/*

 * CDDL HEADER START

 *

 * The contents of this file are subject to the terms of the

 * Common Development and Distribution License (the "License").

 * You may not use this file except in compliance with the License.

 *

 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE

 * or http://www.opensolaris.org/os/licensing.

 * See the License for the specific language governing permissions

 * and limitations under the License.

 *

 * When distributing Covered Code, include this CDDL HEADER in each

 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.

 * If applicable, add the following below this CDDL HEADER, with the

 * fields enclosed by brackets "[]" replaced with your own identifying

 * information: Portions Copyright [yyyy] [name of copyright owner]

 *

 * CDDL HEADER END

 */

/*

 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.

 * Copyright (c) 2013 by Delphix. All rights reserved.

 * Copyright (c) 2013, Joyent, Inc. All rights reserved.

 */

#include <sys/zfs_context.h>

#include <sys/spa.h>

#include <sys/vdev_impl.h>

#include <sys/vdev_disk.h>

#include <sys/vdev_file.h>

#include <sys/vdev_raidz.h>

#include <sys/zio.h>

#include <sys/zio_checksum.h>

#include <sys/fs/zfs.h>

#include <sys/fm/fs/zfs.h>

/*

 * Virtual device vector for RAID-Z.

 *

 * This vdev supports single, double, and triple parity. For single parity,

 * we use a simple XOR of all the data columns. For double or triple parity,

 * we use a special case of Reed-Solomon coding. This extends the

 * technique described in "The mathematics of RAID-6" by H. Peter Anvin by

 * drawing on the system described in "A Tutorial on Reed-Solomon Coding for

 * Fault-Tolerance in RAID-like Systems" by James S. Plank on which the

 * former is also based. The latter is designed to provide higher performance

 * for writes.

 *

 * Note that the Plank paper claimed to support arbitrary N+M, but was then

 * amended six years later identifying a critical flaw that invalidates its

 * claims. Nevertheless, the technique can be adapted to work for up to

 * triple parity. For additional parity, the amendment "Note: Correction to

 * the 1997 Tutorial on Reed-Solomon Coding" by James S. Plank and Ying Ding

 * is viable, but the additional complexity means that write performance will

 * suffer.

 *

 * All of the methods above operate on a Galois field, defined over the

 * integers mod 2^N. In our case we choose N=8 for GF(8) so that all elements

 * can be expressed with a single byte. Briefly, the operations on the

 * field are defined as follows:

 *

 *   o addition (+) is represented by a bitwise XOR

 *   o subtraction (-) is therefore identical to addition: A + B = A - B

 *   o multiplication of A by 2 is defined by the following bitwise expression:

 *

 *	(A * 2)_7 = A_6

 *	(A * 2)_6 = A_5

 *	(A * 2)_5 = A_4

 *	(A * 2)_4 = A_3 + A_7

 *	(A * 2)_3 = A_2 + A_7

 *	(A * 2)_2 = A_1 + A_7

 *	(A * 2)_1 = A_0

 *	(A * 2)_0 = A_7

 *

 * In C, multiplying by 2 is therefore ((a << 1) ^ ((a & 0x80) ? 0x1d : 0)).

 * As an aside, this multiplication is derived from the error correcting

 * primitive polynomial x^8 + x^4 + x^3 + x^2 + 1.

 *

 * Observe that any number in the field (except for 0) can be expressed as a

 * power of 2 -- a generator for the field. We store a table of the powers of

 * 2 and logs base 2 for quick look ups, and exploit the fact that A * B can

 * be rewritten as 2^(log_2(A) + log_2(B)) (where '+' is normal addition rather

 * than field addition). The inverse of a field element A (A^-1) is therefore

 * A ^ (255 - 1) = A^254.

