[apple/xnu.git] / bsd / kern / kern_symfile.c

/*
 * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 * 
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 * 
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 * 
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 * 
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */
/* Copyright (c) 1998 Apple Computer, Inc.  All rights reserved.
 *
 *	File:	bsd/kern/kern_symfile.c
 *
 * HISTORY
 */

#include <mach/vm_param.h>

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/signalvar.h>
#include <sys/resourcevar.h>
#include <sys/namei.h>
#include <sys/vnode_internal.h>
#include <sys/proc_internal.h>
#include <sys/kauth.h>
#include <sys/timeb.h>
#include <sys/times.h>
#include <sys/acct.h>
#include <sys/file_internal.h>
#include <sys/uio.h>
#include <sys/kernel.h>
#include <sys/stat.h>
#include <sys/disk.h>
#include <sys/conf.h>

#include <mach-o/loader.h>
#include <mach-o/nlist.h>

#include <kern/kalloc.h>
#include <vm/vm_kern.h>
#include <pexpert/pexpert.h>
#include <IOKit/IOHibernatePrivate.h>

/* This function is called from kern_sysctl in the current process context;
 * it is exported with the System6.0.exports, but this appears to be a legacy
 * export, as there are no internal consumers.
 */
int
get_kernel_symfile(__unused proc_t p, __unused char const **symfile);
int
get_kernel_symfile(__unused proc_t p, __unused char const **symfile)
{
    return KERN_FAILURE;
}

struct kern_direct_file_io_ref_t
{
    vfs_context_t  ctx;
    struct vnode * vp;
    dev_t          device;
    uint32_t	   blksize;
    off_t          filelength;
    char           pinned;
};


static int file_ioctl(void * p1, void * p2, u_long theIoctl, caddr_t result)
{
    dev_t device = *(dev_t*) p1;

    return ((*bdevsw[major(device)].d_ioctl)
		    (device, theIoctl, result, S_IFBLK, p2));
}

static int device_ioctl(void * p1, __unused void * p2, u_long theIoctl, caddr_t result)
{
    return (VNOP_IOCTL(p1, theIoctl, result, 0, p2));
}

static int
kern_ioctl_file_extents(struct kern_direct_file_io_ref_t * ref, u_long theIoctl, off_t offset, off_t end)
{
    int error;
    int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
    void * p1;
    void * p2;
    uint64_t    fileblk;
    size_t      filechunk;
    dk_extent_t  extent;
    dk_unmap_t   unmap;
    _dk_cs_pin_t pin;

    bzero(&extent, sizeof(dk_extent_t));
    bzero(&unmap, sizeof(dk_unmap_t));
    bzero(&pin, sizeof(pin));
    if (ref->vp->v_type == VREG)
    {
	 p1 = &ref->device;
	 p2 = kernproc;
	 do_ioctl = &file_ioctl;
    }
    else
    {
	/* Partition. */
	p1 = ref->vp;
	p2 = ref->ctx;
	do_ioctl = &device_ioctl;
    }
    while (offset < end) 
    {
        if (ref->vp->v_type == VREG)
        {
            daddr64_t blkno;
	    filechunk = 1*1024*1024*1024;
	    if (filechunk > (size_t)(end - offset))
	    filechunk = (size_t)(end - offset);
            error = VNOP_BLOCKMAP(ref->vp, offset, filechunk, &blkno, &filechunk, NULL, 0, NULL);
			if (error) break;
            fileblk = blkno * ref->blksize;
        }
        else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR))
        {
            fileblk = offset;
            filechunk = ref->filelength;
        }

	if (DKIOCUNMAP == theIoctl)
	{
	    extent.offset = fileblk;
	    extent.length = filechunk;
	    unmap.extents = &extent;
	    unmap.extentsCount = 1;
	    error = do_ioctl(p1, p2, theIoctl, (caddr_t)&unmap);
// 	    printf("DKIOCUNMAP(%d) 0x%qx, 0x%qx\n", error, extent.offset, extent.length);
	}
	else if (_DKIOCCSPINEXTENT == theIoctl)
	{
	    pin.cp_extent.offset = fileblk;
	    pin.cp_extent.length = filechunk;
	    pin.cp_flags = _DKIOCSPINDISCARDDATA;
	    error = do_ioctl(p1, p2, theIoctl, (caddr_t)&pin);
	    if (error && (ENOTTY != error))
	    {
		printf("_DKIOCCSPINEXTENT(%d) 0x%qx, 0x%qx\n", 
			error, pin.cp_extent.offset, pin.cp_extent.length);
	    }
	}
	else error = EINVAL;

