static void
dumpchunkhash(char *name, char *buf, int chunkno, int checkindex)
{
unsigned char hash[16];
MD5((unsigned char *)buf, CHUNKSIZE, hash);
printf("%s %s %d\n", spewhash(hash, 16), name, chunkno);
}
static void
dumphash(struct hashinfo *hinfo)
{
uint32_t i, total = 0;
struct hashregion *reg;
for (i = 0; i < hinfo->nregions; i++) {
reg = &hinfo->regions[i];
printf("[%u-%u]: chunk %d, hash %s\n", reg->region.start,
reg->region.start + reg->region.size - 1,
reg->chunkno, spewhash(reg->hash));
total += reg->region.size;
}
printf("TOTAL = %u\n", total);
}
static void
dumphash(char *name, struct hashinfo *hinfo)
{
uint32_t i;
struct hashregion *reg;
if (detail > 1) {
for (i = 0; i < hinfo->nregions; i++) {
reg = &hinfo->regions[i];
printf("[%u-%u]: chunk %d, hash %s\n",
reg->region.start,
reg->region.start + reg->region.size - 1,
reg->chunkno, spewhash(reg->hash));
}
}
}
/*
* Read from infd, hash the contents and compare with the hash from sig file.
* Optionally (READ_CACHE), read-ahead and cache the blocks
*/
static int
hash_and_cmp(int infd,
unsigned char *(*hashfunc)(const unsigned char *, size_t,
unsigned char *),
int hashlen, struct hashregion *hashreg, int num_reg)
{
unsigned char *bp;
size_t count, byte_size;
ssize_t cc;
off_t byte_start, retval;
unsigned char hash[HASH_MAXSIZE];
struct region hreg = hashreg->region;
int iretval;
//printf("hash_and_cmp: in -- start = %u, size = %x, num_reg = %d.\n",
// hreg.start, hreg.size, num_reg);
#ifdef READ_CACHE
static struct range cache = { 0, 0, NULL, NULL };
static char *odata = NULL;
/*
* We read the blocks here. try to optimize here by reading
* as many contguous blocks as possible (by looking thru the
* hashregions) and store the cached data's range.
* all subsequent calls that can be served from this cache are served.
* when the first request outside this data comes, we purge the cache
* (since request comes sequentially), and fetch the next bunch of
* consecutive blocks....
*/
if (hreg.start + hreg.size <= cache.start + cache.size) {
/*
* serve the request from the cache
*/
buf = cache.data + sectobytes((hreg.start - cache.start));
//printf("hash_and_cmp: fetching from cache start = %d...\n",
// sectobytes((hreg.start - cache.start)));
} else {
int i;
/*
* bad luck ! gotta hit the disk...
*/
//printf("hash_and_cmp: NOT in cache...\n");
/*
* find the contiguous blocks
*/
cache.start = hreg.start;
cache.size = hreg.size;
for (i = 0; i < num_reg - 1; i++) {
/*
* since there are NO overlaps in hashed blocks
* just check end points..
*/
if (hashreg[i].region.start + hashreg[i].region.size
!= hashreg[i+1].region.start) {
break;
}
/*
* voila ! contiguous...
*/
cache.size += hashreg[i+1].region.size;
}
byte_size = sectobytes(cache.size);
byte_start = sectobytes(cache.start);
if (cache.data) {
free(cache.data);
}
cache.data = (unsigned char *) malloc(byte_size);
if (!cache.data) {
fprintf(stderr, "hash_and_cmp: unable to malloc !\n:");
goto error;
}
bzero(cache.data, byte_size);
//printf("hash_and_cmp: gonna fetch start = %d, size = %d\n",
// cache.start, cache.size);
/*
* go fetch the blocks.
*/
retval = lseek(infd, byte_start, SEEK_SET);
// printf("BUG_DBG: hash_and_cmp(): retval = %ld,"
// " byte_start = %ld\n", retval, byte_start);
if (retval < 0) {
fprintf(stderr, "hash_and_cmp: lseek error !\n:");
goto free_error;
}
count = byte_size;
bp = cache.data;
while (count) {
TIMEOP(cc = read(infd, bp, count), time_curr_read);
if (cc < 0) {
perror("hash_and_cmp: read error -- ");
goto free_error;
}
count -= cc;
//printf("looping...%d %d\n", cc, count);
bp += cc;
}
buf = cache.data;
}
#else
/*
* Read from the disk !
