1 /* 2 * An async IO implementation for Linux 3 * Written by Benjamin LaHaise <bcrl@kvack.org> 4 * 5 * Implements an efficient asynchronous io interface. 6 * 7 * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved. 8 * 9 * See ../COPYING for licensing terms. 10 */ 11 #define pr_fmt(fmt) "%s: " fmt, __func__ 12 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> 18 #include <linux/export.h> 19 #include <linux/syscalls.h> 20 #include <linux/backing-dev.h> 21 #include <linux/uio.h> 22 23 #include <linux/sched.h> 24 #include <linux/fs.h> 25 #include <linux/file.h> 26 #include <linux/mm.h> 27 #include <linux/mman.h> 28 #include <linux/mmu_context.h> 29 #include <linux/percpu.h> 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> 36 #include <linux/eventfd.h> 37 #include <linux/blkdev.h> 38 #include <linux/compat.h> 39 #include <linux/migrate.h> 40 #include <linux/ramfs.h> 41 #include <linux/percpu-refcount.h> 42 #include <linux/mount.h> 43 44 #include <asm/kmap_types.h> 45 #include <linux/uaccess.h> 46 47 #include "internal.h" 48 49 #define AIO_RING_MAGIC 0xa10a10a1 50 #define AIO_RING_COMPAT_FEATURES 1 51 #define AIO_RING_INCOMPAT_FEATURES 0 52 struct aio_ring { 53 unsigned id; /* kernel internal index number */ 54 unsigned nr; /* number of io_events */ 55 unsigned head; /* Written to by userland or under ring_lock 56 * mutex by aio_read_events_ring(). */ 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 69 70 struct kioctx_table { 71 struct rcu_head rcu; 72 unsigned nr; 73 struct kioctx *table[]; 74 }; 75 76 struct kioctx_cpu { 77 unsigned reqs_available; 78 }; 79 80 struct ctx_rq_wait { 81 struct completion comp; 82 atomic_t count; 83 }; 84 85 struct kioctx { 86 struct percpu_ref users; 87 atomic_t dead; 88 89 struct percpu_ref reqs; 90 91 unsigned long user_id; 92 93 struct __percpu kioctx_cpu *cpu; 94 95 /* 96 * For percpu reqs_available, number of slots we move to/from global 97 * counter at a time: 98 */ 99 unsigned req_batch; 100 /* 101 * This is what userspace passed to io_setup(), it's not used for 102 * anything but counting against the global max_reqs quota. 103 * 104 * The real limit is nr_events - 1, which will be larger (see 105 * aio_setup_ring()) 106 */ 107 unsigned max_reqs; 108 109 /* Size of ringbuffer, in units of struct io_event */ 110 unsigned nr_events; 111 112 unsigned long mmap_base; 113 unsigned long mmap_size; 114 115 struct page **ring_pages; 116 long nr_pages; 117 118 struct work_struct free_work; 119 120 /* 121 * signals when all in-flight requests are done 122 */ 123 struct ctx_rq_wait *rq_wait; 124 125 struct { 126 /* 127 * This counts the number of available slots in the ringbuffer, 128 * so we avoid overflowing it: it's decremented (if positive) 129 * when allocating a kiocb and incremented when the resulting 130 * io_event is pulled off the ringbuffer. 131 * 132 * We batch accesses to it with a percpu version. 133 */ 134 atomic_t reqs_available; 135 } ____cacheline_aligned_in_smp; 136 137 struct { 138 spinlock_t ctx_lock; 139 struct list_head active_reqs; /* used for cancellation */ 140 } ____cacheline_aligned_in_smp; 141 142 struct { 143 struct mutex ring_lock; 144 wait_queue_head_t wait; 145 } ____cacheline_aligned_in_smp; 146 147 struct { 148 unsigned tail; 149 unsigned completed_events; 150 spinlock_t completion_lock; 151 } ____cacheline_aligned_in_smp; 152 153 struct page *internal_pages[AIO_RING_PAGES]; 154 struct file *aio_ring_file; 155 156 unsigned id; 157 }; 158 159 /* 160 * We use ki_cancel == KIOCB_CANCELLED to indicate that a kiocb has been either 161 * cancelled or completed (this makes a certain amount of sense because 162 * successful cancellation - io_cancel() - does deliver the completion to 163 * userspace). 164 * 165 * And since most things don't implement kiocb cancellation and we'd really like 166 * kiocb completion to be lockless when possible, we use ki_cancel to 167 * synchronize cancellation and completion - we only set it to KIOCB_CANCELLED 168 * with xchg() or cmpxchg(), see batch_complete_aio() and kiocb_cancel(). 169 */ 170 #define KIOCB_CANCELLED ((void *) (~0ULL)) 171 172 struct aio_kiocb { 173 struct kiocb common; 174 175 struct kioctx *ki_ctx; 176 kiocb_cancel_fn *ki_cancel; 177 178 struct iocb __user *ki_user_iocb; /* user's aiocb */ 179 __u64 ki_user_data; /* user's data for completion */ 180 181 struct list_head ki_list; /* the aio core uses this 182 * for cancellation */ 183 184 /* 185 * If the aio_resfd field of the userspace iocb is not zero, 186 * this is the underlying eventfd context to deliver events to. 187 */ 188 struct eventfd_ctx *ki_eventfd; 189 }; 190 191 /*------ sysctl variables----*/ 192 static DEFINE_SPINLOCK(aio_nr_lock); 193 unsigned long aio_nr; /* current system wide number of aio requests */ 194 unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ 195 /*----end sysctl variables---*/ 196 197 static struct kmem_cache *kiocb_cachep; 198 static struct kmem_cache *kioctx_cachep; 199 200 static struct vfsmount *aio_mnt; 201 202 static const struct file_operations aio_ring_fops; 203 static const struct address_space_operations aio_ctx_aops; 204 205 static struct file *aio_private_file(struct kioctx *ctx, loff_t nr_pages) 206 { 207 struct qstr this = QSTR_INIT("[aio]", 5); 208 struct file *file; 209 struct path path; 210 struct inode *inode = alloc_anon_inode(aio_mnt->mnt_sb); 211 if (IS_ERR(inode)) 212 return ERR_CAST(inode); 213 214 inode->i_mapping->a_ops = &aio_ctx_aops; 215 inode->i_mapping->private_data = ctx; 216 inode->i_size = PAGE_SIZE * nr_pages; 217 218 path.dentry = d_alloc_pseudo(aio_mnt->mnt_sb, &this); 219 if (!path.dentry) { 220 iput(inode); 221 return ERR_PTR(-ENOMEM); 222 } 223 path.mnt = mntget(aio_mnt); 224 225 d_instantiate(path.dentry, inode); 226 file = alloc_file(&path, FMODE_READ | FMODE_WRITE, &aio_ring_fops); 227 if (IS_ERR(file)) { 228 path_put(&path); 229 return file; 230 } 231 232 file->f_flags = O_RDWR; 233 return file; 234 } 235 236 static struct dentry *aio_mount(struct file_system_type *fs_type, 237 int flags, const char *dev_name, void *data) 238 { 239 static const struct dentry_operations ops = { 240 .d_dname = simple_dname, 241 }; 242 struct dentry *root = mount_pseudo(fs_type, "aio:", NULL, &ops, 243 AIO_RING_MAGIC); 244 245 if (!IS_ERR(root)) 246 root->d_sb->s_iflags |= SB_I_NOEXEC; 247 return root; 248 } 249 250 /* aio_setup 251 * Creates the slab caches used by the aio routines, panic on 252 * failure as this is done early during the boot sequence. 