	if (error) break;
        offset += filechunk;
    }
    return (error);
}

int
kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, caddr_t addr, vm_size_t len);

struct kern_direct_file_io_ref_t *
kern_open_file_for_direct_io(const char * name, 
			     kern_get_file_extents_callback_t callback, 
			     void * callback_ref,
			     dev_t * partition_device_result,
			     dev_t * image_device_result,
                             uint64_t * partitionbase_result,
                             uint64_t * maxiocount_result,
                             uint32_t * oflags,
                             off_t offset,
                             caddr_t addr,
                             vm_size_t len)
{
    struct kern_direct_file_io_ref_t * ref;

    proc_t			p;
    struct vnode_attr		va;
    int				error;
    off_t			f_offset;
    uint64_t                    fileblk;
    size_t                      filechunk;
    uint64_t                    physoffset;
    dev_t			device;
    dev_t			target = 0;
    int			        isssd = 0;
    uint32_t                    flags = 0;
    uint32_t			blksize;
    off_t 			maxiocount, count;
    boolean_t                   locked = FALSE;

    int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
    void * p1 = NULL;
    void * p2 = NULL;

    error = EFAULT;

    ref = (struct kern_direct_file_io_ref_t *) kalloc(sizeof(struct kern_direct_file_io_ref_t));
    if (!ref)
    {
	error = EFAULT;
    	goto out;
    }

    bzero(ref, sizeof(*ref));
    p = kernproc;
    ref->ctx = vfs_context_create(vfs_context_current());

    if ((error = vnode_open(name, (O_CREAT | FWRITE), (0), 0, &ref->vp, ref->ctx)))
        goto out;

    if (addr && len)
    {
	if ((error = kern_write_file(ref, offset, addr, len)))
	    goto out;
    }

    VATTR_INIT(&va);
    VATTR_WANTED(&va, va_rdev);
    VATTR_WANTED(&va, va_fsid);
    VATTR_WANTED(&va, va_data_size);
    VATTR_WANTED(&va, va_nlink);
    error = EFAULT;
    if (vnode_getattr(ref->vp, &va, ref->ctx))
    	goto out;

    kprintf("vp va_rdev major %d minor %d\n", major(va.va_rdev), minor(va.va_rdev));
    kprintf("vp va_fsid major %d minor %d\n", major(va.va_fsid), minor(va.va_fsid));
    kprintf("vp size %qd\n", va.va_data_size);

    if (ref->vp->v_type == VREG)
    {
	/* Don't dump files with links. */
	if (va.va_nlink != 1)
	    goto out;

        device = va.va_fsid;
        p1 = &device;
        p2 = p;
        do_ioctl = &file_ioctl;
    }
    else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR))
    {
	/* Partition. */
        device = va.va_rdev;

        p1 = ref->vp;
        p2 = ref->ctx;
        do_ioctl = &device_ioctl;
    }
    else
    {
	/* Don't dump to non-regular files. */
	error = EFAULT;
        goto out;
    }
    ref->device = device;

    // get block size

    error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &ref->blksize);
    if (error)
        goto out;

    if (ref->vp->v_type == VREG)
        ref->filelength = va.va_data_size;
    else
    {
        error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &fileblk);
        if (error)
            goto out;
	ref->filelength = fileblk * ref->blksize;    
    }

    // pin logical extents

    error = kern_ioctl_file_extents(ref, _DKIOCCSPINEXTENT, 0, ref->filelength);
    if (error && (ENOTTY != error)) goto out;
    ref->pinned = (error == 0);

