diff -urN /md0/kernels/2.4/v2.4.9-ac9/@ aio-v2.4.9-ac9/@ --- /md0/kernels/2.4/v2.4.9-ac9/@ Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/@ Sat Sep 22 18:18:45 2001 @@ -0,0 +1,397 @@ +/* + md_k.h : kernel internal structure of the Linux MD driver + Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + You should have received a copy of the GNU General Public License + (for example /usr/src/linux/COPYING); if not, write to the Free + Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. +*/ + +#ifndef _MD_K_H +#define _MD_K_H + +#include // for panic() + +#define MD_RESERVED 0UL +#define LINEAR 1UL +#define STRIPED 2UL +#define RAID0 STRIPED +#define RAID1 3UL +#define RAID5 4UL +#define TRANSLUCENT 5UL +#define HSM 6UL +#define MAX_PERSONALITY 7UL + +static inline int pers_to_level (int pers) +{ + switch (pers) { + case HSM: return -3; + case TRANSLUCENT: return -2; + case LINEAR: return -1; + case RAID0: return 0; + case RAID1: return 1; + case RAID5: return 5; + } + panic("pers_to_level()"); + return 0; +} + +static inline int level_to_pers (int level) +{ + switch (level) { + case -3: return HSM; + case -2: return TRANSLUCENT; + case -1: return LINEAR; + case 0: return RAID0; + case 1: return RAID1; + case 4: + case 5: return RAID5; + } + return MD_RESERVED; +} + +typedef struct mddev_s mddev_t; +typedef struct mdk_rdev_s mdk_rdev_t; + +#if (MINORBITS != 8) +#error MD doesnt handle bigger kdev yet +#endif + +#define MAX_MD_DEVS (1<state & (1 << MD_DISK_FAULTY); +} + +static inline int disk_active(mdp_disk_t * d) +{ + return d->state & (1 << MD_DISK_ACTIVE); +} + +static inline int disk_sync(mdp_disk_t * d) +{ + return d->state & (1 << MD_DISK_SYNC); +} + +static inline int disk_spare(mdp_disk_t * d) +{ + return !disk_sync(d) && !disk_active(d) && !disk_faulty(d); +} + +static inline int disk_removed(mdp_disk_t * d) +{ + return d->state & (1 << MD_DISK_REMOVED); +} + +static inline void mark_disk_faulty(mdp_disk_t * d) +{ + d->state |= (1 << MD_DISK_FAULTY); +} + +static inline void mark_disk_active(mdp_disk_t * d) +{ + d->state |= (1 << MD_DISK_ACTIVE); +} + +static inline void mark_disk_sync(mdp_disk_t * d) +{ + d->state |= (1 << MD_DISK_SYNC); +} + +static inline void mark_disk_spare(mdp_disk_t * d) +{ + d->state = 0; +} + +static inline void mark_disk_removed(mdp_disk_t * d) +{ + d->state = (1 << MD_DISK_FAULTY) | (1 << MD_DISK_REMOVED); +} + +static inline void mark_disk_inactive(mdp_disk_t * d) +{ + d->state &= ~(1 << MD_DISK_ACTIVE); +} + +static inline void mark_disk_nonsync(mdp_disk_t * d) +{ + d->state &= ~(1 << MD_DISK_SYNC); +} + +/* + * MD's 'extended' device + */ +struct mdk_rdev_s +{ + struct md_list_head same_set; /* RAID devices within the same set */ + struct md_list_head all; /* all RAID devices */ + struct md_list_head pending; /* undetected RAID devices */ + + kdev_t dev; /* Device number */ + kdev_t old_dev; /* "" when it was last imported */ + unsigned long size; /* Device size (in blocks) */ + mddev_t *mddev; /* RAID array if running */ + unsigned long last_events; /* IO event timestamp */ + + struct block_device *bdev; /* block device handle */ + + mdp_super_t *sb; + unsigned long sb_offset; + + int faulty; /* if faulty do not issue IO requests */ + int desc_nr; /* descriptor index in the superblock */ +}; + + +/* + * disk operations in a working array: + */ +#define DISKOP_SPARE_INACTIVE 0 +#define DISKOP_SPARE_WRITE 1 +#define DISKOP_SPARE_ACTIVE 2 +#define DISKOP_HOT_REMOVE_DISK 3 +#define DISKOP_HOT_ADD_DISK 4 + +typedef struct mdk_personality_s mdk_personality_t; + +struct mddev_s +{ + void *private; + mdk_personality_t *pers; + int __minor; + mdp_super_t *sb; + int nb_dev; + struct md_list_head disks; + int sb_dirty; + mdu_param_t param; + int ro; + unsigned long curr_resync; /* blocks scheduled */ + unsigned long resync_mark; /* a recent timestamp */ + unsigned long resync_mark_cnt;/* blocks written at resync_mark */ + char *name; + int recovery_running; + struct semaphore reconfig_sem; + struct semaphore recovery_sem; + struct semaphore resync_sem; + atomic_t active; + + atomic_t recovery_active; /* blocks scheduled, but not written */ + md_wait_queue_head_t recovery_wait; + + struct md_list_head all_mddevs; +}; + +struct mdk_personality_s +{ + char *name; + int (*make_request)(mddev_t *mddev, int rw, struct buffer_head * bh); + int (*run)(mddev_t *mddev); + int (*stop)(mddev_t *mddev); + int (*status)(char *page, mddev_t *mddev); + int (*error_handler)(mddev_t *mddev, kdev_t dev); + +/* + * Some personalities (RAID-1, RAID-5) can have disks hot-added and + * hot-removed. Hot removal is different from failure. (failure marks + * a disk inactive, but the disk is still part of the array) The interface + * to such operations is the 'pers->diskop()' function, can be NULL. + * + * the diskop function can change the pointer pointing to the incoming + * descriptor, but must do so very carefully. (currently only + * SPARE_ACTIVE expects such a change) + */ + int (*diskop) (mddev_t *mddev, mdp_disk_t **descriptor, int state); + + int (*stop_resync)(mddev_t *mddev); + int (*restart_resync)(mddev_t *mddev); + int (*sync_request)(mddev_t *mddev, unsigned long block_nr); +}; + + +/* + * Currently we index md_array directly, based on the minor + * number. This will have to change to dynamic allocation + * once we start supporting partitioning of md devices. + */ +static inline int mdidx (mddev_t * mddev) +{ + return mddev->__minor; +} + +static inline kdev_t mddev_to_kdev(mddev_t * mddev) +{ + return MKDEV(MD_MAJOR, mdidx(mddev)); +} + +extern mdk_rdev_t * find_rdev(mddev_t * mddev, kdev_t dev); +extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr); + +/* + * iterates through some rdev ringlist. It's safe to remove the + * current 'rdev'. Dont touch 'tmp' though. + */ +#define ITERATE_RDEV_GENERIC(head,field,rdev,tmp) \ + \ + for (tmp = head.next; \ + rdev = md_list_entry(tmp, mdk_rdev_t, field), \ + tmp = tmp->next, tmp->prev != &head \ + ; ) +/* + * iterates through the 'same array disks' ringlist + */ +#define ITERATE_RDEV(mddev,rdev,tmp) \ + ITERATE_RDEV_GENERIC((mddev)->disks,same_set,rdev,tmp) + +/* + * Same as above, but assumes that the device has rdev->desc_nr numbered + * from 0 to mddev->nb_dev, and iterates through rdevs in ascending order. + */ +#define ITERATE_RDEV_ORDERED(mddev,rdev,i) \ + for (i = 0; rdev = find_rdev_nr(mddev, i), i < mddev->nb_dev; i++) + + +/* + * Iterates through all 'RAID managed disks' + */ +#define ITERATE_RDEV_ALL(rdev,tmp) \ + ITERATE_RDEV_GENERIC(all_raid_disks,all,rdev,tmp) + +/* + * Iterates through 'pending RAID disks' + */ +#define ITERATE_RDEV_PENDING(rdev,tmp) \ + ITERATE_RDEV_GENERIC(pending_raid_disks,pending,rdev,tmp) + +/* + * iterates through all used mddevs in the system. + */ +#define ITERATE_MDDEV(mddev,tmp) \ + \ + for (tmp = all_mddevs.next; \ + mddev = md_list_entry(tmp, mddev_t, all_mddevs), \ + tmp = tmp->next, tmp->prev != &all_mddevs \ + ; ) + +static inline int lock_mddev (mddev_t * mddev) +{ + return down_interruptible(&mddev->reconfig_sem); +} + +static inline void unlock_mddev (mddev_t * mddev) +{ + up(&mddev->reconfig_sem); +} + +#define xchg_values(x,y) do { __typeof__(x) __tmp = x; \ + x = y; y = __tmp; } while (0) + +typedef struct mdk_thread_s { + void (*run) (void *data); + void *data; + md_wait_queue_head_t wqueue; + unsigned long flags; + struct completion *event; + struct task_struct *tsk; + const char *name; +} mdk_thread_t; + +#define THREAD_WAKEUP 0 + +#define MAX_DISKNAME_LEN 64 + +typedef struct dev_name_s { + struct md_list_head list; + kdev_t dev; + char namebuf [MAX_DISKNAME_LEN]; + char *name; +} dev_name_t; + + +#define __wait_event_lock_irq(wq, condition, lock) \ +do { \ + wait_queue_t __wait; \ + init_waitqueue_entry(&__wait, current); \ + \ + add_wait_queue(&wq, &__wait); \ + for (;;) { \ + set_current_state(TASK_UNINTERRUPTIBLE); \ + if (condition) \ + break; \ + spin_unlock_irq(&lock); \ + run_task_queue(&tq_disk); \ + schedule(); \ + spin_lock_irq(&lock); \ + } \ + current->state = TASK_RUNNING; \ + remove_wait_queue(&wq, &__wait); \ +} while (0) + +#define wait_event_lock_irq(wq, condition, lock) \ +do { \ + if (condition) \ + break; \ + __wait_event_lock_irq(wq, condition, lock); \ +} while (0) + + +#define __wait_disk_event(wq, condition) \ +do { \ + wait_queue_t __wait; \ + init_waitqueue_entry(&__wait, current); \ + \ + add_wait_queue(&wq, &__wait); \ + for (;;) { \ + set_current_state(TASK_UNINTERRUPTIBLE); \ + if (condition) \ + break; \ + run_task_queue(&tq_disk); \ + schedule(); \ + } \ + current->state = TASK_RUNNING; \ + remove_wait_queue(&wq, &__wait); \ +} while (0) + +#define wait_disk_event(wq, condition) \ +do { \ + if (condition) \ + break; \ + __wait_disk_event(wq, condition); \ +} while (0) + +#endif + diff -urN /md0/kernels/2.4/v2.4.9-ac9/MAINTAINERS aio-v2.4.9-ac9/MAINTAINERS --- /md0/kernels/2.4/v2.4.9-ac9/MAINTAINERS Mon Sep 10 20:35:49 2001 +++ aio-v2.4.9-ac9/MAINTAINERS Fri Sep 7 00:15:16 2001 @@ -201,6 +201,12 @@ L: linux-net@vger.kernel.org S: Maintained +ASYNC IO +P: Benjamin LaHaise +M: bcrl@redhat.com +L: linux-aio@kvack.org +S: Maintained + AX.25 NETWORK LAYER P: Matthias Welwarsky M: dg2fef@afthd.tu-darmstadt.de diff -urN /md0/kernels/2.4/v2.4.9-ac9/Makefile aio-v2.4.9-ac9/Makefile --- /md0/kernels/2.4/v2.4.9-ac9/Makefile Mon Sep 10 20:35:49 2001 +++ aio-v2.4.9-ac9/Makefile Fri Sep 7 00:15:35 2001 @@ -1,7 +1,7 @@ VERSION = 2 PATCHLEVEL = 4 SUBLEVEL = 9 -EXTRAVERSION = -ac9 +EXTRAVERSION = -ac9-aio KERNELRELEASE=$(VERSION).$(PATCHLEVEL).$(SUBLEVEL)$(EXTRAVERSION) diff -urN /md0/kernels/2.4/v2.4.9-ac9/arch/i386/kernel/entry.S aio-v2.4.9-ac9/arch/i386/kernel/entry.S --- /md0/kernels/2.4/v2.4.9-ac9/arch/i386/kernel/entry.S Mon Sep 10 20:35:49 2001 +++ aio-v2.4.9-ac9/arch/i386/kernel/entry.S Mon Sep 17 14:29:38 2001 @@ -626,6 +626,12 @@ .long SYMBOL_NAME(sys_getdents64) /* 220 */ .long SYMBOL_NAME(sys_fcntl64) .long SYMBOL_NAME(sys_ni_syscall) /* reserved for TUX */ + .long SYMBOL_NAME(sys___io_setup) /* 223 */ + .long SYMBOL_NAME(sys___io_destroy) + .long SYMBOL_NAME(sys___io_getevents) + .long SYMBOL_NAME(sys___io_submit) + .long SYMBOL_NAME(sys___io_cancel) + .long SYMBOL_NAME(sys___io_wait) .rept NR_syscalls-(.