 | if (unlikely(ctx || nr_events == 0)) { |
| 1176 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", |
| 1177 | ctx, nr_events); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | goto out; |
| 1179 | } |
| 1180 | |
| 1181 | ioctx = ioctx_alloc(nr_events); |
| 1182 | ret = PTR_ERR(ioctx); |
| 1183 | if (!IS_ERR(ioctx)) { |
| 1184 | ret = put_user(ioctx->user_id, ctxp); |
Al Viro | a2e1859 | 2012-03-20 16:27:57 -0400 | [diff] [blame] | 1185 | if (ret) |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 1186 | kill_ioctx(current->mm, ioctx, NULL); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1187 | percpu_ref_put(&ioctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | } |
| 1189 | |
| 1190 | out: |
| 1191 | return ret; |
| 1192 | } |
| 1193 | |
| 1194 | /* sys_io_destroy: |
| 1195 | * Destroy the aio_context specified. May cancel any outstanding |
| 1196 | * AIOs and block on completion. Will fail with -ENOSYS if not |
Satoru Takeuchi | 642b512 | 2010-08-05 11:23:11 -0700 | [diff] [blame] | 1197 | * implemented. May fail with -EINVAL if the context pointed to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | * is invalid. |
| 1199 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1200 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1201 | { |
| 1202 | struct kioctx *ioctx = lookup_ioctx(ctx); |
| 1203 | if (likely(NULL != ioctx)) { |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 1204 | struct completion requests_done = |
| 1205 | COMPLETION_INITIALIZER_ONSTACK(requests_done); |
Benjamin LaHaise | fb2d448 | 2014-04-29 12:45:17 -0400 | [diff] [blame] | 1206 | int ret; |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 1207 | |
| 1208 | /* Pass requests_done to kill_ioctx() where it can be set |
| 1209 | * in a thread-safe way. If we try to set it here then we have |
| 1210 | * a race condition if two io_destroy() called simultaneously. |
| 1211 | */ |
Benjamin LaHaise | fb2d448 | 2014-04-29 12:45:17 -0400 | [diff] [blame] | 1212 | ret = kill_ioctx(current->mm, ioctx, &requests_done); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1213 | percpu_ref_put(&ioctx->users); |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 1214 | |
| 1215 | /* Wait until all IO for the context are done. Otherwise kernel |
| 1216 | * keep using user-space buffers even if user thinks the context |
| 1217 | * is destroyed. |
| 1218 | */ |
Benjamin LaHaise | fb2d448 | 2014-04-29 12:45:17 -0400 | [diff] [blame] | 1219 | if (!ret) |
| 1220 | wait_for_completion(&requests_done); |
Anatol Pomozov | e02ba72 | 2014-04-15 11:31:33 -0700 | [diff] [blame] | 1221 | |
Benjamin LaHaise | fb2d448 | 2014-04-29 12:45:17 -0400 | [diff] [blame] | 1222 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1223 | } |
| 1224 | pr_debug("EINVAL: io_destroy: invalid context id\n"); |
| 1225 | return -EINVAL; |
| 1226 | } |
| 1227 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1228 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
| 1229 | unsigned long, loff_t); |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1230 | typedef ssize_t (rw_iter_op)(struct kiocb *, struct iov_iter *); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1231 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1232 | static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb, |
| 1233 | int rw, char __user *buf, |
| 1234 | unsigned long *nr_segs, |
| 1235 | struct iovec **iovec, |
| 1236 | bool compat) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1237 | { |
| 1238 | ssize_t ret; |
| 1239 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1240 | *nr_segs = kiocb->ki_nbytes; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1241 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1242 | #ifdef CONFIG_COMPAT |
| 1243 | if (compat) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1244 | ret = compat_rw_copy_check_uvector(rw, |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1245 | (struct compat_iovec __user *)buf, |
| 1246 | *nr_segs, 1, *iovec, iovec); |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1247 | else |
| 1248 | #endif |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1249 | ret = rw_copy_check_uvector(rw, |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1250 | (struct iovec __user *)buf, |
| 1251 | *nr_segs, 1, *iovec, iovec); |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1252 | if (ret < 0) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1253 | return ret; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1254 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1255 | /* ki_nbytes now reflect bytes instead of segs */ |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1256 | kiocb->ki_nbytes = ret; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1257 | return 0; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1258 | } |