    // generate the block list

    error = do_ioctl(p1, p2, DKIOCLOCKPHYSICALEXTENTS, NULL);
    if (error)
        goto out;
    locked = TRUE;

    f_offset = 0;
    while (f_offset < ref->filelength) 
    {
        if (ref->vp->v_type == VREG)
        {
            filechunk = 1*1024*1024*1024;
            daddr64_t blkno;

            error = VNOP_BLOCKMAP(ref->vp, f_offset, filechunk, &blkno, &filechunk, NULL, 0, NULL);
            if (error)
                goto out;

            fileblk = blkno * ref->blksize;
        }
        else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR))
        {
            fileblk = f_offset;
            filechunk = f_offset ? 0 : ref->filelength;
        }

        physoffset = 0;
        while (physoffset < filechunk)
        {
            dk_physical_extent_t getphysreq;
            bzero(&getphysreq, sizeof(getphysreq));

            getphysreq.offset = fileblk + physoffset;
            getphysreq.length = (filechunk - physoffset);
            error = do_ioctl(p1, p2, DKIOCGETPHYSICALEXTENT, (caddr_t) &getphysreq);
            if (error)
                goto out;
            if (!target)
            {
                target = getphysreq.dev;
            }
            else if (target != getphysreq.dev)
            {
                error = ENOTSUP;
                goto out;
            }
            callback(callback_ref, getphysreq.offset, getphysreq.length);
            physoffset += getphysreq.length;
        }
        f_offset += filechunk;
    }
    callback(callback_ref, 0ULL, 0ULL);

    if (ref->vp->v_type == VREG)
        p1 = &target;

    // get partition base

    error = do_ioctl(p1, p2, DKIOCGETBASE, (caddr_t) partitionbase_result);
    if (error)
        goto out;

    // get block size & constraints

    error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &blksize);
    if (error)
        goto out;

    maxiocount = 1*1024*1024*1024;

    error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTREAD, (caddr_t) &count);
    if (error)
        count = 0;
    count *= blksize;
    if (count && (count < maxiocount))
        maxiocount = count;

    error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTWRITE, (caddr_t) &count);
    if (error)
        count = 0;
    count *= blksize;
    if (count && (count < maxiocount))
        maxiocount = count;

    error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTREAD, (caddr_t) &count);
    if (error)
        count = 0;
    if (count && (count < maxiocount))
        maxiocount = count;

    error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTWRITE, (caddr_t) &count);
    if (error)
        count = 0;
    if (count && (count < maxiocount))
        maxiocount = count;

    error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTREAD, (caddr_t) &count);
    if (error)
        count = 0;
    if (count && (count < maxiocount))
        maxiocount = count;

    error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTWRITE, (caddr_t) &count);
    if (error)
        count = 0;
    if (count && (count < maxiocount))
        maxiocount = count;

    kprintf("max io 0x%qx bytes\n", maxiocount);
    if (maxiocount_result)
        *maxiocount_result = maxiocount;

    error = do_ioctl(p1, p2, DKIOCISSOLIDSTATE, (caddr_t)&isssd);
    if (!error && isssd)
        flags |= kIOHibernateOptionSSD;

    if (partition_device_result)
        *partition_device_result = device;
    if (image_device_result)
        *image_device_result = target;
    if (flags)
        *oflags = flags;

out:
    kprintf("kern_open_file_for_direct_io(%d)\n", error);

    if (error && locked)
    {
        p1 = &device;
        (void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
    }

    if (error && ref)
    {
	if (ref->vp)
	{
	    vnode_close(ref->vp, FWRITE, ref->ctx);
	    ref->vp = NULLVP;
	}
	vfs_context_rele(ref->ctx);
	kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
	ref = NULL;
    }
    return(ref);
}

int
kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, caddr_t addr, vm_size_t len)
{
    return (vn_rdwr(UIO_WRITE, ref->vp,
			addr, len, offset,
			UIO_SYSSPACE, IO_SYNC|IO_NODELOCKED|IO_UNIT, 
                        vfs_context_ucred(ref->ctx), (int *) 0,
			vfs_context_proc(ref->ctx)));
}


void
kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref,
			      off_t write_offset, caddr_t addr, vm_size_t write_length,
			      off_t discard_offset, off_t discard_end)
{
    int error;
    kprintf("kern_close_file_for_direct_io\n");

