hptraid.c

来自「linux-2.4.29操作系统的源码」· C语言 代码 · 共 921 行 · 第 1/2 页

/*
   hptraid.c  Copyright (C) 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, 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.  
   
   Authors: 	Arjan van de Ven <[e-mail]>

   Based on work
   	Copyleft  (C) 2001 by Wilfried Weissmann <[e-mail]>
	Copyright (C) 1994-96 Marc ZYNGIER <[e-mail]>
   Based on work done by S鴕en Schmidt for FreeBSD

   Changelog:
   19.08.2003 v0.03 [e-mail]
   	* register the raid volume only if all disks are available
	* print a warning that raid-(0+)1 failover is not supported

   15.06.2003 v0.02 [e-mail]
   	* correct values of raid-1 superbock
	* re-add check for availability of all disks
	* fix offset bug in raid-1 (introduced in raid 0+1 implementation)

   14.06.2003 [e-mail]
   	* superblock has wrong "disks" value on raid-1
   	* fixup for raid-1 disknumbering
	* do _NOT_ align size to 255*63 boundary
		I WILL NOT USE FDISK TO DETERMINE THE VOLUME SIZE.
		I WILL NOT USE FDISK TO DETERMINE THE VOLUME SIZE.
		I WILL NOT USE FDISK TO DETERMINE THE VOLUME SIZE.
		I WILL NOT ...

   13.06.2003 [e-mail]
   	* raid 0+1 support
	* check if all disks of an array are available
	* bump version number

   29.05.2003 [e-mail]
   	* release no more devices than available on unload
	* remove static variables in raid-1 read path

*/

#include <linux/module.h>
#include <linux/init.h>
#include <linux/sched.h>
#include <linux/smp_lock.h>
#include <linux/kernel.h>
#include <linux/blkdev.h>
#include <linux/blkpg.h>
#include <linux/genhd.h>
#include <linux/ioctl.h>

#include <linux/ide.h>
#include <asm/uaccess.h>

#include "ataraid.h"
#include "hptraid.h"


static int hptraid_open(struct inode * inode, struct file * filp);
static int hptraid_release(struct inode * inode, struct file * filp);
static int hptraid_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg);
static int hptraidspan_make_request (request_queue_t *q, int rw, struct buffer_head * bh);
static int hptraid0_make_request (request_queue_t *q, int rw, struct buffer_head * bh);
static int hptraid1_make_request (request_queue_t *q, int rw, struct buffer_head * bh);
static int hptraid01_make_request (request_queue_t *q, int rw, struct buffer_head * bh);



struct hptdisk {
	kdev_t	device;		/* disk-ID/raid 0+1 volume-ID */
	unsigned long sectors;
	struct block_device *bdev;
	unsigned long last_pos;
};

struct hptraid {
	unsigned int stride;	/* stripesize */
	unsigned int disks;	/* number of disks in array */
	unsigned long sectors;	/* disksize in sectors */
        u_int32_t magic_0;
        u_int32_t magic_1;
	struct geom geom;
	
	int previous;		/* most recently accessed disk in mirror */
	struct hptdisk disk[8];
	unsigned long cutoff[8];	/* raid 0 cutoff */
	unsigned int cutoff_disks[8];	
	struct hptraid * raid01;	/* sub arrays for raid 0+1 */
};

struct hptraid_dev {
	int major;
	int minor;
	int device;
};

static struct hptraid_dev devlist[]=
{

	{IDE0_MAJOR,  0, -1},
	{IDE0_MAJOR, 64, -1},
	{IDE1_MAJOR,  0, -1},
	{IDE1_MAJOR, 64, -1},
	{IDE2_MAJOR,  0, -1},
	{IDE2_MAJOR, 64, -1},
	{IDE3_MAJOR,  0, -1},
	{IDE3_MAJOR, 64, -1},
	{IDE4_MAJOR,  0, -1},
	{IDE4_MAJOR, 64, -1},
	{IDE5_MAJOR,  0, -1},
	{IDE5_MAJOR, 64, -1},
	{IDE6_MAJOR,  0, -1},
	{IDE6_MAJOR, 64, -1}
};

static struct raid_device_operations hptraidspan_ops = {
	open:                   hptraid_open,
	release:                hptraid_release,
	ioctl:			hptraid_ioctl,
	make_request:		hptraidspan_make_request
};

static struct raid_device_operations hptraid0_ops = {
	open:                   hptraid_open,
	release:                hptraid_release,
	ioctl:			hptraid_ioctl,
	make_request:		hptraid0_make_request
};

static struct raid_device_operations hptraid1_ops = {
	open:                   hptraid_open,
	release:                hptraid_release,
	ioctl:			hptraid_ioctl,
	make_request:		hptraid1_make_request
};


static struct raid_device_operations hptraid01_ops = {
	open:                   hptraid_open,
	release:                hptraid_release,
	ioctl:			hptraid_ioctl,
	make_request:		hptraid01_make_request
};