253 */ 254 static int __init aio_setup(void) 255 { 256 static struct file_system_type aio_fs = { 257 .name = "aio", 258 .mount = aio_mount, 259 .kill_sb = kill_anon_super, 260 }; 261 aio_mnt = kern_mount(&aio_fs); 262 if (IS_ERR(aio_mnt)) 263 panic("Failed to create aio fs mount."); 264 265 kiocb_cachep = KMEM_CACHE(aio_kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); 266 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); 267 268 pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); 269 270 return 0; 271 } 272 __initcall(aio_setup); 273 274 static void put_aio_ring_file(struct kioctx *ctx) 275 { 276 struct file *aio_ring_file = ctx->aio_ring_file; 277 struct address_space *i_mapping; 278 279 if (aio_ring_file) { 280 truncate_setsize(file_inode(aio_ring_file), 0); 281 282 /* Prevent further access to the kioctx from migratepages */ 283 i_mapping = aio_ring_file->f_mapping; 284 spin_lock(&i_mapping->private_lock); 285 i_mapping->private_data = NULL; 286 ctx->aio_ring_file = NULL; 287 spin_unlock(&i_mapping->private_lock); 288 289 fput(aio_ring_file); 290 } 291 } 292 293 static void aio_free_ring(struct kioctx *ctx) 294 { 295 int i; 296 297 /* Disconnect the kiotx from the ring file. This prevents future 298 * accesses to the kioctx from page migration. 299 */ 300 put_aio_ring_file(ctx); 301 302 for (i = 0; i < ctx->nr_pages; i++) { 303 struct page *page; 304 pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, 305 page_count(ctx->ring_pages[i])); 306 page = ctx->ring_pages[i]; 307 if (!page) 308 continue; 309 ctx->ring_pages[i] = NULL; 310 put_page(page); 311 } 312 313 if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) { 314 kfree(ctx->ring_pages); 315 ctx->ring_pages = NULL; 316 } 317 } 318 319 static int aio_ring_mremap(struct vm_area_struct *vma) 320 { 321 struct file *file = vma->vm_file; 322 struct mm_struct *mm = vma->vm_mm; 323 struct kioctx_table *table; 324 int i, res = -EINVAL; 325 326 spin_lock(&mm->ioctx_lock); 327 rcu_read_lock(); 328 table = rcu_dereference(mm->ioctx_table); 329 for (i = 0; i < table->nr; i++) { 330 struct kioctx *ctx; 331 332 ctx = table->table[i]; 333 if (ctx && ctx->aio_ring_file == file) { 334 if (!atomic_read(&ctx->dead)) { 335 ctx->user_id = ctx->mmap_base = vma->vm_start; 336 res = 0; 337 } 338 break; 339 } 340 } 341 342 rcu_read_unlock(); 343 spin_unlock(&mm->ioctx_lock); 344 return res; 345 } 346 347 static const struct vm_operations_struct aio_ring_vm_ops = { 348 .mremap = aio_ring_mremap, 349 #if IS_ENABLED(CONFIG_MMU) 350 .fault = filemap_fault, 351 .map_pages = filemap_map_pages, 352 .page_mkwrite = filemap_page_mkwrite, 353 #endif 354 }; 355 356 static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) 357 { 358 vma->vm_flags |= VM_DONTEXPAND; 359 vma->vm_ops = &aio_ring_vm_ops; 360 return 0; 361 } 362 363 static const struct file_operations aio_ring_fops = { 364 .mmap = aio_ring_mmap, 365 }; 366 367 #if IS_ENABLED(CONFIG_MIGRATION) 368 static int aio_migratepage(struct address_space *mapping, struct page *new, 369 struct page *old, enum migrate_mode mode) 370 { 371 struct kioctx *ctx; 372 unsigned long flags; 373 pgoff_t idx; 374 int rc; 375 376 rc = 0; 377 378 /* mapping->private_lock here protects against the kioctx teardown. */ 379 spin_lock(&mapping->private_lock); 380 ctx = mapping->private_data; 381 if (!ctx) { 382 rc = -EINVAL; 383 goto out; 384 } 385 386 /* The ring_lock mutex. The prevents aio_read_events() from writing 387 * to the ring's head, and prevents page migration from mucking in 388 * a partially initialized kiotx. 389 */ 390 if (!mutex_trylock(&ctx->ring_lock)) { 391 rc = -EAGAIN; 392 goto out; 393 } 394 395 idx = old->index; 396 if (idx < (pgoff_t)ctx->nr_pages) { 397 /* Make sure the old page hasn't already been changed */ 398 if (ctx->ring_pages[idx] != old) 399 rc = -EAGAIN; 400 } else 401 rc = -EINVAL; 402 403 if (rc != 0) 404 goto out_unlock; 405 406 /* Writeback must be complete */ 407 BUG_ON(PageWriteback(old)); 408 get_page(new); 409 410 rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1); 411 if (rc != MIGRATEPAGE_SUCCESS) { 412 put_page(new); 413 goto out_unlock; 414 } 415 416 /* Take completion_lock to prevent other writes to the ring buffer 417 * while the old page is copied to the new. This prevents new 418 * events from being lost. 419 */ 420 spin_lock_irqsave(&ctx->completion_lock, flags); 421 migrate_page_copy(new, old); 422 BUG_ON(ctx->ring_pages[idx] != old); 423 ctx->ring_pages[idx] = new; 424 spin_unlock_irqrestore(&ctx->completion_lock, flags); 425 426 /* The old page is no longer accessible. */ 427 put_page(old); 428 429 out_unlock: 430 mutex_unlock(&ctx->ring_lock); 431 out: 432 spin_unlock(&mapping->private_lock); 433 return rc; 434 } 435 #endif 436 437 static const struct address_space_operations aio_ctx_aops = { 438 .set_page_dirty = __set_page_dirty_no_writeback, 439 #if IS_ENABLED(CONFIG_MIGRATION) 440 .migratepage = aio_migratepage, 441 #endif 442 }; 443 444 static int aio_setup_ring(struct kioctx *ctx) 445 { 446 struct aio_ring *ring; 447 unsigned nr_events = ctx->max_reqs; 448 struct mm_struct *mm = current->mm; 449 unsigned long size, unused; 450 int nr_pages; 451 int i; 452 struct file *file; 453 454 /* Compensate for the ring buffer's head/tail overlap entry */ 455 nr_events += 2; /* 1 is required, 2 for good luck */ 456 457 size = sizeof(struct aio_ring); 458 size += sizeof(struct io_event) * nr_events; 459 460 nr_pages = PFN_UP(size); 461 if (nr_pages < 0) 462 return -EINVAL; 463 464 file = aio_private_file(ctx, nr_pages); 465 if (IS_ERR(file)) { 466 ctx->aio_ring_file = NULL; 467 return -ENOMEM; 468 } 469 470 ctx->aio_ring_file = file; 471 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) 472 / sizeof(struct io_event); 473 474 ctx->ring_pages = ctx->internal_pages; 475 if (nr_pages > AIO_RING_PAGES) { 476 ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), 477 GFP_KERNEL); 478 if (!ctx->ring_pages) { 479 put_aio_ring_file(ctx); 480 return -ENOMEM; 481 } 482 } 483 484 for (i = 0; i < nr_pages; i++) { 485 struct page *page; 486 page = find_or_create_page(file->f_mapping, 487 i, GFP_HIGHUSER | __GFP_ZERO); 488 if (!page) 489 break; 490 pr_debug("pid(%d) page[%d]->count=%d\n", 491 current->pid, i, page_count(page)); 492 SetPageUptodate(page); 493 unlock_page(page); 494 495 ctx->ring_pages[i] = page; 496 } 497 ctx->nr_pages = i; 498 499 if (unlikely(i != nr_pages)) { 500 