    if (!ref) return;

    if (ref->vp)
    {
        int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
        void * p1;
        void * p2;

        if (ref->vp->v_type == VREG)
        {
            p1 = &ref->device;
            p2 = kernproc;
            do_ioctl = &file_ioctl;
        }
        else
        {
            /* Partition. */
            p1 = ref->vp;
            p2 = ref->ctx;
            do_ioctl = &device_ioctl;
        }
        (void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
        
        if (addr && write_length)
        {
            (void) kern_write_file(ref, write_offset, addr, write_length);
        }
        if (discard_offset && discard_end && !ref->pinned)
        {
            (void) kern_ioctl_file_extents(ref, DKIOCUNMAP, discard_offset, discard_end);
        }

        error = vnode_close(ref->vp, FWRITE, ref->ctx);

        ref->vp = NULLVP;
        kprintf("vnode_close(%d)\n", error);
    }
    vfs_context_rele(ref->ctx);
    ref->ctx = NULL;
    kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
}
Commit	Line	Data
1c79356b	1	/*
2d21ac55	2	* Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
5d5c5d0d	3	*
2d21ac55	4	* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
1c79356b	5	*
2d21ac55 A	6	* This file contains Original Code and/or Modifications of Original Code
	7	* as defined in and that are subject to the Apple Public Source License
	8	* Version 2.0 (the 'License'). You may not use this file except in
	9	* compliance with the License. The rights granted to you under the License
	10	* may not be used to create, or enable the creation or redistribution of,
	11	* unlawful or unlicensed copies of an Apple operating system, or to
	12	* circumvent, violate, or enable the circumvention or violation of, any
	13	* terms of an Apple operating system software license agreement.
8f6c56a5	14	*
2d21ac55 A	15	* Please obtain a copy of the License at
	16	* http://www.opensource.apple.com/apsl/ and read it before using this file.
	17	*
	18	* The Original Code and all software distributed under the License are
	19	* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
8f6c56a5 A	20	* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
8f6c56a5 A	21	* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
2d21ac55 A	22	* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
	23	* Please see the License for the specific language governing rights and
	24	* limitations under the License.
8f6c56a5	25	*
2d21ac55	26	* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
1c79356b A	27	*/
	28	/* Copyright (c) 1998 Apple Computer, Inc. All rights reserved.
	29	*
	30	* File: bsd/kern/kern_symfile.c
	31	*
1c79356b	32	* HISTORY
1c79356b A	33	*/
	34
	35	#include <mach/vm_param.h>
	36
	37	#include <sys/param.h>
	38	#include <sys/systm.h>
	39	#include <sys/signalvar.h>
	40	#include <sys/resourcevar.h>
	41	#include <sys/namei.h>
91447636 A	42	#include <sys/vnode_internal.h>
	43	#include <sys/proc_internal.h>
	44	#include <sys/kauth.h>
1c79356b A	45	#include <sys/timeb.h>
1c79356b A	46	#include <sys/times.h>
1c79356b	47	#include <sys/acct.h>
91447636	48	#include <sys/file_internal.h>
1c79356b A	49	#include <sys/uio.h>
	50	#include <sys/kernel.h>
	51	#include <sys/stat.h>
