diff options
Diffstat (limited to 'fs/aio.c')
| -rw-r--r-- | fs/aio.c | 2406 |
1 files changed, 1766 insertions, 640 deletions
@@ -5,6 +5,7 @@ * Implements an efficient asynchronous io interface. * * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved. + * Copyright 2018 Christoph Hellwig. * * See ../COPYING for licensing terms. */ @@ -18,14 +19,15 @@ #include <linux/export.h> #include <linux/syscalls.h> #include <linux/backing-dev.h> +#include <linux/refcount.h> #include <linux/uio.h> -#include <linux/sched.h> +#include <linux/sched/signal.h> #include <linux/fs.h> #include <linux/file.h> #include <linux/mm.h> #include <linux/mman.h> -#include <linux/mmu_context.h> +#include <linux/percpu.h> #include <linux/slab.h> #include <linux/timer.h> #include <linux/aio.h> @@ -35,19 +37,27 @@ #include <linux/eventfd.h> #include <linux/blkdev.h> #include <linux/compat.h> +#include <linux/migrate.h> +#include <linux/ramfs.h> +#include <linux/percpu-refcount.h> +#include <linux/mount.h> +#include <linux/pseudo_fs.h> -#include <asm/kmap_types.h> -#include <asm/uaccess.h> +#include <linux/uaccess.h> +#include <linux/nospec.h> #include "internal.h" +#define KIOCB_KEY 0 + #define AIO_RING_MAGIC 0xa10a10a1 #define AIO_RING_COMPAT_FEATURES 1 #define AIO_RING_INCOMPAT_FEATURES 0 struct aio_ring { unsigned id; /* kernel internal index number */ unsigned nr; /* number of io_events */ - unsigned head; + unsigned head; /* Written to by userland or under ring_lock + * mutex by aio_read_events_ring(). */ unsigned tail; unsigned magic; @@ -56,20 +66,48 @@ struct aio_ring { unsigned header_length; /* size of aio_ring */ - struct io_event io_events[0]; + struct io_event io_events[]; }; /* 128 bytes + ring size */ +/* + * Plugging is meant to work with larger batches of IOs. If we don't + * have more than the below, then don't bother setting up a plug. + */ +#define AIO_PLUG_THRESHOLD 2 + #define AIO_RING_PAGES 8 +struct kioctx_table { + struct rcu_head rcu; + unsigned nr; + struct kioctx __rcu *table[] __counted_by(nr); +}; + +struct kioctx_cpu { + unsigned reqs_available; +}; + +struct ctx_rq_wait { + struct completion comp; + atomic_t count; +}; + struct kioctx { - atomic_t users; + struct percpu_ref users; atomic_t dead; - /* This needs improving */ + struct percpu_ref reqs; + unsigned long user_id; - struct hlist_node list; + + struct kioctx_cpu __percpu *cpu; /* + * For percpu reqs_available, number of slots we move to/from global + * counter at a time: + */ + unsigned req_batch; + /* * This is what userspace passed to io_setup(), it's not used for * anything but counting against the global max_reqs quota. * @@ -84,14 +122,26 @@ struct kioctx { unsigned long mmap_base; unsigned long mmap_size; - struct page **ring_pages; + struct folio **ring_folios; long nr_pages; - struct rcu_head rcu_head; - struct work_struct rcu_work; + struct rcu_work free_rwork; /* see free_ioctx() */ + + /* + * signals when all in-flight requests are done + */ + struct ctx_rq_wait *rq_wait; struct { - atomic_t reqs_active; + /* + * This counts the number of available slots in the ringbuffer, + * so we avoid overflowing it: it's decremented (if positive) + * when allocating a kiocb and incremented when the resulting + * io_event is pulled off the ringbuffer. + * + * We batch accesses to it with a percpu version. + */ + atomic_t reqs_available; } ____cacheline_aligned_in_smp; struct { @@ -106,114 +156,423 @@ struct kioctx { struct { unsigned tail; + unsigned completed_events; spinlock_t completion_lock; } ____cacheline_aligned_in_smp; - struct page *internal_pages[AIO_RING_PAGES]; + struct folio *internal_folios[AIO_RING_PAGES]; + struct file *aio_ring_file; + + unsigned id; +}; + +/* + * First field must be the file pointer in all the + * iocb unions! See also 'struct kiocb' in <linux/fs.h> + */ +struct fsync_iocb { + struct file *file; + struct work_struct work; + bool datasync; + struct cred *creds; +}; + +struct poll_iocb { + struct file *file; + struct wait_queue_head *head; + __poll_t events; + bool cancelled; + bool work_scheduled; + bool work_need_resched; + struct wait_queue_entry wait; + struct work_struct work; +}; + +/* + * NOTE! Each of the iocb union members has the file pointer + * as the first entry in their struct definition. So you can + * access the file pointer through any of the sub-structs, + * or directly as just 'ki_filp' in this struct. + */ +struct aio_kiocb { + union { + struct file *ki_filp; + struct kiocb rw; + struct fsync_iocb fsync; + struct poll_iocb poll; + }; + + struct kioctx *ki_ctx; + kiocb_cancel_fn *ki_cancel; + + struct io_event ki_res; + + struct list_head ki_list; /* the aio core uses this + * for cancellation */ + refcount_t ki_refcnt; + + /* + * If the aio_resfd field of the userspace iocb is not zero, + * this is the underlying eventfd context to deliver events to. + */ + struct eventfd_ctx *ki_eventfd; }; /*------ sysctl variables----*/ static DEFINE_SPINLOCK(aio_nr_lock); -unsigned long aio_nr; /* current system wide number of aio requests */ -unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ +static unsigned long aio_nr; /* current system wide number of aio requests */ +static unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ /*----end sysctl variables---*/ +#ifdef CONFIG_SYSCTL +static const struct ctl_table aio_sysctls[] = { + { + .procname = "aio-nr", + .data = &aio_nr, + .maxlen = sizeof(aio_nr), + .mode = 0444, + .proc_handler = proc_doulongvec_minmax, + }, + { + .procname = "aio-max-nr", + .data = &aio_max_nr, + .maxlen = sizeof(aio_max_nr), + .mode = 0644, + .proc_handler = proc_doulongvec_minmax, + }, +}; + +static void __init aio_sysctl_init(void) +{ + register_sysctl_init("fs", aio_sysctls); +} +#else +#define aio_sysctl_init() do { } while (0) +#endif static struct kmem_cache *kiocb_cachep; static struct kmem_cache *kioctx_cachep; +static struct vfsmount *aio_mnt; + +static const struct file_operations aio_ring_fops; +static const struct address_space_operations aio_ctx_aops; + +static struct file *aio_private_file(struct kioctx *ctx, loff_t nr_pages) +{ + struct file *file; + struct inode *inode = alloc_anon_inode(aio_mnt->mnt_sb); + if (IS_ERR(inode)) + return ERR_CAST(inode); + + inode->i_mapping->a_ops = &aio_ctx_aops; + inode->i_mapping->i_private_data = ctx; + inode->i_size = PAGE_SIZE * nr_pages; + + file = alloc_file_pseudo(inode, aio_mnt, "[aio]", + O_RDWR, &aio_ring_fops); + if (IS_ERR(file)) + iput(inode); + return file; +} + +static int aio_init_fs_context(struct fs_context *fc) +{ + if (!init_pseudo(fc, AIO_RING_MAGIC)) + return -ENOMEM; + fc->s_iflags |= SB_I_NOEXEC; + return 0; +} + /* aio_setup * Creates the slab caches used by the aio routines, panic on * failure as this is done early during the boot sequence. */ static int __init aio_setup(void) { - kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); + static struct file_system_type aio_fs = { + .name = "aio", + .init_fs_context = aio_init_fs_context, + .kill_sb = kill_anon_super, + }; + aio_mnt = kern_mount(&aio_fs); + if (IS_ERR(aio_mnt)) + panic("Failed to create aio fs mount."); + + kiocb_cachep = KMEM_CACHE(aio_kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); - - pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); - + aio_sysctl_init(); return 0; } __initcall(aio_setup); +static void put_aio_ring_file(struct kioctx *ctx) +{ + struct file *aio_ring_file = ctx->aio_ring_file; + struct address_space *i_mapping; + + if (aio_ring_file) { + truncate_setsize(file_inode(aio_ring_file), 0); + + /* Prevent further access to the kioctx from migratepages */ + i_mapping = aio_ring_file->f_mapping; + spin_lock(&i_mapping->i_private_lock); + i_mapping->i_private_data = NULL; + ctx->aio_ring_file = NULL; + spin_unlock(&i_mapping->i_private_lock); + + fput(aio_ring_file); + } +} + static void aio_free_ring(struct kioctx *ctx) { - long i; + int i; + + /* Disconnect the kiotx from the ring file. This prevents future + * accesses to the kioctx from page migration. + */ + put_aio_ring_file(ctx); + + for (i = 0; i < ctx->nr_pages; i++) { + struct folio *folio = ctx->ring_folios[i]; - for (i = 0; i < ctx->nr_pages; i++) - put_page(ctx->ring_pages[i]); + if (!folio) + continue; - if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) - kfree(ctx->ring_pages); + pr_debug("pid(%d) [%d] folio->count=%d\n", current->pid, i, + folio_ref_count(folio)); + ctx->ring_folios[i] = NULL; + folio_put(folio); + } + + if (ctx->ring_folios && ctx->ring_folios != ctx->internal_folios) { + kfree(ctx->ring_folios); + ctx->ring_folios = NULL; + } } -static int aio_setup_ring(struct kioctx *ctx) +static int aio_ring_mremap(struct vm_area_struct *vma) +{ + struct file *file = vma->vm_file; + struct mm_struct *mm = vma->vm_mm; + struct kioctx_table *table; + int i, res = -EINVAL; + + spin_lock(&mm->ioctx_lock); + rcu_read_lock(); + table = rcu_dereference(mm->ioctx_table); + if (!table) + goto out_unlock; + + for (i = 0; i < table->nr; i++) { + struct kioctx *ctx; + + ctx = rcu_dereference(table->table[i]); + if (ctx && ctx->aio_ring_file == file) { + if (!atomic_read(&ctx->dead)) { + ctx->user_id = ctx->mmap_base = vma->vm_start; + res = 0; + } + break; + } + } + +out_unlock: + rcu_read_unlock(); + spin_unlock(&mm->ioctx_lock); + return res; +} + +static const struct vm_operations_struct aio_ring_vm_ops = { + .mremap = aio_ring_mremap, +#if IS_ENABLED(CONFIG_MMU) + .fault = filemap_fault, + .map_pages = filemap_map_pages, + .page_mkwrite = filemap_page_mkwrite, +#endif +}; + +static int aio_ring_mmap_prepare(struct vm_area_desc *desc) +{ + desc->vm_flags |= VM_DONTEXPAND; + desc->vm_ops = &aio_ring_vm_ops; + return 0; +} + +static const struct file_operations aio_ring_fops = { + .mmap_prepare = aio_ring_mmap_prepare, +}; + +#if IS_ENABLED(CONFIG_MIGRATION) +static int aio_migrate_folio(struct address_space *mapping, struct folio *dst, + struct folio *src, enum migrate_mode mode) +{ + struct kioctx *ctx; + unsigned long flags; + pgoff_t idx; + int rc = 0; + + /* mapping->i_private_lock here protects against the kioctx teardown. */ + spin_lock(&mapping->i_private_lock); + ctx = mapping->i_private_data; + if (!ctx) { + rc = -EINVAL; + goto out; + } + + /* The ring_lock mutex. The prevents aio_read_events() from writing + * to the ring's head, and prevents page migration from mucking in + * a partially initialized kiotx. + */ + if (!mutex_trylock(&ctx->ring_lock)) { + rc = -EAGAIN; + goto out; + } + + idx = src->index; + if (idx < (pgoff_t)ctx->nr_pages) { + /* Make sure the old folio hasn't already been changed */ + if (ctx->ring_folios[idx] != src) + rc = -EAGAIN; + } else + rc = -EINVAL; + + if (rc != 0) + goto out_unlock; + + /* Writeback must be complete */ + BUG_ON(folio_test_writeback(src)); + folio_get(dst); + + rc = folio_migrate_mapping(mapping, dst, src, 1); + if (rc) { + folio_put(dst); + goto out_unlock; + } + + /* Take completion_lock to prevent other writes to the ring buffer + * while the old folio is copied to the new. This prevents new + * events from being lost. + */ + spin_lock_irqsave(&ctx->completion_lock, flags); + folio_copy(dst, src); + folio_migrate_flags(dst, src); + BUG_ON(ctx->ring_folios[idx] != src); + ctx->ring_folios[idx] = dst; + spin_unlock_irqrestore(&ctx->completion_lock, flags); + + /* The old folio is no longer accessible. */ + folio_put(src); + +out_unlock: + mutex_unlock(&ctx->ring_lock); +out: + spin_unlock(&mapping->i_private_lock); + return rc; +} +#else +#define aio_migrate_folio NULL +#endif + +static const struct address_space_operations aio_ctx_aops = { + .dirty_folio = noop_dirty_folio, + .migrate_folio = aio_migrate_folio, +}; + +static int aio_setup_ring(struct kioctx *ctx, unsigned int nr_events) { struct aio_ring *ring; - unsigned nr_events = ctx->max_reqs; struct mm_struct *mm = current->mm; - unsigned long size, populate; + unsigned long size, unused; int nr_pages; + int i; + struct file *file; /* Compensate for the ring buffer's head/tail overlap entry */ nr_events += 2; /* 1 is required, 2 for good luck */ size = sizeof(struct aio_ring); size += sizeof(struct io_event) * nr_events; - nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; + nr_pages = PFN_UP(size); if (nr_pages < 0) return -EINVAL; - nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); + file = aio_private_file(ctx, nr_pages); + if (IS_ERR(file)) { + ctx->aio_ring_file = NULL; + return -ENOMEM; + } + + ctx->aio_ring_file = file; + nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) + / sizeof(struct io_event); - ctx->nr_events = 0; - ctx->ring_pages = ctx->internal_pages; + ctx->ring_folios = ctx->internal_folios; if (nr_pages > AIO_RING_PAGES) { - ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), - GFP_KERNEL); - if (!ctx->ring_pages) + ctx->ring_folios = kcalloc(nr_pages, sizeof(struct folio *), + GFP_KERNEL); + if (!ctx->ring_folios) { + put_aio_ring_file(ctx); return -ENOMEM; + } + } + + for (i = 0; i < nr_pages; i++) { + struct folio *folio; + + folio = __filemap_get_folio(file->f_mapping, i, + FGP_LOCK | FGP_ACCESSED | FGP_CREAT, + GFP_USER | __GFP_ZERO); + if (IS_ERR(folio)) + break; + + pr_debug("pid(%d) [%d] folio->count=%d\n", current->pid, i, + folio_ref_count(folio)); + folio_end_read(folio, true); + + ctx->ring_folios[i] = folio; + } + ctx->nr_pages = i; + + if (unlikely(i != nr_pages)) { + aio_free_ring(ctx); + return -ENOMEM; } ctx->mmap_size = nr_pages * PAGE_SIZE; pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); - down_write(&mm->mmap_sem); - ctx->mmap_base = do_mmap_pgoff(NULL, 0, ctx->mmap_size, - PROT_READ|PROT_WRITE, - MAP_ANONYMOUS|MAP_PRIVATE, 0, &populate); - if (IS_ERR((void *)ctx->mmap_base)) { - up_write(&mm->mmap_sem); + + if (mmap_write_lock_killable(mm)) { ctx->mmap_size = 0; aio_free_ring(ctx); - return -EAGAIN; + return -EINTR; } - pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); - ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, - 1, 0, ctx->ring_pages, NULL); - up_write(&mm->mmap_sem); - - if (unlikely(ctx->nr_pages != nr_pages)) { + ctx->mmap_base = do_mmap(ctx->aio_ring_file, 0, ctx->mmap_size, + PROT_READ | PROT_WRITE, + MAP_SHARED, 0, 0, &unused, NULL); + mmap_write_unlock(mm); + if (IS_ERR((void *)ctx->mmap_base)) { + ctx->mmap_size = 0; aio_free_ring(ctx); - return -EAGAIN; + return -ENOMEM; } - if (populate) - mm_populate(ctx->mmap_base, populate); + + pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); ctx->user_id = ctx->mmap_base; ctx->nr_events = nr_events; /* trusted copy */ - ring = kmap_atomic(ctx->ring_pages[0]); + ring = folio_address(ctx->ring_folios[0]); ring->nr = nr_events; /* user copy */ - ring->id = ctx->user_id; + ring->id = ~0U; ring->head = ring->tail = 0; ring->magic = AIO_RING_MAGIC; ring->compat_features = AIO_RING_COMPAT_FEATURES; ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; ring->header_length = sizeof(struct aio_ring); - kunmap_atomic(ring); - flush_dcache_page(ctx->ring_pages[0]); + flush_dcache_folio(ctx->ring_folios[0]); return 0; } @@ -222,59 +581,62 @@ static int aio_setup_ring(struct kioctx *ctx) #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) -void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) +void kiocb_set_cancel_fn(struct kiocb *iocb, kiocb_cancel_fn *cancel) { - struct kioctx *ctx = req->ki_ctx; + struct aio_kiocb *req; + struct kioctx *ctx; unsigned long flags; - spin_lock_irqsave(&ctx->ctx_lock, flags); + /* + * kiocb didn't come from aio or is neither a read nor a write, hence + * ignore it. + */ + if (!(iocb->ki_flags & IOCB_AIO_RW)) + return; - if (!req->ki_list.next) - list_add(&req->ki_list, &ctx->active_reqs); + req = container_of(iocb, struct aio_kiocb, rw); - req->ki_cancel = cancel; + if (WARN_ON_ONCE(!list_empty(&req->ki_list))) + return; + ctx = req->ki_ctx; + + spin_lock_irqsave(&ctx->ctx_lock, flags); + list_add_tail(&req->ki_list, &ctx->active_reqs); + req->ki_cancel = cancel; spin_unlock_irqrestore(&ctx->ctx_lock, flags); } EXPORT_SYMBOL(kiocb_set_cancel_fn); -static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, - struct io_event *res) +/* + * free_ioctx() should be RCU delayed to synchronize against the RCU + * protected lookup_ioctx() and also needs process context to call + * aio_free_ring(). Use rcu_work. + */ +static void free_ioctx(struct work_struct *work) { - kiocb_cancel_fn *old, *cancel; - int ret = -EINVAL; - - /* - * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it - * actually has a cancel function, hence the cmpxchg() - */ - - cancel = ACCESS_ONCE(kiocb->ki_cancel); - do { - if (!cancel || cancel == KIOCB_CANCELLED) - return ret; - - old = cancel; - cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); - } while (cancel != old); - - atomic_inc(&kiocb->ki_users); - spin_unlock_irq(&ctx->ctx_lock); - - memset(res, 0, sizeof(*res)); - res->obj = (u64)(unsigned long)kiocb->ki_obj.user; - res->data = kiocb->ki_user_data; - ret = cancel(kiocb, res); - - spin_lock_irq(&ctx->ctx_lock); + struct kioctx *ctx = container_of(to_rcu_work(work), struct kioctx, + free_rwork); + pr_debug("freeing %p\n", ctx); - return ret; + aio_free_ring(ctx); + free_percpu(ctx->cpu); + percpu_ref_exit(&ctx->reqs); + percpu_ref_exit(&ctx->users); + kmem_cache_free(kioctx_cachep, ctx); } -static void free_ioctx_rcu(struct rcu_head *head) +static void free_ioctx_reqs(struct percpu_ref *ref) { - struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); - kmem_cache_free(kioctx_cachep, ctx); + struct kioctx *ctx = container_of(ref, struct kioctx, reqs); + + /* At this point we know that there are no any in-flight requests */ + if (ctx->rq_wait && atomic_dec_and_test(&ctx->rq_wait->count)) + complete(&ctx->rq_wait->comp); + + /* Synchronize against RCU protected table->table[] dereferences */ + INIT_RCU_WORK(&ctx->free_rwork, free_ioctx); + queue_rcu_work(system_percpu_wq, &ctx->free_rwork); } /* @@ -282,61 +644,87 @@ static void free_ioctx_rcu(struct rcu_head *head) * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - * now it's safe to cancel any that need to be. */ -static void free_ioctx(struct kioctx *ctx) +static void free_ioctx_users(struct percpu_ref *ref) { - struct aio_ring *ring; - struct io_event res; - struct kiocb *req; - unsigned head, avail; + struct kioctx *ctx = container_of(ref, struct kioctx, users); + struct aio_kiocb *req; spin_lock_irq(&ctx->ctx_lock); while (!list_empty(&ctx->active_reqs)) { req = list_first_entry(&ctx->active_reqs, - struct kiocb, ki_list); - + struct aio_kiocb, ki_list); + req->ki_cancel(&req->rw); list_del_init(&req->ki_list); - kiocb_cancel(ctx, req, &res); } spin_unlock_irq(&ctx->ctx_lock); - ring = kmap_atomic(ctx->ring_pages[0]); - head = ring->head; - kunmap_atomic(ring); - - while (atomic_read(&ctx->reqs_active) > 0) { - wait_event(ctx->wait, - head != ctx->tail || - atomic_read(&ctx->reqs_active) <= 0); + percpu_ref_kill(&ctx->reqs); + percpu_ref_put(&ctx->reqs); +} - avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; +static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) +{ + unsigned i, new_nr; + struct kioctx_table *table, *old; + struct aio_ring *ring; - atomic_sub(avail, &ctx->reqs_active); - head += avail; - head %= ctx->nr_events; - } + spin_lock(&mm->ioctx_lock); + table = rcu_dereference_raw(mm->ioctx_table); + + while (1) { + if (table) + for (i = 0; i < table->nr; i++) + if (!rcu_access_pointer(table->table[i])) { + ctx->id = i; + rcu_assign_pointer(table->table[i], ctx); + spin_unlock(&mm->ioctx_lock); + + /* While kioctx setup is in progress, + * we are protected from page migration + * changes ring_folios by ->ring_lock. + */ + ring = folio_address(ctx->ring_folios[0]); + ring->id = ctx->id; + return 0; + } + + new_nr = (table ? table->nr : 1) * 4; + spin_unlock(&mm->ioctx_lock); + + table = kzalloc(struct_size(table, table, new_nr), GFP_KERNEL); + if (!table) + return -ENOMEM; - WARN_ON(atomic_read(&ctx->reqs_active) < 