 *

 * The up-to-three parity columns, P, Q, R over several data columns,

 * D_0, ... D_n-1, can be expressed by field operations:

 *

 *	P = D_0 + D_1 + ... + D_n-2 + D_n-1

 *	Q = 2^n-1 * D_0 + 2^n-2 * D_1 + ... + 2^1 * D_n-2 + 2^0 * D_n-1

 *	  = ((...((D_0) * 2 + D_1) * 2 + ...) * 2 + D_n-2) * 2 + D_n-1

 *	R = 4^n-1 * D_0 + 4^n-2 * D_1 + ... + 4^1 * D_n-2 + 4^0 * D_n-1

 *	  = ((...((D_0) * 4 + D_1) * 4 + ...) * 4 + D_n-2) * 4 + D_n-1

 *

 * We chose 1, 2, and 4 as our generators because 1 corresponds to the trival

 * XOR operation, and 2 and 4 can be computed quickly and generate linearly-

 * independent coefficients. (There are no additional coefficients that have

 * this property which is why the uncorrected Plank method breaks down.)

 *

 * See the reconstruction code below for how P, Q and R can used individually

 * or in concert to recover missing data columns.

 */

typedef struct raidz_col {

	uint64_t rc_devidx;		/* child device index for I/O */

	uint64_t rc_offset;		/* device offset */

	uint64_t rc_size;		/* I/O size */

	void *rc_data;			/* I/O data */

	void *rc_gdata;			/* used to store the "good" version */

	int rc_error;			/* I/O error for this device */

	uint8_t rc_tried;		/* Did we attempt this I/O column? */

	uint8_t rc_skipped;		/* Did we skip this I/O column? */

} raidz_col_t;

typedef struct raidz_map {

	uint64_t rm_cols;		/* Regular column count */

	uint64_t rm_scols;		/* Count including skipped columns */

	uint64_t rm_bigcols;		/* Number of oversized columns */

	uint64_t rm_asize;		/* Actual total I/O size */

	uint64_t rm_missingdata;	/* Count of missing data devices */

	uint64_t rm_missingparity;	/* Count of missing parity devices */

	uint64_t rm_firstdatacol;	/* First data column/parity count */

	uint64_t rm_nskip;		/* Skipped sectors for padding */

	uint64_t rm_skipstart;		/* Column index of padding start */

	void *rm_datacopy;		/* rm_asize-buffer of copied data */

	uintptr_t rm_reports;		/* # of referencing checksum reports */

	uint8_t	rm_freed;		/* map no longer has referencing ZIO */

	uint8_t	rm_ecksuminjected;	/* checksum error was injected */

	raidz_col_t rm_col[1];		/* Flexible array of I/O columns */

} raidz_map_t;

#define	VDEV_RAIDZ_P		0

#define	VDEV_RAIDZ_Q		1

#define	VDEV_RAIDZ_R		2

#define	VDEV_RAIDZ_MUL_2(x)	(((x) << 1) ^ (((x) & 0x80) ? 0x1d : 0))

#define	VDEV_RAIDZ_MUL_4(x)	(VDEV_RAIDZ_MUL_2(VDEV_RAIDZ_MUL_2(x)))

/*

 * We provide a mechanism to perform the field multiplication operation on a

 * 64-bit value all at once rather than a byte at a time. This works by

 * creating a mask from the top bit in each byte and using that to

 * conditionally apply the XOR of 0x1d.

 */

#define	VDEV_RAIDZ_64MUL_2(x, mask) \

{ \

	(mask) = (x) & 0x8080808080808080ULL; \

	(mask) = ((mask) << 1) - ((mask) >> 7); \

	(x) = (((x) << 1) & 0xfefefefefefefefeULL) ^ \

	    ((mask) & 0x1d1d1d1d1d1d1d1d); \

}

#define	VDEV_RAIDZ_64MUL_4(x, mask) \

{ \

	VDEV_RAIDZ_64MUL_2((x), mask); \

	VDEV_RAIDZ_64MUL_2((x), mask); \

}

#define	VDEV_LABEL_OFFSET(x)	(x + VDEV_LABEL_START_SIZE)

/*

 * Force reconstruction to use the general purpose method.