static __initdata struct {
	struct raid_device_operations *op;
	u_int8_t type;
	char label[8];
} oplist[] = {
	{&hptraid0_ops, HPT_T_RAID_0, "RAID 0"},
	{&hptraid1_ops, HPT_T_RAID_1, "RAID 1"},
	{&hptraidspan_ops, HPT_T_SPAN, "SPAN"},
	{&hptraid01_ops, HPT_T_RAID_01_RAID_0, "RAID 0+1"},
	{0, 0}
};

static struct hptraid raid[14];

static int hptraid_ioctl(struct inode *inode, struct file *file,
		unsigned int cmd, unsigned long arg)
{
	unsigned int minor;
	unsigned char val;
	unsigned long sectors;
	
	if (!inode || !inode->i_rdev) 	
		return -EINVAL;

	minor = MINOR(inode->i_rdev)>>SHIFT;
	
	switch (cmd) {
         	case BLKGETSIZE:   /* Return device size */
			if (!arg)  return -EINVAL;
			sectors = ataraid_gendisk.part[MINOR(inode->i_rdev)].nr_sects;
			if (MINOR(inode->i_rdev)&15)
				return put_user(sectors, (unsigned long *) arg);
			return put_user(raid[minor].sectors , (unsigned long *) arg);
			break;
			

		case HDIO_GETGEO:
		{
			struct hd_geometry *loc = (struct hd_geometry *) arg;
			unsigned short bios_cyl;
			
			if (!loc) return -EINVAL;
			val = 255;
			if (put_user(val, (byte *) &loc->heads)) return -EFAULT;
			val=63;
			if (put_user(val, (byte *) &loc->sectors)) return -EFAULT;
			bios_cyl = raid[minor].sectors/63/255;
			if (put_user(bios_cyl, (unsigned short *) &loc->cylinders)) return -EFAULT;
			if (put_user((unsigned)ataraid_gendisk.part[MINOR(inode->i_rdev)].start_sect,
				(unsigned long *) &loc->start)) return -EFAULT;
			return 0;
		}

		case HDIO_GETGEO_BIG:
		{
			struct hd_big_geometry *loc = (struct hd_big_geometry *) arg;
			unsigned int bios_cyl;
			if (!loc) return -EINVAL;
			val = 255;
			if (put_user(val, (byte *) &loc->heads)) return -EFAULT;
			val = 63;
			if (put_user(val, (byte *) &loc->sectors)) return -EFAULT;
			bios_cyl = raid[minor].sectors/63/255;
			if (put_user(bios_cyl, (unsigned int *) &loc->cylinders)) return -EFAULT;
			if (put_user((unsigned)ataraid_gendisk.part[MINOR(inode->i_rdev)].start_sect,
				(unsigned long *) &loc->start)) return -EFAULT;
			return 0;
		}
			
		default:
			return blk_ioctl(inode->i_rdev, cmd, arg);
	};

	return 0;
}


static int hptraidspan_make_request (request_queue_t *q, int rw, struct buffer_head * bh)
{
	unsigned long rsect;
	unsigned int disk;
	int device;
	struct hptraid *thisraid;

	rsect = bh->b_rsector;

	device = (bh->b_rdev >> SHIFT)&MAJOR_MASK;
	thisraid = &raid[device];

	/*
	 * Partitions need adding of the start sector of the partition to the
	 * requested sector
	 */
	
	rsect += ataraid_gendisk.part[MINOR(bh->b_rdev)].start_sect;

	for (disk=0;disk<thisraid->disks;disk++) {
		if (disk==1)
			rsect+=10;
			// the "on next disk" contition check is a bit odd
		if (thisraid->disk[disk].sectors > rsect+1)
			break;
		rsect-=thisraid->disk[disk].sectors-(disk?11:1);
	}

		// request spans over 2 disks => request must be split
	if(rsect+bh->b_size/512 >= thisraid->disk[disk].sectors)
		return -1;
	
	/*
	 * The new BH_Lock semantics in ll_rw_blk.c guarantee that this
	 * is the only IO operation happening on this bh.
	 */
	 
	bh->b_rdev = thisraid->disk[disk].device;
	bh->b_rsector = rsect;

	/*
	 * Let the main block layer submit the IO and resolve recursion:
	 */
	return 1;
}

static int hptraid0_compute_request (struct hptraid *thisraid,
		request_queue_t *q,
		int rw, struct buffer_head * bh)
{
	unsigned long rsect_left,rsect_accum = 0;
	unsigned long block;
	unsigned int disk=0,real_disk=0;
	int i;

	/* Ok. We need to modify this sector number to a new disk + new sector
	 * number. 
	 * If there are disks of different sizes, this gets tricky. 
	 * Example with 3 disks (1Gb, 4Gb and 5 GB):
	 * The first 3 Gb of the "RAID" are evenly spread over the 3 disks.
	 * Then things get interesting. The next 2Gb (RAID view) are spread
	 * across disk 2 and 3 and the last 1Gb is disk 3 only.
	 *
	 * the way this is solved is like this: We have a list of "cutoff"
	 * points where everytime a disk falls out of the "higher" count, we
	 * mark the max sector. So once we pass a cutoff point, we have to
	 * divide by one less.
	 */
	
	if (thisraid->stride==0)
		thisraid->stride=1;

