[PATCH 03/35] ocfs2: Add the underlying blockcheck code.

From: Mark Fasheh
Date: Thu Dec 25 2008 - 13:06:25 EST


From: Joel Becker <joel.becker@xxxxxxxxxx>

This is the code that computes crc32 and ecc for ocfs2 metadata blocks.
There are high-level functions that check whether the filesystem has the
ecc feature, mid-level functions that work on a single block or array of
buffer_heads, and the low-level ecc hamming code that can handle
multiple buffers like crc32_le().

It's not hooked up to the filesystem yet.

Signed-off-by: Joel Becker <joel.becker@xxxxxxxxxx>
Cc: Christoph Hellwig <hch@xxxxxx>
Signed-off-by: Mark Fasheh <mfasheh@xxxxxxxx>
---
fs/ocfs2/Makefile | 1 +
fs/ocfs2/blockcheck.c | 480 +++++++++++++++++++++++++++++++++++++++++++++++++
fs/ocfs2/blockcheck.h | 82 +++++++++
fs/ocfs2/ocfs2.h | 8 +
4 files changed, 571 insertions(+), 0 deletions(-)
create mode 100644 fs/ocfs2/blockcheck.c
create mode 100644 fs/ocfs2/blockcheck.h

diff --git a/fs/ocfs2/Makefile b/fs/ocfs2/Makefile
index 7e4b361..0159607 100644
--- a/fs/ocfs2/Makefile
+++ b/fs/ocfs2/Makefile
@@ -12,6 +12,7 @@ obj-$(CONFIG_OCFS2_FS_USERSPACE_CLUSTER) += ocfs2_stack_user.o
ocfs2-objs := \
alloc.o \
aops.o \
+ blockcheck.o \
buffer_head_io.o \
dcache.o \
dir.o \
diff --git a/fs/ocfs2/blockcheck.c b/fs/ocfs2/blockcheck.c
new file mode 100644
index 0000000..2bf3d7f
--- /dev/null
+++ b/fs/ocfs2/blockcheck.c
@@ -0,0 +1,480 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * blockcheck.c
+ *
+ * Checksum and ECC codes for the OCFS2 userspace library.
+ *
+ * Copyright (C) 2006, 2008 Oracle. 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, version 2, as published by the Free Software Foundation.
+ *
+ * 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.
+ */
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/crc32.h>
+#include <linux/buffer_head.h>
+#include <linux/bitops.h>
+#include <asm/byteorder.h>
+
+#include "ocfs2.h"
+
+#include "blockcheck.h"
+
+
+
+/*
+ * We use the following conventions:
+ *
+ * d = # data bits
+ * p = # parity bits
+ * c = # total code bits (d + p)
+ */
+static int calc_parity_bits(unsigned int d)
+{
+ unsigned int p;
+
+ /*
+ * Bits required for Single Error Correction is as follows:
+ *
+ * d + p + 1 <= 2^p
+ *
+ * We're restricting ourselves to 31 bits of parity, that should be
+ * sufficient.
+ */
+ for (p = 1; p < 32; p++)
+ {
+ if ((d + p + 1) <= (1 << p))
+ return p;
+ }
+
+ return 0;
+}
+
+/*
+ * Calculate the bit offset in the hamming code buffer based on the bit's
+ * offset in the data buffer. Since the hamming code reserves all
+ * power-of-two bits for parity, the data bit number and the code bit
+ * number are offest by all the parity bits beforehand.
+ *
+ * Recall that bit numbers in hamming code are 1-based. This function
+ * takes the 0-based data bit from the caller.
+ *
+ * An example. Take bit 1 of the data buffer. 1 is a power of two (2^0),
+ * so it's a parity bit. 2 is a power of two (2^1), so it's a parity bit.
+ * 3 is not a power of two. So bit 1 of the data buffer ends up as bit 3
+ * in the code buffer.
+ */
+static unsigned int calc_code_bit(unsigned int i)
+{
+ unsigned int b, p;
+
+ /*
+ * Data bits are 0-based, but we're talking code bits, which
+ * are 1-based.
+ */
+ b = i + 1;
+
+ /*
+ * For every power of two below our bit number, bump our bit.
+ *
+ * We compare with (b + 1) becuase we have to compare with what b
+ * would be _if_ it were bumped up by the parity bit. Capice?
+ */
+ for (p = 0; (1 << p) < (b + 1); p++)
+ b++;
+
+ return b;
+}
+
+/*