 */

int vdev_raidz_default_to_general;

/* Powers of 2 in the Galois field defined above. */

static const uint8_t vdev_raidz_pow2[256] = {

	0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,

	0x1d, 0x3a, 0x74, 0xe8, 0xcd, 0x87, 0x13, 0x26,

	0x4c, 0x98, 0x2d, 0x5a, 0xb4, 0x75, 0xea, 0xc9,

	0x8f, 0x03, 0x06, 0x0c, 0x18, 0x30, 0x60, 0xc0,

	0x9d, 0x27, 0x4e, 0x9c, 0x25, 0x4a, 0x94, 0x35,

	0x6a, 0xd4, 0xb5, 0x77, 0xee, 0xc1, 0x9f, 0x23,

	0x46, 0x8c, 0x05, 0x0a, 0x14, 0x28, 0x50, 0xa0,

	0x5d, 0xba, 0x69, 0xd2, 0xb9, 0x6f, 0xde, 0xa1,

	0x5f, 0xbe, 0x61, 0xc2, 0x99, 0x2f, 0x5e, 0xbc,

	0x65, 0xca, 0x89, 0x0f, 0x1e, 0x3c, 0x78, 0xf0,

	0xfd, 0xe7, 0xd3, 0xbb, 0x6b, 0xd6, 0xb1, 0x7f,

	0xfe, 0xe1, 0xdf, 0xa3, 0x5b, 0xb6, 0x71, 0xe2,

	0xd9, 0xaf, 0x43, 0x86, 0x11, 0x22, 0x44, 0x88,

	0x0d, 0x1a, 0x34, 0x68, 0xd0, 0xbd, 0x67, 0xce,

	0x81, 0x1f, 0x3e, 0x7c, 0xf8, 0xed, 0xc7, 0x93,

	0x3b, 0x76, 0xec, 0xc5, 0x97, 0x33, 0x66, 0xcc,

	0x85, 0x17, 0x2e, 0x5c, 0xb8, 0x6d, 0xda, 0xa9,

	0x4f, 0x9e, 0x21, 0x42, 0x84, 0x15, 0x2a, 0x54,

	0xa8, 0x4d, 0x9a, 0x29, 0x52, 0xa4, 0x55, 0xaa,

	0x49, 0x92, 0x39, 0x72, 0xe4, 0xd5, 0xb7, 0x73,

	0xe6, 0xd1, 0xbf, 0x63, 0xc6, 0x91, 0x3f, 0x7e,

	0xfc, 0xe5, 0xd7, 0xb3, 0x7b, 0xf6, 0xf1, 0xff,

	0xe3, 0xdb, 0xab, 0x4b, 0x96, 0x31, 0x62, 0xc4,

	0x95, 0x37, 0x6e, 0xdc, 0xa5, 0x57, 0xae, 0x41,

	0x82, 0x19, 0x32, 0x64, 0xc8, 0x8d, 0x07, 0x0e,

	0x1c, 0x38, 0x70, 0xe0, 0xdd, 0xa7, 0x53, 0xa6,

	0x51, 0xa2, 0x59, 0xb2, 0x79, 0xf2, 0xf9, 0xef,

	0xc3, 0x9b, 0x2b, 0x56, 0xac, 0x45, 0x8a, 0x09,

	0x12, 0x24, 0x48, 0x90, 0x3d, 0x7a, 0xf4, 0xf5,

	0xf7, 0xf3, 0xfb, 0xeb, 0xcb, 0x8b, 0x0b, 0x16,

	0x2c, 0x58, 0xb0, 0x7d, 0xfa, 0xe9, 0xcf, 0x83,

	0x1b, 0x36, 0x6c, 0xd8, 0xad, 0x47, 0x8e, 0x01

};