| 1259 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1260 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb, |
| 1261 | int rw, char __user *buf, |
| 1262 | unsigned long *nr_segs, |
| 1263 | struct iovec *iovec) |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1264 | { |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1265 | if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes))) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1266 | return -EFAULT; |
Linus Torvalds | a70b52e | 2012-05-21 16:06:20 -0700 | [diff] [blame] | 1267 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1268 | iovec->iov_base = buf; |
| 1269 | iovec->iov_len = kiocb->ki_nbytes; |
| 1270 | *nr_segs = 1; |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1271 | return 0; |
| 1272 | } |
| 1273 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | /* |
Gu Zheng | 2be4e7d | 2014-07-23 18:03:53 +0800 | [diff] [blame^] | 1275 | * aio_run_iocb: |
| 1276 | * Performs the initial checks and io submission. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | */ |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1278 | static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode, |
| 1279 | char __user *buf, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | { |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1281 | struct file *file = req->ki_filp; |
| 1282 | ssize_t ret; |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1283 | unsigned long nr_segs; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1284 | int rw; |
| 1285 | fmode_t mode; |
| 1286 | aio_rw_op *rw_op; |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1287 | rw_iter_op *iter_op; |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1288 | struct iovec inline_vec, *iovec = &inline_vec; |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1289 | struct iov_iter iter; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1291 | switch (opcode) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | case IOCB_CMD_PREAD: |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1293 | case IOCB_CMD_PREADV: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1294 | mode = FMODE_READ; |
| 1295 | rw = READ; |
| 1296 | rw_op = file->f_op->aio_read; |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1297 | iter_op = file->f_op->read_iter; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1298 | goto rw_common; |
| 1299 | |
| 1300 | case IOCB_CMD_PWRITE: |
Badari Pulavarty | eed4e51 | 2006-09-30 23:28:49 -0700 | [diff] [blame] | 1301 | case IOCB_CMD_PWRITEV: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1302 | mode = FMODE_WRITE; |
| 1303 | rw = WRITE; |
| 1304 | rw_op = file->f_op->aio_write; |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1305 | iter_op = file->f_op->write_iter; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1306 | goto rw_common; |
| 1307 | rw_common: |
| 1308 | if (unlikely(!(file->f_mode & mode))) |
| 1309 | return -EBADF; |
| 1310 | |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1311 | if (!rw_op && !iter_op) |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1312 | return -EINVAL; |
| 1313 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1314 | ret = (opcode == IOCB_CMD_PREADV || |
| 1315 | opcode == IOCB_CMD_PWRITEV) |
| 1316 | ? aio_setup_vectored_rw(req, rw, buf, &nr_segs, |
| 1317 | &iovec, compat) |
| 1318 | : aio_setup_single_vector(req, rw, buf, &nr_segs, |
| 1319 | iovec); |
Leon Yu | 754320d | 2014-05-01 03:31:28 +0000 | [diff] [blame] | 1320 | if (!ret) |
| 1321 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1322 | if (ret < 0) { |
| 1323 | if (iovec != &inline_vec) |
| 1324 | kfree(iovec); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1325 | return ret; |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1326 | } |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1327 | |
| 1328 | req->ki_nbytes = ret; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1329 | |
Kent Overstreet | 73a7075 | 2013-05-09 15:03:42 -0700 | [diff] [blame] | 1330 | /* XXX: move/kill - rw_verify_area()? */ |
| 1331 | /* This matches the pread()/pwrite() logic */ |
| 1332 | if (req->ki_pos < 0) { |
| 1333 | ret = -EINVAL; |
| 1334 | break; |
| 1335 | } |
| 1336 | |
| 1337 | if (rw == WRITE) |
| 1338 | file_start_write(file); |
| 1339 | |
Al Viro | 293bc98 | 2014-02-11 18:37:41 -0500 | [diff] [blame] | 1340 | if (iter_op) { |
| 1341 | iov_iter_init(&iter, rw, iovec, nr_segs, req->ki_nbytes); |
| 1342 | ret = iter_op(req, &iter); |
| 1343 | } else { |
| 1344 | ret = rw_op(req, iovec, nr_segs, req->ki_pos); |
| 1345 | } |
Kent Overstreet | 73a7075 | 2013-05-09 15:03:42 -0700 | [diff] [blame] | 1346 | |
| 1347 | if (rw == WRITE) |