-sys_call_table)/4 .long SYMBOL_NAME(sys_ni_syscall) diff -urN /md0/kernels/2.4/v2.4.9-ac9/drivers/char/raw.c aio-v2.4.9-ac9/drivers/char/raw.c --- /md0/kernels/2.4/v2.4.9-ac9/drivers/char/raw.c Mon Sep 10 20:35:51 2001 +++ aio-v2.4.9-ac9/drivers/char/raw.c Mon Sep 24 00:24:20 2001 @@ -16,6 +16,8 @@ #include #include #include +#include +#include #define dprintk(x...) @@ -36,13 +38,18 @@ int raw_open(struct inode *, struct file *); int raw_release(struct inode *, struct file *); int raw_ctl_ioctl(struct inode *, struct file *, unsigned int, unsigned long); - +int raw_kvec_read(struct file *filp, kvec_cb_t cb, size_t size, loff_t pos); +int raw_kvec_write(struct file *filp, kvec_cb_t cb, size_t size, loff_t pos); static struct file_operations raw_fops = { read: raw_read, write: raw_write, open: raw_open, release: raw_release, + aio_read: generic_file_aio_read, + aio_write: generic_file_aio_write, + kvec_read: raw_kvec_read, + kvec_write: raw_kvec_write, }; static struct file_operations raw_ctl_fops = { @@ -130,7 +137,7 @@ * the blocksize on a device which is already mounted. */ - sector_size = 512; + sector_size = 2048; if (is_mounted(rdev)) { if (blksize_size[MAJOR(rdev)]) sector_size = blksize_size[MAJOR(rdev)][MINOR(rdev)]; @@ -260,7 +267,6 @@ } - ssize_t raw_read(struct file *filp, char * buf, size_t size, loff_t *offp) { @@ -393,3 +399,81 @@ out: return err; } + +static int raw_kvec_rw(struct file *filp, int rw, kvec_cb_t cb, size_t size, loff_t pos); +int raw_kvec_read(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + return raw_kvec_rw(file, READ, cb, size, pos); +} + +int raw_kvec_write(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + return raw_kvec_rw(file, WRITE, cb, size, pos); +} + +int raw_kvec_rw(struct file *filp, int rw, kvec_cb_t cb, size_t size, loff_t pos) +{ + int err; + unsigned minor; + kdev_t dev; + unsigned long limit, blocknr, blocks; + + unsigned sector_size, sector_bits, sector_mask; + unsigned max_sectors; + + pr_debug("raw_rw_kiovec: %p %d %d %p %d %d %Lu\n", filp, rw, nr, kiovec, flags, size, pos); + /* + * First, a few checks on device size limits + */ + + minor = MINOR(filp->f_dentry->d_inode->i_rdev); + dev = to_kdev_t(raw_devices[minor].binding->bd_dev); + sector_size = raw_devices[minor].sector_size; + sector_bits = raw_devices[minor].sector_bits; + sector_mask = sector_size- 1; + max_sectors = 25000; //KIO_MAX_SECTORS >> (sector_bits - 9); + + if (blk_size[MAJOR(dev)]) + limit = (((loff_t) blk_size[MAJOR(dev)][MINOR(dev)]) << BLOCK_SIZE_BITS) >> sector_bits; + else + limit = INT_MAX; + dprintk ("rw_raw_dev_async: dev %d:%d (+%d)\n", + MAJOR(dev), MINOR(dev), limit); + + err = -EINVAL; + if ((pos < 0) || (pos & sector_mask) || (size & sector_mask)) { + printk("pos/size wrong\n"); + goto out; + } + + err = -ENXIO; + if ((pos >> sector_bits) >= limit) { + printk("raw: %Lu > %lu, %d\n", pos >> sector_bits, limit, sector_bits); + goto out; + } + + /* + * Split the IO into KIO_MAX_SECTORS chunks, mapping and + * unmapping the single kiobuf as we go to perform each chunk of + * IO. + */ + + blocknr = pos >> sector_bits; + blocks = size >> sector_bits; + if (blocks > max_sectors) + blocks = max_sectors; + if (blocks > limit - blocknr) + blocks = limit - blocknr; + err = -ENXIO; + pr_debug("raw: !blocks %d %ld %ld\n", max_sectors, limit, blocknr); + if (!blocks) + goto out; + + err = brw_kvec_async(rw, cb, dev, blocks, blocknr, sector_bits); + pr_debug("brw_kiovec_async: %d\n", err); + +out: + pr_debug("brw_kiovec_async: ret is %d\n", err); + return err; +} + diff -urN /md0/kernels/2.4/v2.4.9-ac9/fs/Makefile aio-v2.4.9-ac9/fs/Makefile --- /md0/kernels/2.4/v2.4.9-ac9/fs/Makefile Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/fs/Makefile Fri Sep 7 00:15:17 2001 @@ -12,7 +12,7 @@ obj-y := open.o read_write.o devices.o file_table.o buffer.o \ super.o block_dev.o char_dev.o stat.o exec.o pipe.o namei.o \ - fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \ + fcntl.o ioctl.o readdir.o select.o fifo.o locks.o aio.o \ dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \ filesystems.o jbd-kernel.o namespace.o diff -urN /md0/kernels/2.4/v2.4.9-ac9/fs/aio.c aio-v2.4.9-ac9/fs/aio.c --- /md0/kernels/2.4/v2.4.9-ac9/fs/aio.c Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/fs/aio.c Mon Sep 24 19:08:20 2001 @@ -0,0 +1,711 @@ +//#define DEBUG 1 +/* drivers/char/aio.c + * An async IO implementation for Linux + * Written by Benjamin LaHaise + * + * Implements /dev/aio, something on top of which it should be possible + * to write a POSIX AIO library. + * + * Copyright 2000, 2001 Red Hat, Inc. All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#undef KERN_DEBUG +#define KERN_DEBUG "" +#define MAX_IOCTXS 0x800 +#define dprintk(x...) do { ; } while (0) + +static spinlock_t aio_read_lock = SPIN_LOCK_UNLOCKED; +static spinlock_t aio_req_lock = SPIN_LOCK_UNLOCKED; + +static kmem_cache_t *kiocb_cachep; +static kmem_cache_t *kioctx_cachep; + +/* Lockless for reads. Needs replacement rsn. */ +static struct kioctx *ioctx_list; +static unsigned long new_ioctx_id; + +/* tunable. Needs to be added to sysctl. */ +int max_aio_reqs = 0x10000; + +/* 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_create("kiocb", sizeof(struct kiocb), + 0, SLAB_HWCACHE_ALIGN, NULL, NULL); + if (!kiocb_cachep) + panic("unable to create kiocb cache\n"); + + kioctx_cachep = kmem_cache_create("kioctx", sizeof(struct kioctx), + 0, SLAB_HWCACHE_ALIGN, NULL, NULL); + if (!kioctx_cachep) + panic("unable to create kioctx cache"); + + printk(KERN_NOTICE "aio_setup: okay!\n"); + printk(KERN_NOTICE "aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); + + return 0; +} + +/* ioctx_alloc + * Allocates and initializes an aioctx. Returns an ERR_PTR if it failed. + */ +static struct kioctx *ioctx_alloc(unsigned nr_reqs) +{ + struct kioctx *ctx; + unsigned i; + long size; + + if (nr_reqs > (0x70000000U / sizeof(struct io_event))) { + pr_debug("ENOMEM: nr_reqs too high\n"); + return ERR_PTR(-ENOMEM); + } + + /* Round off to a power of 2. Needed for cheap mask operations */ + for (i=1; imax_reqs = nr_reqs; + + atomic_set(&ctx->users, 1); + spin_lock_init(&ctx->lock); + init_waitqueue_head(&ctx->wait); + + size = sizeof(struct kiocb) * nr_reqs; + ctx->reqs = kmalloc(size, GFP_KERNEL); + if (!ctx->reqs) + goto out_freectx; + + memset(ctx->reqs, 0, size); + for (i=0; ireqs[i].ctx = ctx; + ctx->reqs[i].user_obj = ctx->reqs + i + 1; + } + ctx->reqs[nr_reqs-1].user_obj = NULL; + ctx->free_req = ctx->reqs; + size = sizeof(struct aio_ring); + size += sizeof(struct io_event) * nr_reqs; + /* This limits things somewhat for now. */ + ctx->ring = kmalloc(size, GFP_KERNEL); + if (!ctx->ring) + goto out_freereqs; + + memset(ctx->ring, 0, size); + ctx->mm = current->mm; + ctx->ring_mask = nr_reqs - 1; /* trusted copy */ + ctx->ring->mask = ctx->ring_mask; /* user copy */ + + /* now link into global list. kludge. FIXME */ + spin_lock(&aio_req_lock); /* FIXME */ + ctx->ring->id = ctx->user_id = new_ioctx_id++; /* FIXME */ + ctx->next = ioctx_list; /* FIXME */ + ioctx_list = ctx; /* FIXME */ + spin_unlock(&aio_req_lock); /* FIXME */ + + printk("aio: allocated aioctx %p[%ld]: mm=%p mask=0x%x\n", ctx, ctx->user_id, ctx->mm, ctx->ring->mask); + return ctx; + +out_freereqs: + kfree(ctx->reqs); +out_freectx: + kmem_cache_free(kioctx_cachep, ctx); + ctx = ERR_PTR(-ENOMEM); + + printk("aio: error allocating aioctx %p\n", ctx); + return ctx; +} + +/* Is this safe from irq context? */ +void kiocb_free(struct kiocb *iocb) +{ +#if 0 + int i; + for (i=0; inr_kiovec; i++) + unmap_kiobuf(iocb->kiovec[i]); + + free_kiovec_sz(iocb->nr_kiovec, iocb->kiovec, iocb->kiovec_sz); + iocb->nr_kiovec = 0; +#endif + //FIXME: gross hack. + //fput(iocb->filp); + if (atomic_dec_and_test(&iocb->filp->f_count)) + BUG(); /* not really, but... */ + + iocb->filp = NULL; + kmem_cache_free(kiocb_cachep, iocb); +} + +/* aio_complete + * Called when the io request on the given iocb is complete. + */ +void aio_complete(struct kiocb *iocb, long res, long res2) +{ + struct kioctx *ctx = iocb->ctx; + struct aio_ring *ring = ctx->ring; + struct io_event *event; + unsigned long flags; + unsigned long tail; + + /* add a completion event to the ring buffer. + * must be done holding ctx->lock to prevent + * other code from messing with the tail + * pointer since we might be called from irq + * context. + */ + spin_lock_irqsave(&ctx->lock, flags); + + tail = ring->tail; + event = &ring->io_events[tail]; + tail = (tail + 1) & ring->mask; + + event->obj = (u64)(unsigned long)iocb->user_obj; + event->data = iocb->user_data; + event->res = res; + event->res2 = res2; + + dprintk("aio_complete: %p[%lu]: %p: %Lx %Lx %lx %lx\n", + ctx, tail, iocb, iocb->user_obj, iocb->user_data, res, res2); + + /* after flagging the request as done, we + * must never even look at it again + */ + barrier(); + + ring->tail = tail; + + wmb(); + if (!ring->woke) + ring->woke = 1; + + spin_unlock_irqrestore(&ctx->lock, flags); + + pr_debug("added to ring %p at [%lu]\n", iocb, tail); +#if 0 + if (!wake) { + printk("kio_complete: should send user of %p a signal...\n", ctx); + } +#endif + + wake_up(&ctx->wait); + + /* everything turned out well, dispose of the aiocb. */ + kiocb_free(iocb); +} + +/* aio_read_evt + * Pull an event off of the aioctx's event ring. + * FIXME: make this use cmpxchg. + * TODO: make the ringbuffer user mmap()able (requires FIXME). + */ +static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) +{ + struct aio_ring *ring = ioctx->ring; + unsigned long head; + int ret = -EAGAIN; + + pr_debug("in aio_read_evt h%lu t%lu\n", ring->head, ring->tail); + barrier(); + if (ring->head == ring->tail) + goto out; + + spin_lock(&aio_read_lock); /* investigate the value of making this per-ctx */ + + head = ring->head; + if (head != ring->tail) { + *ent = ring->io_events[head]; + head = (head + 1) & ioctx->ring_mask; + barrier(); + ring->head = head; + ret = 0; + } + spin_unlock(&aio_read_lock); + +out: + pr_debug("leaving aio_read_evt: %d h%lu t%lu\n", ret, ring->head, ring->tail); + return ret; +} + +struct timeout { + struct timer_list timer; + int timed_out; + wait_queue_head_t wait; +}; + +static void timeout_func(unsigned long data) +{ + struct timeout *to = (struct timeout *)data; + + to->timed_out = 1; + wake_up(&to->wait); +} + +static inline void init_timeout(struct timeout *to) +{ + init_timer(&to->timer); + to->timer.data = (unsigned long)to; + to->timer.function = timeout_func; + to->timed_out = 0; + init_waitqueue_head(&to->wait); +} + +static inline void set_timeout(struct timeout *to, struct timespec *ts) +{ + unsigned long how_long; + + if (!ts->tv_sec && !ts->tv_nsec) { + to->timed_out = 1; + return; + } + + how_long = ts->tv_sec * HZ; +#define HZ_NS (1000000000 / HZ) + how_long += (ts->tv_nsec + HZ_NS - 1) / HZ_NS; + + to->timer.expires = jiffies + how_long; + add_timer(&to->timer); +} + +static inline void clear_timeout(struct timeout *to) +{ + del_timer_sync(&to->timer); +} + +static int read_events(struct kioctx *ctx, int nr, struct io_event *event, + struct timespec *timeout) +{ + struct task_struct *tsk = current; + DECLARE_WAITQUEUE(wait, tsk); + DECLARE_WAITQUEUE(to_wait, tsk); + int ret; + int i = 0; + struct io_event ent; + struct timespec ts; + struct timeout to; + + init_timeout(&to); + + if (timeout) { + ret = -EFAULT; + if (copy_from_user(&ts, timeout, sizeof(ts))) + goto out; + + set_timeout(&to, &ts); + } + + memset(&ent, 0, sizeof(ent)); + ret = 0; + + while (i < nr) { + ret = aio_read_evt(ctx, &ent); + if (ret) { + if (i) + break; + + ret = 0; + if (!i && !timeout) + break; + + add_wait_queue(&ctx->wait, &wait); + add_wait_queue(&to.wait, &to_wait); + do { + set_task_state(tsk, TASK_INTERRUPTIBLE); + + ret = aio_read_evt(ctx, &ent); + if (!ret) + break; + ret = -ETIMEDOUT; + if (to.timed_out) + break; + schedule(); + if (to.timed_out) + break; + if (signal_pending(tsk)) { + ret = -EINTR; + break; + } + ret = aio_read_evt(ctx, &ent); + } while (ret) ; + + set_task_state(tsk, TASK_RUNNING); + remove_wait_queue(&ctx->wait, &wait); + remove_wait_queue(&to.wait, &to_wait); + } + + if (ret) + break; + + pr_debug("read event: %Lx %Lx %Lx %Lx\n", + ent.data, ent.obj, ent.res, ent.res2); + + /* FIXME: split checks in two */ + ret = -EFAULT; + if (copy_to_user(event, &ent, sizeof(ent))) { + /* FIXME: we lose an event here. */ + printk(KERN_DEBUG "aio: lost an event due to EFAULT.