0); + table->nr = new_nr; - aio_free_ring(ctx); + spin_lock(&mm->ioctx_lock); + old = rcu_dereference_raw(mm->ioctx_table); - pr_debug("freeing %p\n", ctx); + if (!old) { + rcu_assign_pointer(mm->ioctx_table, table); + } else if (table->nr > old->nr) { + memcpy(table->table, old->table, + old->nr * sizeof(struct kioctx *)); - /* - * Here the call_rcu() is between the wait_event() for reqs_active to - * hit 0, and freeing the ioctx. - * - * aio_complete() decrements reqs_active, but it has to touch the ioctx - * after to issue a wakeup so we use rcu. - */ - call_rcu(&ctx->rcu_head, free_ioctx_rcu); + rcu_assign_pointer(mm->ioctx_table, table); + kfree_rcu(old, rcu); + } else { + kfree(table); + table = old; + } + } } -static void put_ioctx(struct kioctx *ctx) +static void aio_nr_sub(unsigned nr) { - if (unlikely(atomic_dec_and_test(&ctx->users))) - free_ioctx(ctx); + spin_lock(&aio_nr_lock); + if (WARN_ON(aio_nr - nr > aio_nr)) + aio_nr = 0; + else + aio_nr -= nr; + spin_unlock(&aio_nr_lock); } /* ioctx_alloc @@ -348,123 +736,150 @@ static struct kioctx *ioctx_alloc(unsigned nr_events) struct kioctx *ctx; int err = -ENOMEM; + /* + * Store the original nr_events -- what userspace passed to io_setup(), + * for counting against the global limit -- before it changes. + */ + unsigned int max_reqs = nr_events; + + /* + * We keep track of the number of available ringbuffer slots, to prevent + * overflow (reqs_available), and we also use percpu counters for this. + * + * So since up to half the slots might be on other cpu's percpu counters + * and unavailable, double nr_events so userspace sees what they + * expected: additionally, we move req_batch slots to/from percpu + * counters at a time, so make sure that isn't 0: + */ + nr_events = max(nr_events, num_possible_cpus() * 4); + nr_events *= 2; + /* Prevent overflows */ - if ((nr_events > (0x10000000U / sizeof(struct io_event))) || - (nr_events > (0x10000000U / sizeof(struct kiocb)))) { + if (nr_events > (0x10000000U / sizeof(struct io_event))) { pr_debug("ENOMEM: nr_events too high\n"); return ERR_PTR(-EINVAL); } - if (!nr_events || (unsigned long)nr_events > aio_max_nr) + if (!nr_events || (unsigned long)max_reqs > aio_max_nr) return ERR_PTR(-EAGAIN); ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); if (!ctx) return ERR_PTR(-ENOMEM); - ctx->max_reqs = nr_events; + ctx->max_reqs = max_reqs; - atomic_set(&ctx->users, 2); - atomic_set(&ctx->dead, 0); spin_lock_init(&ctx->ctx_lock); spin_lock_init(&ctx->completion_lock); mutex_init(&ctx->ring_lock); + /* Protect against page migration throughout kiotx setup by keeping + * the ring_lock mutex held until setup is complete. */ + mutex_lock(&ctx->ring_lock); init_waitqueue_head(&ctx->wait); INIT_LIST_HEAD(&ctx->active_reqs); - if (aio_setup_ring(ctx) < 0) - goto out_freectx; + if (percpu_ref_init(&ctx->users, free_ioctx_users, 0, GFP_KERNEL)) + goto err; + + if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs, 0, GFP_KERNEL)) + goto err; + + ctx->cpu = alloc_percpu(struct kioctx_cpu); + if (!ctx->cpu) + goto err; + + err = aio_setup_ring(ctx, nr_events); + if (err < 0) + goto err; + + atomic_set(&ctx->reqs_available, ctx->nr_events - 1); + ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); + if (ctx->req_batch < 1) + ctx->req_batch = 1; /* limit the number of system wide aios */ spin_lock(&aio_nr_lock); - if (aio_nr + nr_events > aio_max_nr || - aio_nr + nr_events < aio_nr) { + if (aio_nr + ctx->max_reqs > aio_max_nr || + aio_nr + ctx->max_reqs < aio_nr) { spin_unlock(&aio_nr_lock); - goto out_cleanup; + err = -EAGAIN; + goto err_ctx; } aio_nr += ctx->max_reqs; spin_unlock(&aio_nr_lock); - /* now link into global list. */ - spin_lock(&mm->ioctx_lock); - hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); - spin_unlock(&mm->ioctx_lock); + percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ + percpu_ref_get(&ctx->reqs); /* free_ioctx_users() will drop this */ + + err = ioctx_add_table(ctx, mm); + if (err) + goto err_cleanup; + + /* Release the ring_lock mutex now that all setup is complete. */ + mutex_unlock(&ctx->ring_lock); pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", ctx, ctx->user_id, mm, ctx->nr_events); return ctx; -out_cleanup: - err = -EAGAIN; +err_cleanup: + aio_nr_sub(ctx->max_reqs); +err_ctx: + atomic_set(&ctx->dead, 1); + if (ctx->mmap_size) + vm_munmap(ctx->mmap_base, ctx->mmap_size); aio_free_ring(ctx); -out_freectx: +err: + mutex_unlock(&ctx->ring_lock); + free_percpu(ctx->cpu); + percpu_ref_exit(&ctx->reqs); + percpu_ref_exit(&ctx->users); kmem_cache_free(kioctx_cachep, ctx); pr_debug("error allocating ioctx %d\n", err); return ERR_PTR(err); } -static void kill_ioctx_work(struct work_struct *work) -{ - struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); - - wake_up_all(&ctx->wait); - put_ioctx(ctx); -} - -static void kill_ioctx_rcu(struct rcu_head *head) -{ - struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); - - INIT_WORK(&ctx->rcu_work, kill_ioctx_work); - schedule_work(&ctx->rcu_work); -} - /* kill_ioctx * Cancels all outstanding aio requests on an aio context. Used * when the processes owning a context have all exited to encourage * the rapid destruction of the kioctx. */ -static void kill_ioctx(struct kioctx *ctx) +static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx, + struct ctx_rq_wait *wait) { - if (!atomic_xchg(&ctx->dead, 1)) { - hlist_del_rcu(&ctx->list); + struct kioctx_table *table; - /* - * It'd be more correct to do this in free_ioctx(), after all - * the outstanding kiocbs have finished - but by then io_destroy - * has already returned, so io_setup() could potentially return - * -EAGAIN with no ioctxs actually in use (as far as userspace - * could tell). - */ - spin_lock(&aio_nr_lock); - BUG_ON(aio_nr - ctx->max_reqs > aio_nr); - aio_nr -= ctx->max_reqs; - spin_unlock(&aio_nr_lock); + spin_lock(&mm->ioctx_lock); + if (atomic_xchg(&ctx->dead, 1)) { + spin_unlock(&mm->ioctx_lock); + return -EINVAL; + } - if (ctx->mmap_size) - vm_munmap(ctx->mmap_base, ctx->mmap_size); + table = rcu_dereference_raw(mm->ioctx_table); + WARN_ON(ctx != rcu_access_pointer(table->table[ctx->id])); + RCU_INIT_POINTER(table->table[ctx->id], NULL); + spin_unlock(&mm->ioctx_lock); - /* Between hlist_del_rcu() and dropping the initial ref */ - call_rcu(&ctx->rcu_head, kill_ioctx_rcu); - } -} + /* free_ioctx_reqs() will do the necessary RCU synchronization */ + wake_up_all(&ctx->wait); -/* wait_on_sync_kiocb: - * Waits on the given sync kiocb to complete. - */ -ssize_t wait_on_sync_kiocb(struct kiocb *iocb) -{ - while (atomic_read(&iocb->ki_users)) { - set_current_state(TASK_UNINTERRUPTIBLE); - if (!atomic_read(&iocb->ki_users)) - break; - io_schedule(); - } - __set_current_state(TASK_RUNNING); - return iocb->ki_user_data; + /* + * It'd be more correct to do this in free_ioctx(), after all + * the outstanding kiocbs have finished - but by then io_destroy + * has already returned, so io_setup() could potentially return + * -EAGAIN with no ioctxs actually in use (as far as userspace + * could tell). + */ + aio_nr_sub(ctx->max_reqs); + + if (ctx->mmap_size) + vm_munmap(ctx->mmap_base, ctx->mmap_size); + + ctx->rq_wait = wait; + percpu_ref_kill(&ctx->users); + return 0; } -EXPORT_SYMBOL(wait_on_sync_kiocb); /* * exit_aio: called when the last user of mm goes away. At this point, there is @@ -476,152 +891,240 @@ EXPORT_SYMBOL(wait_on_sync_kiocb); */ void exit_aio(struct mm_struct *mm) { - struct kioctx *ctx; - struct hlist_node *n; - - hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { - if (1 != atomic_read(&ctx->users)) - printk(KERN_DEBUG - "exit_aio:ioctx still alive: %d %d %d\n", - atomic_read(&ctx->users), - atomic_read(&ctx->dead), - atomic_read(&ctx->reqs_active)); + struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table); + struct ctx_rq_wait wait; + int i, skipped; + + if (!table) + return; + + atomic_set(&wait.count, table->nr); + init_completion(&wait.comp); + + skipped = 0; + for (i = 0; i < table->nr; ++i) { + struct kioctx *ctx = + rcu_dereference_protected(table->table[i], true); + + if (!ctx) { + skipped++; + continue; + } + /* - * We don't need to bother with munmap() here - - * exit_mmap(mm) is coming and it'll unmap everything. - * Since aio_free_ring() uses non-zero ->mmap_size - * as indicator that it needs to unmap the area, - * just set it to 0; aio_free_ring() is the only - * place that uses ->mmap_size, so it's safe. + * We don't need to bother with munmap() here - exit_mmap(mm) + * is coming and it'll unmap everything. And we simply can't, + * this is not necessarily our ->mm. + * Since kill_ioctx() uses non-zero ->mmap_size as indicator + * that it needs to unmap the area, just set it to 0. */ ctx->mmap_size = 0; + kill_ioctx(mm, ctx, &wait); + } - kill_ioctx(ctx); + if (!atomic_sub_and_test(skipped, &wait.count)) { + /* Wait until all IO for the context are done. */ + wait_for_completion(&wait.comp); } + + RCU_INIT_POINTER(mm->ioctx_table, NULL); + kfree(table); } -/* aio_get_req - * Allocate a slot for an aio request. Increments the ki_users count - * of the kioctx so that the kioctx stays around until all requests are - * complete. Returns NULL if no requests are free. - * - * Returns with kiocb->ki_users set to 2. The io submit code path holds - * an extra reference while submitting the i/o. - * This prevents races between the aio code path referencing the - * req (after submitting it) and aio_complete() freeing the req. +static void put_reqs_available(struct kioctx *ctx, unsigned nr) +{ + struct kioctx_cpu *kcpu; + unsigned long flags; + + local_irq_save(flags); + kcpu = this_cpu_ptr(ctx->cpu); + kcpu->reqs_available += nr; + + while (kcpu->reqs_available >= ctx->req_batch * 2) { + kcpu->reqs_available -= ctx->req_batch; + atomic_add(ctx->req_batch, &ctx->reqs_available); + } + + local_irq_restore(flags); +} + +static bool __get_reqs_available(struct kioctx *ctx) +{ + struct kioctx_cpu *kcpu; + bool ret = false; + unsigned long flags; + + local_irq_save(flags); + kcpu = this_cpu_ptr(ctx->cpu); + if (!kcpu->reqs_available) { + int avail = atomic_read(&ctx->reqs_available); + + do { + if (avail < ctx->req_batch) + goto out; + } while (!atomic_try_cmpxchg(&ctx->reqs_available, + &avail, avail - ctx->req_batch)); + + kcpu->reqs_available += ctx->req_batch; + } + + ret = true; + kcpu->reqs_available--; +out: + local_irq_restore(flags); + return ret; +} + +/* refill_reqs_available + * Updates the reqs_available reference counts used for tracking the + * number of free slots in the completion ring. This can be called + * from aio_complete() (to optimistically update reqs_available) or + * from aio_get_req() (the we're out of events case). It must be + * called holding ctx->completion_lock. */ -static inline struct kiocb *aio_get_req(struct kioctx *ctx) +static void refill_reqs_available(struct kioctx *ctx, unsigned head, + unsigned tail) { - struct kiocb *req; + unsigned events_in_ring, completed; - if (atomic_read(&ctx->reqs_active) >= ctx->nr_events) - return NULL; + /* Clamp head since userland can write to it. */ + head %= ctx->nr_events; + if (head <= tail) + events_in_ring = tail - head; + else + events_in_ring = ctx->nr_events - (head - tail); - if (atomic_inc_return(&ctx->reqs_active) > ctx->nr_events - 1) - goto out_put; + completed = ctx->completed_events; + if (events_in_ring < completed) + completed -= events_in_ring; + else + completed = 0; - req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); - if (unlikely(!req)) - goto out_put; + if (!completed) + return; - atomic_set(&req->ki_users, 2); - req->ki_ctx = ctx; + ctx->completed_events -= completed; + put_reqs_available(ctx, completed); +} - return req; -out_put: - atomic_dec(&ctx->reqs_active); - return NULL; +/* user_refill_reqs_available + * Called to refill reqs_available when aio_get_req() encounters an + * out of space in the completion ring. + */ +static void user_refill_reqs_available(struct kioctx *ctx) +{ + spin_lock_irq(&ctx->completion_lock); + if (ctx->completed_events) { + struct aio_ring *ring; + unsigned head; + + /* Access of ring->head may race with aio_read_events_ring() + * here, but that's okay since whether we read the old version + * or the new version, and either will be valid. The important + * part is that head cannot pass tail since we prevent + * aio_complete() from updating tail by holding + * ctx->completion_lock. Even if head is invalid, the check + * against ctx->completed_events below will make sure we do the + * safe/right thing. + */ + ring = folio_address(ctx->ring_folios[0]); + head = ring->head; + + refill_reqs_available(ctx, head, ctx->tail); + } + + spin_unlock_irq(&ctx->completion_lock); } -static void kiocb_free(struct kiocb *req) +static bool get_reqs_available(struct kioctx *ctx) { - if (req->ki_filp) - fput(req->ki_filp); - if (req->ki_eventfd != NULL) - eventfd_ctx_put(req->ki_eventfd); - if (req->ki_dtor) - req->ki_dtor(req); - if (req->ki_iovec != &req->ki_inline_vec) - kfree(req->ki_iovec); - kmem_cache_free(kiocb_cachep, req); + if (__get_reqs_available(ctx)) + return true; + user_refill_reqs_available(ctx); + return __get_reqs_available(ctx); } -void aio_put_req(struct kiocb *req) +/* aio_get_req + * Allocate a slot for an aio request. + * Returns NULL if no requests are free. + * + * The refcount is initialized to 2 - one for the async op completion, + * one for the synchronous code that does this. + */ +static inline struct aio_kiocb *aio_get_req(struct kioctx *ctx) { - if (atomic_dec_and_test(&req->ki_users)) - kiocb_free(req); + struct aio_kiocb *req; + + req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); + if (unlikely(!req)) + return NULL; + + if (unlikely(!get_reqs_available(ctx))) { + kmem_cache_free(kiocb_cachep, req); + return NULL; + } + + percpu_ref_get(&ctx->reqs); + req->ki_ctx = ctx; + INIT_LIST_HEAD(&req->ki_list); + refcount_set(&req->ki_refcnt, 2); + req->ki_eventfd = NULL; + return req; } -EXPORT_SYMBOL(aio_put_req); static struct kioctx *lookup_ioctx(unsigned long ctx_id) { + struct aio_ring __user *ring = (void __user *)ctx_id; struct mm_struct *mm = current->mm; struct kioctx *ctx, *ret = NULL; + struct kioctx_table *table; + unsigned id; + + if (get_user(id, &ring->id)) + return NULL; rcu_read_lock(); + table = rcu_dereference(mm->ioctx_table); - hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { - if (ctx->user_id == ctx_id) { - atomic_inc(&ctx->users); + if (!table || id >= table->nr) + goto out; + + id = array_index_nospec(id, table->nr); + ctx = rcu_dereference(table->table[id]); + if (ctx && ctx->user_id == ctx_id) { + if (percpu_ref_tryget_live(&ctx->users)) ret = ctx; - break; - } } - +out: rcu_read_unlock(); return ret; } +static inline void iocb_destroy(struct aio_kiocb *iocb) +{ + if (iocb->ki_eventfd) + eventfd_ctx_put(iocb->ki_eventfd); + if (iocb->ki_filp) + fput(iocb->ki_filp); + percpu_ref_put(&iocb->ki_ctx->reqs); + kmem_cache_free(kiocb_cachep, iocb); +} + +struct aio_waiter { + struct wait_queue_entry w; + size_t min_nr; +}; + /* aio_complete * Called when the io request on the given iocb is complete. */ -void aio_complete(struct kiocb *iocb, long res, long res2) +static void aio_complete(struct aio_kiocb *iocb) { struct kioctx *ctx = iocb->ki_ctx; struct aio_ring *ring; struct io_event *ev_page, *event; + unsigned tail, pos, head, avail; unsigned long flags; - unsigned tail, pos; - - /* - * Special case handling for sync iocbs: - * - events go directly into the iocb for fast handling - * - the sync task with the iocb in its stack holds the single iocb - * ref, no other paths have a way to get another ref - * - the sync task helpfully left a reference to itself in the iocb - */ - if (is_sync_kiocb(iocb)) { - BUG_ON(atomic_read(&iocb->ki_users) != 1); - iocb->ki_user_data = res; - atomic_set(&iocb->ki_users, 0); - wake_up_process(iocb->ki_obj.tsk); - return; - } - - /* - * Take rcu_read_lock() in case the kioctx is being destroyed, as we - * need to issue a wakeup after decrementing reqs_active. - */ - rcu_read_lock(); - - if (iocb->ki_list.next) { - unsigned long flags; - - spin_lock_irqsave(&ctx->ctx_lock, flags); - list_del(&iocb->ki_list); - spin_unlock_irqrestore(&ctx->ctx_lock, flags); - } - - /* - * cancelled requests don't get events, userland was given one - * when the event got cancelled. - */ - if (unlikely(xchg(&iocb->ki_cancel, - KIOCB_CANCELLED) == KIOCB_CANCELLED)) { - atomic_dec(&ctx->reqs_active); - /* Still need the wake_up in case free_ioctx is waiting */ - goto put_rq; - } /* * Add a completion event to the ring buffer. Must be done holding @@ -636,20 +1139,16 @@ void aio_complete(struct kiocb *iocb, long res, long res2) if (++tail >= ctx->nr_events) tail = 0; - ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); + ev_page = folio_address(ctx->ring_folios[pos / AIO_EVENTS_PER_PAGE]); event = ev_page + pos % AIO_EVENTS_PER_PAGE; - event->obj = (u64)(unsigned long)iocb->ki_obj.user; - event->data = iocb->ki_user_data; - event->res = res; - event->res2 = res2; + *event = iocb->ki_res; - kunmap_atomic(ev_page); - flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); + flush_dcache_folio(ctx->ring_folios[pos / AIO_EVENTS_PER_PAGE]); - pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", - ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, - res, res2); + pr_debug("%p[%u]: %p: %p %Lx %Lx %Lx\n", ctx, tail, iocb, + (void __user *)(unsigned long)iocb->ki_res.obj, + iocb->ki_res.data, iocb->ki_res.res, iocb->ki_res.res2); /* after flagging the request as done, we * must never even look at it again @@ -658,11 +1157,18 @@ void aio_complete(struct kiocb *iocb, long res, long res2) ctx->tail = tail; - ring = kmap_atomic(ctx->ring_pages[0]); + ring = folio_address(ctx->ring_folios[0]); + head = ring->head; ring->tail = tail; - kunmap_atomic(ring); - flush_dcache_page(ctx->ring_pages[0]); + flush_dcache_folio(ctx->ring_folios[0]); + ctx->completed_events++; + if (ctx->completed_events > 1) + refill_reqs_available(ctx, head, tail); + + avail = tail > head + ? tail - head + : tail + ctx->nr_events - head; spin_unlock_irqrestore(&ctx->completion_lock, flags); pr_debug("added to ring %p at [%u]\n", iocb, tail); @@ -672,12 +1178,8 @@ void aio_complete(struct kiocb *iocb, long res, long res2) * eventfd. The eventfd_signal() function is safe to be called * from IRQ context. */ - if (iocb->ki_eventfd != NULL) - eventfd_signal(iocb->ki_eventfd, 1); - -put_rq: - /* everything turned out well, dispose of the aiocb. */ - aio_put_req(iocb); + if (iocb->ki_eventfd) + eventfd_signal(iocb->ki_eventfd); /* * We have to order our ring_info tail store above and test @@ -687,14 +1189,29 @@ put_rq: */ smp_mb(); - if (waitqueue_active(&ctx->wait)) - wake_up(&ctx->wait); + if (waitqueue_active(&ctx->wait)) { + struct aio_waiter *curr, *next; + unsigned long flags; - rcu_read_unlock(); + spin_lock_irqsave(&ctx->wait.lock, flags); + list_for_each_entry_safe(curr, next, &ctx->wait.head, w.entry) + if (avail >= curr->min_nr) { + wake_up_process(curr->w.private); + list_del_init_careful(&curr->w.entry); + } + spin_unlock_irqrestore(&ctx->wait.lock, flags); + } +} + +static inline void iocb_put(struct aio_kiocb *iocb) +{ + if (refcount_dec_and_test(&iocb->ki_refcnt)) { + aio_complete(iocb); + iocb_destroy(iocb); + } } -EXPORT_SYMBOL(aio_complete); -/* aio_read_events +/* aio_read_events_ring * Pull an event off of the ioctx's event ring. Returns the number of * events fetched */ @@ -702,42 +1219,57 @@ static long aio_read_events_ring(struct kioctx *ctx, struct io_event __user *event, long nr) { struct aio_ring *ring; - unsigned head, pos; + unsigned head, tail, pos; long ret = 0; int copy_ret; + /* + * The mutex can block and wake us up and that will cause + * wait_event_interruptible_hrtimeout() to schedule without sleeping + * and repeat. This should be rare enough that