+ * This is the low level encoder function. It can be called across
+ * multiple hunks just like the crc32 code. 'd' is the number of bits
+ * _in_this_hunk_. nr is the bit offset of this hunk. So, if you had
+ * two 512B buffers, you would do it like so:
+ *
+ * parity = ocfs2_hamming_encode(0, buf1, 512 * 8, 0);
+ * parity = ocfs2_hamming_encode(parity, buf2, 512 * 8, 512 * 8);
+ *
+ * If you just have one buffer, use ocfs2_hamming_encode_block().
+ */
+u32 ocfs2_hamming_encode(u32 parity, void *data, unsigned int d, unsigned int nr)
+{
+ unsigned int p = calc_parity_bits(nr + d);
+ unsigned int i, j, b;
+
+ BUG_ON(!p);
+
+ /*
+ * b is the hamming code bit number. Hamming code specifies a
+ * 1-based array, but C uses 0-based. So 'i' is for C, and 'b' is
+ * for the algorithm.
+ *
+ * The i++ in the for loop is so that the start offset passed
+ * to ocfs2_find_next_bit_set() is one greater than the previously
+ * found bit.
+ */
+ for (i = 0; (i = ocfs2_find_next_bit(data, d, i)) < d; i++)
+ {
+ /*
+ * i is the offset in this hunk, nr + i is the total bit
+ * offset.
+ */
+ b = calc_code_bit(nr + i);
+
+ for (j = 0; j < p; j++)
+ {
+ /*
+ * Data bits in the resultant code are checked by
+ * parity bits that are part of the bit number
+ * representation. Huh?
+ *
+ * <wikipedia href="http://en.wikipedia.org/wiki/Hamming_code";>
+ * In other words, the parity bit at position 2^k
+ * checks bits in positions having bit k set in
+ * their binary representation. Conversely, for
+ * instance, bit 13, i.e. 1101(2), is checked by
+ * bits 1000(2) = 8, 0100(2)=4 and 0001(2) = 1.
+ * </wikipedia>
+ *
+ * Note that 'k' is the _code_ bit number. 'b' in
+ * our loop.
+ */
+ if (b & (1 << j))
+ parity ^= (1 << j);
+ }
+ }
+
+ /* While the data buffer was treated as little endian, the
+ * return value is in host endian. */
+ return parity;
+}
+
+u32 ocfs2_hamming_encode_block(void *data, unsigned int blocksize)
+{
+ return ocfs2_hamming_encode(0, data, blocksize * 8, 0);
+}
+
+/*
+ * Like ocfs2_hamming_encode(), this can handle hunks. nr is the bit
+ * offset of the current hunk. If bit to be fixed is not part of the
+ * current hunk, this does nothing.
+ *
+ * If you only have one hunk, use ocfs2_hamming_fix_block().
+ */
+void ocfs2_hamming_fix(void *data, unsigned int d, unsigned int nr,
+ unsigned int fix)
+{
+ unsigned int p = calc_parity_bits(nr + d);
+ unsigned int i, b;
+
+ BUG_ON(!p);
+
+ /*
+ * If the bit to fix has an hweight of 1, it's a parity bit. One
+ * busted parity bit is its own error. Nothing to do here.
+ */
+ if (hweight32(fix) == 1)
+ return;
+
+ /*
+ * nr + d is the bit right past the data hunk we're looking at.
+ * If fix after that, nothing to do
+ */
+ if (fix >= calc_code_bit(nr + d))
+ return;
+
+ /*
+ * nr is the offset in the data hunk we're starting at. Let's
+ * start b at the offset in the code buffer. See hamming_encode()
+ * for a more detailed description of 'b'.
+ */
+ b = calc_code_bit(nr);
+ /* If the fix is before this hunk, nothing to do */
+ if (fix < b)
+ return;
+
+ for (i = 0; i < d; i++, b++)
+ {
+ /* Skip past parity bits */
+ while (hweight32(b) == 1)
+ b++;
+
+ /*
+ * i is the offset in this data hunk.
+ * nr + i is the offset in the total data buffer.
+ * b is the offset in the total code buffer.
+ *
+ * Thus, when b == fix, bit i in the current hunk needs
+ * fixing.
+ */
+ if (b == fix)
+ {
+ if (ocfs2_test_bit(i, data))
+ ocfs2_clear_bit(i, data);
+ else
+ ocfs2_set_bit(i, data);
+ break;
+ }
+ }
+}
+
+void ocfs2_hamming_fix_block(void *data, unsigned int blocksize,