\n"); + break; + } + + /* Now complete the aio request and copy the result codes to userland. */ + event ++; + i ++; + } + + if (timeout) + clear_timeout(&to); +out: + return i ? i : ret; +} + +/* __aioctx_put + * Called when the last user of an aio context has gone away, + * and the struct needs to be freed. + */ +void __aioctx_put(struct kioctx *ctx) +{ + printk("aio: free aioctx %p\n", ctx); + + kfree(ctx->ring); + kfree(ctx->reqs); + kmem_cache_free(kioctx_cachep, ctx); +} + +/* aio_get_req + * Allocate a slot for an aio request. Increments the users count + * of the kioctx so that the kioctx stays around until all requests are + * complete. Returns -EAGAIN if no requests are free. + */ +static inline struct kiocb *aio_get_req(struct kioctx *ctx) +{ + struct kiocb *req; + + /* FIXME: use cmpxchg instead of spin_lock? */ + spin_lock_irq(&ctx->lock); + req = ctx->free_req; + if (req) { + ctx->free_req = req->user_obj; + spin_unlock_irq(&ctx->lock); + req->user_obj = NULL; + + atomic_inc(&ctx->users); + return req; + } + spin_unlock_irq(&ctx->lock); + + return NULL; +} + +static void aio_put_req(struct kioctx *ctx, struct kiocb *req) +{ + req = ctx->reqs; + /* FIXME: use cmpxchg instead of spin_lock? */ + spin_lock_irq(&ctx->lock); + req->cancel = NULL; + req->user_obj = ctx->free_req; + ctx->free_req = req; + spin_unlock_irq(&ctx->lock); +} + +/* Lookup an ioctx id. ioctx_list is lockless for reads. + * FIXME: this is O(n) and is only suitable for development. + */ +static inline struct kioctx *get_ioctx(unsigned long ctx_id) +{ + struct kioctx *ioctx = ioctx_list; + struct mm_struct *mm = current->mm; + + do { + if (ioctx->user_id == ctx_id && ioctx->mm == mm) + return ioctx; + ioctx = ioctx->next; + } while (ioctx); + + return NULL; +} + +static inline void put_ioctx(struct kioctx *ctx) +{ + // FIXME!!! + //aioctx_put(ctx); +} + +asmlinkage long sys___io_setup(unsigned nr_reqs, aio_context_t *ctxp) +{ + struct kioctx *ioctx = NULL; + unsigned long ctx; + long ret; + + ret = get_user(ctx, ctxp); + if (ret) + goto out; + + ret = -EINVAL; + if (ctx || nr_reqs > max_aio_reqs) { + pr_debug("EINVAL: io_setup: !ctx or nr_reqs > max\n"); + goto out; + } + + ioctx = ioctx_alloc(nr_reqs); + ret = PTR_ERR(ioctx); + if (!IS_ERR(ioctx)) { + ret = put_user(ioctx->user_id, ctxp); + if (!ret) + return 0; + aioctx_put(ioctx); + } + +out: + return ret; +} + +/* aio_release + * Free the aioctx associated with the file. FIXME! + */ +asmlinkage long sys___io_destroy(aio_context_t ctx) +{ + struct kioctx *ioctx = get_ioctx(ctx); + if (ioctx) { + dprintk("aio_release(%p)\n", filp->private_data); + aioctx_put(ioctx); + return 0; + } + pr_debug("EINVAL: io_destroy: invalid context id\n"); + return -EINVAL; +} + +/* sys___io_submit + * Copy an aiocb from userspace into kernel space, then convert it to + * a kiocb, submit and repeat until done. Error codes on copy/submit + * only get returned for the first aiocb copied as otherwise the size + * of aiocbs copied is returned (standard write sematics). + */ +asmlinkage long sys___io_submit(aio_context_t ctx_id, int nr, struct iocb **iocbpp) +{ + struct kioctx *ctx; + long ret = 0; + int i; + + ctx = get_ioctx(ctx_id); + if (!ctx) P + pr_debug("EINVAL: io_submit: invalid context id\n"); + return -EINVAL; + } + + for (i=0; ireqs; + ret = put_user(tmp.aio_key, &iocbp->aio_key); + if (ret) + goto out_put_req; + + switch (tmp.aio_lio_opcode) { + case IOCB_CMD_PREAD: op = file->f_op->aio_read; break; + case IOCB_CMD_PREADX: op = file->f_op->aio_readx; break; + case IOCB_CMD_PWRITE: op = file->f_op->aio_write; break; + case IOCB_CMD_FSYNC: op = file->f_op->aio_fsync; break; + default: op = NULL; break; + } + ret = -EINVAL; + if (!op) { + pr_debug("EINVAL: io_submit: no operation provided\n"); + goto out_put_req; + } + + ret = op(file, req, tmp); + if (!ret) + continue; + + pr_debug("io_submit: op returned %ld\n", ret); + + out_put_req: + aio_put_req(ctx, req); + out_fput: + fput(file); + break; + } + + put_ioctx(ctx); + run_task_queue(&tq_disk); + return i ? i : ret; +} + +void generic_aio_complete(void *_iocb, struct kvec *vec, ssize_t res) +{ + struct kiocb *iocb = _iocb; + + aio_complete(iocb, res, 0); +} + +ssize_t generic_aio_read(struct file *file, struct kiocb *req, struct iocb iocb, size_t min_size) +{ + unsigned long buf = iocb.aio_buf; + size_t size = iocb.aio_nbytes; + ssize_t nr_read; + kvec_cb_t cb; + + nr_read = file->f_op->new_read(file, (void *)buf, size, + &iocb.aio_offset, F_ATOMIC); + if ((nr_read >= min_size) || + (nr_read < 0 && nr_read != -EAGAIN)) + return nr_read; + + size -= nr_read; + buf += nr_read; + cb.vec = map_user_kvec(WRITE, buf, size); + cb.fn = generic_aio_complete; + cb.data = req; + + return file->f_op->kvec_read(file, cb, size, iocb.aio_offset); +} + +ssize_t generic_file_aio_read(struct file *file, struct kiocb *req, struct iocb iocb) +{ + return generic_aio_read(file, req, iocb, iocb.aio_nbytes); +} + +ssize_t generic_aio_write(struct file *file, struct kiocb *req, struct iocb iocb, size_t min_size) +{ + unsigned long buf = iocb.aio_buf; + size_t size = iocb.aio_nbytes; + ssize_t nr_written; + kvec_cb_t cb; + + nr_written = file->f_op->new_write(file, (void *)buf, size, + &iocb.aio_offset, F_ATOMIC); + if ((nr_written >= min_size) || + (nr_written < 0 && nr_written != -EAGAIN)) + return nr_written; + + size -= nr_written; + buf += nr_written; + cb.vec = map_user_kvec(READ, buf, size); + cb.fn = generic_aio_complete; + cb.data = req; + + return file->f_op->kvec_write(file, cb, size, iocb.aio_offset); +} + +ssize_t generic_file_aio_write(struct file *file, struct kiocb *req, struct iocb iocb) +{ + return generic_aio_write(file, req, iocb, iocb.aio_nbytes); +} + +asmlinkage long sys___io_cancel(aio_context_t ctx, struct iocb *iocb) +{ + return -ENOSYS; +} + +asmlinkage long sys___io_wait(aio_context_t ctx_id, struct iocb *iocb, struct timespec *timeout) +{ +#if 0 /* FIXME. later. */ + struct kioctx *ioctx; + long ret = -EINVAL; + unsigned key; + long obj = (long)iocb; + + ioctx = get_ioctx(ctx_id); + if (!ioctx) + goto out; + + ret = get_user(key, &iocb->aio_key); + if (ret) + goto out; + + ret = __aio_complete(ioctx, key, obj, !!timeout); + put_ioctx(ioctx); + +out: + return ret; +#endif + return -ENOSYS; +} + +asmlinkage long sys___io_getevents(int ctx_id, int nr, struct io_event *events, + struct timespec *timeout) +{ + struct kioctx *ioctx = get_ioctx(ctx_id); + long ret = -EINVAL; + + if (ioctx) { + ret = read_events(ioctx, nr, events, timeout); + put_ioctx(ioctx); + } + + return ret; +} + +__initcall(aio_setup); diff -urN /md0/kernels/2.4/v2.4.9-ac9/fs/buffer.c aio-v2.4.9-ac9/fs/buffer.c --- /md0/kernels/2.4/v2.4.9-ac9/fs/buffer.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/fs/buffer.c Mon Sep 24 00:22:27 2001 @@ -140,8 +140,7 @@ { clear_bit(BH_Lock, &bh->b_state); smp_mb__after_clear_bit(); - if (waitqueue_active(&bh->b_wait)) - wake_up(&bh->b_wait); + wake_up(&bh->b_wait); } /* @@ -2061,6 +2060,7 @@ return tmp.b_blocknr; } +#if 1 /* * IO completion routine for a buffer_head being used for kiobuf IO: we * can't dispatch the kiobuf callback until io_count reaches 0. @@ -2237,6 +2237,7 @@ return transferred; return err; } +#endif /* * Start I/O on a page. @@ -2868,3 +2869,222 @@ module_init(bdflush_init) +/* async kio interface */ +struct brw_cb { + kvec_cb_t cb; + atomic_t io_count; + int nr; + struct buffer_head *bh[1]; +}; + +static inline void brw_cb_put(struct brw_cb *brw_cb) +{ + if (atomic_dec_and_test(&brw_cb->io_count)) { + ssize_t res = 0, err = 0; + int nr; + + /* Walk the buffer heads associated with this kiobuf + * checking for errors and freeing them as we go. + */ + for (nr=0; nr < brw_cb->nr; nr++) { + struct buffer_head *bh = brw_cb->bh[nr]; + if (!err && buffer_uptodate(bh)) + res += bh->b_size; + else + err = -EIO; + kmem_cache_free(bh_cachep, bh); + } + + if (!res) + res = err; + + brw_cb->cb.fn(brw_cb->cb.data, brw_cb->cb.vec, res); + + kfree(brw_cb); + } +} + +/* + * IO completion routine for a buffer_head being used for kiobuf IO: we + * can't dispatch the kiobuf callback until io_count reaches 0. + */ + +static void end_buffer_io_kiobuf_async(struct buffer_head *bh, int uptodate) +{ + struct brw_cb *brw_cb; + + mark_buffer_uptodate(bh, uptodate); + + brw_cb = bh->b_private; + unlock_buffer(bh); + + brw_cb_put(brw_cb); +} + + +/* + * Start I/O on a physical range of kernel memory, defined by a vector + * of kiobuf structs (much like a user-space iovec list). + * + * The kiobuf must already be locked for IO. IO is submitted + * asynchronously: you need to check page->locked, page->uptodate, and + * maybe wait on page->wait. + * + * It is up to the caller to make sure that there are enough blocks + * passed in to completely map the iobufs to disk. + */ + +int brw_kvec_async(int rw, kvec_cb_t cb, kdev_t dev, unsigned blocks, unsigned long blknr, int sector_shift) +{ + struct kvec *vec = cb.vec; + struct kveclet *veclet; + int err; + int length; + unsigned sector_size = 1 << sector_shift; + int i; + + struct brw_cb *brw_cb; + + if (!vec->nr) + BUG(); + + /* + * First, do some alignment and validity checks + */ + length = 0; + for (veclet=vec->veclet, i=0; i < vec->nr; i++,veclet++) { + length += veclet->length; + if ((veclet->offset & (sector_size-1)) || + (veclet->length & (sector_size-1))) { + printk("brw_kiovec_async: tuple[%d]->offset=0x%x length=0x%x sector_size: 0x%x\n", i, veclet->offset, veclet->length, sector_size); + return -EINVAL; + } + } + + if (length < (blocks << sector_shift)) + BUG(); + + /* + * OK to walk down the iovec doing page IO on each page we find. + */ + err = 0; + + if (!blocks) { + printk("brw_kiovec_async: !i\n"); + return -EINVAL; + } + + /* FIXME: tie into userbeans here */ + brw_cb = kmalloc(sizeof(*brw_cb) + (blocks * sizeof(struct buffer_head *)), GFP_KERNEL); + if (!brw_cb) + return -ENOMEM; + + brw_cb->cb = cb; + brw_cb->nr = 0; + + /* This is ugly. FIXME. */ + for (i=0, veclet=vec->veclet; inr; i++,veclet++) { + struct page *page = veclet->page; + unsigned offset = veclet->offset; + unsigned length = veclet->length; + + if (!page) + BUG(); + + while (length > 0) { + struct buffer_head *tmp; + tmp = kmem_cache_alloc(bh_cachep, GFP_NOIO); + err = -ENOMEM; + if (!tmp) + goto error; + + memset(tmp, 0, sizeof(*tmp)); + init_waitqueue_head(&tmp->b_wait); + tmp->b_dev = B_FREE; + tmp->b_size = sector_size; + set_bh_page(tmp, page, offset); + tmp->b_this_page = tmp; + + init_buffer(tmp, end_buffer_io_kiobuf_async, NULL); + tmp->b_dev = dev; + tmp->b_blocknr = blknr++; + tmp->b_state = (1 << BH_Mapped) | (1 << BH_Lock) + | (1 << BH_Req); + tmp->b_private = brw_cb; + + if (rw == WRITE) { + set_bit(BH_Uptodate, &tmp->b_state); + clear_bit(BH_Dirty, &tmp->b_state); + } + + brw_cb->bh[brw_cb->nr++] = tmp; + length -= sector_size; + offset += sector_size; + + if (offset >= PAGE_SIZE) { + offset = 0; + break; + } + + if (brw_cb->nr >= blocks) + goto submit; + } /* End of block loop */ + } /* End of page loop */ + +submit: + atomic_set(&brw_cb->io_count, brw_cb->nr+1); + /* okay, we've setup all our io requests, now fire them off! */ + for (i=0; inr; i++) + submit_bh(rw, brw_cb->bh[i]); + brw_cb_put(brw_cb); + + return 0; + +error: + /* Walk brw_cb_table freeing all the goop associated with each kiobuf */ + if (brw_cb) { + /* We got an error allocating the bh'es. Just free the current + buffer_heads and exit. */ + for (i = brw_cb->nr-1; i--; ) + kmem_cache_free(bh_cachep, brw_cb->bh[i]); + kfree(brw_cb); + } + + return err; +} +#if 0 +int brw_kiovec(int rw, int nr, struct kiobuf *iovec[], + kdev_t dev, int nr_blocks, unsigned long b[], int sector_size) +{ + int i; + int transferred = 0; + int err = 0; + + if (!nr) + return 0; + + /* queue up and trigger the io */ + err = brw_kiovec_async(rw, nr, iovec, dev, nr_blocks, b, sector_size); + if (err) + goto out; + + /* wait on the last iovec first -- it's more likely to finish last */ + for (i=nr; --i >= 0; ) + kiobuf_wait_for_io(iovec[i]); + + run_task_queue(&tq_disk); + + /* okay, how much data actually got through? */ + for (i=0; ierrno) { + if (!err) + err = iovec[i]->errno; + break; + } + transferred += iovec[i]->length; + } + +out: + return transferred ? transferred : err; +} +#endif diff -urN /md0/kernels/2.4/v2.4.9-ac9/fs/ext2/file.c aio-v2.4.9-ac9/fs/ext2/file.c --- /md0/kernels/2.4/v2.4.9-ac9/fs/ext2/file.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/fs/ext2/file.c Sat Sep 22 18:16:02 2001 @@ -42,6 +42,7 @@ llseek: generic_file_llseek, read: generic_file_read, write: generic_file_write, + //rw_kiovec: generic_file_rw_kiovec, ioctl: ext2_ioctl, mmap: generic_file_mmap, open: generic_file_open, diff -urN /md0/kernels/2.4/v2.4.9-ac9/fs/nfs/file.c aio-v2.4.9-ac9/fs/nfs/file.c --- /md0/kernels/2.4/v2.4.9-ac9/fs/nfs/file.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/fs/nfs/file.c Sat Sep 22 18:22:46 2001 @@ -50,6 +50,7 @@ release: nfs_release, fsync: nfs_fsync, lock: nfs_lock, + //rw_kiovec: generic_file_rw_kiovec, }; struct inode_operations nfs_file_inode_operations = { diff -urN /md0/kernels/2.4/v2.4.9-ac9/fs/select.c aio-v2.4.9-ac9/fs/select.c --- /md0/kernels/2.4/v2.4.9-ac9/fs/select.c Thu Aug 16 16:58:49 2001 +++ aio-v2.4.9-ac9/fs/select.c Sun Sep 9 00:10:40 2001 @@ -12,22 +12,31 @@ * 24 January 2000 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian). + * + * June 2001 + * Added async_poll implementation. -ben */ +#include #include #include #include #include +#include +#include #include #define ROUND_UP(x,y) (((x)+(y)-1)/(y)) #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM) +static kmem_cache_t *poll_table_cache; + struct poll_table_entry { - struct file * filp; - wait_queue_t wait; - wait_queue_head_t * wait_address; + wait_queue_t wait; + wait_queue_head_t *wait_address; + struct file *filp; + poll_table *p; }; struct poll_table_page { @@ -71,6 +80,72 @@ } } +void async_poll_complete(void *data) +{ + poll_table *p = data, *pwait; + struct kiocb *iocb = p->iocb; + unsigned int mask; + + pwait = p; + p->wake = 0; + wmb(); + do { + mask = iocb->filp->f_op->poll(iocb->filp, p); + mask &= p->events | POLLERR | POLLHUP; + if (mask) { + poll_freewait(p); + aio_complete(iocb, mask, 0); + return; + } + p->sync = 0; + wmb(); + } while (p->wake); + +} + +static void async_poll_waiter(wait_queue_t *wait) +{ + struct poll_table_entry *entry = (struct poll_table_entry *)wait; + poll_table *p = entry->p; + + /* avoid writes to the cacheline if possible for SMP */ + if (!p->wake) { + p->wake = 1; + /* ensure only one wake up queues the wtd */ + if (!p->sync && !test_and_set_bit(0, &p->sync)) + wtd_queue(&p->wtd); + } +} + +int async_poll(struct kiocb *iocb, int events) +{ + unsigned int mask; + poll_table *p, *pwait; + + p = kmem_cache_alloc(poll_table_cache, SLAB_KERNEL); + if (!p) + return -ENOMEM; + + poll_initwait(p); + wtd_set_action(&p->wtd, async_poll_complete, p); + p->iocb = iocb; + p->wake = 0; + p->sync = 0; + p->events = events; + pwait = p; + + mask = DEFAULT_POLLMASK; + if (iocb->filp->f_op && iocb->filp->f_op->poll) + mask = iocb->filp->f_op->poll(iocb->filp, p); + mask &= events | POLLERR | POLLHUP; + if (mask) { + poll_freewait(p); + aio_complete(iocb, mask, 0); + } + + return 0; +} + void __pollwait(struct file * filp, wait_queue_head_t * wait_address, poll_table *p) { struct poll_table_page *table = p->table; @@ -97,7 +172,11 @@ get_file(filp); entry->filp = filp; entry->wait_address = wait_address; - init_waitqueue_entry(&entry->wait, current); + entry->p = p; + if (p->iocb) + init_waitqueue_func_entry(&entry->wait, async_poll_waiter); + else + init_waitqueue_entry(&entry->wait, current); add_wait_queue(wait_address,&entry->wait); } } @@ -493,3 +572,14 @@ poll_freewait(&table); return err; } + +static int __init poll_init(void) +{ + poll_table_cache = kmem_cache_create("poll table", + sizeof(poll_table), 0, 0, NULL, NULL); + if (!poll_table_cache) + panic("unable to alloc poll_table_cache"); + return 0; +} + +module_init(poll_init); diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/asm-i386/errno.h aio-v2.4.9-ac9/include/asm-i386/errno.h --- /md0/kernels/2.4/v2.4.9-ac9/include/asm-i386/errno.h Mon Feb 26 10:20:14 2001 +++ aio-v2.4.9-ac9/include/asm-i386/errno.h Sat Sep 22 18:17:53 2001 @@ -128,5 +128,6 @@ #define ENOMEDIUM 123 /* No medium found */ #define EMEDIUMTYPE 124 /* Wrong medium type */ +#define ENOAIO 125 /* fd does not support aio */ #endif diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/asm-i386/uaccess.h aio-v2.4.9-ac9/include/asm-i386/uaccess.h --- /md0/kernels/2.4/v2.4.9-ac9/include/asm-i386/uaccess.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/asm-i386/uaccess.h Mon Sep 24 16:11:47 2001 @@ -127,6 +127,7 @@ extern void __put_user_1(void); extern void __put_user_2(void); extern void __put_user_4(void); +extern void __put_user_8(void); extern void __put_user_bad(void); @@ -162,6 +163,13 @@ case 1: __put_user_asm(x,ptr,retval,"b","b","iq"); break; \ case 2: __put_user_asm(x,ptr,retval,"w","w","ir"); break; \ case 4: __put_user_asm(x,ptr,retval,"l","","ir"); break; \ + case 8: { \ + u32 *__put_ptr = (void*)(ptr); \ + u64 __put_val = (x); \ + __put_user_asm((u32)__put_val,__put_ptr,retval,"l","","ir"); \ + __put_user_asm((u32)(__put_val>>32),__put_ptr+1,retval,"l","","ir"); \ + break; \ + } \ default: __put_user_bad(); \ } \ } while (0) diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/asm-i386/unistd.h aio-v2.4.9-ac9/include/asm-i386/unistd.h --- /md0/kernels/2.4/v2.4.9-ac9/include/asm-i386/unistd.h Fri Aug 11 17:39:23 2000 +++ aio-v2.4.9-ac9/include/asm-i386/unistd.h Tue Sep 18 05:07:19 2001 @@ -227,9 +227,18 @@ #define __NR_madvise1 219 /* delete when C lib stub is removed */ #define __NR_getdents64 220 #define __NR_fcntl64 221 +/* reserved for tux 222 */ +#define __NR___io_setup 223 +#define __NR___io_destroy 224 +#define __NR___io_getevents 225 +#define __NR___io_submit 226 +#define __NR___io_cancel 227 +#define __NR___io_wait 228 /* user-visible error numbers are in the range -1 - -124: see */ - +#ifdef NO_SYSCALL_ERRNO +#define __syscall_return(type, res) return (type)(res) +#else #define __syscall_return(type, res) \ do { \ if ((unsigned long)(res) >= (unsigned long)(-125)) { \ @@ -238,6 +247,7 @@ } \ return (type) (res); \ } while (0) +#endif /* XXX - _foo needs to be __foo, while __NR_bar could be _NR_bar. */ #define _syscall0(type,name) \ diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/aio.h aio-v2.4.9-ac9/include/linux/aio.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/aio.h Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/include/linux/aio.h Sun Sep 23 22:36:48 2001 @@ -0,0 +1,129 @@ +/* linux/aio.h + * Written by Benjamin LaHaise + */ +#ifndef __LINUX__AIO_H +#define __LINUX__AIO_H + +#include + +typedef unsigned long aio_context_t; + +enum { + IOCB_CMD_PREAD = 0, + IOCB_CMD_PWRITE = 1, + IOCB_CMD_FSYNC = 2, + IOCB_CMD_FDSYNC = 3, + IOCB_CMD_PREADX = 4, +}; + +/* read() from /dev/aio returns these structures. */ +struct io_event { + __u64 data; /* the data field from the iocb */ + __u64 obj; /* what iocb this event came from */ + __s64 res; /* result code for this event */ + __s64 res2; /* secondary result */ +}; + +struct aio_ring { + __u32 id; /* kernel internal index number */ + __u32 mask; /* number of io_events - 1 */ + __u32 head; + __u32 tail; + + __u32 woke; /* set when a wakeup was sent */ + __u32 pad1; + __u32 pad2; + __u32 pad3; + + __u32 pad4[24]; /* pad out to 128 bytes */ + + struct io_event io_events[0]; +}; /* 128 bytes + ring size */ + +#if defined(__LITTLE_ENDIAN) +#define PADDED(x,y) x, y +#elif defined(__BIG_ENDIAN) +#define PADDED(x,y) y, x +#else +#error edit for your odd byteorder. +#endif + +/* + * we always use a 64bit off_t when communicating + * with userland. its up to libraries to do the + * proper padding and aio_error abstraction + */ + +struct iocb { + /* these are internal to the kernel/libc. */ + __u64 aio_data; /* data to be returned in event's data */ + __u32 PADDED(aio_key, aio_reserved1); + /* the kernel sets aio_key to the req # */ + + /* common fields */ + __u16 aio_lio_opcode; /* see IOCB_CMD_ above */ + __s16 aio_reqprio; + __u32 aio_fildes; + + __u64 aio_buf; + __u64 aio_nbytes; + __s64 aio_offset; + + /* extra parameters */ + __u64 aio_reserved2; + __u64 aio_reserved3; +}; /* 64 bytes */ + +#undef IFBIG +#undef IFLITTLE + +#ifdef __KERNEL__ +#ifndef __LINUX__KIOVEC_H +#include +#endif + +#define AIO_MAXSEGS 4 +#define AIO_KIOGRP_NR_ATOMIC 8 + +struct kioctx; + +struct kiocb { + void (*cancel)(void *data, struct kioctx *ctx, int idx); + struct file *filp; + struct kioctx *ctx; + void *user_obj; + __u64 user_data; +}; + +struct kioctx { + atomic_t users; + + /* This needs improving */ + unsigned long user_id; + struct kioctx *next; + struct mm_struct *mm; + + wait_queue_head_t wait; + + spinlock_t lock; + + struct kiocb *reqs; + struct kiocb *free_req; + + unsigned max_reqs; + unsigned ring_mask; + struct aio_ring *ring; +}; + +extern struct file_operations aio_fops; + +extern void aio_complete(struct kiocb *iocb, long res, long res2); +extern void __aioctx_put(struct kioctx *ctx); + +#define aioctx_get(kioctx) atomic_inc(&(kioctx)->users) +#define aioctx_put(kioctx) do { if (atomic_dec_and_test(&(kioctx)->users)) __aioctx_put(kioctx); } while (0) + +#endif /*__KERNEL__*/ + +#endif /* __AIO_H__ */ + diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/blkdev.h aio-v2.4.9-ac9/include/linux/blkdev.