it doesn't cause + * peformance issues. See the comment in read_events() for more detail. + */ + sched_annotate_sleep(); mutex_lock(&ctx->ring_lock); - ring = kmap_atomic(ctx->ring_pages[0]); + /* Access to ->ring_folios here is protected by ctx->ring_lock. */ + ring = folio_address(ctx->ring_folios[0]); head = ring->head; - kunmap_atomic(ring); + tail = ring->tail; + + /* + * Ensure that once we've read the current tail pointer, that + * we also see the events that were stored up to the tail. + */ + smp_rmb(); - pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events); + pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); - if (head == ctx->tail) + if (head == tail) goto out; + head %= ctx->nr_events; + tail %= ctx->nr_events; + while (ret < nr) { long avail; struct io_event *ev; - struct page *page; + struct folio *folio; - avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; - if (head == ctx->tail) + avail = (head <= tail ? tail : ctx->nr_events) - head; + if (head == tail) break; - avail = min(avail, nr - ret); - avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - - ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); - pos = head + AIO_EVENTS_OFFSET; - page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; + folio = ctx->ring_folios[pos / AIO_EVENTS_PER_PAGE]; pos %= AIO_EVENTS_PER_PAGE; - ev = kmap(page); + avail = min(avail, nr - ret); + avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - pos); + + ev = folio_address(folio); copy_ret = copy_to_user(event + ret, ev + pos, sizeof(*ev) * avail); - kunmap(page); if (unlikely(copy_ret)) { ret = -EFAULT; @@ -749,14 +1281,11 @@ static long aio_read_events_ring(struct kioctx *ctx, head %= ctx->nr_events; } - ring = kmap_atomic(ctx->ring_pages[0]); + ring = folio_address(ctx->ring_folios[0]); ring->head = head; - kunmap_atomic(ring); - flush_dcache_page(ctx->ring_pages[0]); - - pr_debug("%li h%u t%u\n", ret, head, ctx->tail); + flush_dcache_folio(ctx->ring_folios[0]); - atomic_sub(ret, &ctx->reqs_active); + pr_debug("%li h%u t%u\n", ret, head, tail); out: mutex_unlock(&ctx->ring_lock); @@ -782,19 +1311,11 @@ static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, static long read_events(struct kioctx *ctx, long min_nr, long nr, struct io_event __user *event, - struct timespec __user *timeout) + ktime_t until) { - ktime_t until = { .tv64 = KTIME_MAX }; - long ret = 0; - - if (timeout) { - struct timespec ts; - - if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) - return -EFAULT; - - until = timespec_to_ktime(ts); - } + struct hrtimer_sleeper t; + struct aio_waiter w; + long ret = 0, ret2 = 0; /* * Note that aio_read_events() is being called as the conditional - i.e. @@ -810,11 +1331,37 @@ static long read_events(struct kioctx *ctx, long min_nr, long nr, * the ringbuffer empty. So in practice we should be ok, but it's * something to be aware of when touching this code. */ - wait_event_interruptible_hrtimeout(ctx->wait, - aio_read_events(ctx, min_nr, nr, event, &ret), until); + aio_read_events(ctx, min_nr, nr, event, &ret); + if (until == 0 || ret < 0 || ret >= min_nr) + return ret; - if (!ret && signal_pending(current)) - ret = -EINTR; + hrtimer_setup_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + if (until != KTIME_MAX) { + hrtimer_set_expires_range_ns(&t.timer, until, current->timer_slack_ns); + hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_REL); + } + + init_wait(&w.w); + + while (1) { + unsigned long nr_got = ret; + + w.min_nr = min_nr - ret; + + ret2 = prepare_to_wait_event(&ctx->wait, &w.w, TASK_INTERRUPTIBLE); + if (!ret2 && !t.task) + ret2 = -ETIME; + + if (aio_read_events(ctx, min_nr, nr, event, &ret) || ret2) + break; + + if (nr_got == ret) + schedule(); + } + + finish_wait(&ctx->wait, &w.w); + hrtimer_cancel(&t.timer); + destroy_hrtimer_on_stack(&t.timer); return ret; } @@ -844,7 +1391,7 @@ SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) ret = -EINVAL; if (unlikely(ctx || nr_events == 0)) { - pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", + pr_debug("EINVAL: ctx %lu nr_events %u\n", ctx, nr_events); goto out; } @@ -854,14 +1401,47 @@ SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) if (!IS_ERR(ioctx)) { ret = put_user(ioctx->user_id, ctxp); if (ret) - kill_ioctx(ioctx); - put_ioctx(ioctx); + kill_ioctx(current->mm, ioctx, NULL); + percpu_ref_put(&ioctx->users); } out: return ret; } +#ifdef CONFIG_COMPAT +COMPAT_SYSCALL_DEFINE2(io_setup, unsigned, nr_events, u32 __user *, ctx32p) +{ + struct kioctx *ioctx = NULL; + unsigned long ctx; + long ret; + + ret = get_user(ctx, ctx32p); + if (unlikely(ret)) + goto out; + + ret = -EINVAL; + if (unlikely(ctx || nr_events == 0)) { + pr_debug("EINVAL: ctx %lu nr_events %u\n", + ctx, nr_events); + goto out; + } + + ioctx = ioctx_alloc(nr_events); + ret = PTR_ERR(ioctx); + if (!IS_ERR(ioctx)) { + /* truncating is ok because it's a user address */ + ret = put_user((u32)ioctx->user_id, ctx32p); + if (ret) + kill_ioctx(current->mm, ioctx, NULL); + percpu_ref_put(&ioctx->users); + } + +out: + return ret; +} +#endif + /* sys_io_destroy: * Destroy the aio_context specified. May cancel any outstanding * AIOs and block on completion. Will fail with -ENOSYS if not @@ -872,282 +1452,635 @@ SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) { struct kioctx *ioctx = lookup_ioctx(ctx); if (likely(NULL != ioctx)) { - kill_ioctx(ioctx); - put_ioctx(ioctx); - return 0; + struct ctx_rq_wait wait; + int ret; + + init_completion(&wait.comp); + atomic_set(&wait.count, 1); + + /* Pass requests_done to kill_ioctx() where it can be set + * in a thread-safe way. If we try to set it here then we have + * a race condition if two io_destroy() called simultaneously. + */ + ret = kill_ioctx(current->mm, ioctx, &wait); + percpu_ref_put(&ioctx->users); + + /* Wait until all IO for the context are done. Otherwise kernel + * keep using user-space buffers even if user thinks the context + * is destroyed. + */ + if (!ret) + wait_for_completion(&wait.comp); + + return ret; } - pr_debug("EINVAL: io_destroy: invalid context id\n"); + pr_debug("EINVAL: invalid context id\n"); return -EINVAL; } -static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) +static void aio_remove_iocb(struct aio_kiocb *iocb) { - struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; + struct kioctx *ctx = iocb->ki_ctx; + unsigned long flags; - BUG_ON(ret <= 0); + spin_lock_irqsave(&ctx->ctx_lock, flags); + list_del(&iocb->ki_list); + spin_unlock_irqrestore(&ctx->ctx_lock, flags); +} + +static void aio_complete_rw(struct kiocb *kiocb, long res) +{ + struct aio_kiocb *iocb = container_of(kiocb, struct aio_kiocb, rw); + + if (!list_empty_careful(&iocb->ki_list)) + aio_remove_iocb(iocb); + + if (kiocb->ki_flags & IOCB_WRITE) { + struct inode *inode = file_inode(kiocb->ki_filp); - while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { - ssize_t this = min((ssize_t)iov->iov_len, ret); - iov->iov_base += this; - iov->iov_len -= this; - iocb->ki_left -= this; - ret -= this; - if (iov->iov_len == 0) { - iocb->ki_cur_seg++; - iov++; + if (S_ISREG(inode->i_mode)) + kiocb_end_write(kiocb); + } + + iocb->ki_res.res = res; + iocb->ki_res.res2 = 0; + iocb_put(iocb); +} + +static int aio_prep_rw(struct kiocb *req, const struct iocb *iocb, int rw_type) +{ + int ret; + + req->ki_write_stream = 0; + req->ki_complete = aio_complete_rw; + req->private = NULL; + req->ki_pos = iocb->aio_offset; + req->ki_flags = req->ki_filp->f_iocb_flags | IOCB_AIO_RW; + if (iocb->aio_flags & IOCB_FLAG_RESFD) + req->ki_flags |= IOCB_EVENTFD; + if (iocb->aio_flags & IOCB_FLAG_IOPRIO) { + /* + * If the IOCB_FLAG_IOPRIO flag of aio_flags is set, then + * aio_reqprio is interpreted as an I/O scheduling + * class and priority. + */ + ret = ioprio_check_cap(iocb->aio_reqprio); + if (ret) { + pr_debug("aio ioprio check cap error: %d\n", ret); + return ret; } + + req->ki_ioprio = iocb->aio_reqprio; + } else + req->ki_ioprio = get_current_ioprio(); + + ret = kiocb_set_rw_flags(req, iocb->aio_rw_flags, rw_type); + if (unlikely(ret)) + return ret; + + req->ki_flags &= ~IOCB_HIPRI; /* no one is going to poll for this I/O */ + return 0; +} + +static ssize_t aio_setup_rw(int rw, const struct iocb *iocb, + struct iovec **iovec, bool vectored, bool compat, + struct iov_iter *iter) +{ + void __user *buf = (void __user *)(uintptr_t)iocb->aio_buf; + size_t len = iocb->aio_nbytes; + + if (!vectored) { + ssize_t ret = import_ubuf(rw, buf, len, iter); + *iovec = NULL; + return ret; } - /* the caller should not have done more io than what fit in - * the remaining iovecs */ - BUG_ON(ret > 0 && iocb->ki_left == 0); + return __import_iovec(rw, buf, len, UIO_FASTIOV, iovec, iter, compat); } -typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, - unsigned long, loff_t); +static inline void aio_rw_done(struct kiocb *req, ssize_t ret) +{ + switch (ret) { + case -EIOCBQUEUED: + break; + case -ERESTARTSYS: + case -ERESTARTNOINTR: + case -ERESTARTNOHAND: + case -ERESTART_RESTARTBLOCK: + /* + * There's no easy way to restart the syscall since other AIO's + * may be already running. Just fail this IO with EINTR. + */ + ret = -EINTR; + fallthrough; + default: + req->ki_complete(req, ret); + } +} -static ssize_t aio_rw_vect_retry(struct kiocb *iocb, int rw, aio_rw_op *rw_op) +static int aio_read(struct kiocb *req, const struct iocb *iocb, + bool vectored, bool compat) { - struct file *file = iocb->ki_filp; - struct address_space *mapping = file->f_mapping; - struct inode *inode = mapping->host; - ssize_t ret = 0; + struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; + struct iov_iter iter; + struct file *file; + int ret; - /* This matches the pread()/pwrite() logic */ - if (iocb->ki_pos < 0) + ret = aio_prep_rw(req, iocb, READ); + if (ret) + return ret; + file = req->ki_filp; + if (unlikely(!