+ unsigned int fix)
+{
+ ocfs2_hamming_fix(data, blocksize * 8, 0, fix);
+}
+
+/*
+ * This function generates check information for a block.
+ * data is the block to be checked. bc is a pointer to the
+ * ocfs2_block_check structure describing the crc32 and the ecc.
+ *
+ * bc should be a pointer inside data, as the function will
+ * take care of zeroing it before calculating the check information. If
+ * bc does not point inside data, the caller must make sure any inline
+ * ocfs2_block_check structures are zeroed.
+ *
+ * The data buffer must be in on-disk endian (little endian for ocfs2).
+ * bc will be filled with little-endian values and will be ready to go to
+ * disk.
+ */
+void ocfs2_block_check_compute(void *data, size_t blocksize,
+ struct ocfs2_block_check *bc)
+{
+ u32 crc;
+ u32 ecc;
+
+ memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+ crc = crc32_le(~0, data, blocksize);
+ ecc = ocfs2_hamming_encode_block(data, blocksize);
+
+ /*
+ * No ecc'd ocfs2 structure is larger than 4K, so ecc will be no
+ * larger than 16 bits.
+ */
+ BUG_ON(ecc > USHORT_MAX);
+
+ bc->bc_crc32e = cpu_to_le32(crc);
+ bc->bc_ecc = cpu_to_le16((u16)ecc);
+}
+
+/*
+ * This function validates existing check information. Like _compute,
+ * the function will take care of zeroing bc before calculating check codes.
+ * If bc is not a pointer inside data, the caller must have zeroed any
+ * inline ocfs2_block_check structures.
+ *
+ * Again, the data passed in should be the on-disk endian.
+ */
+int ocfs2_block_check_validate(void *data, size_t blocksize,
+ struct ocfs2_block_check *bc)
+{
+ int rc = 0;
+ struct ocfs2_block_check check;
+ u32 crc, ecc;
+
+ check.bc_crc32e = le32_to_cpu(bc->bc_crc32e);
+ check.bc_ecc = le16_to_cpu(bc->bc_ecc);
+
+ memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+ /* Fast path - if the crc32 validates, we're good to go */
+ crc = crc32_le(~0, data, blocksize);
+ if (crc == check.bc_crc32e)
+ goto out;
+
+ /* Ok, try ECC fixups */
+ ecc = ocfs2_hamming_encode_block(data, blocksize);
+ ocfs2_hamming_fix_block(data, blocksize, ecc ^ check.bc_ecc);
+
+ /* And check the crc32 again */
+ crc = crc32_le(~0, data, blocksize);
+ if (crc == check.bc_crc32e)
+ goto out;
+
+ rc = -EIO;
+
+out:
+ bc->bc_crc32e = cpu_to_le32(check.bc_crc32e);
+ bc->bc_ecc = cpu_to_le16(check.bc_ecc);
+
+ return rc;
+}
+
+/*
+ * This function generates check information for a list of buffer_heads.
+ * bhs is the blocks to be checked. bc is a pointer to the
+ * ocfs2_block_check structure describing the crc32 and the ecc.
+ *
+ * bc should be a pointer inside data, as the function will
+ * take care of zeroing it before calculating the check information. If
+ * bc does not point inside data, the caller must make sure any inline
+ * ocfs2_block_check structures are zeroed.
+ *
+ * The data buffer must be in on-disk endian (little endian for ocfs2).
+ * bc will be filled with little-endian values and will be ready to go to
+ * disk.
+ */
+void ocfs2_block_check_compute_bhs(struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc)
+{
+ int i;
+ u32 crc, ecc;
+
+ BUG_ON(nr < 0);
+
+ if (!nr)
+ return;
+
+ memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+ for (i = 0, crc = ~0, ecc = 0; i < nr; i++) {
+ crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
+ /*
+ * The number of bits in a buffer is obviously b_size*8.
+ * The offset of this buffer is b_size*i, so the bit offset
+ * of this buffer is b_size*8*i.
+ */
+ ecc = (u16)ocfs2_hamming_encode(ecc, bhs[i]->b_data,
