You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

553 lines
16 KiB

/*
* Copyright (C) 2015 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <fcntl.h>
#include <stdlib.h>
#include <sys/mman.h>
extern "C" {
#include <fec.h>
}
#include "fec_private.h"
using rs_unique_ptr = std::unique_ptr<void, decltype(&free_rs_char)>;
/* prints a hexdump of `data' using warn(...) */
static void dump(const char *name, uint64_t value, const uint8_t *data,
size_t size)
{
const int bytes_per_line = 16;
char hex[bytes_per_line * 3 + 1];
char prn[bytes_per_line + 1];
warn("%s (%" PRIu64 ") (%zu bytes):", name ? name : "", value, size);
if (!data) {
warn(" (null)");
return;
}
for (size_t n = 0; n < size; n += bytes_per_line) {
memset(hex, 0, sizeof(hex));
memset(prn, 0, sizeof(prn));
for (size_t m = 0; m < bytes_per_line; ++m) {
if (n + m < size) {
ptrdiff_t offset = &hex[m * 3] - hex;
snprintf(hex + offset, sizeof(hex) - offset, "%02x ",
data[n + m]);
if (isprint(data[n + m])) {
prn[m] = data[n + m];
} else {
prn[m] = '.';
}
} else {
strcpy(&hex[m * 3], " ");
}
}
warn(" %04zu %s %s", n, hex, prn);
}
}
/* checks if `offset' is within a corrupted block */
static inline bool is_erasure(fec_handle *f, uint64_t offset,
const uint8_t *data)
{
if (unlikely(offset >= f->data_size)) {
return false;
}
/* ideally, we would like to know if a specific byte on this block has
been corrupted, but knowing whether any of them is can be useful as
well, because often the entire block is corrupted */
uint64_t n = offset / FEC_BLOCKSIZE;
return !f->hashtree().check_block_hash_with_index(n, data);
}
/* check if `offset' is within a block expected to contain zeros */
static inline bool is_zero(fec_handle *f, uint64_t offset)
{
auto hashtree = f->hashtree();
if (hashtree.hash_data.empty() || unlikely(offset >= f->data_size)) {
return false;
}
uint64_t hash_offset = (offset / FEC_BLOCKSIZE) * SHA256_DIGEST_LENGTH;
if (unlikely(hash_offset >
hashtree.hash_data.size() - SHA256_DIGEST_LENGTH)) {
return false;
}
return !memcmp(hashtree.zero_hash.data(), &hashtree.hash_data[hash_offset],
SHA256_DIGEST_LENGTH);
}
/* reads and decodes a single block starting from `offset', returns the number
of bytes corrected in `errors' */
static int __ecc_read(fec_handle *f, void *rs, uint8_t *dest, uint64_t offset,
bool use_erasures, uint8_t *ecc_data, size_t *errors)
{
check(offset % FEC_BLOCKSIZE == 0);
ecc_info *e = &f->ecc;
/* reverse interleaving: calculate the RS block that includes the requested
offset */
uint64_t rsb = offset - (offset / (e->rounds * FEC_BLOCKSIZE)) *
e->rounds * FEC_BLOCKSIZE;
int data_index = -1;
int erasures[e->rsn];
int neras = 0;
/* verity is required to check for erasures */
check(!use_erasures || !f->hashtree().hash_data.empty());
for (int i = 0; i < e->rsn; ++i) {
uint64_t interleaved = fec_ecc_interleave(rsb * e->rsn + i, e->rsn,
e->rounds);
if (interleaved == offset) {
data_index = i;
}
/* to improve our chances of correcting IO errors, initialize the
buffer to zeros even if we are going to read to it later */
uint8_t bbuf[FEC_BLOCKSIZE] = {0};
if (likely(interleaved < e->start) && !is_zero(f, interleaved)) {
/* copy raw data to reconstruct the RS block */
if (!raw_pread(f->fd, bbuf, FEC_BLOCKSIZE, interleaved)) {
warn("failed to read: %s", strerror(errno));
/* treat errors as corruption */
if (use_erasures && neras <= e->roots) {
erasures[neras++] = i;
}
} else if (use_erasures && neras <= e->roots &&
is_erasure(f, interleaved, bbuf)) {
erasures[neras++] = i;
}
}
for (int j = 0; j < FEC_BLOCKSIZE; ++j) {
ecc_data[j * FEC_RSM + i] = bbuf[j];
}
}
check(data_index >= 0);
size_t nerrs = 0;
uint8_t copy[FEC_RSM];
for (int i = 0; i < FEC_BLOCKSIZE; ++i) {
/* copy parity data */
if (!raw_pread(f->fd, &ecc_data[i * FEC_RSM + e->rsn], e->roots,
e->start + (i + rsb) * e->roots)) {
error("failed to read ecc data: %s", strerror(errno));
return -1;
}
/* for debugging decoding failures, because decode_rs_char can mangle
ecc_data */
if (unlikely(use_erasures)) {
memcpy(copy, &ecc_data[i * FEC_RSM], FEC_RSM);
}
/* decode */
int rc = decode_rs_char(rs, &ecc_data[i * FEC_RSM], erasures, neras);