| 1348 | file_end_write(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1350 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | case IOCB_CMD_FDSYNC: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1352 | if (!file->f_op->aio_fsync) |
| 1353 | return -EINVAL; |
| 1354 | |
| 1355 | ret = file->f_op->aio_fsync(req, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1357 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1358 | case IOCB_CMD_FSYNC: |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1359 | if (!file->f_op->aio_fsync) |
| 1360 | return -EINVAL; |
| 1361 | |
| 1362 | ret = file->f_op->aio_fsync(req, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | break; |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1364 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | default: |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1366 | pr_debug("EINVAL: no operation provided\n"); |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1367 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | } |
| 1369 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1370 | if (iovec != &inline_vec) |
| 1371 | kfree(iovec); |
| 1372 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1373 | if (ret != -EIOCBQUEUED) { |
| 1374 | /* |
| 1375 | * There's no easy way to restart the syscall since other AIO's |
| 1376 | * may be already running. Just fail this IO with EINTR. |
| 1377 | */ |
| 1378 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || |
| 1379 | ret == -ERESTARTNOHAND || |
| 1380 | ret == -ERESTART_RESTARTBLOCK)) |
| 1381 | ret = -EINTR; |
| 1382 | aio_complete(req, ret, 0); |
| 1383 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1384 | |
| 1385 | return 0; |
| 1386 | } |
| 1387 | |
Adrian Bunk | d5470b5 | 2008-04-29 00:58:57 -0700 | [diff] [blame] | 1388 | 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] | 1389 | struct iocb *iocb, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1390 | { |
| 1391 | struct kiocb *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | ssize_t ret; |
| 1393 | |
| 1394 | /* enforce forwards compatibility on users */ |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1395 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1396 | pr_debug("EINVAL: reserve field set\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | return -EINVAL; |
| 1398 | } |
| 1399 | |
| 1400 | /* prevent overflows */ |
| 1401 | if (unlikely( |
| 1402 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || |
| 1403 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || |
| 1404 | ((ssize_t)iocb->aio_nbytes < 0) |
| 1405 | )) { |
| 1406 | pr_debug("EINVAL: io_submit: overflow check\n"); |
| 1407 | return -EINVAL; |
| 1408 | } |
| 1409 | |
Kent Overstreet | 41ef4eb | 2013-05-07 16:19:11 -0700 | [diff] [blame] | 1410 | req = aio_get_req(ctx); |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1411 | if (unlikely(!req)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1412 | return -EAGAIN; |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1413 | |
| 1414 | req->ki_filp = fget(iocb->aio_fildes); |
| 1415 | if (unlikely(!req->ki_filp)) { |
| 1416 | ret = -EBADF; |
| 1417 | goto out_put_req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | } |
Kent Overstreet | 1d98ebf | 2013-05-07 16:18:37 -0700 | [diff] [blame] | 1419 | |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1420 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
| 1421 | /* |
| 1422 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an |
| 1423 | * instance of the file* now. The file descriptor must be |
| 1424 | * an eventfd() fd, and will be signaled for each completed |
| 1425 | * event using the eventfd_signal() function. |
| 1426 | */ |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 1427 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
Hirofumi Nakagawa | 801678c | 2008-04-29 01:03:09 -0700 | [diff] [blame] | 1428 | if (IS_ERR(req->ki_eventfd)) { |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1429 | ret = PTR_ERR(req->ki_eventfd); |
Davide Libenzi | 87c3a86 | 2009-03-18 17:04:19 -0700 | [diff] [blame] | 1430 | req->ki_eventfd = NULL; |
Davide Libenzi | 9c3060b | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 1431 | goto out_put_req; |
| 1432 | } |
| 1433 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1434 | |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1435 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1436 | if (unlikely(ret)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1437 | pr_debug("EFAULT: aio_key\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | goto out_put_req; |
| 1439 | } |
| 1440 | |
| 1441 | req->ki_obj.user = user_iocb; |
| 1442 | req->ki_user_data = iocb->aio_data; |
| 1443 | req->ki_pos = iocb->aio_offset; |