+ bhs[i]->b_size * 8,
+ bhs[i]->b_size * 8 * i);
+ }
+
+ /*
+ * No ecc'd ocfs2 structure is larger than 4K, so ecc will be no
+ * larger than 16 bits.
+ */
+ BUG_ON(ecc > USHORT_MAX);
+
+ bc->bc_crc32e = cpu_to_le32(crc);
+ bc->bc_ecc = cpu_to_le16((u16)ecc);
+}
+
+/*
+ * This function validates existing check information on a list of
+ * buffer_heads. Like _compute_bhs, the function will take care of
+ * zeroing bc before calculating check codes. If bc is not a pointer
+ * inside data, the caller must have zeroed any inline
+ * ocfs2_block_check structures.
+ *
+ * Again, the data passed in should be the on-disk endian.
+ */
+int ocfs2_block_check_validate_bhs(struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc)
+{
+ int i, rc = 0;
+ struct ocfs2_block_check check;
+ u32 crc, ecc, fix;
+
+ BUG_ON(nr < 0);
+
+ if (!nr)
+ return 0;
+
+ check.bc_crc32e = le32_to_cpu(bc->bc_crc32e);
+ check.bc_ecc = le16_to_cpu(bc->bc_ecc);
+
+ memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+ /* Fast path - if the crc32 validates, we're good to go */
+ for (i = 0, crc = ~0; i < nr; i++)
+ crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
+ if (crc == check.bc_crc32e)
+ goto out;
+
+ mlog(ML_ERROR,
+ "CRC32 failed: stored: %u, computed %u. Applying ECC.\n",
+ (unsigned int)check.bc_crc32e, (unsigned int)crc);
+
+ /* Ok, try ECC fixups */
+ for (i = 0, ecc = 0; i < nr; i++) {
+ /*
+ * The number of bits in a buffer is obviously b_size*8.
+ * The offset of this buffer is b_size*i, so the bit offset
+ * of this buffer is b_size*8*i.
+ */
+ ecc = (u16)ocfs2_hamming_encode(ecc, bhs[i]->b_data,
+ bhs[i]->b_size * 8,
+ bhs[i]->b_size * 8 * i);
+ }
+ fix = ecc ^ check.bc_ecc;
+ for (i = 0; i < nr; i++) {
+ /*
+ * Try the fix against each buffer. It will only affect
+ * one of them.
+ */
+ ocfs2_hamming_fix(bhs[i]->b_data, bhs[i]->b_size * 8,
+ bhs[i]->b_size * 8 * i, fix);
+ }
+
+ /* And check the crc32 again */
+ for (i = 0, crc = ~0; i < nr; i++)
+ crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
+ if (crc == check.bc_crc32e)
+ goto out;
+
+ mlog(ML_ERROR, "Fixed CRC32 failed: stored: %u, computed %u\n",
+ (unsigned int)check.bc_crc32e, (unsigned int)crc);
+
+ rc = -EIO;
+
+out:
+ bc->bc_crc32e = cpu_to_le32(check.bc_crc32e);
+ bc->bc_ecc = cpu_to_le16(check.bc_ecc);
+
+ return rc;
+}
+
+/*
+ * These are the main API. They check the superblock flag before
+ * calling the underlying operations.
+ *
+ * They expect the buffer(s) to be in disk format.
+ */
+void ocfs2_compute_meta_ecc(struct super_block *sb, void *data,
+ struct ocfs2_block_check *bc)
+{
+ if (ocfs2_meta_ecc(OCFS2_SB(sb)))
+ ocfs2_block_check_compute(data, sb->s_blocksize, bc);
+}
+
+int ocfs2_validate_meta_ecc(struct super_block *sb, void *data,
+ struct ocfs2_block_check *bc)
+{
+ int rc = 0;
+
+ if (ocfs2_meta_ecc(OCFS2_SB(sb)))
+ rc = ocfs2_block_check_validate(data, sb->s_blocksize, bc);
+
+ return rc;
+}
+
+void ocfs2_compute_meta_ecc_bhs(struct super_block *sb,
+ struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc)
+{
+ if (ocfs2_meta_ecc(OCFS2_SB(sb)))
+ ocfs2_block_check_compute_bhs(bhs, nr, bc);
+}
+
+int ocfs2_validate_meta_ecc_bhs(struct super_block *sb,
+ struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc)
+{
+ int rc = 0;
+
+ if (ocfs2_meta_ecc(OCFS2_SB(sb)))
+ rc = ocfs2_block_check_validate_bhs(bhs, nr, bc);
+
+ return rc;
+}
+
diff --git a/fs/ocfs2/blockcheck.h b/fs/ocfs2/blockcheck.h
new file mode 100644
index 0000000..70ec3fe
--- /dev/null
+++ b/fs/ocfs2/blockcheck.h