aio_free_ring(ctx); 501 return -ENOMEM; 502 } 503 504 ctx->mmap_size = nr_pages * PAGE_SIZE; 505 pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); 506 507 if (down_write_killable(&mm->mmap_sem)) { 508 ctx->mmap_size = 0; 509 aio_free_ring(ctx); 510 return -EINTR; 511 } 512 513 ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, 514 PROT_READ | PROT_WRITE, 515 MAP_SHARED, 0, &unused); 516 up_write(&mm->mmap_sem); 517 if (IS_ERR((void *)ctx->mmap_base)) { 518 ctx->mmap_size = 0; 519 aio_free_ring(ctx); 520 return -ENOMEM; 521 } 522 523 pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); 524 525 ctx->user_id = ctx->mmap_base; 526 ctx->nr_events = nr_events; /* trusted copy */ 527 528 ring = kmap_atomic(ctx->ring_pages[0]); 529 ring->nr = nr_events; /* user copy */ 530 ring->id = ~0U; 531 ring->head = ring->tail = 0; 532 ring->magic = AIO_RING_MAGIC; 533 ring->compat_features = AIO_RING_COMPAT_FEATURES; 534 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; 535 ring->header_length = sizeof(struct aio_ring); 536 kunmap_atomic(ring); 537 flush_dcache_page(ctx->ring_pages[0]); 538 539 return 0; 540 } 541 542 #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) 543 #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) 544 #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) 545 546 void kiocb_set_cancel_fn(struct kiocb *iocb, kiocb_cancel_fn *cancel) 547 { 548 struct aio_kiocb *req = container_of(iocb, struct aio_kiocb, common); 549 struct kioctx *ctx = req->ki_ctx; 550 unsigned long flags; 551 552 spin_lock_irqsave(&ctx->ctx_lock, flags); 553 554 if (!req->ki_list.next) 555 list_add(&req->ki_list, &ctx->active_reqs); 556 557 req->ki_cancel = cancel; 558 559 spin_unlock_irqrestore(&ctx->ctx_lock, flags); 560 } 561 EXPORT_SYMBOL(kiocb_set_cancel_fn); 562 563 static int kiocb_cancel(struct aio_kiocb *kiocb) 564 { 565 kiocb_cancel_fn *old, *cancel; 566 567 /* 568 * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it 569 * actually has a cancel function, hence the cmpxchg() 570 */ 571 572 cancel = ACCESS_ONCE(kiocb->ki_cancel); 573 do { 574 if (!cancel || cancel == KIOCB_CANCELLED) 575 return -EINVAL; 576 577 old = cancel; 578 cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); 579 } while (cancel != old); 580 581 return cancel(&kiocb->common); 582 } 583 584 static void free_ioctx(struct work_struct *work) 585 { 586 struct kioctx *ctx = container_of(work, struct kioctx, free_work); 587 588 pr_debug("freeing %p\n", ctx); 589 590 aio_free_ring(ctx); 591 free_percpu(ctx->cpu); 592 percpu_ref_exit(&ctx->reqs); 593 percpu_ref_exit(&ctx->users); 594 kmem_cache_free(kioctx_cachep, ctx); 595 } 596 597 static void free_ioctx_reqs(struct percpu_ref *ref) 598 { 599 struct kioctx *ctx = container_of(ref, struct kioctx, reqs); 600 601 /* At this point we know that there are no any in-flight requests */ 602 if (ctx->rq_wait && atomic_dec_and_test(&ctx->rq_wait->count)) 603 complete(&ctx->rq_wait->comp); 604 605 INIT_WORK(&ctx->free_work, free_ioctx); 606 schedule_work(&ctx->free_work); 607 } 608 609 /* 610 * When this function runs, the kioctx has been removed from the "hash table" 611 * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - 612 * now it's safe to cancel any that need to be. 613 */ 614 static void free_ioctx_users(struct percpu_ref *ref) 615 { 616 struct kioctx *ctx = container_of(ref, struct kioctx, users); 617 struct aio_kiocb *req; 618 619 spin_lock_irq(&ctx->ctx_lock); 620 621 while (!list_empty(&ctx->active_reqs)) { 622 req = list_first_entry(&ctx->active_reqs, 623 struct aio_kiocb, ki_list); 624 625 list_del_init(&req->ki_list); 626 kiocb_cancel(req); 627 } 628 629 spin_unlock_irq(&ctx->ctx_lock); 630 631 percpu_ref_kill(&ctx->reqs); 632 percpu_ref_put(&ctx->reqs); 633 } 634 635 static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) 636 { 637 unsigned i, new_nr; 638 struct kioctx_table *table, *old; 639 struct aio_ring *ring; 640 641 spin_lock(&mm->ioctx_lock); 642 table = rcu_dereference_raw(mm->ioctx_table); 643 644 while (1) { 645 if (table) 646 for (i = 0; i < table->nr; i++) 647 if (!table->table[i]) { 648 ctx->id = i; 649 table->table[i] = ctx; 650 spin_unlock(&mm->ioctx_lock); 651 652 /* While kioctx setup is in progress, 653 * we are protected from page migration 654 * changes ring_pages by ->ring_lock. 655 */ 656 ring = kmap_atomic(ctx->ring_pages[0]); 657 ring->id = ctx->id; 658 kunmap_atomic(ring); 659 return 0; 660 } 661 662 new_nr = (table ? table->nr : 1) * 4; 663 spin_unlock(&mm->ioctx_lock); 664 665 table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * 666 new_nr, GFP_KERNEL); 667 if (!table) 668 return -ENOMEM; 669 670 table->nr = new_nr; 671 672 spin_lock(&mm->ioctx_lock); 673 old = rcu_dereference_raw(mm->ioctx_table); 674 675 if (!old) { 676 rcu_assign_pointer(mm->ioctx_table, table); 677 } else if (table->nr > old->nr) { 678 memcpy(table->table, old->table, 679 old->nr * sizeof(struct kioctx *)); 680 681 rcu_assign_pointer(mm->ioctx_table, table); 682 kfree_rcu(old, rcu); 683 } else { 684 kfree(table); 685 table = old; 686 } 687 } 688 } 689 690 static void aio_nr_sub(unsigned nr) 691 { 692 spin_lock(&aio_nr_lock); 693 if (WARN_ON(aio_nr - nr > aio_nr)) 694 aio_nr = 0; 695 else 696 aio_nr -= nr; 697 spin_unlock(&aio_nr_lock); 698 } 699 700 /* ioctx_alloc 701 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. 702 */ 703 static struct kioctx *ioctx_alloc(unsigned nr_events) 704 { 705 struct mm_struct *mm = current->mm; 706 struct kioctx *ctx; 707 int err = -ENOMEM; 708 709 /* 710 * We keep track of the number of available ringbuffer slots, to prevent 711 * overflow (reqs_available), and we also use percpu counters for this. 712 * 713 * So since up to half the slots might be on other cpu's percpu counters 714 * and unavailable, double nr_events so userspace sees what they 715 * expected: additionally, we move req_batch slots to/from percpu 716 * counters at a time, so make sure that isn't 0: 717 */ 718 nr_events = max(nr_events, num_possible_cpus() * 4); 719 nr_events *= 2; 720 721 /* Prevent overflows */ 722 if (nr_events > (0x10000000U / sizeof(struct io_event))) { 723 pr_debug("ENOMEM: nr_events too high\n"); 724 return ERR_PTR(-EINVAL); 725 } 726 727 if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) 728 return ERR_PTR(-EAGAIN); 729 730 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); 731 if (!ctx) 732 return ERR_PTR(-ENOMEM); 733 734 ctx->max_reqs = nr_events; 735 736 