(file->f_mode & FMODE_READ))) + return -EBADF; + if (unlikely(!file->f_op->read_iter)) return -EINVAL; - if (rw == WRITE) - file_start_write(file); - do { - ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], - iocb->ki_nr_segs - iocb->ki_cur_seg, - iocb->ki_pos); - if (ret > 0) - aio_advance_iovec(iocb, ret); - - /* retry all partial writes. retry partial reads as long as its a - * regular file. */ - } while (ret > 0 && iocb->ki_left > 0 && - (rw == WRITE || - (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); - if (rw == WRITE) - file_end_write(file); - - /* This means we must have transferred all that we could */ - /* No need to retry anymore */ - if ((ret == 0) || (iocb->ki_left == 0)) - ret = iocb->ki_nbytes - iocb->ki_left; - - /* If we managed to write some out we return that, rather than - * the eventual error. */ - if (rw == WRITE - && ret < 0 && ret != -EIOCBQUEUED - && iocb->ki_nbytes - iocb->ki_left) - ret = iocb->ki_nbytes - iocb->ki_left; - + ret = aio_setup_rw(ITER_DEST, iocb, &iovec, vectored, compat, &iter); + if (ret < 0) + return ret; + ret = rw_verify_area(READ, file, &req->ki_pos, iov_iter_count(&iter)); + if (!ret) + aio_rw_done(req, file->f_op->read_iter(req, &iter)); + kfree(iovec); return ret; } -static ssize_t aio_setup_vectored_rw(int rw, struct kiocb *kiocb, bool compat) +static int aio_write(struct kiocb *req, const struct iocb *iocb, + bool vectored, bool compat) { - ssize_t ret; + struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; + struct iov_iter iter; + struct file *file; + int ret; - kiocb->ki_nr_segs = kiocb->ki_nbytes; + ret = aio_prep_rw(req, iocb, WRITE); + if (ret) + return ret; + file = req->ki_filp; -#ifdef CONFIG_COMPAT - if (compat) - ret = compat_rw_copy_check_uvector(rw, - (struct compat_iovec __user *)kiocb->ki_buf, - kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, - &kiocb->ki_iovec); - else -#endif - ret = rw_copy_check_uvector(rw, - (struct iovec __user *)kiocb->ki_buf, - kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, - &kiocb->ki_iovec); + if (unlikely(!(file->f_mode & FMODE_WRITE))) + return -EBADF; + if (unlikely(!file->f_op->write_iter)) + return -EINVAL; + + ret = aio_setup_rw(ITER_SOURCE, iocb, &iovec, vectored, compat, &iter); if (ret < 0) return ret; + ret = rw_verify_area(WRITE, file, &req->ki_pos, iov_iter_count(&iter)); + if (!ret) { + if (S_ISREG(file_inode(file)->i_mode)) + kiocb_start_write(req); + req->ki_flags |= IOCB_WRITE; + aio_rw_done(req, file->f_op->write_iter(req, &iter)); + } + kfree(iovec); + return ret; +} - /* ki_nbytes now reflect bytes instead of segs */ - kiocb->ki_nbytes = ret; - return 0; +static void aio_fsync_work(struct work_struct *work) +{ + struct aio_kiocb *iocb = container_of(work, struct aio_kiocb, fsync.work); + + scoped_with_creds(iocb->fsync.creds) + iocb->ki_res.res = vfs_fsync(iocb->fsync.file, iocb->fsync.datasync); + + put_cred(iocb->fsync.creds); + iocb_put(iocb); } -static ssize_t aio_setup_single_vector(int rw, struct kiocb *kiocb) +static int aio_fsync(struct fsync_iocb *req, const struct iocb *iocb, + bool datasync) { - if (unlikely(!access_ok(!rw, kiocb->ki_buf, kiocb->ki_nbytes))) - return -EFAULT; + if (unlikely(iocb->aio_buf || iocb->aio_offset || iocb->aio_nbytes || + iocb->aio_rw_flags)) + return -EINVAL; - kiocb->ki_iovec = &kiocb->ki_inline_vec; - kiocb->ki_iovec->iov_base = kiocb->ki_buf; - kiocb->ki_iovec->iov_len = kiocb->ki_nbytes; - kiocb->ki_nr_segs = 1; + if (unlikely(!req->file->f_op->fsync)) + return -EINVAL; + + req->creds = prepare_creds(); + if (!req->creds) + return -ENOMEM; + + req->datasync = datasync; + INIT_WORK(&req->work, aio_fsync_work); + schedule_work(&req->work); return 0; } +static void aio_poll_put_work(struct work_struct *work) +{ + struct poll_iocb *req = container_of(work, struct poll_iocb, work); + struct aio_kiocb *iocb = container_of(req, struct aio_kiocb, poll); + + iocb_put(iocb); +} + /* - * aio_setup_iocb: - * Performs the initial checks and aio retry method - * setup for the kiocb at the time of io submission. + * Safely lock the waitqueue which the request is on, synchronizing with the + * case where the ->poll() provider decides to free its waitqueue early. + * + * Returns true on success, meaning that req->head->lock was locked, req->wait + * is on req->head, and an RCU read lock was taken. Returns false if the + * request was already removed from its waitqueue (which might no longer exist). */ -static ssize_t aio_run_iocb(struct kiocb *req, bool compat) +static bool poll_iocb_lock_wq(struct poll_iocb *req) { - struct file *file = req->ki_filp; - ssize_t ret; - int rw; - fmode_t mode; - aio_rw_op *rw_op; + wait_queue_head_t *head; - switch (req->ki_opcode) { - case IOCB_CMD_PREAD: - case IOCB_CMD_PREADV: - mode = FMODE_READ; - rw = READ; - rw_op = file->f_op->aio_read; - goto rw_common; + /* + * While we hold the waitqueue lock and the waitqueue is nonempty, + * wake_up_pollfree() will wait for us. However, taking the waitqueue + * lock in the first place can race with the waitqueue being freed. + * + * We solve this as eventpoll does: by taking advantage of the fact that + * all users of wake_up_pollfree() will RCU-delay the actual free. If + * we enter rcu_read_lock() and see that the pointer to the queue is + * non-NULL, we can then lock it without the memory being freed out from + * under us, then check whether the request is still on the queue. + * + * Keep holding rcu_read_lock() as long as we hold the queue lock, in + * case the caller deletes the entry from the queue, leaving it empty. + * In that case, only RCU prevents the queue memory from being freed. + */ + rcu_read_lock(); + head = smp_load_acquire(&req->head); + if (head) { + spin_lock(&head->lock); + if (!list_empty(&req->wait.entry)) + return true; + spin_unlock(&head->lock); + } + rcu_read_unlock(); + return false; +} - case IOCB_CMD_PWRITE: - case IOCB_CMD_PWRITEV: - mode = FMODE_WRITE; - rw = WRITE; - rw_op = file->f_op->aio_write; - goto rw_common; -rw_common: - if (unlikely(!(file->f_mode & mode))) - return -EBADF; - - if (!rw_op) - return -EINVAL; - - ret = (req->ki_opcode == IOCB_CMD_PREADV || - req->ki_opcode == IOCB_CMD_PWRITEV) - ? aio_setup_vectored_rw(rw, req, compat) - : aio_setup_single_vector(rw, req); - if (ret) - return ret; +static void poll_iocb_unlock_wq(struct poll_iocb *req) +{ + spin_unlock(&req->head->lock); + rcu_read_unlock(); +} - ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); - if (ret < 0) - return ret; +static void aio_poll_complete_work(struct work_struct *work) +{ + struct poll_iocb *req = container_of(work, struct poll_iocb, work); + struct aio_kiocb *iocb = container_of(req, struct aio_kiocb, poll); + struct poll_table_struct pt = { ._key = req->events }; + struct kioctx *ctx = iocb->ki_ctx; + __poll_t mask = 0; - req->ki_nbytes = ret; - req->ki_left = ret; + if (!READ_ONCE(req->cancelled)) + mask = vfs_poll(req->file, &pt) & req->events; - ret = aio_rw_vect_retry(req, rw, rw_op); - break; + /* + * Note that ->ki_cancel callers also delete iocb from active_reqs after + * calling ->ki_cancel. We need the ctx_lock roundtrip here to + * synchronize with them. In the cancellation case the list_del_init + * itself is not actually needed, but harmless so we keep it in to + * avoid further branches in the fast path. + */ + spin_lock_irq(&ctx->ctx_lock); + if (poll_iocb_lock_wq(req)) { + if (!mask && !READ_ONCE(req->cancelled)) { + /* + * The request isn't actually ready to be completed yet. + * Reschedule completion if another wakeup came in. + */ + if (req->work_need_resched) { + schedule_work(&req->work); + req->work_need_resched = false; + } else { + req->work_scheduled = false; + } + poll_iocb_unlock_wq(req); + spin_unlock_irq(&ctx->ctx_lock); + return; + } + list_del_init(&req->wait.entry); + poll_iocb_unlock_wq(req); + } /* else, POLLFREE has freed the waitqueue, so we must complete */ + list_del_init(&iocb->ki_list); + iocb->ki_res.res = mangle_poll(mask); + spin_unlock_irq(&ctx->ctx_lock); - case IOCB_CMD_FDSYNC: - if (!file->f_op->aio_fsync) - return -EINVAL; + iocb_put(iocb); +} - ret = file->f_op->aio_fsync(req, 1); - break; +/* assumes we are called with irqs disabled */ +static int aio_poll_cancel(struct kiocb *iocb) +{ + struct aio_kiocb *aiocb = container_of(iocb, struct aio_kiocb, rw); + struct poll_iocb *req = &aiocb->poll; + + if (poll_iocb_lock_wq(req)) { + WRITE_ONCE(req->cancelled, true); + if (!req->work_scheduled) { + schedule_work(&aiocb->poll.work); + req->work_scheduled = true; + } + poll_iocb_unlock_wq(req); + } /* else, the request was force-cancelled by POLLFREE already */ - case IOCB_CMD_FSYNC: - if (!file->f_op->aio_fsync) - return -EINVAL; + return 0; +} - ret = file->f_op->aio_fsync(req, 0); - break; +static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, + void *key) +{ + struct poll_iocb *req = container_of(wait, struct poll_iocb, wait); + struct aio_kiocb *iocb = container_of(req, struct aio_kiocb, poll); + __poll_t mask = key_to_poll(key); + unsigned long flags; - default: - pr_debug("EINVAL: no operation provided\n"); - return -EINVAL; - } + /* for instances that support it check for an event match first: */ + if (mask && !