91447636 A	52	#include <sys/disk.h>
91447636 A	53	#include <sys/conf.h>
1c79356b A	54
	55	#include <mach-o/loader.h>
	56	#include <mach-o/nlist.h>
	57
91447636	58	#include <kern/kalloc.h>
1c79356b	59	#include <vm/vm_kern.h>
91447636	60	#include <pexpert/pexpert.h>
3a60a9f5	61	#include <IOKit/IOHibernatePrivate.h>
1c79356b	62
2d21ac55 A	63	/* This function is called from kern_sysctl in the current process context;
	64	* it is exported with the System6.0.exports, but this appears to be a legacy
	65	* export, as there are no internal consumers.
1c79356b	66	*/
2d21ac55	67	int
7ddcb079 A	68	get_kernel_symfile(__unused proc_t p, __unused char const **symfile);
7ddcb079 A	69	int
2d21ac55	70	get_kernel_symfile(__unused proc_t p, __unused char const **symfile)
1c79356b	71	{
2d21ac55	72	return KERN_FAILURE;
0b4e3aa0	73	}
91447636	74
3a60a9f5 A	75	struct kern_direct_file_io_ref_t
3a60a9f5 A	76	{
6d2010ae A	77	vfs_context_t ctx;
	78	struct vnode * vp;
	79	dev_t device;
7ddcb079	80	uint32_t blksize;
316670eb A	81	off_t filelength;
316670eb A	82	char pinned;
3a60a9f5 A	83	};
	84
	85
6d2010ae	86	static int file_ioctl(void * p1, void * p2, u_long theIoctl, caddr_t result)
3a60a9f5	87	{
b0d623f7	88	dev_t device = (dev_t) p1;
3a60a9f5 A	89
	90	return ((*bdevsw[major(device)].d_ioctl)
	91	(device, theIoctl, result, S_IFBLK, p2));
	92	}
	93
6d2010ae	94	static int device_ioctl(void * p1, __unused void * p2, u_long theIoctl, caddr_t result)
3a60a9f5 A	95	{
	96	return (VNOP_IOCTL(p1, theIoctl, result, 0, p2));
	97	}
	98
316670eb A	99	static int
	100	kern_ioctl_file_extents(struct kern_direct_file_io_ref_t * ref, u_long theIoctl, off_t offset, off_t end)
	101	{
	102	int error;
	103	int (do_ioctl)(void p1, void * p2, u_long theIoctl, caddr_t result);
	104	void * p1;
	105	void * p2;
	106	uint64_t fileblk;
	107	size_t filechunk;
	108	dk_extent_t extent;
	109	dk_unmap_t unmap;
	110	_dk_cs_pin_t pin;
	111
	112	bzero(&extent, sizeof(dk_extent_t));
	113	bzero(&unmap, sizeof(dk_unmap_t));
	114	bzero(&pin, sizeof(pin));
	115	if (ref->vp->v_type == VREG)
	116	{
	117	p1 = &ref->device;
	118	p2 = kernproc;
	119	do_ioctl = &file_ioctl;
	120	}
	121	else
	122	{
	123	/* Partition. */
	124	p1 = ref->vp;
	125	p2 = ref->ctx;
	126	do_ioctl = &device_ioctl;
	127	}
	128	while (offset < end)
	129	{
	130	if (ref->vp->v_type == VREG)
	131	{
	132	daddr64_t blkno;
	133	filechunk = 110241024*1024;
	134	if (filechunk > (size_t)(end - offset))
	135	filechunk = (size_t)(end - offset);
	136	error = VNOP_BLOCKMAP(ref->vp, offset, filechunk, &blkno, &filechunk, NULL, 0, NULL);
	137	if (error) break;
	138	fileblk = blkno * ref->blksize;
	139	}
	140	else if ((ref->vp->v_type == VBLK) \|\| (ref->vp->v_type == VCHR))
	141	{
	142	fileblk = offset;
	143	filechunk = ref->filelength;
	144	}
	145
	146	if (DKIOCUNMAP == theIoctl)
	147	{
	148	extent.offset = fileblk;
	149	extent.length = filechunk;
	150	unmap.extents = &extent;
	151	unmap.extentsCount = 1;
	152	error = do_ioctl(p1, p2, theIoctl, (caddr_t)&unmap);
	153	// printf("DKIOCUNMAP(%d) 0x%qx, 0x%qx\n", error, extent.offset, extent.length);
	154	}
	155	else if (_DKIOCCSPINEXTENT == theIoctl)
	156	{
	157	pin.cp_extent.offset = fileblk;
	158	pin.cp_extent.length = filechunk;
	159	pin.cp_flags = _DKIOCSPINDISCARDDATA;