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/blkdev.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/blkdev.h Mon Sep 24 16:11:47 2001 @@ -149,7 +149,7 @@ extern struct blk_dev_struct blk_dev[MAX_BLKDEV]; extern void grok_partitions(struct gendisk *dev, int drive, unsigned minors, long size); extern void register_disk(struct gendisk *dev, kdev_t first, unsigned minors, struct block_device_operations *ops, long size); -extern void generic_make_request(int rw, struct buffer_head * bh); +extern void generic_make_request(int rw, struct buffer_head *bh); extern request_queue_t *blk_get_queue(kdev_t dev); extern inline request_queue_t *__blk_get_queue(kdev_t dev); extern void blkdev_release_request(struct request *); diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/brlock.h aio-v2.4.9-ac9/include/linux/brlock.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/brlock.h Mon Sep 17 22:29:35 2001 +++ aio-v2.4.9-ac9/include/linux/brlock.h Sun Sep 23 19:05:00 2001 @@ -34,6 +34,7 @@ enum brlock_indices { BR_GLOBALIRQ_LOCK, BR_NETPROTO_LOCK, + BR_AIO_LOCK, __BR_END }; diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/event.h aio-v2.4.9-ac9/include/linux/event.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/event.h Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/include/linux/event.h Fri Sep 7 00:15:17 2001 @@ -0,0 +1,21 @@ +#ifndef _LINUX_KEVENTQ_H +#define _LINUX_KEVENTQ_H + +typedef struct file *keventq_t; + +keventq_t keventq_get(int qid); +#define keventq_put(evq) fput(evq) + +keventq_t keventq_get(int qid) +{ + struct file *filp = fget(qid); + if (filp) { + if (&keventq_fops == filp->f_op) + return filp; + fput(filp); + } + return NULL; +} + + +#endif diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/fs.h aio-v2.4.9-ac9/include/linux/fs.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/fs.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/fs.h Mon Sep 24 16:11:47 2001 @@ -20,7 +20,6 @@ #include #include #include -#include #include #include @@ -802,7 +801,21 @@ * NOTE: * read, write, poll, fsync, readv, writev can be called * without the big kernel lock held in all filesystems. - */ + * + * rw_kiovec returns the number of bytes that will actually + * be transferred into the kiovec, or an error that occurred + * during queueing. + */ +struct iocb; +struct kioctx; +struct kiocb; +struct kiobuf; +#include /* FIXME */ +#include + +#define F_ATOMIC 0x0001 +#define F_OFFSETOK 0x0002 + struct file_operations { struct module *owner; loff_t (*llseek) (struct file *, loff_t, int); @@ -822,6 +835,20 @@ ssize_t (*writev) (struct file *, const struct iovec *, unsigned long, loff_t *); ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int); unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long); + + + /* this will replace read/write ops above in 2.5 */ + ssize_t (*new_read) (struct file *, char *, size_t, loff_t *, int); + ssize_t (*new_write) (struct file *, char *, size_t, loff_t *, int); + + ssize_t (*aio_read)(struct file *, struct kiocb *, struct iocb); + ssize_t (*aio_readx)(struct file *, struct kiocb *, struct iocb); + ssize_t (*aio_write)(struct file *, struct kiocb *, struct iocb); + ssize_t (*aio_fsync)(struct file *, struct kiocb *, struct iocb); + + /* in-kernel async api */ + int (*kvec_read)(struct file *, kvec_cb_t, size_t, loff_t); + int (*kvec_write)(struct file *, kvec_cb_t, size_t, loff_t); }; struct inode_operations { @@ -1401,6 +1428,10 @@ unsigned long *); extern int block_sync_page(struct page *); +extern int generic_aio_read(struct file *, struct kiocb *, struct iocb, size_t); +extern int generic_aio_write(struct file *, struct kiocb *, struct iocb, size_t); +extern int generic_file_aio_read(struct file *, struct kiocb *, struct iocb); +extern int generic_file_aio_write(struct file *, struct kiocb *, struct iocb); int generic_block_bmap(struct address_space *, long, get_block_t *); int generic_commit_write(struct file *, struct page *, unsigned, unsigned); int block_truncate_page(struct address_space *, loff_t, get_block_t *); @@ -1411,6 +1442,7 @@ extern ssize_t generic_file_read(struct file *, char *, size_t, loff_t *); extern ssize_t generic_file_write(struct file *, const char *, size_t, loff_t *); extern void do_generic_file_read(struct file *, loff_t *, read_descriptor_t *, read_actor_t); +extern int generic_file_rw_kiovec(struct file *filp, int rw, int nr, struct kiobuf **kiovec, int flags, size_t size, loff_t pos); extern ssize_t generic_read_dir(struct file *, char *, size_t, loff_t *); extern loff_t generic_file_llseek(struct file *, loff_t, int); diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/iobuf.h aio-v2.4.9-ac9/include/linux/iobuf.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/iobuf.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/iobuf.h Mon Sep 24 16:11:47 2001 @@ -53,8 +53,10 @@ /* Dynamic state for IO completion: */ atomic_t io_count; /* IOs still in progress */ + int transferred; /* Number of bytes of completed IO at the beginning of the buffer */ int errno; /* Status of completed IO */ void (*end_io) (struct kiobuf *); /* Completion callback */ + void *end_io_data; wait_queue_head_t wait_queue; }; @@ -80,6 +82,8 @@ /* fs/buffer.c */ +int brw_kiovec_async(int rw, int nr, struct kiobuf *iovec[], + kdev_t dev, int nr_blocks, unsigned long b[], int size); int brw_kiovec(int rw, int nr, struct kiobuf *iovec[], kdev_t dev, unsigned long b[], int size); diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/kiovec.h aio-v2.4.9-ac9/include/linux/kiovec.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/kiovec.h Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/include/linux/kiovec.h Mon Sep 24 00:00:07 2001 @@ -0,0 +1,36 @@ +#ifndef __LINUX__IOBUF_H +#define __LINUX__IOBUF_H + +struct page; + +struct kveclet { + struct page *page; + unsigned offset; + unsigned length; +}; + +struct kvec { + unsigned max_nr; + unsigned nr; + struct kveclet veclet[0]; +}; + +struct kvec_cb { + struct kvec *vec; + void (*fn)(void *data, struct kvec *vec, ssize_t res); + void *data; +}; + +#ifndef _LINUX_TYPES_H +#include +#endif +#ifndef _LINUX_KDEV_T_H +#include +#endif + +extern struct kvec *map_user_kvec(int rw, unsigned long va, size_t len); +extern void unmap_kvec(struct kvec *); +extern int brw_kvec_async(int rw, kvec_cb_t cb, kdev_t dev, unsigned count, + unsigned long blknr, int sector_shift); + +#endif diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/lib_lio.h aio-v2.4.9-ac9/include/linux/lib_lio.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/lib_lio.h Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/include/linux/lib_lio.h Fri Sep 7 00:15:17 2001 @@ -0,0 +1,108 @@ +#ifndef __LIB_LIO_H +#define __LIB_LIO_H + +struct timespec; +struct sockaddr; +struct iovec; + + +typedef enum lio_iocb_cmd { + + LIO_CMD_PREAD, + LIO_CMD_PWRITE, + LIO_CMD_ACCEPT, + LIO_CMD_CONNECT, + LIO_CMD_SENDTO, + LIO_CMD_RECVFROM, + + LIO_CMD_POLL, +} lio_iocb_cmd_t; + +struct lio_iocb_sendto { + void *msg; + int len; + int flags; + struct sockaddr *addr; +}; + +struct lio_iocb_poll { + int events; +}; /* result code is the set of result flags or -'ve errno */ + +struct lio_iocb_sockaddr { + struct sockaddr *addr; + int len; +}; /* result code is the length of the sockaddr, or -'ve errno */ + +struct lio_iocb_common { + void *buf; + long nbytes; + long long offset; +}; /* result code is the amount read or -'ve errno */ + +struct lio_iocb_vector { + const struct iovec *vec; + int nr; + long long offset; +}; /* result code is the amount read or -'ve errno */ + +typedef struct lio_iocb { + long key; /* For use in identifying io requests */ + void *data; /* Return in the io completion event */ + int aio_fildes; + short aio_reqprio; + short aio_lio_opcode; + + union { + struct lio_iocb_common c; + struct lio_iocb_vector v; + struct lio_iocb_poll poll; + struct lio_iocb_sockaddr saddr; + } u; +} lio_iocb_t; + +typedef void (*lio_callback_t)(int qid, lio_iocb_t *iocb, long result); + +extern int lio_queue_init(int maxevents); +extern int lio_queue_grow(int qid, int new_maxevents); +extern int lio_queue_release(int qid); +extern int lio_queue_wait(int qid, struct timespec *timeout); +extern int lio_queue_run(int qid); +extern int lio_submit(int qid, int nr, lio_iocb_t *ios[]); + +static inline void lio_prep_accept(lio_iocb_t *iocb, int s, struct sockaddr *addr, int addrlen) +{ + iocb->aio_fildes = s; + iocb->aio_lio_opcode = LIO_CMD_ACCEPT; + iocb->aio_reqprio = 0; + iocb->u.c.buf = addr; + iocb->u.c.nbytes = addrlen; + iocb->u.c.offset = 0; +} + +static inline void lio_prep_pread(lio_iocb_t *iocb, int fd, void *buf, long count, long long offset) +{ + iocb->aio_fildes = fd; + iocb->aio_lio_opcode = LIO_CMD_PREAD; + iocb->aio_reqprio = 0; + iocb->u.c.buf = buf; + iocb->u.c.nbytes = count; + iocb->u.c.offset = offset; +} + +static inline void lio_prep_poll(lio_iocb_t *iocb, lio_callback_t *cb, int fd, int events) +{ + iocb->data = cb; + iocb->aio_fildes = fd; + iocb->aio_lio_opcode = LIO_CMD_POLL; + iocb->aio_reqprio = 0; + iocb->u.poll.events = events; +} + +static inline int lio_poll(int qid, lio_iocb_t *iocb, lio_callback_t *cb, int fd, int events) +{ + lio_prep_poll(iocb, cb, fd, events); + return lio_submit(qid, 1, &iocb); +} + +#endif diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/mm.h aio-v2.4.9-ac9/include/linux/mm.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/mm.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/mm.h Mon Sep 24 16:11:47 2001 @@ -322,8 +322,7 @@ smp_mb__before_clear_bit(); \ if (!test_and_clear_bit(PG_locked, &(page)->flags)) BUG(); \ smp_mb__after_clear_bit(); \ - if (waitqueue_active(&(page)->wait)) \ - wake_up(&(page)->wait); \ + wake_up(&(page)->wait); \ } while (0) #define PageError(page) test_bit(PG_error, &(page)->flags) #define SetPageError(page) set_bit(PG_error, &(page)->flags) diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/net.h aio-v2.4.9-ac9/include/linux/net.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/net.h Mon Sep 24 01:51:28 2001 +++ aio-v2.4.9-ac9/include/linux/net.h Sun Sep 23 19:01:36 2001 @@ -83,6 +83,9 @@ struct scm_cookie; struct vm_area_struct; struct page; +struct iocb; +struct kioctx; +#include /* shut gcc up */ struct proto_ops { int family; @@ -110,6 +113,7 @@ int (*recvmsg) (struct socket *sock, struct msghdr *m, int total_len, int flags, struct scm_cookie *scm); int (*mmap) (struct file *file, struct socket *sock, struct vm_area_struct * vma); ssize_t (*sendpage) (struct socket *sock, struct page *page, int offset, size_t size, int flags); + int (*begin_read) (struct socket *sock, struct kioctx *ctx, struct iocb iocb, struct iocb *iocbptr); }; struct net_proto_family diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/poll.h aio-v2.4.9-ac9/include/linux/poll.