Kent Overstreet | 73a7075 | 2013-05-09 15:03:42 -0700 | [diff] [blame] | 1444 | req->ki_nbytes = iocb->aio_nbytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | |
Kent Overstreet | 8bc92af | 2013-02-25 16:36:27 -0800 | [diff] [blame] | 1446 | ret = aio_run_iocb(req, iocb->aio_lio_opcode, |
| 1447 | (char __user *)(unsigned long)iocb->aio_buf, |
| 1448 | compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1449 | if (ret) |
| 1450 | goto out_put_req; |
| 1451 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | out_put_req: |
Kent Overstreet | e1bdd5f | 2013-04-26 10:58:39 +1000 | [diff] [blame] | 1454 | put_reqs_available(ctx, 1); |
Kent Overstreet | e34ecee | 2013-10-10 19:31:47 -0700 | [diff] [blame] | 1455 | percpu_ref_put(&ctx->reqs); |
Kent Overstreet | 57282d8 | 2013-05-13 13:42:52 -0700 | [diff] [blame] | 1456 | kiocb_free(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | return ret; |
| 1458 | } |
| 1459 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1460 | long do_io_submit(aio_context_t ctx_id, long nr, |
| 1461 | struct iocb __user *__user *iocbpp, bool compat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | { |
| 1463 | struct kioctx *ctx; |
| 1464 | long ret = 0; |
Jeff Moyer | 080d676 | 2011-11-02 13:40:10 -0700 | [diff] [blame] | 1465 | int i = 0; |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1466 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | |
| 1468 | if (unlikely(nr < 0)) |
| 1469 | return -EINVAL; |
| 1470 | |
Jeff Moyer | 75e1c70 | 2010-09-10 14:16:00 -0700 | [diff] [blame] | 1471 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
| 1472 | nr = LONG_MAX/sizeof(*iocbpp); |
| 1473 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1474 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
| 1475 | return -EFAULT; |
| 1476 | |
| 1477 | ctx = lookup_ioctx(ctx_id); |
| 1478 | if (unlikely(!ctx)) { |
Kent Overstreet | caf4167 | 2013-05-07 16:18:35 -0700 | [diff] [blame] | 1479 | pr_debug("EINVAL: invalid context id\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | return -EINVAL; |
| 1481 | } |
| 1482 | |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1483 | blk_start_plug(&plug); |
| 1484 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | /* |
| 1486 | * AKPM: should this return a partial result if some of the IOs were |
| 1487 | * successfully submitted? |
| 1488 | */ |
| 1489 | for (i=0; i<nr; i++) { |
| 1490 | struct iocb __user *user_iocb; |
| 1491 | struct iocb tmp; |
| 1492 | |
| 1493 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { |
| 1494 | ret = -EFAULT; |
| 1495 | break; |
| 1496 | } |
| 1497 | |
| 1498 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { |
| 1499 | ret = -EFAULT; |
| 1500 | break; |
| 1501 | } |
| 1502 | |
Kent Overstreet | a1c8eae | 2013-05-07 16:18:53 -0700 | [diff] [blame] | 1503 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | if (ret) |
| 1505 | break; |
| 1506 | } |
Shaohua Li | 9f5b942 | 2010-07-01 07:55:01 +0200 | [diff] [blame] | 1507 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1509 | percpu_ref_put(&ctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | return i ? i : ret; |
| 1511 | } |
| 1512 | |
Jeff Moyer | 9d85cba | 2010-05-26 14:44:26 -0700 | [diff] [blame] | 1513 | /* sys_io_submit: |
| 1514 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns |
| 1515 | * the number of iocbs queued. May return -EINVAL if the aio_context |
| 1516 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at |
| 1517 | * *iocbpp[0] is not properly initialized, if the operation specified |
| 1518 | * is invalid for the file descriptor in the iocb. May fail with |
| 1519 | * -EFAULT if any of the data structures point to invalid data. May |
| 1520 | * fail with -EBADF if the file descriptor specified in the first |
| 1521 | * iocb is invalid. May fail with -EAGAIN if insufficient resources |
| 1522 | * are available to queue any iocbs. Will return 0 if nr is 0. Will |
| 1523 | * fail with -ENOSYS if not implemented. |
| 1524 | */ |
| 1525 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, |
| 1526 | struct iocb __user * __user *, iocbpp) |
| 1527 | { |
| 1528 | return do_io_submit(ctx_id, nr, iocbpp, 0); |
| 1529 | } |
| 1530 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | /* lookup_kiocb |
| 1532 | * Finds a given iocb for cancellation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1533 | */ |
Adrian Bunk | 25ee7e3 | 2005-04-25 08:18:14 -0700 | [diff] [blame] | 1534 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
| 1535 | u32 key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | { |
| 1537 | struct list_head *pos; |
Zach Brown | d00689a | 2005-11-13 16:07:34 -0800 | [diff] [blame] | 1538 | |
| 1539 | assert_spin_locked(&ctx->ctx_lock); |