	160	error = do_ioctl(p1, p2, theIoctl, (caddr_t)&pin);
	161	if (error && (ENOTTY != error))
	162	{
163	printf("_DKIOCCSPINEXTENT(%d) 0x%qx, 0x%qx\n",
164	error, pin.cp_extent.offset, pin.cp_extent.length);
165	}
166	}
167	else error = EINVAL;
168
169	if (error) break;
170	offset += filechunk;
171	}
172	return (error);
173	}
174
7ddcb079 A	175	int
	176	kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, caddr_t addr, vm_size_t len);
	177
3a60a9f5 A	178	struct kern_direct_file_io_ref_t *
	179	kern_open_file_for_direct_io(const char * name,
	180	kern_get_file_extents_callback_t callback,
	181	void * callback_ref,
6d2010ae A	182	dev_t * partition_device_result,
6d2010ae A	183	dev_t * image_device_result,
3a60a9f5	184	uint64_t * partitionbase_result,
0b4c1975	185	uint64_t * maxiocount_result,
6d2010ae A	186	uint32_t * oflags,
	187	off_t offset,
	188	caddr_t addr,
	189	vm_size_t len)
3a60a9f5 A	190	{
	191	struct kern_direct_file_io_ref_t * ref;
	192
2d21ac55	193	proc_t p;
3a60a9f5 A	194	struct vnode_attr va;
	195	int error;
	196	off_t f_offset;
6d2010ae A	197	uint64_t fileblk;
	198	size_t filechunk;
	199	uint64_t physoffset;
3a60a9f5	200	dev_t device;
6d2010ae A	201	dev_t target = 0;
	202	int isssd = 0;
	203	uint32_t flags = 0;
	204	uint32_t blksize;
3a60a9f5	205	off_t maxiocount, count;
6d2010ae	206	boolean_t locked = FALSE;
3a60a9f5	207
6d2010ae A	208	int (do_ioctl)(void p1, void * p2, u_long theIoctl, caddr_t result);
	209	void * p1 = NULL;
	210	void * p2 = NULL;
3a60a9f5 A	211
	212	error = EFAULT;
	213
	214	ref = (struct kern_direct_file_io_ref_t *) kalloc(sizeof(struct kern_direct_file_io_ref_t));
	215	if (!ref)
	216	{
	217	error = EFAULT;
	218	goto out;
	219	}
	220
316670eb	221	bzero(ref, sizeof(*ref));
6d2010ae	222	p = kernproc;
2d21ac55	223	ref->ctx = vfs_context_create(vfs_context_current());
3a60a9f5	224
2d21ac55	225	if ((error = vnode_open(name, (O_CREAT \| FWRITE), (0), 0, &ref->vp, ref->ctx)))
3a60a9f5 A	226	goto out;
3a60a9f5 A	227
6d2010ae A	228	if (addr && len)
	229	{
	230	if ((error = kern_write_file(ref, offset, addr, len)))
	231	goto out;
	232	}
	233
3a60a9f5 A	234	VATTR_INIT(&va);
	235	VATTR_WANTED(&va, va_rdev);
	236	VATTR_WANTED(&va, va_fsid);
	237	VATTR_WANTED(&va, va_data_size);
	238	VATTR_WANTED(&va, va_nlink);
	239	error = EFAULT;
2d21ac55	240	if (vnode_getattr(ref->vp, &va, ref->ctx))
3a60a9f5 A	241	goto out;
	242
	243	kprintf("vp va_rdev major %d minor %d\n", major(va.va_rdev), minor(va.va_rdev));
	244	kprintf("vp va_fsid major %d minor %d\n", major(va.va_fsid), minor(va.va_fsid));
	245	kprintf("vp size %qd\n", va.va_data_size);
	246
	247	if (ref->vp->v_type == VREG)
	248	{
	249	/* Don't dump files with links. */
	250	if (va.va_nlink != 1)
	251	goto out;
	252
	253	device = va.va_fsid;
b0d623f7	254	p1 = &device;
3a60a9f5 A	255	p2 = p;
	256	do_ioctl = &file_ioctl;
	257	}
	258	else if ((ref->vp->v_type == VBLK) \|\| (ref->vp->v_type == VCHR))
	259	{
	260	/* Partition. */
	261	device = va.va_rdev;
	262
	263	p1 = ref->vp;
2d21ac55	264	p2 = ref->ctx;
3a60a9f5 A	265	do_ioctl = &device_ioctl;
	266	}
	267	else
	268	{
	269	/* Don't dump to non-regular files. */
	270	error = EFAULT;
	271	goto out;
	272	}
6d2010ae A	273	ref->device = device;
6d2010ae A	274