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/poll.h Mon Sep 17 22:27:32 2001 +++ aio-v2.4.9-ac9/include/linux/poll.h Mon Sep 24 16:11:47 2001 @@ -7,14 +7,25 @@ #include #include +#ifndef __LINUX__MM_H #include +#endif #include +#ifndef __LINUX__WORKTODO_H +#include +#endif struct poll_table_page; +struct kiocb; typedef struct poll_table_struct { - int error; - struct poll_table_page * table; + struct worktodo wtd; + int error; + struct poll_table_page *table; + struct kiocb *iocb; /* iocb for async poll */ + int events; /* event mask for async poll */ + int wake; + long sync; } poll_table; extern void __pollwait(struct file * filp, wait_queue_head_t * wait_address, poll_table *p); @@ -29,7 +40,9 @@ { pt->error = 0; pt->table = NULL; + pt->iocb = NULL; } + extern void poll_freewait(poll_table* pt); diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/raid/md_k.h aio-v2.4.9-ac9/include/linux/raid/md_k.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/raid/md_k.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/raid/md_k.h Sun Sep 23 19:01:36 2001 @@ -38,6 +38,7 @@ case RAID5: return 5; } panic("pers_to_level()"); + return 0; } static inline int level_to_pers (int level) diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/sched.h aio-v2.4.9-ac9/include/linux/sched.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/sched.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/sched.h Mon Sep 24 16:11:47 2001 @@ -763,6 +763,7 @@ extern void FASTCALL(add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)); extern void FASTCALL(add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)); +extern void FASTCALL(add_wait_queue_exclusive_lifo(wait_queue_head_t *q, wait_queue_t * wait)); extern void FASTCALL(remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)); #define __wait_event(wq, condition) \ diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/tqueue.h aio-v2.4.9-ac9/include/linux/tqueue.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/tqueue.h Mon Sep 24 01:51:28 2001 +++ aio-v2.4.9-ac9/include/linux/tqueue.h Sun Sep 23 19:01:36 2001 @@ -67,6 +67,7 @@ #define TQ_ACTIVE(q) (!list_empty(&q)) extern task_queue tq_timer, tq_immediate, tq_disk; +extern struct tq_struct run_disk_tq; /* * To implement your own list of active bottom halfs, use the following diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/types.h aio-v2.4.9-ac9/include/linux/types.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/types.h Mon Sep 17 22:27:32 2001 +++ aio-v2.4.9-ac9/include/linux/types.h Sun Sep 23 18:59:38 2001 @@ -127,4 +127,9 @@ char f_fpack[6]; }; +/* kernel typedefs -- they belong here. */ +#ifdef __KERNEL__ +typedef struct kvec_cb kvec_cb_t; +#endif /* __KERNEL__ */ + #endif /* _LINUX_TYPES_H */ diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/wait.h aio-v2.4.9-ac9/include/linux/wait.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/wait.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/linux/wait.h Sun Sep 23 19:01:36 2001 @@ -28,17 +28,20 @@ #define WAITQUEUE_DEBUG 0 #endif +typedef struct __wait_queue wait_queue_t; +typedef void (*wait_queue_func_t)(wait_queue_t *wait); + struct __wait_queue { unsigned int flags; #define WQ_FLAG_EXCLUSIVE 0x01 struct task_struct * task; struct list_head task_list; + wait_queue_func_t func; #if WAITQUEUE_DEBUG long __magic; long __waker; #endif }; -typedef struct __wait_queue wait_queue_t; /* * 'dual' spinlock architecture. Can be switched between spinlock_t and @@ -137,6 +140,7 @@ #endif #define __WAITQUEUE_INITIALIZER(name, tsk) { \ + func: NULL, \ task: tsk, \ task_list: { NULL, NULL }, \ __WAITQUEUE_DEBUG_INIT(name)} @@ -174,6 +178,22 @@ #endif q->flags = 0; q->task = p; + q->func = NULL; +#if WAITQUEUE_DEBUG + q->__magic = (long)&q->__magic; +#endif +} + +static inline void init_waitqueue_func_entry(wait_queue_t *q, + wait_queue_func_t func) +{ +#if WAITQUEUE_DEBUG + if (!q || !p) + WQ_BUG(); +#endif + q->flags = 0; + q->task = NULL; + q->func = func; #if WAITQUEUE_DEBUG q->__magic = (long)&q->__magic; #endif @@ -231,6 +251,19 @@ list_del(&old->task_list); } +#define add_wait_queue_cond(q, wait, cond, fail) \ + do { \ + unsigned long flags; \ + wq_write_lock_irqsave(&(q)->lock, flags); \ + (wait)->flags = 0; \ + if (cond) \ + __add_wait_queue((q), (wait)); \ + else { \ + fail; \ + } \ + wq_write_unlock_irqrestore(&(q)->lock, flags); \ + } while (0) + #endif /* __KERNEL__ */ #endif diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/linux/worktodo.h aio-v2.4.9-ac9/include/linux/worktodo.h --- /md0/kernels/2.4/v2.4.9-ac9/include/linux/worktodo.h Wed Dec 31 19:00:00 1969 +++ aio-v2.4.9-ac9/include/linux/worktodo.h Sun Sep 23 19:01:36 2001 @@ -0,0 +1,39 @@ +#ifndef __LINUX__WORKTODO_H +#define __LINUX__WORKTODO_H + +#ifndef _LINUX_WAIT_H +#include +#endif +#ifndef _LINUX_TQUEUE_H +#include +#endif + +struct worktodo { + wait_queue_t wait; + struct tq_struct tq; + + void *data; /* for use by the wtd_ primatives */ +}; + +/* FIXME NOTE: factor from kernel/context.c */ +#define wtd_queue(wtd) schedule_task(&(wtd)->tq) + +#define wtd_set_action(wtd, action, wtddata) \ + do { \ + (wtd)->tq.routine = (action); \ + (wtd)->tq.data = (wtddata); \ + } while (0) + +struct page; +extern void wtd_wait_page(struct worktodo *wtd, struct page *page); +extern void wtd_lock_page(struct worktodo *wtd, struct page *page); +struct buffer_head; +extern void wtd_wait_on_buffer(struct worktodo *wtd, struct buffer_head *bh); + +#if 0 /* not implemented yet */ +extern void wtd_down(struct worktodo *wtd, struct semaphore *sem); +extern void wtd_down_write(struct worktodo *wtd, struct rw_semaphore *sem); +extern void wtd_down_read(struct worktodo *wtd, struct rw_semaphore *sem); +#endif + +#endif /* __LINUX__WORKTODO_H */ diff -urN /md0/kernels/2.4/v2.4.9-ac9/include/net/sock.h aio-v2.4.9-ac9/include/net/sock.h --- /md0/kernels/2.4/v2.4.9-ac9/include/net/sock.h Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/include/net/sock.h Mon Sep 24 16:11:47 2001 @@ -692,6 +692,10 @@ (__skb)->next = NULL; \ } while(0) +struct kioctx; +struct iocb; +#include /* FIXME */ + /* IP protocol blocks we attach to sockets. * socket layer -> transport layer interface * transport -> network interface is defined by struct inet_proto @@ -721,6 +725,8 @@ int (*recvmsg)(struct sock *sk, struct msghdr *msg, int len, int noblock, int flags, int *addr_len); + int (*begin_read)(struct sock *, struct kioctx *, + struct iocb, struct iocb *); int (*bind)(struct sock *sk, struct sockaddr *uaddr, int addr_len); diff -urN /md0/kernels/2.4/v2.4.9-ac9/kernel/context.c aio-v2.4.9-ac9/kernel/context.c --- /md0/kernels/2.4/v2.4.9-ac9/kernel/context.c Fri May 25 22:48:10 2001 +++ aio-v2.4.9-ac9/kernel/context.c Fri Sep 7 00:15:17 2001 @@ -91,12 +91,18 @@ */ for (;;) { set_task_state(curtask, TASK_INTERRUPTIBLE); - add_wait_queue(&context_task_wq, &wait); - if (TQ_ACTIVE(tq_context)) + add_wait_queue_exclusive_lifo(&context_task_wq, &wait); + if (spin_is_locked(&tqueue_lock) || TQ_ACTIVE(tq_context)) set_task_state(curtask, TASK_RUNNING); - schedule(); + else + schedule(); remove_wait_queue(&context_task_wq, &wait); run_task_queue(&tq_context); + while (TQ_ACTIVE(tq_context)) { + if (current->need_resched) + schedule(); + run_task_queue(&tq_context); + } wake_up(&context_task_done); if (signal_pending(curtask)) { while (waitpid(-1, (unsigned int *)0, __WALL|WNOHANG) > 0) diff -urN /md0/kernels/2.4/v2.4.9-ac9/kernel/fork.c aio-v2.4.9-ac9/kernel/fork.c --- /md0/kernels/2.4/v2.4.9-ac9/kernel/fork.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/kernel/fork.c Fri Sep 7 00:15:17 2001 @@ -45,6 +45,16 @@ wq_write_unlock_irqrestore(&q->lock, flags); } +void add_wait_queue_exclusive_lifo(wait_queue_head_t *q, wait_queue_t * wait) +{ + unsigned long flags; + + wq_write_lock_irqsave(&q->lock, flags); + wait->flags = WQ_FLAG_EXCLUSIVE; + __add_wait_queue(q, wait); + wq_write_unlock_irqrestore(&q->lock, flags); +} + void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait) { unsigned long flags; diff -urN /md0/kernels/2.4/v2.4.9-ac9/kernel/sched.c aio-v2.4.9-ac9/kernel/sched.c --- /md0/kernels/2.4/v2.4.9-ac9/kernel/sched.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/kernel/sched.c Fri Sep 7 00:15:17 2001 @@ -714,13 +714,13 @@ } /* - * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just wake everything - * up. If it's an exclusive wakeup (nr_exclusive == small +ve number) then we wake all the - * non-exclusive tasks and one exclusive task. + * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just + * wake everything up. If it's an exclusive wakeup (nr_exclusive == small + * +ve number) then we wake all the non-exclusive tasks and one exclusive task. * * There are circumstances in which we can try to wake a task which has already - * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns zero - * in this (rare) case, and we handle it by contonuing to scan the queue. + * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns + * zero in this (rare) case, and we handle it by contonuing to scan the queue. */ static inline void __wake_up_common (wait_queue_head_t *q, unsigned int mode, int nr_exclusive, const int sync) @@ -733,14 +733,25 @@ list_for_each(tmp,&q->task_list) { unsigned int state; - wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); + wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); + wait_queue_func_t func; CHECK_MAGIC(curr->__magic); + func = curr->func; + if (func) { + unsigned flags = curr->flags; + func(curr); + if ((flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) + break; + continue; + } p = curr->task; state = p->state; if (state & mode) { WQ_NOTE_WAKER(curr); - if (try_to_wake_up(p, sync) && (curr->flags&WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) + if (try_to_wake_up(p, sync) && + (curr->flags & WQ_FLAG_EXCLUSIVE) && + !--nr_exclusive) break; } } diff -urN /md0/kernels/2.4/v2.4.9-ac9/kernel/softirq.c aio-v2.4.9-ac9/kernel/softirq.c --- /md0/kernels/2.4/v2.4.9-ac9/kernel/softirq.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/kernel/softirq.c Fri Sep 7 00:15:17 2001 @@ -354,6 +354,7 @@ data = p->data; wmb(); p->sync = 0; + smp_mb(); if (f) f(data); } diff -urN /md0/kernels/2.4/v2.4.9-ac9/mm/filemap.c aio-v2.4.9-ac9/mm/filemap.c --- /md0/kernels/2.4/v2.4.9-ac9/mm/filemap.