spin_lock_init(&ctx->ctx_lock); 737 spin_lock_init(&ctx->completion_lock); 738 mutex_init(&ctx->ring_lock); 739 /* Protect against page migration throughout kiotx setup by keeping 740 * the ring_lock mutex held until setup is complete. */ 741 mutex_lock(&ctx->ring_lock); 742 init_waitqueue_head(&ctx->wait); 743 744 INIT_LIST_HEAD(&ctx->active_reqs); 745 746 if (percpu_ref_init(&ctx->users, free_ioctx_users, 0, GFP_KERNEL)) 747 goto err; 748 749 if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs, 0, GFP_KERNEL)) 750 goto err; 751 752 ctx->cpu = alloc_percpu(struct kioctx_cpu); 753 if (!ctx->cpu) 754 goto err; 755 756 err = aio_setup_ring(ctx); 757 if (err < 0) 758 goto err; 759 760 atomic_set(&ctx->reqs_available, ctx->nr_events - 1); 761 ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); 762 if (ctx->req_batch < 1) 763 ctx->req_batch = 1; 764 765 /* limit the number of system wide aios */ 766 spin_lock(&aio_nr_lock); 767 if (aio_nr + nr_events > (aio_max_nr * 2UL) || 768 aio_nr + nr_events < aio_nr) { 769 spin_unlock(&aio_nr_lock); 770 err = -EAGAIN; 771 goto err_ctx; 772 } 773 aio_nr += ctx->max_reqs; 774 spin_unlock(&aio_nr_lock); 775 776 percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ 777 percpu_ref_get(&ctx->reqs); /* free_ioctx_users() will drop this */ 778 779 err = ioctx_add_table(ctx, mm); 780 if (err) 781 goto err_cleanup; 782 783 /* Release the ring_lock mutex now that all setup is complete. */ 784 mutex_unlock(&ctx->ring_lock); 785 786 pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", 787 ctx, ctx->user_id, mm, ctx->nr_events); 788 return ctx; 789 790 err_cleanup: 791 aio_nr_sub(ctx->max_reqs); 792 err_ctx: 793 atomic_set(&ctx->dead, 1); 794 if (ctx->mmap_size) 795 vm_munmap(ctx->mmap_base, ctx->mmap_size); 796 aio_free_ring(ctx); 797 err: 798 mutex_unlock(&ctx->ring_lock); 799 free_percpu(ctx->cpu); 800 percpu_ref_exit(&ctx->reqs); 801 percpu_ref_exit(&ctx->users); 802 kmem_cache_free(kioctx_cachep, ctx); 803 pr_debug("error allocating ioctx %d\n", err); 804 return ERR_PTR(err); 805 } 806 807 /* kill_ioctx 808 * Cancels all outstanding aio requests on an aio context. Used 809 * when the processes owning a context have all exited to encourage 810 * the rapid destruction of the kioctx. 811 */ 812 static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx, 813 struct ctx_rq_wait *wait) 814 { 815 struct kioctx_table *table; 816 817 spin_lock(&mm->ioctx_lock); 818 if (atomic_xchg(&ctx->dead, 1)) { 819 spin_unlock(&mm->ioctx_lock); 820 return -EINVAL; 821 } 822 823 table = rcu_dereference_raw(mm->ioctx_table); 824 WARN_ON(ctx != table->table[ctx->id]); 825 table->table[ctx->id] = NULL; 826 spin_unlock(&mm->ioctx_lock); 827 828 /* percpu_ref_kill() will do the necessary call_rcu() */ 829 wake_up_all(&ctx->wait); 830 831 /* 832 * It'd be more correct to do this in free_ioctx(), after all 833 * the outstanding kiocbs have finished - but by then io_destroy 834 * has already returned, so io_setup() could potentially return 835 * -EAGAIN with no ioctxs actually in use (as far as userspace 836 * could tell). 837 */ 838 aio_nr_sub(ctx->max_reqs); 839 840 if (ctx->mmap_size) 841 vm_munmap(ctx->mmap_base, ctx->mmap_size); 842 843 ctx->rq_wait = wait; 844 percpu_ref_kill(&ctx->users); 845 return 0; 846 } 847 848 /* 849 * exit_aio: called when the last user of mm goes away. At this point, there is 850 * no way for any new requests to be submited or any of the io_* syscalls to be 851 * called on the context. 852 * 853 * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on 854 * them. 855 */ 856 void exit_aio(struct mm_struct *mm) 857 { 858 struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table); 859 struct ctx_rq_wait wait; 860 int i, skipped; 861 862 if (!table) 863 return; 864 865 atomic_set(&wait.count, table->nr); 866 init_completion(&wait.comp); 867 868 skipped = 0; 869 for (i = 0; i < table->nr; ++i) { 870 struct kioctx *ctx = table->table[i]; 871 872 if (!ctx) { 873 skipped++; 874 continue; 875 } 876 877 /* 878 * We don't need to bother with munmap() here - exit_mmap(mm) 879 * is coming and it'll unmap everything. And we simply can't, 880 * this is not necessarily our ->mm. 881 * Since kill_ioctx() uses non-zero ->mmap_size as indicator 882 * that it needs to unmap the area, just set it to 0. 883 */ 884 ctx->mmap_size = 0; 885 kill_ioctx(mm, ctx, &wait); 886 } 887 888 if (!atomic_sub_and_test(skipped, &wait.count)) { 889 /* Wait until all IO for the context are done. */ 890 wait_for_completion(&wait.comp); 891 } 892 893 RCU_INIT_POINTER(mm->ioctx_table, NULL); 894 kfree(table); 895 } 896 897 static void put_reqs_available(struct kioctx *ctx, unsigned nr) 898 { 899 struct kioctx_cpu *kcpu; 900 unsigned long flags; 901 902 local_irq_save(flags); 903 kcpu = this_cpu_ptr(ctx->cpu); 904 kcpu->reqs_available += nr; 905 906 while (kcpu->reqs_available >= ctx->req_batch * 2) { 907 kcpu->reqs_available -= ctx->req_batch; 908 atomic_add(ctx->req_batch, &ctx->reqs_available); 909 } 910 911 local_irq_restore(flags); 912 } 913 914 static bool get_reqs_available(struct kioctx *ctx) 915 { 916 struct kioctx_cpu *kcpu; 917 bool ret = false; 918 unsigned long flags; 919 920 local_irq_save(flags); 921 kcpu = this_cpu_ptr(ctx->cpu); 922 if (!kcpu->reqs_available) { 923 int old, avail = atomic_read(&ctx->reqs_available); 924 925 do { 926 if (avail < ctx->req_batch) 927 goto out; 928 929 old = avail; 930 avail = atomic_cmpxchg(&ctx->reqs_available, 931 avail, avail - ctx->req_batch); 932 } while (avail != old); 933 934 kcpu->reqs_available += ctx->req_batch; 935 } 936 937 ret = true; 938 kcpu->reqs_available--; 939 out: 940 local_irq_restore(flags); 941 return ret; 942 } 943 944 /* refill_reqs_available 945 * Updates the reqs_available reference counts used for tracking the 946 * number of free slots in the completion ring. This can be called 947 * from aio_complete() (to optimistically update reqs_available) or 948 * from aio_get_req() (the we're out of events case). It must be 949 * called holding ctx->completion_lock. 950 */ 951 static void refill_reqs_available(struct kioctx *ctx, unsigned head, 952 unsigned tail) 953 { 954 unsigned events_in_ring, completed; 955 956 /* Clamp head since userland can write to it. */ 957 head %= ctx->nr_events; 958 if (head <= tail) 959 events_in_ring = tail - head; 960 else 961 events_in_ring = ctx->nr_events - (head - tail); 962 963 completed = ctx->completed_events; 964 if (events_in_ring < completed) 965 