/* Logs of 2 in the Galois field defined above. */

static const uint8_t vdev_raidz_log2[256] = {

	0x00, 0x00, 0x01, 0x19, 0x02, 0x32, 0x1a, 0xc6,

	0x03, 0xdf, 0x33, 0xee, 0x1b, 0x68, 0xc7, 0x4b,

	0x04, 0x64, 0xe0, 0x0e, 0x34, 0x8d, 0xef, 0x81,

	0x1c, 0xc1, 0x69, 0xf8, 0xc8, 0x08, 0x4c, 0x71,

	0x05, 0x8a, 0x65, 0x2f, 0xe1, 0x24, 0x0f, 0x21,

	0x35, 0x93, 0x8e, 0xda, 0xf0, 0x12, 0x82, 0x45,

	0x1d, 0xb5, 0xc2, 0x7d, 0x6a, 0x27, 0xf9, 0xb9,

	0xc9, 0x9a, 0x09, 0x78, 0x4d, 0xe4, 0x72, 0xa6,

	0x06, 0xbf, 0x8b, 0x62, 0x66, 0xdd, 0x30, 0xfd,

	0xe2, 0x98, 0x25, 0xb3, 0x10, 0x91, 0x22, 0x88,

	0x36, 0xd0, 0x94, 0xce, 0x8f, 0x96, 0xdb, 0xbd,

	0xf1, 0xd2, 0x13, 0x5c, 0x83, 0x38, 0x46, 0x40,

	0x1e, 0x42, 0xb6, 0xa3, 0xc3, 0x48, 0x7e, 0x6e,

	0x6b, 0x3a, 0x28, 0x54, 0xfa, 0x85, 0xba, 0x3d,

	0xca, 0x5e, 0x9b, 0x9f, 0x0a, 0x15, 0x79, 0x2b,

	0x4e, 0xd4, 0xe5, 0xac, 0x73, 0xf3, 0xa7, 0x57,

	0x07, 0x70, 0xc0, 0xf7, 0x8c, 0x80, 0x63, 0x0d,

	0x67, 0x4a, 0xde, 0xed, 0x31, 0xc5, 0xfe, 0x18,

	0xe3, 0xa5, 0x99, 0x77, 0x26, 0xb8, 0xb4, 0x7c,

	0x11, 0x44, 0x92, 0xd9, 0x23, 0x20, 0x89, 0x2e,

	0x37, 0x3f, 0xd1, 0x5b, 0x95, 0xbc, 0xcf, 0xcd,

	0x90, 0x87, 0x97, 0xb2, 0xdc, 0xfc, 0xbe, 0x61,

	0xf2, 0x56, 0xd3, 0xab, 0x14, 0x2a, 0x5d, 0x9e,

	0x84, 0x3c, 0x39, 0x53, 0x47, 0x6d, 0x41, 0xa2,

	0x1f, 0x2d, 0x43, 0xd8, 0xb7, 0x7b, 0xa4, 0x76,

	0xc4, 0x17, 0x49, 0xec, 0x7f, 0x0c, 0x6f, 0xf6,

	0x6c, 0xa1, 0x3b, 0x52, 0x29, 0x9d, 0x55, 0xaa,

	0xfb, 0x60, 0x86, 0xb1, 0xbb, 0xcc, 0x3e, 0x5a,

	0xcb, 0x59, 0x5f, 0xb0, 0x9c, 0xa9, 0xa0, 0x51,

	0x0b, 0xf5, 0x16, 0xeb, 0x7a, 0x75, 0x2c, 0xd7,

	0x4f, 0xae, 0xd5, 0xe9, 0xe6, 0xe7, 0xad, 0xe8,

	0x74, 0xd6, 0xf4, 0xea, 0xa8, 0x50, 0x58, 0xaf,

};

static void vdev_raidz_generate_parity(raidz_map_t *rm);

/*

 * Multiply a given number by 2 raised to the given power.