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/mm/filemap.c Fri Sep 7 00:15:17 2001 @@ -22,12 +22,14 @@ #include #include #include +#include #include #include #include #include +#include /* * Shared mappings implemented 30.11.1994. It's not fully working yet, @@ -2762,3 +2764,729 @@ panic("Failed to allocate page hash table\n"); memset((void *)page_hash_table, 0, PAGE_HASH_SIZE * sizeof(struct page *)); } + +/* address_space_map + * Maps a series of pages from the page cache into the given array. + */ +static int address_space_map(struct address_space *as, unsigned long index, + int nr, struct page **pages, + int *nr_newp, struct page **new_pages) +{ + struct page *cached_page = NULL; + int nr_new = 0; + int ret; + + ret = -EINVAL; + if (nr <= 0) + goto out; + + ret = 0; + + spin_lock(&pagecache_lock); + + while (nr > 0) { + struct page **hash = page_hash(as, index); + struct page *page; + + page = __find_page_nolock(as, index, *hash); + if (page) { + page_cache_get(page); +got_page: + pages[ret++] = page; + index++; + nr--; + continue; + } + + if (cached_page) { + __add_to_page_cache(cached_page, as, index, hash); + nr_new++; + *new_pages++ = page = cached_page; + cached_page = NULL; + goto got_page; + } + spin_unlock(&pagecache_lock); + + cached_page = page_cache_alloc(as); + if (!cached_page) + goto out; + + /* Okay, we now have an allocated page. Retry + * the search and add. */ + spin_lock(&pagecache_lock); + } + + spin_unlock(&pagecache_lock); + +out: + if (cached_page) + page_cache_free(cached_page); + + *nr_newp = nr_new; + return ret ? ret : -ENOMEM; +} + +struct iodesc { + struct worktodo wtd; + + struct page *good_page; /* the highest Uptodate page */ + int good_idx; + int err; + int did_read; + int rw; + + struct page **pages; + struct page **new_pages; + struct page **cur_pagep; + struct page **src_pagep; + int nr_pages; + int nr_new_pages; + + struct address_space *as; + struct file *file; + struct kiobuf *kiovec[8]; + int kio_nr; + + size_t size; + unsigned long transferred; + unsigned offset; + unsigned src_offset; + struct kiobuf *iobuf; + + int sync; + +#define READDESC_NR_DEF 3 + struct page *def_pages[READDESC_NR_DEF]; + struct page *def_new_pages[READDESC_NR_DEF]; +}; + +static void __iodesc_free(struct iodesc *io) +{ + int i; + + for (i=0; inr_pages; i++) + page_cache_release(io->pages[i]); + + if (io->new_pages != io->def_new_pages) + kfree(io->new_pages); + if (io->pages != io->def_pages) + kfree(io->pages); + kfree(io); +} + +/* By the time this function is called, all of the pages prior to + * the current good_idx have been released appropriately. The remaining + * duties are to release any remaining pages and to honour O_SYNC. + */ +static void __iodesc_finish_write(struct iodesc *io) +{ + int i; + + pr_debug("__iodesc_finish_write(%p)\n", io); + + if (WRITE == io->rw) + for (i=0; inr_pages; i++) { + struct page *page = io->pages[i]; + UnlockPage(page); + deactivate_page(page); + //page_cache_release(page); + } + + /* FIXME: this is buggy */ + { + struct kiobuf *iobuf = io->kiovec[0]; + iobuf->transferred = io->transferred; + iobuf->errno = io->err; + iobuf->end_io(iobuf); + } + + __iodesc_free(io); +} + +/* This is mostly ripped from generic_file_write */ +static int __iodesc_write_page(struct iodesc *io, struct page *page) +{ + unsigned long bytes; + unsigned long offset, src_offset; + struct page *src_page; + long status; + char *kaddr; + int src_bytes; + char *src; + int done = 0; + unsigned left; + + src_bytes = PAGE_CACHE_SIZE - io->src_offset; + src_page = *io->src_pagep; + src = kmap(src_page) + io->src_offset; + + offset = io->offset; + src_offset = io->src_offset; + kaddr = kmap(page); + kaddr += offset; + + bytes = PAGE_CACHE_SIZE - offset; + if (io->size < bytes) + bytes = io->size; + + pr_debug("__iodesc_write_page(%p (%lu), %lu %lu %lu)\n", page, page->index, offset, bytes, src_offset); + + io->err = io->as->a_ops->prepare_write(io->file, page, + offset, offset + bytes); + if (io->err) { +printk("prepare_write: %d\n", io->err); + goto unlock; + } + + left = bytes; + for (;;) { + if (left < src_bytes) + src_bytes = left; + + memcpy(kaddr, src, src_bytes); + kaddr += src_bytes; + src += src_bytes; + left -= src_bytes; + src_offset += src_bytes; + src_offset &= PAGE_SIZE - 1; + if (!src_offset) + io->src_pagep++; + + if (left <= 0) + break; + + if (!src_offset) { + kunmap(src_page); + src_page = *io->src_pagep; + src = kmap(src_page); + src_bytes = PAGE_SIZE; + } + } + flush_dcache_page(page); + status = io->as->a_ops->commit_write(io->file, page, + offset, offset+bytes); + + /* We don't handle short writes */ + if (status > 0 && status != bytes) + done = 1; + + if (!status) + status = bytes; +else +printk("commit_write: %ld\n", status); + + if (status > 0) { + io->transferred += status; + io->size -= status; + io->offset = (offset + status) & (PAGE_CACHE_SIZE - 1); + + if (io->offset) + done = 1; + + io->src_offset += status; + io->src_offset &= PAGE_CACHE_SIZE - 1; + } else { + io->err = status; + done = 1; + } + +unlock: + kunmap(page); + kunmap(src_page); + + //UnlockPage(page); + //deactivate_page(page); + //page_cache_release(page); + + return done; +} + +void __iodesc_sync_wait_page(void *data) +{ + struct iodesc *io = data; + + do { + struct buffer_head *bh, *head = io->pages[io->good_idx]->buffers; + + if (!head) + continue; + + bh = head; + do { + if (buffer_locked(bh)) { +//printk("waiting on bh=%pi io=%p\n", bh, io); + wtd_wait_on_buffer(&io->wtd, bh); + return; + } + if (buffer_req(bh) && !buffer_uptodate(bh)) { +//printk("io err bh=%p (%p)\n", bh, io); + io->err = -EIO; + break; + } + } while ((bh = bh->b_this_page) != head); + } while (!io->err && ++io->good_idx < io->nr_pages) ; + +//printk("finish_write(%p)\n", io); + __iodesc_finish_write(io); +} + +static void __iodesc_do_write(void *data) +{ + struct iodesc *io = data; + unsigned i; + + up(&io->file->f_dentry->d_inode->i_sem); + + for (i=0; inr_pages; i++) + if (__iodesc_write_page(io, io->pages[i])) + break; + + if (io->sync) { + io->good_idx = 0; + +//printk("writing out pages(%p)\n", io); + for (i=0; inr_pages; i++) { + if (io->pages[i]->buffers) + writeout_one_page(io->pages[i]); + } + +//printk("calling __iodesc_sync_wait_page(%p)\n", io); + wtd_set_action(&io->wtd, __iodesc_sync_wait_page, io); + __iodesc_sync_wait_page(io); + return; + } + + __iodesc_finish_write(io); +} + +static void __iodesc_write_lock_next_page(void *data) +{ + struct iodesc *io = data; + pr_debug("__iodesc_write_next_page(%p)\n", io); + + while (io->good_idx < io->nr_pages) { + io->good_page = io->pages[io->good_idx++]; + if (io->good_page == *io->cur_pagep) + io->cur_pagep++; + else { + wtd_lock_page(&io->wtd, io->good_page); + return; + } + } + + //__iodesc_do_write(io); + wtd_set_action(&io->wtd, __iodesc_do_write, io); + wtd_queue(&io->wtd); +} + +static +void __generic_file_write_iodesc(struct iodesc *io) +{ + struct inode *inode = io->file->f_dentry->d_inode; + time_t now = CURRENT_TIME; + + remove_suid(inode); + if (inode->i_ctime != now || inode->i_mtime != now) { + inode->i_ctime = inode->i_mtime = now; + mark_inode_dirty_sync(inode); + } + + wtd_set_action(&io->wtd, __iodesc_write_lock_next_page, io); + io->sync = !!(io->file->f_flags & O_SYNC); + io->good_idx = 0; + io->cur_pagep = io->new_pages; + io->src_offset = io->kiovec[0]->offset; + io->src_pagep = io->kiovec[0]->maplist; + __iodesc_write_lock_next_page(io); +} + +static void __iodesc_read_finish(struct iodesc *io) +{ + char *dst_addr, *src_addr; + int src_off, i; + size_t size; + size_t valid; + + struct page **src_pagep; + + pr_debug("__iodesc_read_finish: good_idx = %d\n", io->good_idx); + if (io->good_idx <= 0) + goto no_data; + + size = io->size; + src_off = io->offset; + src_pagep = io->pages; + src_addr = kmap(*src_pagep); + + valid = (size_t)io->good_idx << PAGE_CACHE_SHIFT; + valid -= src_off; + pr_debug("size=%d valid=%d src_off=%d\n", size, valid, src_off); + + if (valid < size) + size = valid; + + for (i=0; ikio_nr; i++) { + struct kiobuf *iobuf = io->kiovec[i]; + int dst_len = iobuf->length; + int dst_off = iobuf->offset; + struct page **dst_pagep = iobuf->maplist; + + dst_addr = kmap(*dst_pagep); + iobuf->transferred = 0; + + while (size > 0) { + int this = PAGE_CACHE_SIZE - src_off; + if ((PAGE_SIZE - dst_off) < this) + this = PAGE_SIZE - dst_off; + if (size < this) + this = size; + pr_debug("this=%d src_off=%d dst_off=%d dst_len=%d\n", + this, src_off, dst_off, dst_len); + memcpy(dst_addr + dst_off, src_addr + src_off, this); + + src_off += this; + dst_off += this; + dst_len -= this; + size -= this; + iobuf->transferred += this; + pr_debug("read_finish: this=%d transferred=%d\n", this, iobuf->transferred); + + if (dst_len <= 0) + break; + + if (size <= 0) + break; + + if (dst_off >= PAGE_SIZE) { + kunmap(*dst_pagep); + dst_pagep++; + dst_addr = kmap(*dst_pagep); + dst_off = 0; + } + + if (src_off >= PAGE_SIZE) { /* FIXME: PAGE_CACHE_SIZE */ + kunmap(*src_pagep); +pr_debug("page(%lu)->count = %d\n", (*src_pagep)->index, atomic_read(&(*src_pagep)->count)); + src_pagep++; + src_addr = kmap(*src_pagep); + src_off = 0; + } + } + kunmap(*dst_pagep); + + iobuf->errno = iobuf->transferred ? 0 : io->err; + if (iobuf->errno && i) + iobuf->errno = -EAGAIN; + iobuf->end_io(iobuf); + } + + kunmap(*src_pagep); + __iodesc_free(io); + + return; + +no_data: + io->kiovec[0]->errno = io->err; + io->kiovec[0]->transferred = 0; + io->kiovec[0]->end_io(io->kiovec[0]); + + for (i=1; ikio_nr; i++) { + struct kiobuf *iobuf = io->kiovec[i]; + + iobuf->errno = -EAGAIN; + iobuf->transferred = 0; + iobuf->end_io(iobuf); + } + __iodesc_free(io); +} + +static void __iodesc_make_uptodate(void *data) +{ + struct iodesc *io = data; + struct page *page = io->good_page; + int locked = 1; + + pr_debug("__iodesc_make_uptodate: io=%p index=%lu\n", io, page->index); + while (Page_Uptodate(page)) { +again: + pr_debug("page index %lu uptodate\n", page->index); + if (locked) { + UnlockPage(page); + locked = 0; + } + io->did_read = 0; + io->good_idx++; + if (io->good_idx >= io->nr_pages) { + __iodesc_read_finish(io); + return; + } + page = io->good_page = io->pages[io->good_idx]; + pr_debug("__iodesc_make_uptodate: index=%lu\n", page->index); + } + + if (!locked) { + wtd_lock_page(&io->wtd, page); + return; + } + + if (!io->did_read) { + /* We haven't tried reading this page before, give it a go. */ + printk("attempting to read %lu\n", page->index); + io->did_read = 1; + io->err = page->mapping->a_ops->readpage(io->file, page); + if (!io->err) { + if (Page_Uptodate(page)) + goto again; + wtd_lock_page(&io->wtd, page); + return; + } + } + + if (locked) + UnlockPage(page); + + /* We've already read this page before. Set err to EIO and quite */ + if (!io->err) + io->err = -EIO; + __iodesc_read_finish(io); +} + +static void __wtdgeneric_file_read_iodesc(void *data); + +static void __generic_file_read_iodesc(struct iodesc *io, int mayblock) +{ + int (*readpage)(struct file *, struct page *); + int i; + + wtd_set_action(&io->wtd, __iodesc_make_uptodate, io); + readpage = io->as->a_ops->readpage; + for (i=0; inr_new_pages; i++) { + int foo; + if (!mayblock) + goto do_wtd; + foo = readpage(io->file, io->new_pages[i]); + if (foo) + printk(KERN_DEBUG "__generic_file_read_kiovec: readpage(%lu) = %d\n", io->new_pages[i]->index, foo); + } + + for (i=0; inr_pages; i++) { + struct page *page = io->pages[i]; + if (Page_Uptodate(page)) { + pr_debug("__generic_file_read_iodesc: %lu is uptodate\n", page->index); + continue; + } + + if (!mayblock) + goto do_wtd; + if (!TryLockPage(page)) { + int foo = readpage(io->file, page); + if (foo) + printk(KERN_DEBUG "__generic_file_read_iodesc: readpage(%lu): %d\n", page->index, foo); + } + + if (!Page_Uptodate(page) && io->good_idx == -1) { + pr_debug("first good_idx=%d (%lu)\n", i, page->index); + io->good_idx = i; + io->good_page = page; + } + } + + /* Whee, all the pages are uptodate! */ + if (!io->good_page) { + do {static int zoo; if (!mayblock && zoo++ < 5) printk("all uptodate\n");} while(0); + pr_debug("all pages uptodate!