if (unlikely(rc < 0)) {
if (use_erasures) {
error("RS block %" PRIu64 ": decoding failed (%d erasures)",
rsb, neras);
dump("raw RS block", rsb, copy, FEC_RSM);
} else if (f->hashtree().hash_data.empty()) {
warn("RS block %" PRIu64 ": decoding failed", rsb);
} else {
debug("RS block %" PRIu64 ": decoding failed", rsb);
}
errno = EIO;
return -1;
} else if (unlikely(rc > 0)) {
check(rc <= (use_erasures ? e->roots : e->roots / 2));
nerrs += rc;
}
dest[i] = ecc_data[i * FEC_RSM + data_index];
}
if (nerrs) {
warn("RS block %" PRIu64 ": corrected %zu errors", rsb, nerrs);
*errors += nerrs;
}
return FEC_BLOCKSIZE;
}
/* initializes RS decoder and allocates memory for interleaving */
static int ecc_init(fec_handle *f, rs_unique_ptr& rs,
std::unique_ptr<uint8_t[]>& ecc_data)
{
check(f);
rs.reset(init_rs_char(FEC_PARAMS(f->ecc.roots)));
if (unlikely(!rs)) {
error("failed to initialize RS");
errno = ENOMEM;
return -1;
}
ecc_data.reset(new (std::nothrow) uint8_t[FEC_RSM * FEC_BLOCKSIZE]);
if (unlikely(!ecc_data)) {
error("failed to allocate ecc buffer");
errno = ENOMEM;
return -1;
}
return 0;
}
/* reads `count' bytes from `offset' and corrects possible errors without
erasure detection, returning the number of corrected bytes in `errors' */
static ssize_t ecc_read(fec_handle *f, uint8_t *dest, size_t count,
uint64_t offset, size_t *errors)
{
check(f);
check(dest);
check(offset < f->data_size);
check(offset + count <= f->data_size);
check(errors);
debug("[%" PRIu64 ", %" PRIu64 ")", offset, offset + count);
rs_unique_ptr rs(NULL, free_rs_char);
std::unique_ptr<uint8_t[]> ecc_data;
if (ecc_init(f, rs, ecc_data) == -1) {
return -1;
}
uint64_t curr = offset / FEC_BLOCKSIZE;
size_t coff = (size_t)(offset - curr * FEC_BLOCKSIZE);
size_t left = count;
uint8_t data[FEC_BLOCKSIZE];
while (left > 0) {
/* there's no erasure detection without verity metadata */
if (__ecc_read(f, rs.get(), data, curr * FEC_BLOCKSIZE, false,
ecc_data.get(), errors) == -1) {
return -1;
}
size_t copy = FEC_BLOCKSIZE - coff;
if (copy > left) {
copy = left;
}
memcpy(dest, &data[coff], copy);
dest += copy;
left -= copy;
coff = 0;
++curr;
}
return count;
}
/* reads `count' bytes from `offset', corrects possible errors with
erasure detection, and verifies the integrity of read data using
verity hash tree; returns the number of corrections in `errors' */
static ssize_t verity_read(fec_handle *f, uint8_t *dest, size_t count,
uint64_t offset, size_t *errors)
{
check(f);
check(dest);
check(offset < f->data_size);
check(offset + count <= f->data_size);
check(!f->hashtree().hash_data.empty());
check(errors);
debug("[%" PRIu64 ", %" PRIu64 ")", offset, offset + count);
rs_unique_ptr rs(NULL, free_rs_char);
std::unique_ptr<uint8_t[]> ecc_data;
if (f->ecc.start && ecc_init(f, rs, ecc_data) == -1) {
return -1;
}
uint64_t curr = offset / FEC_BLOCKSIZE;
size_t coff = (size_t)(offset - curr * FEC_BLOCKSIZE);
size_t left = count;
uint8_t data[FEC_BLOCKSIZE];
uint64_t max_hash_block =
(f->hashtree().hash_data.size() - SHA256_DIGEST_LENGTH) /
SHA256_DIGEST_LENGTH;
while (left > 0) {
check(curr <= max_hash_block);
uint64_t curr_offset = curr * FEC_BLOCKSIZE;
bool expect_zeros = is_zero(f, curr_offset);
/* if we are in read-only mode and expect to read a zero block,
skip reading and just return zeros */
if ((f->mode & O_ACCMODE) == O_RDONLY && expect_zeros) {
memset(data, 0, FEC_BLOCKSIZE);
goto valid;
}
/* copy raw data without error correction */
if (!raw_pread(f->fd, data, FEC_BLOCKSIZE, curr_offset)) {
error("failed to read: %s", strerror(errno));
return -1;
}
if (likely(f->hashtree().check_block_hash_with_index(curr, data))) {
goto valid;
}
/* we know the block is supposed to contain zeros, so return zeros
instead of trying to correct it */
if (expect_zeros) {
memset(data, 0, FEC_BLOCKSIZE);
goto corrected;
}
if (!f->ecc.start) {
/* fatal error without ecc */
error("[%" PRIu64 ", %" PRIu64 "): corrupted block %" PRIu64,
offset, offset + count, curr);
return -1;
} else {
debug("[%" PRIu64 ", %" PRIu64 "): corrupted block %" PRIu64,
offset, offset + count, curr);
}