completed -= events_in_ring; 966 else 967 completed = 0; 968 969 if (!completed) 970 return; 971 972 ctx->completed_events -= completed; 973 put_reqs_available(ctx, completed); 974 } 975 976 /* user_refill_reqs_available 977 * Called to refill reqs_available when aio_get_req() encounters an 978 * out of space in the completion ring. 979 */ 980 static void user_refill_reqs_available(struct kioctx *ctx) 981 { 982 spin_lock_irq(&ctx->completion_lock); 983 if (ctx->completed_events) { 984 struct aio_ring *ring; 985 unsigned head; 986 987 /* Access of ring->head may race with aio_read_events_ring() 988 * here, but that's okay since whether we read the old version 989 * or the new version, and either will be valid. The important 990 * part is that head cannot pass tail since we prevent 991 * aio_complete() from updating tail by holding 992 * ctx->completion_lock. Even if head is invalid, the check 993 * against ctx->completed_events below will make sure we do the 994 * safe/right thing. 995 */ 996 ring = kmap_atomic(ctx->ring_pages[0]); 997 head = ring->head; 998 kunmap_atomic(ring); 999 1000 refill_reqs_available(ctx, head, ctx->tail); 1001 } 1002 1003 spin_unlock_irq(&ctx->completion_lock); 1004 } 1005 1006 /* aio_get_req 1007 * Allocate a slot for an aio request. 1008 * Returns NULL if no requests are free. 1009 */ 1010 static inline struct aio_kiocb *aio_get_req(struct kioctx *ctx) 1011 { 1012 struct aio_kiocb *req; 1013 1014 if (!get_reqs_available(ctx)) { 1015 user_refill_reqs_available(ctx); 1016 if (!get_reqs_available(ctx)) 1017 return NULL; 1018 } 1019 1020 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); 1021 if (unlikely(!req)) 1022 goto out_put; 1023 1024 percpu_ref_get(&ctx->reqs); 1025 1026 req->ki_ctx = ctx; 1027 return req; 1028 out_put: 1029 put_reqs_available(ctx, 1); 1030 return NULL; 1031 } 1032 1033 static void kiocb_free(struct aio_kiocb *req) 1034 { 1035 if (req->common.ki_filp) 1036 fput(req->common.ki_filp); 1037 if (req->ki_eventfd != NULL) 1038 eventfd_ctx_put(req->ki_eventfd); 1039 kmem_cache_free(kiocb_cachep, req); 1040 } 1041 1042 static struct kioctx *lookup_ioctx(unsigned long ctx_id) 1043 { 1044 struct aio_ring __user *ring = (void __user *)ctx_id; 1045 struct mm_struct *mm = current->mm; 1046 struct kioctx *ctx, *ret = NULL; 1047 struct kioctx_table *table; 1048 unsigned id; 1049 1050 if (get_user(id, &ring->id)) 1051 return NULL; 1052 1053 rcu_read_lock(); 1054 table = rcu_dereference(mm->ioctx_table); 1055 1056 if (!table || id >= table->nr) 1057 goto out; 1058 1059 ctx = table->table[id]; 1060 if (ctx && ctx->user_id == ctx_id) { 1061 percpu_ref_get(&ctx->users); 1062 ret = ctx; 1063 } 1064 out: 1065 rcu_read_unlock(); 1066 return ret; 1067 } 1068 1069 /* aio_complete 1070 * Called when the io request on the given iocb is complete. 1071 */ 1072 static void aio_complete(struct kiocb *kiocb, long res, long res2) 1073 { 1074 struct aio_kiocb *iocb = container_of(kiocb, struct aio_kiocb, common); 1075 struct kioctx *ctx = iocb->ki_ctx; 1076 struct aio_ring *ring; 1077 struct io_event *ev_page, *event; 1078 unsigned tail, pos, head; 1079 unsigned long flags; 1080 1081 if (kiocb->ki_flags & IOCB_WRITE) { 1082 struct file *file = kiocb->ki_filp; 1083 1084 /* 1085 * Tell lockdep we inherited freeze protection from submission 1086 * thread. 1087 */ 1088 __sb_writers_acquired(file_inode(file)->i_sb, SB_FREEZE_WRITE); 1089 file_end_write(file); 1090 } 1091 1092 /* 1093 * Special case handling for sync iocbs: 1094 * - events go directly into the iocb for fast handling 1095 * - the sync task with the iocb in its stack holds the single iocb 1096 * ref, no other paths have a way to get another ref 1097 * - the sync task helpfully left a reference to itself in the iocb 1098 */ 1099 BUG_ON(is_sync_kiocb(kiocb)); 1100 1101 if (iocb->ki_list.next) { 1102 unsigned long flags; 1103 1104 spin_lock_irqsave(&ctx->ctx_lock, flags); 1105 list_del(&iocb->ki_list); 1106 spin_unlock_irqrestore(&ctx->ctx_lock, flags); 1107 } 1108 1109 /* 1110 * Add a completion event to the ring buffer. Must be done holding 1111 * ctx->completion_lock to prevent other code from messing with the tail 1112 * pointer since we might be called from irq context. 1113 */ 1114 spin_lock_irqsave(&ctx->completion_lock, flags); 1115 1116 tail = ctx->tail; 1117 pos = tail + AIO_EVENTS_OFFSET; 1118 1119 if (++tail >= ctx->nr_events) 1120 tail = 0; 1121 1122 ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); 1123 event = ev_page + pos % AIO_EVENTS_PER_PAGE; 1124 1125 event->obj = (u64)(unsigned long)iocb->ki_user_iocb; 1126 event->data = iocb->ki_user_data; 1127 event->res = res; 1128 event->res2 = res2; 1129 1130 kunmap_atomic(ev_page); 1131 flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); 1132 1133 pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", 1134 ctx, tail, iocb, iocb->ki_user_iocb, iocb->ki_user_data, 1135 res, res2); 1136 1137 /* after flagging the request as done, we 1138 * must never even look at it again 1139 */ 1140 smp_wmb(); /* make event visible before updating tail */ 1141 1142 ctx->tail = tail; 1143 1144 ring = kmap_atomic(ctx->ring_pages[0]); 1145 head = ring->head; 1146 ring->tail = tail; 1147 kunmap_atomic(ring); 1148 flush_dcache_page(ctx->ring_pages[0]); 1149 1150 ctx->completed_events++; 1151 if (ctx->completed_events > 1) 1152 refill_reqs_available(ctx, head, tail); 1153 spin_unlock_irqrestore(&ctx->completion_lock, flags); 1154 1155 pr_debug("added to ring %p at [%u]\n", iocb, tail); 1156 1157 /* 1158 * Check if the user asked us to deliver the result through an 1159 * eventfd. The eventfd_signal() function is safe to be called 1160 * from IRQ context. 1161 */ 1162 if (iocb->ki_eventfd != NULL) 1163 eventfd_signal(iocb->ki_eventfd, 1); 1164 1165 /* everything turned out well, dispose of the aiocb. */ 1166 kiocb_free(iocb); 1167 1168 /* 1169 * We have to order our ring_info tail store above and test 1170 * of the wait list below outside the wait lock. This is 1171 * like in wake_up_bit() where clearing a bit has to be 1172 * ordered with the unlocked test. 1173 */ 1174 smp_mb(); 1175 1176 if (waitqueue_active(&ctx->wait)) 1177 wake_up(&ctx->wait); 1178 1179 percpu_ref_put(&ctx->reqs); 1180 } 1181 1182 /* aio_read_events_ring 1183 * Pull an event off of the ioctx's event ring. Returns the number of 1184 * events fetched 1185 */ 1186 static long aio_read_events_ring(struct kioctx *ctx, 1187 struct io_event __user *event, long nr) 1188 { 1189 struct aio_ring *ring; 1190 unsigned head, tail, pos; 1191 long ret = 0; 1192 int copy_ret; 1193 1194 /* 1195 * The mutex can block and wake us up and that will cause 1196 * wait_event_interruptible_hrtimeout() to schedule without sleeping 1197 * and repeat. This should be rare enough that it doesn't cause 1198 * peformance issues. See the comment in read_events() for more detail. 