\n"); + io->good_idx = io->nr_pages; + __iodesc_read_finish(io); + return; + } + + pr_debug("locking good_page\n"); + wtd_lock_page(&io->wtd, io->good_page); + return; + +do_wtd: + do {static int zoo; if (zoo++ < 5) printk("read sleep\n");} while(0); + wtd_set_action(&io->wtd, __wtdgeneric_file_read_iodesc, io); + wtd_queue(&io->wtd); +} + +static void __wtdgeneric_file_read_iodesc(void *data) +{ + struct iodesc *io = data; + __generic_file_read_iodesc(io, 1); +} + +int generic_file_rw_kiovec(struct file *file, int rw, + int kio_nr, struct kiobuf **kiovec, int flags, size_t size, loff_t pos) +{ + struct inode *inode = file->f_dentry->d_inode; + struct address_space *as = inode->i_mapping; + unsigned long index; + unsigned long eindex; + unsigned long nr_pages; + struct iodesc *io = NULL; + int ret; + + ret = -EINVAL; + if (rw != READ && rw != WRITE) + goto out; + + ret = -ENOMEM; + io = kmalloc(sizeof(*io), GFP_KERNEL); + if (!io) + goto out; + + memset(io, 0, sizeof(*io)); + io->size = size; + + if (READ == rw) { + pr_debug("pos=%Ld i_size=%Ld\n", pos, inode->i_size); + + if (pos > inode->i_size) + size = 0; + else if ((pos + size) > inode->i_size) + size = inode->i_size - pos; + + if (io->size < size) + size = io->size; + else if (size < io->size) + io->size = size; + + pr_debug("io->size=%d size=%d\n", io->size, size); + } + + index = pos >> PAGE_CACHE_SHIFT; + eindex = (pos + size - 1) >> PAGE_CACHE_SHIFT; + nr_pages = eindex - index + 1; + + pr_debug("nr_pages: %lu\n", nr_pages); + + io->good_idx = -1; + io->good_page = NULL; + io->did_read = 0; + io->err = 0; + io->rw = rw; + io->as = as; + io->offset = (unsigned long)pos & (PAGE_CACHE_SIZE - 1); + io->file = file; + io->kio_nr = kio_nr; + if (kio_nr > 8) + BUG(); + memcpy(io->kiovec, kiovec, sizeof(struct kiobuf *) * kio_nr); + if (nr_pages < READDESC_NR_DEF) { + io->pages = io->def_pages; + io->new_pages = io->def_new_pages; + } else { + io->pages = kmalloc(sizeof(*io->pages) * (nr_pages + 1), GFP_KERNEL); + if (!io->pages) + goto out_io; + + io->new_pages = kmalloc(sizeof(*io->new_pages) * (nr_pages + 1), GFP_KERNEL); + if (!io->new_pages) + goto out_pages; + } + + /* FIXME: make the down a WTD_op */ + if (rw == WRITE) + down(&io->file->f_dentry->d_inode->i_sem); + + ret = address_space_map(as, index, nr_pages, io->pages, + &io->nr_new_pages, io->new_pages); + pr_debug("as_map: %d (%d new)\n", ret, io->nr_new_pages); + if (ret <= 0) + goto out_new_pages; + + io->nr_pages = ret; + io->pages[io->nr_pages] = NULL; + io->new_pages[io->nr_new_pages] = NULL; + + if (rw == READ) + __generic_file_read_iodesc(io, 0); + else if (rw == WRITE) + __generic_file_write_iodesc(io); + + return 0; + +out_new_pages: + if (io->new_pages != io->def_new_pages) + kfree(io->new_pages); +out_pages: + if (io->pages != io->def_pages) + kfree(io->pages); +out_io: + kfree(io); +out: + return ret; +} + +static void __wtd_lock_page_waiter(wait_queue_t *wait) +{ + struct worktodo *wtd = (struct worktodo *)wait; + struct page *page = (struct page *)wtd->data; + + if (!TryLockPage(page)) { + __remove_wait_queue(&page->wait, &wtd->wait); + wtd_queue(wtd); + } else { + schedule_task(&run_disk_tq); + } +} + +void wtd_lock_page(struct worktodo *wtd, struct page *page) +{ + if (TryLockPage(page)) { + int raced = 0; + wtd->data = page; + init_waitqueue_func_entry(&wtd->wait, __wtd_lock_page_waiter); + add_wait_queue_cond(&page->wait, &wtd->wait, TryLockPage(page), raced = 1); + + if (!raced) { + run_task_queue(&tq_disk); + return; + } + } + + wtd->tq.routine(wtd->tq.data); +} + +static void __wtd_bh_waiter(wait_queue_t *wait) +{ + struct worktodo *wtd = (struct worktodo *)wait; + struct buffer_head *bh = (struct buffer_head *)wtd->data; + + if (!buffer_locked(bh)) { + __remove_wait_queue(&bh->b_wait, &wtd->wait); + wtd_queue(wtd); + } else { + schedule_task(&run_disk_tq); + } +} + +void wtd_wait_on_buffer(struct worktodo *wtd, struct buffer_head *bh) +{ + int raced = 0; + + if (!buffer_locked(bh)) { + wtd->tq.routine(wtd->tq.data); + return; + } + wtd->data = bh; + init_waitqueue_func_entry(&wtd->wait, __wtd_bh_waiter); + add_wait_queue_cond(&bh->b_wait, &wtd->wait, buffer_locked(bh), raced = 1); + + if (raced) + wtd->tq.routine(wtd->tq.data); + else + run_task_queue(&tq_disk); +} + +void do_run_tq_disk(void *data) +{ + run_task_queue(&tq_disk); +} + +struct tq_struct run_disk_tq = { + routine: do_run_tq_disk, + data: NULL +}; + diff -urN /md0/kernels/2.4/v2.4.9-ac9/mm/memory.c aio-v2.4.9-ac9/mm/memory.c --- /md0/kernels/2.4/v2.4.9-ac9/mm/memory.c Mon Sep 10 20:35:52 2001 +++ aio-v2.4.9-ac9/mm/memory.c Mon Sep 24 00:05:02 2001 @@ -44,6 +44,7 @@ #include #include #include +#include #include #include @@ -1474,3 +1475,129 @@ } while (addr < end); return 0; } + +/* + * Force in an entire range of pages from the current process's user VA, + * and pin them in physical memory. + */ + +#define dprintk(x...) +struct kvec *map_user_kvec(int rw, unsigned long ptr, size_t len) +{ + struct kvec *vec; + struct kveclet *veclet; + unsigned long end; + int err; + struct mm_struct * mm; + struct vm_area_struct * vma = 0; + int i; + int datain = (rw == READ); + unsigned nr_pages; + + end = ptr + len; + if (end < ptr) + return ERR_PTR(-EINVAL); + + nr_pages = (ptr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; + nr_pages -= ptr >> PAGE_SHIFT; + vec = kmalloc(sizeof(struct kvec) + nr_pages * sizeof(struct kveclet), + GFP_KERNEL); + if (!vec) + return ERR_PTR(-ENOMEM); + vec->max_nr = nr_pages; + veclet = vec->veclet; + + /* Make sure the iobuf is not already mapped somewhere. */ + mm = current->mm; + dprintk ("map_user_kiobuf: begin\n"); + + down_read(&mm->mmap_sem); + + err = -EFAULT; + + i = 0; + + /* + * First of all, try to fault in all of the necessary pages + */ + while (ptr < end) { + struct page *map; + veclet->offset = ptr & ~PAGE_MASK; + veclet->length = PAGE_SIZE - veclet->offset; + ptr &= PAGE_MASK; + + if (!vma || ptr >= vma->vm_end) { + vma = find_vma(current->mm, ptr); + if (!vma) + goto out_unlock; + if (vma->vm_start > ptr) { + if (!(vma->vm_flags & VM_GROWSDOWN)) + goto out_unlock; + if (expand_stack(vma, ptr)) + goto out_unlock; + } + if (((datain) && (!(vma->vm_flags & VM_WRITE))) || + (!(vma->vm_flags & VM_READ))) { + err = -EACCES; + goto out_unlock; + } + } + spin_lock(&mm->page_table_lock); + while (!(map = follow_page(ptr, datain))) { + int ret; + + spin_unlock(&mm->page_table_lock); + ret = handle_mm_fault(current->mm, vma, ptr, datain); + if (ret <= 0) { + if (!ret) + goto out_unlock; + else { + err = -ENOMEM; + goto out_unlock; + } + } + spin_lock(&mm->page_table_lock); + } + map = get_page_map(map); + if (map) { + flush_dcache_page(map); + atomic_inc(&map->count); + } else + printk (KERN_INFO "Mapped page missing [%d]\n", i); + spin_unlock(&mm->page_table_lock); + veclet->page = map; + veclet++; + + ptr += PAGE_SIZE; + vec->nr = i++; + } + + up_read(&mm->mmap_sem); + dprintk ("map_user_kiobuf: end OK\n"); + return 0; + + out_unlock: + up_read(&mm->mmap_sem); + unmap_kvec(vec); + dprintk ("map_user_kiobuf: end %d\n", err); + kfree(vec); + return ERR_PTR(err); +} + +/* + * Unmap all of the pages referenced by a kiobuf. We release the pages, + * and unlock them if they were locked. + */ + +void unmap_kvec (struct kvec *vec) +{ + struct kveclet *veclet, *end = vec->veclet + vec->nr; + + for (veclet=vec->veclet; vecletpage; + if (map) + __free_page(map); + } + + vec->nr = 0; +} diff -urN /md0/kernels/2.4/v2.4.9-ac9/net/ipv4/af_inet.c aio-v2.4.9-ac9/net/ipv4/af_inet.c --- /md0/kernels/2.4/v2.4.9-ac9/net/ipv4/af_inet.c Mon Aug 13 15:12:09 2001 +++ aio-v2.4.9-ac9/net/ipv4/af_inet.c Thu Sep 20 04:22:29 2001 @@ -732,6 +732,13 @@ } +int inet_begin_read(struct socket *sock, struct kioctx *ctx, + struct iocb iocb, struct iocb *iocbptr) +{ + struct sock *sk = sock->sk; + + return sk->prot->begin_read(sk, ctx, iocb, iocbptr); +} int inet_recvmsg(struct socket *sock, struct msghdr *msg, int size, int flags, struct scm_cookie *scm) @@ -973,6 +980,7 @@ recvmsg: inet_recvmsg, mmap: sock_no_mmap, sendpage: sock_no_sendpage, + begin_read: inet_begin_read, }; struct net_proto_family inet_family_ops = { diff -urN /md0/kernels/2.4/v2.4.9-ac9/net/ipv4/udp.c aio-v2.4.9-ac9/net/ipv4/udp.c --- /md0/kernels/2.4/v2.4.9-ac9/net/ipv4/udp.c Thu May 3 11:22:21 2001 +++ aio-v2.4.9-ac9/net/ipv4/udp.c Thu Sep 20 17:59:24 2001 @@ -619,6 +619,15 @@ __udp_checksum_complete(skb); } + +static int udp_begin_read(struct sock *sk, struct kioctx *ctx, + struct iocb iocb, struct iocb *iocbptr) +{ + struct sk_buff *skb; + printk("udp_begin_read\n"); +} + + /* * This should be easy, if there is something there we * return it, otherwise we block. @@ -1015,6 +1024,7 @@ getsockopt: ip_getsockopt, sendmsg: udp_sendmsg, recvmsg: udp_recvmsg, + begin_read: udp_begin_read, backlog_rcv: udp_queue_rcv_skb, hash: udp_v4_hash, unhash: udp_v4_unhash, diff -urN /md0/kernels/2.4/v2.4.9-ac9/net/socket.c aio-v2.4.9-ac9/net/socket.c --- /md0/kernels/2.4/v2.4.9-ac9/net/socket.c Mon Sep 10 20:35:53 2001 +++ aio-v2.4.9-ac9/net/socket.c Sun Sep 23 00:54:06 2001 @@ -107,6 +107,8 @@ unsigned long count, loff_t *ppos); static ssize_t sock_sendpage(struct file *file, struct page *page, int offset, size_t size, loff_t *ppos, int more); +static int sock_begin_read(struct file *file, struct kioctx *ctx, + struct iocb iocb, struct iocb *iocbptr); /* @@ -126,6 +128,7 @@ fasync: sock_fasync, readv: sock_readv, writev: sock_writev, + //begin_read: sock_begin_read, sendpage: sock_sendpage }; @@ -623,6 +626,16 @@ return sock->ops->sendpage(sock, page, offset, size, flags); } +static int sock_begin_read(struct file *file, struct kioctx *ctx, + struct iocb iocb, struct iocb *iocbptr) +{ + struct socket *sock; + sock = socki_lookup(file->f_dentry->d_inode); + if (sock->ops->begin_read) + sock->ops->begin_read(sock, ctx, iocb, iocbptr); + return -EINVAL; +} + int sock_readv_writev(int type, struct inode * inode, struct file * file, const struct iovec * iov, long count, long size) {