(mask & req->events)) + return 0; - if (ret != -EIOCBQUEUED) { + /* + * Complete the request inline if possible. This requires that three + * conditions be met: + * 1. An event mask must have been passed. If a plain wakeup was done + * instead, then mask == 0 and we have to call vfs_poll() to get + * the events, so inline completion isn't possible. + * 2. The completion work must not have already been scheduled. + * 3. ctx_lock must not be busy. We have to use trylock because we + * already hold the waitqueue lock, so this inverts the normal + * locking order. Use irqsave/irqrestore because not all + * filesystems (e.g. fuse) call this function with IRQs disabled, + * yet IRQs have to be disabled before ctx_lock is obtained. + */ + if (mask && !req->work_scheduled && + spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) { + struct kioctx *ctx = iocb->ki_ctx; + + list_del_init(&req->wait.entry); + list_del(&iocb->ki_list); + iocb->ki_res.res = mangle_poll(mask); + if (iocb->ki_eventfd && !eventfd_signal_allowed()) { + iocb = NULL; + INIT_WORK(&req->work, aio_poll_put_work); + schedule_work(&req->work); + } + spin_unlock_irqrestore(&ctx->ctx_lock, flags); + if (iocb) + iocb_put(iocb); + } else { /* - * There's no easy way to restart the syscall since other AIO's - * may be already running. Just fail this IO with EINTR. + * Schedule the completion work if needed. If it was already + * scheduled, record that another wakeup came in. + * + * Don't remove the request from the waitqueue here, as it might + * not actually be complete yet (we won't know until vfs_poll() + * is called), and we must not miss any wakeups. POLLFREE is an + * exception to this; see below. */ - if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || - ret == -ERESTARTNOHAND || - ret == -ERESTART_RESTARTBLOCK)) - ret = -EINTR; - aio_complete(req, ret, 0); - } + if (req->work_scheduled) { + req->work_need_resched = true; + } else { + schedule_work(&req->work); + req->work_scheduled = true; + } - return 0; + /* + * If the waitqueue is being freed early but we can't complete + * the request inline, we have to tear down the request as best + * we can. That means immediately removing the request from its + * waitqueue and preventing all further accesses to the + * waitqueue via the request. We also need to schedule the + * completion work (done above). Also mark the request as + * cancelled, to potentially skip an unneeded call to ->poll(). + */ + if (mask & POLLFREE) { + WRITE_ONCE(req->cancelled, true); + list_del_init(&req->wait.entry); + + /* + * Careful: this *must* be the last step, since as soon + * as req->head is NULL'ed out, the request can be + * completed and freed, since aio_poll_complete_work() + * will no longer need to take the waitqueue lock. + */ + smp_store_release(&req->head, NULL); + } + } + return 1; } -static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, - struct iocb *iocb, bool compat) +struct aio_poll_table { + struct poll_table_struct pt; + struct aio_kiocb *iocb; + bool queued; + int error; +}; + +static void +aio_poll_queue_proc(struct file *file, struct wait_queue_head *head, + struct poll_table_struct *p) { - struct kiocb *req; - ssize_t ret; + struct aio_poll_table *pt = container_of(p, struct aio_poll_table, pt); - /* enforce forwards compatibility on users */ - if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { - pr_debug("EINVAL: reserve field set\n"); - return -EINVAL; + /* multiple wait queues per file are not supported */ + if (unlikely(pt->queued)) { + pt->error = -EINVAL; + return; } - /* prevent overflows */ - if (unlikely( - (iocb->aio_buf != (unsigned long)iocb->aio_buf) || - (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || - ((ssize_t)iocb->aio_nbytes < 0) - )) { - pr_debug("EINVAL: io_submit: overflow check\n"); + pt->queued = true; + pt->error = 0; + pt->iocb->poll.head = head; + add_wait_queue(head, &pt->iocb->poll.wait); +} + +static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb) +{ + struct kioctx *ctx = aiocb->ki_ctx; + struct poll_iocb *req = &aiocb->poll; + struct aio_poll_table apt; + bool cancel = false; + __poll_t mask; + + /* reject any unknown events outside the normal event mask. */ + if ((u16)iocb->aio_buf != iocb->aio_buf) + return -EINVAL; + /* reject fields that are not defined for poll */ + if (iocb->aio_offset || iocb->aio_nbytes || iocb->aio_rw_flags) return -EINVAL; - } - req = aio_get_req(ctx); - if (unlikely(!req)) - return -EAGAIN; + INIT_WORK(&req->work, aio_poll_complete_work); + req->events = demangle_poll(iocb->aio_buf) | EPOLLERR | EPOLLHUP; - req->ki_filp = fget(iocb->aio_fildes); - if (unlikely(!req->ki_filp)) { - ret = -EBADF; - goto out_put_req; + req->head = NULL; + req->cancelled = false; + req->work_scheduled = false; + req->work_need_resched = false; + + apt.pt._qproc = aio_poll_queue_proc; + apt.pt._key = req->events; + apt.iocb = aiocb; + apt.queued = false; + apt.error = -EINVAL; /* same as no support for IOCB_CMD_POLL */ + + /* initialized the list so that we can do list_empty checks */ + INIT_LIST_HEAD(&req->wait.entry); + init_waitqueue_func_entry(&req->wait, aio_poll_wake); + + mask = vfs_poll(req->file, &apt.pt) & req->events; + spin_lock_irq(&ctx->ctx_lock); + if (likely(apt.queued)) { + bool on_queue = poll_iocb_lock_wq(req); + + if (!on_queue || req->work_scheduled) { + /* + * aio_poll_wake() already either scheduled the async + * completion work, or completed the request inline. + */ + if (apt.error) /* unsupported case: multiple queues */ + cancel = true; + apt.error = 0; + mask = 0; + } + if (mask || apt.error) { + /* Steal to complete synchronously. */ + list_del_init(&req->wait.entry); + } else if (cancel) { + /* Cancel if possible (may be too late though). */ + WRITE_ONCE(req->cancelled, true); + } else if (on_queue) { + /* + * Actually waiting for an event, so add the request to + * active_reqs so that it can be cancelled if needed. + */ + list_add_tail(&aiocb->ki_list, &ctx->active_reqs); + aiocb->ki_cancel = aio_poll_cancel; + } + if (on_queue) + poll_iocb_unlock_wq(req); + } + if (mask) { /* no async, we'd stolen it */ + aiocb->ki_res.res = mangle_poll(mask); + apt.error = 0; } + spin_unlock_irq(&ctx->ctx_lock); + if (mask) + iocb_put(aiocb); + return apt.error; +} + +static int __io_submit_one(struct kioctx *ctx, const struct iocb *iocb, + struct iocb __user *user_iocb, struct aio_kiocb *req, + bool compat) +{ + req->ki_filp = fget(iocb->aio_fildes); + if (unlikely(!req->ki_filp)) + return -EBADF; if (iocb->aio_flags & IOCB_FLAG_RESFD) { + struct eventfd_ctx *eventfd; /* * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an * instance of the file* now. The file descriptor must be * an eventfd() fd, and will be signaled for each completed * event using the eventfd_signal() function. */ - req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); - if (IS_ERR(req->ki_eventfd)) { - ret = PTR_ERR(req->ki_eventfd); - req->ki_eventfd = NULL; - goto out_put_req; - } + eventfd = eventfd_ctx_fdget(iocb->aio_resfd); + if (IS_ERR(eventfd)) + return PTR_ERR(eventfd); + + req->ki_eventfd = eventfd; } - ret = put_user(KIOCB_KEY, &user_iocb->aio_key); - if (unlikely(ret)) { + if (unlikely(put_user(KIOCB_KEY, &user_iocb->aio_key))) { pr_debug("EFAULT: aio_key\n"); - goto out_put_req; + return -EFAULT; } - req->ki_obj.user = user_iocb; - req->ki_user_data = iocb->aio_data; - req->ki_pos = iocb->aio_offset; + req->ki_res.obj = (u64)(unsigned long)user_iocb; + req->ki_res.data = iocb->aio_data; + req->ki_res.res = 0; + req->ki_res.res2 = 0; - req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; - req->ki_left = req->ki_nbytes = iocb->aio_nbytes; - req->ki_opcode = iocb->aio_lio_opcode; + switch (iocb->aio_lio_opcode) { + case IOCB_CMD_PREAD: + return aio_read(&req->rw, iocb, false, compat); + case IOCB_CMD_PWRITE: + return aio_write(&req->rw, iocb, false, compat); + case IOCB_CMD_PREADV: + return aio_read(&req->rw, iocb, true, compat); + case IOCB_CMD_PWRITEV: + return aio_write(&req->rw, iocb, true, compat); + case IOCB_CMD_FSYNC: + return aio_fsync(&req->fsync, iocb, false); + case IOCB_CMD_FDSYNC: + return aio_fsync(&req->fsync, iocb, true); + case IOCB_CMD_POLL: + return aio_poll(req, iocb); + default: + pr_debug("invalid aio operation %d\n", iocb->aio_lio_opcode); + return -EINVAL; + } +} - ret = aio_run_iocb(req, compat); - if (ret) - goto out_put_req; +static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, + bool compat) +{ + struct aio_kiocb *req; + struct iocb iocb; + int err; - aio_put_req(req); /* drop extra ref to req */ - return 0; -out_put_req: - atomic_dec(&ctx->reqs_active); - aio_put_req(req); /* drop extra ref to req */ - aio_put_req(req); /* drop i/o ref to req */ - return ret; + if (unlikely(copy_from_user(&iocb, user_iocb, sizeof(iocb)))) + return -EFAULT; + + /* enforce forwards compatibility on users */ + if (unlikely(iocb.aio_reserved2)) { + pr_debug("EINVAL: reserve field set\n"); + return -EINVAL; + } + + /* prevent overflows */ + if (unlikely( + (iocb.aio_buf != (unsigned long)iocb.aio_buf) || + (iocb.aio_nbytes != (size_t)iocb.aio_nbytes) || + ((ssize_t)iocb.aio_nbytes < 0) + )) { + pr_debug("EINVAL: overflow check\n"); + return -EINVAL; + } + + req = aio_get_req(ctx); + if (unlikely(!req)) + return -EAGAIN; + + err = __io_submit_one(ctx, &iocb, user_iocb, req, compat); + + /* Done with the synchronous reference */ + iocb_put(req); + + /* + * If err is 0, we'd either done aio_complete() ourselves or have + * arranged for that to be done asynchronously. Anything non-zero + * means that we need to destroy req ourselves. + */ + if (unlikely(err)) { + iocb_destroy(req); + put_reqs_available(ctx, 1); + } + return err; } -long do_io_submit(aio_context_t ctx_id, long nr, - struct iocb __user *__user *iocbpp, bool compat) +/* sys_io_submit: + * Queue the nr iocbs pointed to by iocbpp for processing. Returns + * the number of iocbs queued. May return -EINVAL if the aio_context + * specified by ctx_id is invalid, if nr is < 0, if the iocb at + * *iocbpp[0] is not properly initialized, if the operation specified + * is invalid for the file descriptor in the iocb. May fail with + * -EFAULT if any of the data structures point to invalid data. May + * fail with -EBADF if the file descriptor specified in the first + * iocb is invalid. May fail with -EAGAIN if insufficient resources + * are available to queue any iocbs. Will return 0 if nr is 0. Will + * fail with -ENOSYS if not implemented. + */ +SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, + struct iocb __user * __user *, iocbpp) { struct kioctx *ctx; long ret = 0; @@ -1157,87 +2090,78 @@ long do_io_submit(aio_context_t ctx_id, long nr, if (unlikely(nr < 0)) return -EINVAL; - if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) - nr = LONG_MAX/sizeof(*iocbpp); - - if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) - return -EFAULT; - ctx = lookup_ioctx(ctx_id); if (unlikely(!ctx)) { pr_debug("EINVAL: invalid context id\n"); return -EINVAL; } - blk_start_plug(&plug); + if (nr > ctx->nr_events) + nr = ctx->nr_events; - /* - * AKPM: should this return a partial result if some of the IOs were - * successfully submitted? - */ - for (i=0; i<nr; i++) { + if (nr > AIO_PLUG_THRESHOLD) + blk_start_plug(&plug); + for (i = 0; i < nr; i++) { struct iocb __user *user_iocb; - struct iocb tmp; - - if (unlikely(__get_user(user_iocb, iocbpp + i))) { - ret = -EFAULT; - break; - } - if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { + if (unlikely(get_user(user_iocb, iocbpp + i))) { ret = -EFAULT; break; } - ret = io_submit_one(ctx, user_iocb, &tmp, compat); + ret = io_submit_one(ctx, user_iocb, false); if (ret) break; } - blk_finish_plug(&plug); + if (nr > AIO_PLUG_THRESHOLD) + blk_finish_plug(&plug); - put_ioctx(ctx); + percpu_ref_put(&ctx->users); return i ? i : ret; } -/* sys_io_submit: - * Queue the nr iocbs pointed to by iocbpp for processing. Returns - * the number of iocbs queued. May return -EINVAL if the aio_context - * specified by ctx_id is invalid, if nr is < 0, if the iocb at - * *iocbpp[0] is not properly initialized, if the operation specified - * is invalid for the file descriptor in the iocb. May fail with - * -EFAULT if any of the data structures point to invalid data. May - * fail with -EBADF if the file descriptor specified in the first - * iocb is invalid. May fail with -EAGAIN if insufficient resources - * are available to queue any iocbs. Will return 0 if nr is 0. Will - * fail with -ENOSYS if not implemented. - */ -SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, - struct iocb __user * __user *, iocbpp) +#ifdef CONFIG_COMPAT +COMPAT_SYSCALL_DEFINE3(io_submit, compat_aio_context_t, ctx_id, + int, nr, compat_uptr_t __user *, iocbpp) { - return do_io_submit(ctx_id, nr, iocbpp, 0); -} + struct kioctx *ctx; + long ret = 0; + int i = 0; + struct blk_plug plug; -/* lookup_kiocb - * Finds a given iocb for cancellation. - */ -static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, - u32 key) -{ - struct list_head *pos; + if (unlikely(nr < 0)) + return -EINVAL; + + ctx = lookup_ioctx(ctx_id); + if (unlikely(!ctx)) { + pr_debug("EINVAL: invalid context id\n"); + return -EINVAL; + } - assert_spin_locked(&ctx->ctx_lock); + if (nr > ctx->nr_events) + nr = ctx->nr_events; - if (key != KIOCB_KEY) - return NULL; + if (nr > AIO_PLUG_THRESHOLD) + blk_start_plug(&plug); + for (i = 0; i < nr; i++) { + compat_uptr_t user_iocb; - /* TODO: use a hash or array, this sucks. */ - list_for_each(pos, &ctx->active_reqs) { - struct kiocb *kiocb = list_kiocb(pos); - if (kiocb->ki_obj.user == iocb) - return kiocb; + if (unlikely(get_user(user_iocb, iocbpp + i))) { + ret = -EFAULT; + break; + } + + ret = io_submit_one(ctx, compat_ptr(user_iocb), true); + if (ret) + break; } - return NULL; + if (nr > AIO_PLUG_THRESHOLD) + blk_finish_plug(&plug); + + percpu_ref_put(&ctx->users); + return i ? i : ret; } +#endif /* sys_io_cancel: * Attempts to cancel an iocb previously passed to io_submit. If @@ -1252,39 +2176,60 @@ static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, struct io_event __user *, result) { - struct io_event res; struct kioctx *ctx; - struct kiocb *kiocb; + struct aio_kiocb *kiocb; + int ret = -EINVAL; u32 key; - int ret; + u64 obj = (u64)(unsigned long)iocb; - ret = get_user(key, &iocb->aio_key); - if (unlikely(ret)) + if (unlikely(get_user(key, &iocb->aio_key))) return -EFAULT; + if (unlikely(key != KIOCB_KEY)) + return -EINVAL; ctx = lookup_ioctx(ctx_id); if (unlikely(!ctx)) return -EINVAL; spin_lock_irq(&ctx->ctx_lock); - - kiocb = lookup_kiocb(ctx, iocb, key); - if (kiocb) - ret = kiocb_cancel(ctx, kiocb, &res); - else - ret = -EINVAL; - + list_for_each_entry(kiocb, &ctx->active_reqs, ki_list) { + if (kiocb->ki_res.obj == obj) { + ret = kiocb->ki_cancel(&kiocb->rw); + list_del_init(&kiocb->ki_list); + break; + } + } spin_unlock_irq(&ctx->ctx_lock); if (!ret) { - /* Cancellation succeeded -- copy the result - * into the user's buffer. + /* + * The result argument is no longer used - the io_event is + * always delivered via the ring buffer. -EINPROGRESS indicates + * cancellation is progress: */ - if (copy_to_user(result, &res, sizeof(res))) - ret = -EFAULT; + ret = -EINPROGRESS; } - put_ioctx(ctx); + percpu_ref_put(&ctx->users); + + return ret; +} + +static long do_io_getevents(aio_context_t ctx_id, + long min_nr, + long nr, + struct io_event __user *events, + struct timespec64 *ts) +{ + ktime_t until = ts ? timespec64_to_ktime(*ts) : KTIME_MAX; + struct kioctx *ioctx = lookup_ioctx(ctx_id); + long ret = -EINVAL; + + if (likely(ioctx)) { + if (likely(min_nr <= nr && min_nr >= 0)) + ret = read_events(ioctx, min_nr, nr, events, until); + percpu_ref_put(&ioctx->users); + } return ret; } @@ -1301,19 +2246,200 @@ SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, * specifies an infinite timeout. Note that the timeout pointed to by * timeout is relative. Will fail with -ENOSYS if not implemented. */ +#ifdef CONFIG_64BIT + SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, long, min_nr, long, nr, struct io_event __user *, events, - struct timespec __user *, timeout) + struct __kernel_timespec __user *, timeout) { - struct kioctx *ioctx = lookup_ioctx(ctx_id); - long ret = -EINVAL; + struct timespec64 ts; + int ret; - if (likely(ioctx)) { - if (likely(min_nr <= nr && min_nr >= 0)) - ret = read_events(ioctx, min_nr, nr, events, timeout); - put_ioctx(ioctx); - } + if (timeout && unlikely(get_timespec64(&ts, timeout))) + return -EFAULT; + + ret = do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &ts : NULL); + if (!ret && signal_pending(current)) + ret = -EINTR; return ret; } + +#endif + +struct __aio_sigset { + const sigset_t __user *sigmask; + size_t sigsetsize; +}; + +SYSCALL_DEFINE6(io_pgetevents, + aio_context_t, ctx_id, + long, min_nr, + long, nr, + struct io_event __user *, events, + struct __kernel_timespec __user *, timeout, + const struct __aio_sigset __user *, usig) +{ + struct __aio_sigset ksig = { NULL, }; + struct timespec64 ts; + bool interrupted; + int ret; + + if (timeout && unlikely(get_timespec64(&ts, timeout))) + return -EFAULT; + + if (usig && copy_from_user(&ksig, usig, sizeof(ksig))) + return -EFAULT; + + ret = set_user_sigmask(ksig.sigmask, ksig.sigsetsize); + if (ret) + return ret; + + ret = do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &ts : NULL); + + interrupted = signal_pending(current); + restore_saved_sigmask_unless(interrupted); + if (interrupted && !ret) + ret = -ERESTARTNOHAND; + + return ret; +} + +#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) + +SYSCALL_DEFINE6(io_pgetevents_time32, + aio_context_t, ctx_id, + long, min_nr, + long, nr, + struct io_event __user *, events, + struct old_timespec32 __user *, timeout, + const struct __aio_sigset __user *, usig) +{ + struct __aio_sigset ksig = { NULL, }; + struct timespec64 ts; + bool interrupted; + int ret; + + if (timeout && unlikely(get_old_timespec32(&ts, timeout))) + return -EFAULT; + + if (usig && copy_from_user(&ksig, usig, sizeof(ksig))) + return -EFAULT; + + + ret = set_user_sigmask(ksig.sigmask, ksig.sigsetsize); + if (ret) + return ret; + + ret = do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &ts : NULL); + + interrupted = signal_pending(current); + restore_saved_sigmask_unless(interrupted); + if (interrupted && !ret) + ret = -ERESTARTNOHAND; + + return ret; +} + +#endif + +#if defined(CONFIG_COMPAT_32BIT_TIME) + +SYSCALL_DEFINE5(io_getevents_time32, __u32, ctx_id, + __s32, min_nr, + __s32, nr, + struct io_event __user *, events, + struct old_timespec32 __user *, timeout) +{ + struct timespec64 t; + int ret; + + if (timeout && get_old_timespec32(&t, timeout)) + return -EFAULT; + + ret = do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &t : NULL); + if (!ret && signal_pending(current)) + ret = -EINTR; + return ret; +} + +#endif + +#ifdef CONFIG_COMPAT + +struct __compat_aio_sigset { + compat_uptr_t sigmask; + compat_size_t sigsetsize; +}; + +#if defined(CONFIG_COMPAT_32BIT_TIME) + +COMPAT_SYSCALL_DEFINE6(io_pgetevents, + compat_aio_context_t, ctx_id, + compat_long_t, min_nr, + compat_long_t, nr, + struct io_event __user *, events, + struct old_timespec32 __user *, timeout, + const struct __compat_aio_sigset __user *, usig) +{ + struct __compat_aio_sigset ksig = { 0, }; + struct timespec64 t; + bool interrupted; + int ret; + + if (timeout && get_old_timespec32(&t, timeout)) + return -EFAULT; + + if (usig && copy_from_user(&ksig, usig, sizeof(ksig))) + return -EFAULT; + + ret = set_compat_user_sigmask(compat_ptr(ksig.sigmask), ksig.sigsetsize); + if (ret) + return ret; + + ret = do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &t : NULL); + + interrupted = signal_pending(current); + restore_saved_sigmask_unless(interrupted); + if (interrupted && !ret) + ret = -ERESTARTNOHAND; + + return ret; +} + +#endif + +COMPAT_SYSCALL_DEFINE6(io_pgetevents_time64, + compat_aio_context_t, ctx_id, + compat_long_t, min_nr, + compat_long_t, nr, + struct io_event __user *, events, + struct __kernel_timespec __user *, timeout, + const struct __compat_aio_sigset __user *, usig) +{ + struct __compat_aio_sigset ksig = { 0, }; + struct timespec64 t; + bool interrupted; + int ret; + + if (timeout && get_timespec64(&t, timeout)) + return -EFAULT; + + if (usig && copy_from_user(&ksig, usig, sizeof(ksig))) + return -EFAULT; + + ret = set_compat_user_sigmask(compat_ptr(ksig.sigmask), ksig.sigsetsize); + if (ret) + return ret; + + ret = do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &t : NULL); + + interrupted = signal_pending(current); + restore_saved_sigmask_unless(interrupted); + if (interrupted && !ret) + ret = -ERESTARTNOHAND; + + return ret; +} +#endif |