	/*
	 * Woops we need to split the request to avoid crossing a stride
	 * barrier
	 */
	if ((bh->b_rsector/thisraid->stride) !=
			((bh->b_rsector+(bh->b_size/512)-1)/thisraid->stride)) {
		return -1;
	}
			
	rsect_left = bh->b_rsector;;
	
	for (i=0;i<8;i++) {
		if (thisraid->cutoff_disks[i]==0)
			break;
		if (bh->b_rsector > thisraid->cutoff[i]) {
			/* we're in the wrong area so far */
			rsect_left -= thisraid->cutoff[i];
			rsect_accum += thisraid->cutoff[i] /
				thisraid->cutoff_disks[i];
		} else {
			block = rsect_left / thisraid->stride;
			disk = block % thisraid->cutoff_disks[i];
			block = (block / thisraid->cutoff_disks[i]) *
				thisraid->stride;
			bh->b_rsector = rsect_accum +
				(rsect_left % thisraid->stride) + block;
			break;
		}
	}
	
	for (i=0;i<8;i++) {
		if ((disk==0) && (thisraid->disk[i].sectors > rsect_accum)) {
			real_disk = i;
			break;
		}
		if ((disk>0) && (thisraid->disk[i].sectors >= rsect_accum)) {
			disk--;
		}
		
	}
	disk = real_disk;
	
	/* All but the first disk have a 10 sector offset */
	if (i>0)
		bh->b_rsector+=10;
		
	
	/*
	 * The new BH_Lock semantics in ll_rw_blk.c guarantee that this
	 * is the only IO operation happening on this bh.
	 */
	 
	bh->b_rdev = thisraid->disk[disk].device;

	/*
	 * Let the main block layer submit the IO and resolve recursion:
	 */
	return 1;
}

static int hptraid0_make_request (request_queue_t *q, int rw, struct buffer_head * bh)
{
	unsigned long rsect;
	int device;

	/*
	 * save the sector, it must be restored before a request-split
	 * is performed
	 */
	rsect = bh->b_rsector;

	/*
	 * Partitions need adding of the start sector of the partition to the
	 * requested sector
	 */
	
	bh->b_rsector += ataraid_gendisk.part[MINOR(bh->b_rdev)].start_sect;

	device = (bh->b_rdev >> SHIFT)&MAJOR_MASK;
	if( hptraid0_compute_request(raid+device, q, rw, bh) != 1 ) {
			/* request must be split => restore sector */
		bh->b_rsector = rsect;
		return -1;
	}

	return 1;
}

static int hptraid1_read_request (request_queue_t *q, int rw, struct buffer_head * bh)
{
	int device;
	int dist;
	int bestsofar,bestdist,i;

	/* Reads are simple in principle. Pick a disk and go. 
	   Initially I cheat by just picking the one which the last known
	   head position is closest by.
	   Later on, online/offline checking and performance needs adding */
	
	device = (bh->b_rdev >> SHIFT)&MAJOR_MASK;
	bh->b_rsector += ataraid_gendisk.part[MINOR(bh->b_rdev)].start_sect;

	bestsofar = 0; 
	bestdist = raid[device].disk[0].last_pos - bh->b_rsector;
	if (bestdist<0) 
		bestdist=-bestdist;
	if (bestdist>4095)
		bestdist=4095;

	for (i=1 ; i<raid[device].disks; i++) {
		dist = raid[device].disk[i].last_pos - bh->b_rsector;
		if (dist<0) 
			dist = -dist;
		if (dist>4095)
			dist=4095;

		  /* it's a tie; try to do some read balancing */
		if (bestdist==dist) {
			if ( (raid[device].previous>bestsofar) &&
					(raid[device].previous<=i) )  
				bestsofar = i;
			raid[device].previous =
				(raid[device].previous + 1) %
				raid[device].disks;
		} else if (bestdist>dist) {
			bestdist = dist;
			bestsofar = i;
		}
	
	}
	
	bh->b_rdev = raid[device].disk[bestsofar].device; 
	raid[device].disk[bestsofar].last_pos = bh->b_rsector+(bh->b_size>>9);

	/*
	 * Let the main block layer submit the IO and resolve recursion:
	 */
                          	
	return 1;
}

static int hptraid1_write_request(request_queue_t *q, int rw, struct buffer_head * bh)
{
	struct buffer_head *bh1;
	struct ataraid_bh_private *private;
	int device;
	int i;

	device = (bh->b_rdev >> SHIFT)&MAJOR_MASK;
	private = ataraid_get_private();
	if (private==NULL)
		BUG();

	private->parent = bh;
	
	atomic_set(&private->count,raid[device].disks);


	for (i = 0; i< raid[device].disks; i++) { 
		bh1=ataraid_get_bhead();