1199 */ 1200 sched_annotate_sleep(); 1201 mutex_lock(&ctx->ring_lock); 1202 1203 /* Access to ->ring_pages here is protected by ctx->ring_lock. */ 1204 ring = kmap_atomic(ctx->ring_pages[0]); 1205 head = ring->head; 1206 tail = ring->tail; 1207 kunmap_atomic(ring); 1208 1209 /* 1210 * Ensure that once we've read the current tail pointer, that 1211 * we also see the events that were stored up to the tail. 1212 */ 1213 smp_rmb(); 1214 1215 pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); 1216 1217 if (head == tail) 1218 goto out; 1219 1220 head %= ctx->nr_events; 1221 tail %= ctx->nr_events; 1222 1223 while (ret < nr) { 1224 long avail; 1225 struct io_event *ev; 1226 struct page *page; 1227 1228 avail = (head <= tail ? tail : ctx->nr_events) - head; 1229 if (head == tail) 1230 break; 1231 1232 avail = min(avail, nr - ret); 1233 avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - 1234 ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); 1235 1236 pos = head + AIO_EVENTS_OFFSET; 1237 page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; 1238 pos %= AIO_EVENTS_PER_PAGE; 1239 1240 ev = kmap(page); 1241 copy_ret = copy_to_user(event + ret, ev + pos, 1242 sizeof(*ev) * avail); 1243 kunmap(page); 1244 1245 if (unlikely(copy_ret)) { 1246 ret = -EFAULT; 1247 goto out; 1248 } 1249 1250 ret += avail; 1251 head += avail; 1252 head %= ctx->nr_events; 1253 } 1254 1255 ring = kmap_atomic(ctx->ring_pages[0]); 1256 ring->head = head; 1257 kunmap_atomic(ring); 1258 flush_dcache_page(ctx->ring_pages[0]); 1259 1260 pr_debug("%li h%u t%u\n", ret, head, tail); 1261 out: 1262 mutex_unlock(&ctx->ring_lock); 1263 1264 return ret; 1265 } 1266 1267 static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, 1268 struct io_event __user *event, long *i) 1269 { 1270 long ret = aio_read_events_ring(ctx, event + *i, nr - *i); 1271 1272 if (ret > 0) 1273 *i += ret; 1274 1275 if (unlikely(atomic_read(&ctx->dead))) 1276 ret = -EINVAL; 1277 1278 if (!*i) 1279 *i = ret; 1280 1281 return ret < 0 || *i >= min_nr; 1282 } 1283 1284 static long read_events(struct kioctx *ctx, long min_nr, long nr, 1285 struct io_event __user *event, 1286 struct timespec __user *timeout) 1287 { 1288 ktime_t until = KTIME_MAX; 1289 long ret = 0; 1290 1291 if (timeout) { 1292 struct timespec ts; 1293 1294 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) 1295 return -EFAULT; 1296 1297 until = timespec_to_ktime(ts); 1298 } 1299 1300 /* 1301 * Note that aio_read_events() is being called as the conditional - i.e. 1302 * we're calling it after prepare_to_wait() has set task state to 1303 * TASK_INTERRUPTIBLE. 1304 * 1305 * But aio_read_events() can block, and if it blocks it's going to flip 1306 * the task state back to TASK_RUNNING. 1307 * 1308 * This should be ok, provided it doesn't flip the state back to 1309 * TASK_RUNNING and return 0 too much - that causes us to spin. That 1310 * will only happen if the mutex_lock() call blocks, and we then find 1311 * the ringbuffer empty. So in practice we should be ok, but it's 1312 * something to be aware of when touching this code. 1313 */ 1314 if (until == 0) 1315 aio_read_events(ctx, min_nr, nr, event, &ret); 1316 else 1317 wait_event_interruptible_hrtimeout(ctx->wait, 1318 aio_read_events(ctx, min_nr, nr, event, &ret), 1319 until); 1320 1321 if (!ret && signal_pending(current)) 1322 ret = -EINTR; 1323 1324 return ret; 1325 } 1326 1327 /* sys_io_setup: 1328 * Create an aio_context capable of receiving at least nr_events. 1329 * ctxp must not point to an aio_context that already exists, and 1330 * must be initialized to 0 prior to the call. On successful 1331 * creation of the aio_context, *ctxp is filled in with the resulting 1332 * handle. May fail with -EINVAL if *ctxp is not initialized, 1333 * if the specified nr_events exceeds internal limits. May fail 1334 * with -EAGAIN if the specified nr_events exceeds the user's limit 1335 * of available events. May fail with -ENOMEM if insufficient kernel 1336 * resources are available. May fail with -EFAULT if an invalid 1337 * pointer is passed for ctxp. Will fail with -ENOSYS if not 1338 * implemented. 1339 */ 1340 SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) 1341 { 1342 struct kioctx *ioctx = NULL; 1343 unsigned long ctx; 1344 long ret; 1345 1346 ret = get_user(ctx, ctxp); 1347 if (unlikely(ret)) 1348 goto out; 1349 1350 ret = -EINVAL; 1351 if (unlikely(ctx || nr_events == 0)) { 1352 pr_debug("EINVAL: ctx %lu nr_events %u\n", 1353 ctx, nr_events); 1354 goto out; 1355 } 1356 1357 ioctx = ioctx_alloc(nr_events); 1358 ret = PTR_ERR(ioctx); 1359 if (!IS_ERR(ioctx)) { 1360 ret = put_user(ioctx->user_id, ctxp); 1361 if (ret) 1362 kill_ioctx(current->mm, ioctx, NULL); 1363 percpu_ref_put(&ioctx->users); 1364 } 1365 1366 out: 1367 return ret; 1368 } 1369 1370 #ifdef CONFIG_COMPAT 1371 COMPAT_SYSCALL_DEFINE2(io_setup, unsigned, nr_events, u32 __user *, ctx32p) 1372 { 1373 struct kioctx *ioctx = NULL; 1374 unsigned long ctx; 1375 long ret; 1376 1377 ret = get_user(ctx, ctx32p); 1378 if (unlikely(ret)) 1379 goto out; 1380 1381 ret = -EINVAL; 1382 if (unlikely(ctx || nr_events == 0)) { 1383 pr_debug("EINVAL: ctx %lu nr_events %u\n", 1384 ctx, nr_events); 1385 goto out; 1386 } 1387 1388 ioctx = ioctx_alloc(nr_events); 1389 ret = PTR_ERR(ioctx); 1390 if (!IS_ERR(ioctx)) { 1391 /* truncating is ok because it's a user address */ 1392 ret = put_user((u32)ioctx->user_id, ctx32p); 1393 if (ret) 1394 kill_ioctx(current->mm, ioctx, NULL); 1395 percpu_ref_put(&ioctx->users); 1396 } 1397 1398 out: 1399 return ret; 1400 } 1401 #endif 1402 1403 /* sys_io_destroy: 1404 * Destroy the aio_context specified. May cancel any outstanding 1405 * AIOs and block on completion. Will fail with -ENOSYS if not 1406 * implemented. May fail with -EINVAL if the context pointed to 1407 * is invalid. 1408 */ 1409 SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) 1410 { 1411 struct kioctx *ioctx = lookup_ioctx(ctx); 1412 if (likely(NULL != ioctx)) { 1413 struct ctx_rq_wait wait; 1414 int ret; 1415 1416 init_completion(&wait.comp); 1417 atomic_set(&wait.count, 1); 1418 1419 /* Pass requests_done to kill_ioctx() where it can be set 1420 * in a thread-safe way. If we try to set it here then we have 1421 * a race condition if two io_destroy() called simultaneously. 