6d2010ae A	275	// get block size
6d2010ae A	276
7ddcb079	277	error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &ref->blksize);
6d2010ae A	278	if (error)
	279	goto out;
	280
	281	if (ref->vp->v_type == VREG)
7ddcb079	282	ref->filelength = va.va_data_size;
6d2010ae A	283	else
	284	{
	285	error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &fileblk);
	286	if (error)
	287	goto out;
7ddcb079	288	ref->filelength = fileblk * ref->blksize;
6d2010ae A	289	}
6d2010ae A	290
316670eb A	291	// pin logical extents
	292
	293	error = kern_ioctl_file_extents(ref, _DKIOCCSPINEXTENT, 0, ref->filelength);
	294	if (error && (ENOTTY != error)) goto out;
	295	ref->pinned = (error == 0);
	296
	297	// generate the block list
	298
	299	error = do_ioctl(p1, p2, DKIOCLOCKPHYSICALEXTENTS, NULL);
	300	if (error)
	301	goto out;
	302	locked = TRUE;
	303
6d2010ae	304	f_offset = 0;
7ddcb079	305	while (f_offset < ref->filelength)
6d2010ae A	306	{
	307	if (ref->vp->v_type == VREG)
	308	{
	309	filechunk = 110241024*1024;
	310	daddr64_t blkno;
	311
	312	error = VNOP_BLOCKMAP(ref->vp, f_offset, filechunk, &blkno, &filechunk, NULL, 0, NULL);
	313	if (error)
	314	goto out;
	315
7ddcb079	316	fileblk = blkno * ref->blksize;
6d2010ae A	317	}
	318	else if ((ref->vp->v_type == VBLK) \|\| (ref->vp->v_type == VCHR))
	319	{
	320	fileblk = f_offset;
7ddcb079	321	filechunk = f_offset ? 0 : ref->filelength;
6d2010ae A	322	}
	323
	324	physoffset = 0;
	325	while (physoffset < filechunk)
	326	{
	327	dk_physical_extent_t getphysreq;
	328	bzero(&getphysreq, sizeof(getphysreq));
	329
	330	getphysreq.offset = fileblk + physoffset;
	331	getphysreq.length = (filechunk - physoffset);
	332	error = do_ioctl(p1, p2, DKIOCGETPHYSICALEXTENT, (caddr_t) &getphysreq);
	333	if (error)
	334	goto out;
	335	if (!target)
	336	{
	337	target = getphysreq.dev;
	338	}
	339	else if (target != getphysreq.dev)
	340	{
	341	error = ENOTSUP;
	342	goto out;
	343	}
	344	callback(callback_ref, getphysreq.offset, getphysreq.length);
	345	physoffset += getphysreq.length;
	346	}
	347	f_offset += filechunk;
	348	}
	349	callback(callback_ref, 0ULL, 0ULL);
	350
	351	if (ref->vp->v_type == VREG)
	352	p1 = &target;
3a60a9f5 A	353
	354	// get partition base
	355
	356	error = do_ioctl(p1, p2, DKIOCGETBASE, (caddr_t) partitionbase_result);
	357	if (error)
	358	goto out;
	359
	360	// get block size & constraints
	361
	362	error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &blksize);
	363	if (error)
	364	goto out;
	365
	366	maxiocount = 110241024*1024;
	367
	368	error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTREAD, (caddr_t) &count);
	369	if (error)
	370	count = 0;
	371	count *= blksize;
	372	if (count && (count < maxiocount))
	373	maxiocount = count;
	374
	375	error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTWRITE, (caddr_t) &count);
	376	if (error)
	377	count = 0;
	378	count *= blksize;
	379	if (count && (count < maxiocount))
	380	maxiocount = count;
	381
	382	error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTREAD, (caddr_t) &count);
	383	if (error)
	384	count = 0;
	385	if (count && (count < maxiocount))
	386	maxiocount = count;
	387
	388	error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTWRITE, (caddr_t) &count);
	389	if (error)
	390	count = 0;