@@ -0,0 +1,82 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * blockcheck.h
+ *
+ * Checksum and ECC codes for the OCFS2 userspace library.
+ *
+ * Copyright (C) 2004, 2008 Oracle. 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, version 2, as published by the Free Software Foundation.
+ *
+ * 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.
+ */
+
+#ifndef OCFS2_BLOCKCHECK_H
+#define OCFS2_BLOCKCHECK_H
+
+
+/* High level block API */
+void ocfs2_compute_meta_ecc(struct super_block *sb, void *data,
+ struct ocfs2_block_check *bc);
+int ocfs2_validate_meta_ecc(struct super_block *sb, void *data,
+ struct ocfs2_block_check *bc);
+void ocfs2_compute_meta_ecc_bhs(struct super_block *sb,
+ struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc);
+int ocfs2_validate_meta_ecc_bhs(struct super_block *sb,
+ struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc);
+
+/* Lower level API */
+void ocfs2_block_check_compute(void *data, size_t blocksize,
+ struct ocfs2_block_check *bc);
+int ocfs2_block_check_validate(void *data, size_t blocksize,
+ struct ocfs2_block_check *bc);
+void ocfs2_block_check_compute_bhs(struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc);
+int ocfs2_block_check_validate_bhs(struct buffer_head **bhs, int nr,
+ struct ocfs2_block_check *bc);
+
+/*
+ * Hamming code functions
+ */
+
+/*
+ * Encoding hamming code parity bits for a buffer.
+ *
+ * This is the low level encoder function. It can be called across
+ * multiple hunks just like the crc32 code. 'd' is the number of bits
+ * _in_this_hunk_. nr is the bit offset of this hunk. So, if you had
+ * two 512B buffers, you would do it like so:
+ *
+ * parity = ocfs2_hamming_encode(0, buf1, 512 * 8, 0);
+ * parity = ocfs2_hamming_encode(parity, buf2, 512 * 8, 512 * 8);
+ *
+ * If you just have one buffer, use ocfs2_hamming_encode_block().
+ */
+u32 ocfs2_hamming_encode(u32 parity, void *data, unsigned int d,
+ unsigned int nr);
+/*
+ * Fix a buffer with a bit error. The 'fix' is the original parity
+ * xor'd with the parity calculated now.
+ *
+ * Like ocfs2_hamming_encode(), this can handle hunks. nr is the bit
+ * offset of the current hunk. If bit to be fixed is not part of the
+ * current hunk, this does nothing.
+ *
+ * If you only have one buffer, use ocfs2_hamming_fix_block().
+ */
+void ocfs2_hamming_fix(void *data, unsigned int d, unsigned int nr,
+ unsigned int fix);
+
+/* Convenience wrappers for a single buffer of data */
+extern u32 ocfs2_hamming_encode_block(void *data, unsigned int blocksize);
+extern void ocfs2_hamming_fix_block(void *data, unsigned int blocksize,
+ unsigned int fix);
+#endif
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index 5c77798..2bb389f 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -382,6 +382,13 @@ static inline int ocfs2_supports_xattr(struct ocfs2_super *osb)
return 0;
}

+static inline int ocfs2_meta_ecc(struct ocfs2_super *osb)
+{
+ if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_META_ECC)
+ return 1;
+ return 0;
+}
+
/* set / clear functions because cluster events can make these happen
* in parallel so we want the transitions to be atomic. this also
* means that any future flags osb_flags must be protected by spinlock
@@ -615,5 +622,6 @@ static inline s16 ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
#define ocfs2_clear_bit ext2_clear_bit
#define ocfs2_test_bit ext2_test_bit
#define ocfs2_find_next_zero_bit ext2_find_next_zero_bit
+#define ocfs2_find_next_bit ext2_find_next_bit
#endif /* OCFS2_H */

--
1.5.6

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/