/* try to correct without erasures first, because checking for
erasure locations is slower */
if (__ecc_read(f, rs.get(), data, curr_offset, false, ecc_data.get(),
errors) == FEC_BLOCKSIZE &&
f->hashtree().check_block_hash_with_index(curr, data)) {
goto corrected;
}
/* try to correct with erasures */
if (__ecc_read(f, rs.get(), data, curr_offset, true, ecc_data.get(),
errors) == FEC_BLOCKSIZE &&
f->hashtree().check_block_hash_with_index(curr, data)) {
goto corrected;
}
error("[%" PRIu64 ", %" PRIu64 "): corrupted block %" PRIu64
" (offset %" PRIu64 ") cannot be recovered",
offset, offset + count, curr, curr_offset);
dump("decoded block", curr, data, FEC_BLOCKSIZE);
errno = EIO;
return -1;
corrected:
/* update the corrected block to the file if we are in r/w mode */
if (f->mode & O_RDWR &&
!raw_pwrite(f->fd, data, FEC_BLOCKSIZE, curr_offset)) {
error("failed to write: %s", strerror(errno));
return -1;
}
valid:
size_t copy = FEC_BLOCKSIZE - coff;
if (copy > left) {
copy = left;
}
memcpy(dest, &data[coff], copy);
dest += copy;
left -= copy;
coff = 0;
++curr;
}
return count;
}
/* sets the internal file position to `offset' relative to `whence' */
int fec_seek(struct fec_handle *f, int64_t offset, int whence)
{
check(f);
if (whence == SEEK_SET) {
if (offset < 0) {
errno = EOVERFLOW;
return -1;
}
f->pos = offset;
} else if (whence == SEEK_CUR) {
if (offset < 0 && f->pos < (uint64_t)-offset) {
errno = EOVERFLOW;
return -1;
} else if (offset > 0 && (uint64_t)offset > UINT64_MAX - f->pos) {
errno = EOVERFLOW;
return -1;
}
f->pos += offset;
} else if (whence == SEEK_END) {
if (offset >= 0) {
errno = ENXIO;
return -1;
} else if ((uint64_t)-offset > f->size) {
errno = EOVERFLOW;
return -1;
}
f->pos = f->size + offset;
} else {
errno = EINVAL;
return -1;
}
return 0;
}
/* reads up to `count' bytes starting from the internal file position using
error correction and integrity validation, if available */
ssize_t fec_read(struct fec_handle *f, void *buf, size_t count)
{
ssize_t rc = fec_pread(f, buf, count, f->pos);
if (rc > 0) {
check(f->pos < UINT64_MAX - rc);
f->pos += rc;
}
return rc;
}
/* for a file with size `max', returns the number of bytes we can read starting
from `offset', up to `count' bytes */
static inline size_t get_max_count(uint64_t offset, size_t count, uint64_t max)
{
if (offset >= max) {
return 0;
} else if (offset > max - count) {
return (size_t)(max - offset);
}
return count;
}
/* reads `count' bytes from `f->fd' starting from `offset', and copies the
data to `buf' */
bool raw_pread(int fd, void *buf, size_t count, uint64_t offset) {
check(buf);
uint8_t *p = (uint8_t *)buf;
size_t remaining = count;
while (remaining > 0) {
ssize_t n = TEMP_FAILURE_RETRY(pread64(fd, p, remaining, offset));
if (n <= 0) {
return false;
}
p += n;
remaining -= n;
offset += n;
}
return true;
}
/* writes `count' bytes from `buf' to `f->fd' to a file position `offset' */
bool raw_pwrite(int fd, const void *buf, size_t count, uint64_t offset) {
check(buf);
const uint8_t *p = (const uint8_t *)buf;
size_t remaining = count;
while (remaining > 0) {
ssize_t n = TEMP_FAILURE_RETRY(pwrite64(fd, p, remaining, offset));
if (n <= 0) {
return false;
}
p += n;
remaining -= n;
offset += n;
}
return true;
}
/* reads up to `count' bytes starting from `offset' using error correction and
integrity validation, if available */
ssize_t fec_pread(struct fec_handle *f, void *buf, size_t count,
uint64_t offset)
{
check(f);
check(buf);
if (unlikely(offset > UINT64_MAX - count)) {
errno = EOVERFLOW;
return -1;
}
if (!f->hashtree().hash_data.empty()) {
return process(f, (uint8_t *)buf,
get_max_count(offset, count, f->data_size), offset,
verity_read);
} else if (f->ecc.start) {
check(f->ecc.start < f->size);
count = get_max_count(offset, count, f->data_size);
ssize_t rc = process(f, (uint8_t *)buf, count, offset, ecc_read);
if (rc >= 0) {
return rc;
}
/* return raw data if pure ecc read fails; due to interleaving
the specific blocks the caller wants may still be fine */
} else {
count = get_max_count(offset, count, f->size);
}
if (raw_pread(f->fd, buf, count, offset)) {
return count;
}
return -1;
}