 */

static uint8_t

vdev_raidz_exp2(uint_t a, int exp)

{

	if (a == 0)

		return (0);

	ASSERT(exp >= 0);

	ASSERT(vdev_raidz_log2[a] > 0 || a == 1);

	exp += vdev_raidz_log2[a];

	if (exp > 255)

		exp -= 255;

	return (vdev_raidz_pow2[exp]);

}

static void

vdev_raidz_map_free(raidz_map_t *rm)

{

	int c;

	size_t size;

	for (c = 0; c < rm->rm_firstdatacol; c++) {

		zio_buf_free(rm->rm_col[c].rc_data, rm->rm_col[c].rc_size);

		if (rm->rm_col[c].rc_gdata != NULL)

			zio_buf_free(rm->rm_col[c].rc_gdata,

			    rm->rm_col[c].rc_size);

	}

	size = 0;

	for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++)

		size += rm->rm_col[c].rc_size;

	if (rm->rm_datacopy != NULL)

		zio_buf_free(rm->rm_datacopy, size);

	kmem_free(rm, offsetof(raidz_map_t, rm_col[rm->rm_scols]));

}

static void

vdev_raidz_map_free_vsd(zio_t *zio)

{

	raidz_map_t *rm = zio->io_vsd;

	ASSERT0(rm->rm_freed);

	rm->rm_freed = 1;

	if (rm->rm_reports == 0)

		vdev_raidz_map_free(rm);

}

/*ARGSUSED*/

static void

vdev_raidz_cksum_free(void *arg, size_t ignored)

{

	raidz_map_t *rm = arg;

	ASSERT3U(rm->rm_reports, >, 0);

	if (--rm->rm_reports == 0 && rm->rm_freed != 0)

		vdev_raidz_map_free(rm);

}

static void

vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const void *good_data)

{

	raidz_map_t *rm = zcr->zcr_cbdata;

	size_t c = zcr->zcr_cbinfo;

	size_t x;

	const char *good = NULL;

	const char *bad = rm->rm_col[c].rc_data;

	if (good_data == NULL) {

		zfs_ereport_finish_checksum(zcr, NULL, NULL, B_FALSE);

		return;

	}

	if (c < rm->rm_firstdatacol) {

		/*

		 * The first time through, calculate the parity blocks for

		 * the good data (this relies on the fact that the good

		 * data never changes for a given logical ZIO)

		 */

		if (rm->rm_col[0].rc_gdata == NULL) {

			char *bad_parity[VDEV_RAIDZ_MAXPARITY];

			char *buf;

			/*

			 * Set up the rm_col[]s to generate the parity for

			 * good_data, first saving the parity bufs and

			 * replacing them with buffers to hold the result.

			 */

			for (x = 0; x < rm->rm_firstdatacol; x++) {

				bad_parity[x] = rm->rm_col[x].rc_data;

				rm->rm_col[x].rc_data = rm->rm_col[x].rc_gdata =

				    zio_buf_alloc(rm->rm_col[x].rc_size);

			}

			/* fill in the data columns from good_data */

			buf = (char *)good_data;

			for (; x < rm->rm_cols; x++) {

				rm->rm_col[x].rc_data = buf;

				buf += rm->rm_col[x].rc_size;

			}

			/*

			 * Construct the parity from the good data.

			 */

			vdev_raidz_generate_parity(rm);

			/* restore everything back to its original state */

			for (x = 0; x < rm->rm_firstdatacol; x++)

				rm->rm_col[x].rc_data = bad_parity[x];

			buf = rm->rm_datacopy;

			for (x = rm->rm_firstdatacol; x < rm->rm_cols; x++) {

				rm->rm_col[x].rc_data = buf;

				buf += rm->rm_col[x].rc_size;

			}

		}

		ASSERT3P(rm->rm_col[c].rc_gdata, !=, NULL);

		good = rm->rm_col[c].rc_gdata;

	} else {

		/* adjust good_data to point at the start of our column */

		good = good_data;

		for (x = rm->rm_firstdatacol; x < c; x++)

			good += rm->rm_col[x].rc_size;

	}

	/* we drop the ereport if it ends up that the data was good */

	zfs_ereport_finish_checksum(zcr, good, bad, B_TRUE);

}

/*

 * Invoked indirectly by zfs_ereport_start_checksum(), called

 * below when our read operation fails completely.  The main point

 * is to keep a copy of everything we read from disk, so that at

 * vdev_raidz_cksum_finish() time we can compare it with the good data.