1422 */ 1423 ret = kill_ioctx(current->mm, ioctx, &wait); 1424 percpu_ref_put(&ioctx->users); 1425 1426 /* Wait until all IO for the context are done. Otherwise kernel 1427 * keep using user-space buffers even if user thinks the context 1428 * is destroyed. 1429 */ 1430 if (!ret) 1431 wait_for_completion(&wait.comp); 1432 1433 return ret; 1434 } 1435 pr_debug("EINVAL: invalid context id\n"); 1436 return -EINVAL; 1437 } 1438 1439 static int aio_setup_rw(int rw, struct iocb *iocb, struct iovec **iovec, 1440 bool vectored, bool compat, struct iov_iter *iter) 1441 { 1442 void __user *buf = (void __user *)(uintptr_t)iocb->aio_buf; 1443 size_t len = iocb->aio_nbytes; 1444 1445 if (!vectored) { 1446 ssize_t ret = import_single_range(rw, buf, len, *iovec, iter); 1447 *iovec = NULL; 1448 return ret; 1449 } 1450 #ifdef CONFIG_COMPAT 1451 if (compat) 1452 return compat_import_iovec(rw, buf, len, UIO_FASTIOV, iovec, 1453 iter); 1454 #endif 1455 return import_iovec(rw, buf, len, UIO_FASTIOV, iovec, iter); 1456 } 1457 1458 static inline ssize_t aio_ret(struct kiocb *req, ssize_t ret) 1459 { 1460 switch (ret) { 1461 case -EIOCBQUEUED: 1462 return ret; 1463 case -ERESTARTSYS: 1464 case -ERESTARTNOINTR: 1465 case -ERESTARTNOHAND: 1466 case -ERESTART_RESTARTBLOCK: 1467 /* 1468 * There's no easy way to restart the syscall since other AIO's 1469 * may be already running. Just fail this IO with EINTR. 1470 */ 1471 ret = -EINTR; 1472 /*FALLTHRU*/ 1473 default: 1474 aio_complete(req, ret, 0); 1475 return 0; 1476 } 1477 } 1478 1479 static ssize_t aio_read(struct kiocb *req, struct iocb *iocb, bool vectored, 1480 bool compat) 1481 { 1482 struct file *file = req->ki_filp; 1483 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; 1484 struct iov_iter iter; 1485 ssize_t ret; 1486 1487 if (unlikely(!(file->f_mode & FMODE_READ))) 1488 return -EBADF; 1489 if (unlikely(!file->f_op->read_iter)) 1490 return -EINVAL; 1491 1492 ret = aio_setup_rw(READ, iocb, &iovec, vectored, compat, &iter); 1493 if (ret) 1494 return ret; 1495 ret = rw_verify_area(READ, file, &req->ki_pos, iov_iter_count(&iter)); 1496 if (!ret) 1497 ret = aio_ret(req, file->f_op->read_iter(req, &iter)); 1498 kfree(iovec); 1499 return ret; 1500 } 1501 1502 static ssize_t aio_write(struct kiocb *req, struct iocb *iocb, bool vectored, 1503 bool compat) 1504 { 1505 struct file *file = req->ki_filp; 1506 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; 1507 struct iov_iter iter; 1508 ssize_t ret; 1509 1510 if (unlikely(!(file->f_mode & FMODE_WRITE))) 1511 return -EBADF; 1512 if (unlikely(!file->f_op->write_iter)) 1513 return -EINVAL; 1514 1515 ret = aio_setup_rw(WRITE, iocb, &iovec, vectored, compat, &iter); 1516 if (ret) 1517 return ret; 1518 ret = rw_verify_area(WRITE, file, &req->ki_pos, iov_iter_count(&iter)); 1519 if (!ret) { 1520 req->ki_flags |= IOCB_WRITE; 1521 file_start_write(file); 1522 ret = aio_ret(req, file->f_op->write_iter(req, &iter)); 1523 /* 1524 * We release freeze protection in aio_complete(). Fool lockdep 1525 * by telling it the lock got released so that it doesn't 1526 * complain about held lock when we return to userspace. 1527 */ 1528 __sb_writers_release(file_inode(file)->i_sb, SB_FREEZE_WRITE); 1529 } 1530 kfree(iovec); 1531 return ret; 1532 } 1533 1534 static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, 1535 struct iocb *iocb, bool compat) 1536 { 1537 struct aio_kiocb *req; 1538 struct file *file; 1539 ssize_t ret; 1540 1541 /* enforce forwards compatibility on users */ 1542 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { 1543 pr_debug("EINVAL: reserve field set\n"); 1544 return -EINVAL; 1545 } 1546 1547 /* prevent overflows */ 1548 if (unlikely( 1549 (iocb->aio_buf != (unsigned long)iocb->aio_buf) || 1550 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || 1551 ((ssize_t)iocb->aio_nbytes < 0) 1552 )) { 1553 pr_debug("EINVAL: overflow check\n"); 1554 return -EINVAL; 1555 } 1556 1557 req = aio_get_req(ctx); 1558 if (unlikely(!req)) 1559 return -EAGAIN; 1560 1561 req->common.ki_filp = file = fget(iocb->aio_fildes); 1562 if (unlikely(!req->common.ki_filp)) { 1563 ret = -EBADF; 1564 goto out_put_req; 1565 } 1566 req->common.ki_pos = iocb->aio_offset; 1567 req->common.ki_complete = aio_complete; 1568 req->common.ki_flags = iocb_flags(req->common.ki_filp); 1569 1570 if (iocb->aio_flags & IOCB_FLAG_RESFD) { 1571 /* 1572 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an 1573 * instance of the file* now. The file descriptor must be 1574 * an eventfd() fd, and will be signaled for each completed 1575 * event using the eventfd_signal() function. 1576 */ 1577 req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); 1578 if (IS_ERR(req->ki_eventfd)) { 1579 ret = PTR_ERR(req->ki_eventfd); 1580 req->ki_eventfd = NULL; 1581 goto out_put_req; 1582 } 1583 1584 req->common.ki_flags |= IOCB_EVENTFD; 1585 } 1586 1587 ret = put_user(KIOCB_KEY, &user_iocb->aio_key); 1588 if (unlikely(ret)) { 1589 pr_debug("EFAULT: aio_key\n"); 1590 goto out_put_req; 1591 } 1592 1593 req->ki_user_iocb = user_iocb; 1594 req->ki_user_data = iocb->aio_data; 1595 1596 get_file(file); 1597 switch (iocb->aio_lio_opcode) { 1598 case IOCB_CMD_PREAD: 1599 ret = aio_read(&req->common, iocb, false, compat); 1600 break; 1601 case IOCB_CMD_PWRITE: 1602 ret = aio_write(&req->common, iocb, false, compat); 1603 break; 1604 case IOCB_CMD_PREADV: 1605 ret = aio_read(&req->common, iocb, true, compat); 1606 break; 1607 case IOCB_CMD_PWRITEV: 1608 ret = aio_write(&req->common, iocb, true, compat); 1609 break; 1610 default: 1611 pr_debug("invalid aio operation %d\n", iocb->aio_lio_opcode); 1612 ret = -EINVAL; 1613 break; 1614 } 1615 fput(file); 1616 1617 if (ret && ret != -EIOCBQUEUED) 1618 goto out_put_req; 1619 return 0; 1620 out_put_req: 1621 put_reqs_available(ctx, 1); 1622 percpu_ref_put(&ctx->reqs); 1623 kiocb_free(req); 1624 return ret; 1625 } 1626 1627 static long do_io_submit(aio_context_t ctx_id, long nr, 1628 struct iocb __user *__user *iocbpp, bool compat) 1629 { 1630 struct kioctx *ctx; 1631 long ret = 0; 1632 int i = 0; 1633 struct blk_plug plug; 1634 1635 if (unlikely(nr < 0)) 1636 return -EINVAL; 1637 1638 if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) 1639 nr = LONG_MAX/sizeof(*iocbpp); 1640 1641 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) 1642 return -EFAULT; 1643 1644 ctx = lookup_ioctx(ctx_id); 1645 if (unlikely(!ctx)) { 1646 pr_debug("EINVAL: invalid context id\n"); 1647 return -EINVAL; 1648 } 1649 1650 blk_start_plug(&plug); 1651 1652 /* 1653 * AKPM: should this return a partial result if some of the IOs were 1654 * successfully submitted? 