| 1540 | |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1541 | if (key != KIOCB_KEY) |
| 1542 | return NULL; |
| 1543 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1544 | /* TODO: use a hash or array, this sucks. */ |
| 1545 | list_for_each(pos, &ctx->active_reqs) { |
| 1546 | struct kiocb *kiocb = list_kiocb(pos); |
Kent Overstreet | 8a66089 | 2013-05-07 16:19:10 -0700 | [diff] [blame] | 1547 | if (kiocb->ki_obj.user == iocb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1548 | return kiocb; |
| 1549 | } |
| 1550 | return NULL; |
| 1551 | } |
| 1552 | |
| 1553 | /* sys_io_cancel: |
| 1554 | * Attempts to cancel an iocb previously passed to io_submit. If |
| 1555 | * the operation is successfully cancelled, the resulting event is |
| 1556 | * copied into the memory pointed to by result without being placed |
| 1557 | * into the completion queue and 0 is returned. May fail with |
| 1558 | * -EFAULT if any of the data structures pointed to are invalid. |
| 1559 | * May fail with -EINVAL if aio_context specified by ctx_id is |
| 1560 | * invalid. May fail with -EAGAIN if the iocb specified was not |
| 1561 | * cancelled. Will fail with -ENOSYS if not implemented. |
| 1562 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1563 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
| 1564 | struct io_event __user *, result) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1565 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | struct kioctx *ctx; |
| 1567 | struct kiocb *kiocb; |
| 1568 | u32 key; |
| 1569 | int ret; |
| 1570 | |
| 1571 | ret = get_user(key, &iocb->aio_key); |
| 1572 | if (unlikely(ret)) |
| 1573 | return -EFAULT; |
| 1574 | |
| 1575 | ctx = lookup_ioctx(ctx_id); |
| 1576 | if (unlikely(!ctx)) |
| 1577 | return -EINVAL; |
| 1578 | |
| 1579 | spin_lock_irq(&ctx->ctx_lock); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1580 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1581 | kiocb = lookup_kiocb(ctx, iocb, key); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1582 | if (kiocb) |
Fabian Frederick | d52a8f9 | 2014-04-22 07:26:58 +0200 | [diff] [blame] | 1583 | ret = kiocb_cancel(kiocb); |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1584 | else |
| 1585 | ret = -EINVAL; |
| 1586 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1587 | spin_unlock_irq(&ctx->ctx_lock); |
| 1588 | |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1589 | if (!ret) { |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 1590 | /* |
| 1591 | * The result argument is no longer used - the io_event is |
| 1592 | * always delivered via the ring buffer. -EINPROGRESS indicates |
| 1593 | * cancellation is progress: |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1594 | */ |
Kent Overstreet | bec68faa | 2013-05-13 14:45:08 -0700 | [diff] [blame] | 1595 | ret = -EINPROGRESS; |
Kent Overstreet | 906b973 | 2013-05-07 16:18:31 -0700 | [diff] [blame] | 1596 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1597 | |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1598 | percpu_ref_put(&ctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1599 | |
| 1600 | return ret; |
| 1601 | } |
| 1602 | |
| 1603 | /* io_getevents: |
| 1604 | * 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] | 1605 | * the completion queue for the aio_context specified by ctx_id. If |
| 1606 | * it succeeds, the number of read events is returned. May fail with |
| 1607 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is |
| 1608 | * out of range, if timeout is out of range. May fail with -EFAULT |
| 1609 | * if any of the memory specified is invalid. May return 0 or |
| 1610 | * < min_nr if the timeout specified by timeout has elapsed |
| 1611 | * before sufficient events are available, where timeout == NULL |
| 1612 | * specifies an infinite timeout. Note that the timeout pointed to by |
Jeff Moyer | 6900807 | 2013-05-24 15:55:24 -0700 | [diff] [blame] | 1613 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | */ |
Heiko Carstens | 002c897 | 2009-01-14 14:14:18 +0100 | [diff] [blame] | 1615 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
| 1616 | long, min_nr, |
| 1617 | long, nr, |
| 1618 | struct io_event __user *, events, |
| 1619 | struct timespec __user *, timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1620 | { |
| 1621 | struct kioctx *ioctx = lookup_ioctx(ctx_id); |
| 1622 | long ret = -EINVAL; |
| 1623 | |
| 1624 | if (likely(ioctx)) { |
Namhyung Kim | 2e41025 | 2011-01-12 17:01:08 -0800 | [diff] [blame] | 1625 | if (likely(min_nr <= nr && min_nr >= 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
Kent Overstreet | 723be6e | 2013-05-28 15:14:48 -0700 | [diff] [blame] | 1627 | percpu_ref_put(&ioctx->users); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1629 | return ret; |
| 1630 | } |