 */

static void

vdev_raidz_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *arg)

{

	size_t c = (size_t)(uintptr_t)arg;

	caddr_t buf;

	raidz_map_t *rm = zio->io_vsd;

	size_t size;

	/* set up the report and bump the refcount  */

	zcr->zcr_cbdata = rm;

	zcr->zcr_cbinfo = c;

	zcr->zcr_finish = vdev_raidz_cksum_finish;

	zcr->zcr_free = vdev_raidz_cksum_free;

	rm->rm_reports++;

	ASSERT3U(rm->rm_reports, >, 0);

	if (rm->rm_datacopy != NULL)

		return;

	/*

	 * It's the first time we're called for this raidz_map_t, so we need

	 * to copy the data aside; there's no guarantee that our zio's buffer

	 * won't be re-used for something else.

	 *

	 * Our parity data is already in separate buffers, so there's no need

	 * to copy them.

	 */

	size = 0;

	for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++)

		size += rm->rm_col[c].rc_size;

	buf = rm->rm_datacopy = zio_buf_alloc(size);

	for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {

		raidz_col_t *col = &rm->rm_col[c];

		bcopy(col->rc_data, buf, col->rc_size);

		col->rc_data = buf;

		buf += col->rc_size;

	}

	ASSERT3P(buf - (caddr_t)rm->rm_datacopy, ==, size);

}

static const zio_vsd_ops_t vdev_raidz_vsd_ops = {

	vdev_raidz_map_free_vsd,

	vdev_raidz_cksum_report

};

/*

 * Divides the IO evenly across all child vdevs; usually, dcols is

 * the number of children in the target vdev.

 */

static raidz_map_t *

vdev_raidz_map_alloc(caddr_t data, uint64_t size, uint64_t offset,

    uint64_t unit_shift, uint64_t dcols, uint64_t nparity)

{

	raidz_map_t *rm;

	/* The starting RAIDZ (parent) vdev sector of the block. */

	uint64_t b = offset >> unit_shift;

	/* The zio's size in units of the vdev's minimum sector size. */

	uint64_t s = size >> unit_shift;

	/* The first column for this stripe. */

	uint64_t f = b % dcols;

	/* The starting byte offset on each child vdev. */

	uint64_t o = (b / dcols) << unit_shift;

	uint64_t q, r, c, bc, col, acols, scols, coff, devidx, asize, tot;

	/*

	 * "Quotient": The number of data sectors for this stripe on all but

	 * the "big column" child vdevs that also contain "remainder" data.

	 */

	q = s / (dcols - nparity);

	/*

	 * "Remainder": The number of partial stripe data sectors in this I/O.

	 * This will add a sector to some, but not all, child vdevs.

	 */

	r = s - q * (dcols - nparity);

	/* The number of "big columns" - those which contain remainder data. */

	bc = (r == 0 ? 0 : r + nparity);

	/*

	 * The total number of data and parity sectors associated with

	 * this I/O.

	 */

	tot = s + nparity * (q + (r == 0 ? 0 : 1));

	/* acols: The columns that will be accessed. */

	/* scols: The columns that will be accessed or skipped. */

	if (q == 0) {

		/* Our I/O request doesn't span all child vdevs. */

		acols = bc;

		scols = MIN(dcols, roundup(bc, nparity + 1));

	} else {

		acols = dcols;

		scols = dcols;

	}

	ASSERT3U(acols, <=, scols);