	391	if (count && (count < maxiocount))
	392	maxiocount = count;
	393
	394	error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTREAD, (caddr_t) &count);
	395	if (error)
	396	count = 0;
	397	if (count && (count < maxiocount))
	398	maxiocount = count;
	399
	400	error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTWRITE, (caddr_t) &count);
	401	if (error)
	402	count = 0;
	403	if (count && (count < maxiocount))
	404	maxiocount = count;
	405
	406	kprintf("max io 0x%qx bytes\n", maxiocount);
	407	if (maxiocount_result)
	408	*maxiocount_result = maxiocount;
	409
6d2010ae A	410	error = do_ioctl(p1, p2, DKIOCISSOLIDSTATE, (caddr_t)&isssd);
	411	if (!error && isssd)
	412	flags \|= kIOHibernateOptionSSD;
3a60a9f5	413
6d2010ae A	414	if (partition_device_result)
	415	*partition_device_result = device;
	416	if (image_device_result)
	417	*image_device_result = target;
	418	if (flags)
	419	*oflags = flags;
3a60a9f5 A	420
	421	out:
	422	kprintf("kern_open_file_for_direct_io(%d)\n", error);
	423
6d2010ae A	424	if (error && locked)
	425	{
	426	p1 = &device;
	427	(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
	428	}
	429
	430	if (error && ref)
	431	{
	432	if (ref->vp)
	433	{
2d21ac55 A	434	vnode_close(ref->vp, FWRITE, ref->ctx);
	435	ref->vp = NULLVP;
	436	}
2d21ac55 A	437	vfs_context_rele(ref->ctx);
	438	kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
	439	ref = NULL;
3a60a9f5	440	}
3a60a9f5 A	441	return(ref);
	442	}
	443
	444	int
	445	kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, caddr_t addr, vm_size_t len)
	446	{
	447	return (vn_rdwr(UIO_WRITE, ref->vp,
	448	addr, len, offset,
b0d623f7	449	UIO_SYSSPACE, IO_SYNC\|IO_NODELOCKED\|IO_UNIT,
2d21ac55 A	450	vfs_context_ucred(ref->ctx), (int *) 0,
2d21ac55 A	451	vfs_context_proc(ref->ctx)));
3a60a9f5 A	452	}
3a60a9f5 A	453
7ddcb079	454
3a60a9f5	455	void
6d2010ae	456	kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref,
7ddcb079 A	457	off_t write_offset, caddr_t addr, vm_size_t write_length,
7ddcb079 A	458	off_t discard_offset, off_t discard_end)
3a60a9f5	459	{
6d2010ae	460	int error;
3a60a9f5 A	461	kprintf("kern_close_file_for_direct_io\n");
3a60a9f5 A	462
6d2010ae	463	if (!ref) return;
3a60a9f5	464
6d2010ae A	465	if (ref->vp)
	466	{
	467	int (do_ioctl)(void p1, void * p2, u_long theIoctl, caddr_t result);
	468	void * p1;
	469	void * p2;
	470
	471	if (ref->vp->v_type == VREG)
	472	{
	473	p1 = &ref->device;
	474	p2 = kernproc;
	475	do_ioctl = &file_ioctl;
	476	}
	477	else
	478	{
	479	/* Partition. */
	480	p1 = ref->vp;
	481	p2 = ref->ctx;
	482	do_ioctl = &device_ioctl;
	483	}
	484	(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
	485
7ddcb079 A	486	if (addr && write_length)
	487	{
	488	(void) kern_write_file(ref, write_offset, addr, write_length);
	489	}
316670eb	490	if (discard_offset && discard_end && !ref->pinned)
6d2010ae	491	{
316670eb	492	(void) kern_ioctl_file_extents(ref, DKIOCUNMAP, discard_offset, discard_end);
6d2010ae A	493	}
	494
	495	error = vnode_close(ref->vp, FWRITE, ref->ctx);
	496
	497	ref->vp = NULLVP;
	498	kprintf("vnode_close(%d)\n", error);
3a60a9f5	499	}
6d2010ae A	500	vfs_context_rele(ref->ctx);
	501	ref->ctx = NULL;
	502	kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
3a60a9f5	503	}
2d21ac55	504