1655 */ 1656 for (i=0; i<nr; i++) { 1657 struct iocb __user *user_iocb; 1658 struct iocb tmp; 1659 1660 if (unlikely(__get_user(user_iocb, iocbpp + i))) { 1661 ret = -EFAULT; 1662 break; 1663 } 1664 1665 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { 1666 ret = -EFAULT; 1667 break; 1668 } 1669 1670 ret = io_submit_one(ctx, user_iocb, &tmp, compat); 1671 if (ret) 1672 break; 1673 } 1674 blk_finish_plug(&plug); 1675 1676 percpu_ref_put(&ctx->users); 1677 return i ? i : ret; 1678 } 1679 1680 /* sys_io_submit: 1681 * Queue the nr iocbs pointed to by iocbpp for processing. Returns 1682 * the number of iocbs queued. May return -EINVAL if the aio_context 1683 * specified by ctx_id is invalid, if nr is < 0, if the iocb at 1684 * *iocbpp[0] is not properly initialized, if the operation specified 1685 * is invalid for the file descriptor in the iocb. May fail with 1686 * -EFAULT if any of the data structures point to invalid data. May 1687 * fail with -EBADF if the file descriptor specified in the first 1688 * iocb is invalid. May fail with -EAGAIN if insufficient resources 1689 * are available to queue any iocbs. Will return 0 if nr is 0. Will 1690 * fail with -ENOSYS if not implemented. 1691 */ 1692 SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, 1693 struct iocb __user * __user *, iocbpp) 1694 { 1695 return do_io_submit(ctx_id, nr, iocbpp, 0); 1696 } 1697 1698 #ifdef CONFIG_COMPAT 1699 static inline long 1700 copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64) 1701 { 1702 compat_uptr_t uptr; 1703 int i; 1704 1705 for (i = 0; i < nr; ++i) { 1706 if (get_user(uptr, ptr32 + i)) 1707 return -EFAULT; 1708 if (put_user(compat_ptr(uptr), ptr64 + i)) 1709 return -EFAULT; 1710 } 1711 return 0; 1712 } 1713 1714 #define MAX_AIO_SUBMITS (PAGE_SIZE/sizeof(struct iocb *)) 1715 1716 COMPAT_SYSCALL_DEFINE3(io_submit, compat_aio_context_t, ctx_id, 1717 int, nr, u32 __user *, iocb) 1718 { 1719 struct iocb __user * __user *iocb64; 1720 long ret; 1721 1722 if (unlikely(nr < 0)) 1723 return -EINVAL; 1724 1725 if (nr > MAX_AIO_SUBMITS) 1726 nr = MAX_AIO_SUBMITS; 1727 1728 iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64)); 1729 ret = copy_iocb(nr, iocb, iocb64); 1730 if (!ret) 1731 ret = do_io_submit(ctx_id, nr, iocb64, 1); 1732 return ret; 1733 } 1734 #endif 1735 1736 /* lookup_kiocb 1737 * Finds a given iocb for cancellation. 1738 */ 1739 static struct aio_kiocb * 1740 lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, u32 key) 1741 { 1742 struct aio_kiocb *kiocb; 1743 1744 assert_spin_locked(&ctx->ctx_lock); 1745 1746 if (key != KIOCB_KEY) 1747 return NULL; 1748 1749 /* TODO: use a hash or array, this sucks. */ 1750 list_for_each_entry(kiocb, &ctx->active_reqs, ki_list) { 1751 if (kiocb->ki_user_iocb == iocb) 1752 return kiocb; 1753 } 1754 return NULL; 1755 } 1756 1757 /* sys_io_cancel: 1758 * Attempts to cancel an iocb previously passed to io_submit. If 1759 * the operation is successfully cancelled, the resulting event is 1760 * copied into the memory pointed to by result without being placed 1761 * into the completion queue and 0 is returned. May fail with 1762 * -EFAULT if any of the data structures pointed to are invalid. 1763 * May fail with -EINVAL if aio_context specified by ctx_id is 1764 * invalid. May fail with -EAGAIN if the iocb specified was not 1765 * cancelled. Will fail with -ENOSYS if not implemented. 1766 */ 1767 SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, 1768 struct io_event __user *, result) 1769 { 1770 struct kioctx *ctx; 1771 struct aio_kiocb *kiocb; 1772 u32 key; 1773 int ret; 1774 1775 ret = get_user(key, &iocb->aio_key); 1776 if (unlikely(ret)) 1777 return -EFAULT; 1778 1779 ctx = lookup_ioctx(ctx_id); 1780 if (unlikely(!ctx)) 1781 return -EINVAL; 1782 1783 spin_lock_irq(&ctx->ctx_lock); 1784 1785 kiocb = lookup_kiocb(ctx, iocb, key); 1786 if (kiocb) 1787 ret = kiocb_cancel(kiocb); 1788 else 1789 ret = -EINVAL; 1790 1791 spin_unlock_irq(&ctx->ctx_lock); 1792 1793 if (!ret) { 1794 /* 1795 * The result argument is no longer used - the io_event is 1796 * always delivered via the ring buffer. -EINPROGRESS indicates 1797 * cancellation is progress: 1798 */ 1799 ret = -EINPROGRESS; 1800 } 1801 1802 percpu_ref_put(&ctx->users); 1803 1804 return ret; 1805 } 1806 1807 /* io_getevents: 1808 * Attempts to read at least min_nr events and up to nr events from 1809 * the completion queue for the aio_context specified by ctx_id. If 1810 * it succeeds, the number of read events is returned. May fail with 1811 * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is 1812 * out of range, if timeout is out of range. May fail with -EFAULT 1813 * if any of the memory specified is invalid. May return 0 or 1814 * < min_nr if the timeout specified by timeout has elapsed 1815 * before sufficient events are available, where timeout == NULL 1816 * specifies an infinite timeout. Note that the timeout pointed to by 1817 * timeout is relative. Will fail with -ENOSYS if not implemented. 1818 */ 1819 SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, 1820 long, min_nr, 1821 long, nr, 1822 struct io_event __user *, events, 1823 struct timespec __user *, timeout) 1824 { 1825 struct kioctx *ioctx = lookup_ioctx(ctx_id); 1826 long ret = -EINVAL; 1827 1828 if (likely(ioctx)) { 1829 if (likely(min_nr <= nr && min_nr >= 0)) 1830 ret = read_events(ioctx, min_nr, nr, events, timeout); 1831 percpu_ref_put(&ioctx->users); 1832 } 1833 return ret; 1834 } 1835 1836 #ifdef CONFIG_COMPAT 1837 COMPAT_SYSCALL_DEFINE5(io_getevents, compat_aio_context_t, ctx_id, 1838 compat_long_t, min_nr, 1839 compat_long_t, nr, 1840 struct io_event __user *, events, 1841 struct compat_timespec __user *, timeout) 1842 { 1843 struct timespec t; 1844 struct timespec __user *ut = NULL; 1845 1846 if (timeout) { 1847 if (compat_get_timespec(&t, timeout)) 1848 return -EFAULT; 1849 1850 ut = compat_alloc_user_space(sizeof(*ut)); 1851 if (copy_to_user(ut, &t, sizeof(t))) 1852 return -EFAULT; 1853 } 1854 return sys_io_getevents(ctx_id, min_nr, nr, events, ut); 1855 } 1856 #endif 1857