*/
byte_size = sectobytes(hreg.size);
byte_start = sectobytes(hreg.start);
assert(hreg.size <= hashdatasize);
retval = lseek(infd, byte_start, SEEK_SET);
if (retval < 0) {
perror("hash_and_cmp: lseek error");
return -1;
}
count = byte_size;
bp = hashdata;
while (count > 0) {
TIMEOP(cc = read(infd, bp, count), time_curr_read);
if (cc < 0) {
perror("hash_and_cmp: read error");
return -1;
}
if (cc == 0) {
fprintf(stderr, "hash_and_cmp: unexpected EOF\n");
return -1;
}
count -= cc;
bp += cc;
}
#endif
/*
* now caculate the hash and compare it.
*/
TIMEOP(
(void)(*hashfunc)(hashdata, byte_size, hash),
time_hash);
#if 0
fprintf(stderr, "disk: %s\n", spewhash(hash));
fprintf(stderr, "sig: %s\n", spewhash(hashreg->hash));
#endif
iretval = (memcmp(hashreg->hash, hash, hashlen) != 0);
#ifdef HASHSTATS
hashstats.hash_compares++;
hashstats.hash_scompares += hreg.size;
if (!iretval) {
hashstats.hash_identical++;
hashstats.hash_sidentical += hreg.size;
}
#endif
return iretval;
#ifdef READ_CACHE
free_error:
free(cache.data);
cache.data = NULL;
error:
cache.start = 0;
cache.size = 0;
#endif
return -1;
}
static int
checkhash(char *name, struct hashinfo *hinfo)
{
uint32_t i, inbad, badstart, badsize, reportbad;
uint32_t badhashes, badchunks, lastbadchunk;
uint64_t badhashdata;
struct hashregion *reg;
int hashlen, chunkno;
unsigned char hash[HASH_MAXSIZE];
unsigned char *(*hashfunc)(const unsigned char *, unsigned long,
unsigned char *);
char *hashstr;
readbuf_t *rbuf;
size_t size;
#ifdef TIMEIT
u_int64_t sstamp, estamp;
#endif
if (startreader(name, hinfo))
return -1;
chunkno = lastbadchunk = -1;
badhashes = badchunks = inbad = reportbad = 0;
badhashdata = 0;
badstart = badsize = ~0;
switch (hinfo->hashtype) {
case HASH_TYPE_MD5:
default:
hashlen = 16;
hashfunc = MD5;
hashstr = "MD5";
break;
case HASH_TYPE_SHA1:
hashlen = 20;
hashfunc = SHA1;
hashstr = "SHA1";
break;
}
fprintf(stderr, "Checking disk contents using %s digest\n", hashstr);
for (i = 0, reg = hinfo->regions; i < hinfo->nregions; i++, reg++) {
if (chunkno != reg->chunkno) {
nchunks++;
chunkno = reg->chunkno;
}
size = sectobytes(reg->region.size);
rbuf = getblock(reg);
#ifdef TIMEIT
sstamp = rdtsc();
#endif
(void)(*hashfunc)(rbuf->data, size, hash);
#ifdef TIMEIT
estamp = rdtsc();
hcycles += (estamp - sstamp);
#endif
putblock(rbuf);
ndatabytes += size;
if (detail > 2) {
printf("[%u-%u]:\n", reg->region.start,
reg->region.start + reg->region.size - 1);
printf(" sig %s\n", spewhash(reg->hash));
printf(" disk %s\n", spewhash(hash));
}
if (memcmp(reg->hash, hash, hashlen) == 0) {
/*
* Hash is good.
* If we were in a bad stretch, be sure to dump info
*/
if (inbad)
reportbad = 1;
} else {
/*
* Hash is bad.
* If not already in a bad stretch, start one.
* If in a bad stretch, lengthen it if contig.
* Otherwise, dump the info.
*/
badhashes++;
if (chunkno != lastbadchunk) {
badchunks++;
lastbadchunk = chunkno;
}
badhashdata += size;
if (!inbad) {
inbad = 1;
badstart = reg->region.start;
badsize = reg->region.size;
} else {
if (badstart + badsize == reg->region.start)
badsize += reg->region.size;
else
reportbad = 1;
}
}
#ifdef TIMEIT
sstamp = rdtsc();
ccycles += (sstamp - estamp);
#endif
/*
* Report on a bad stretch
*/
if (reportbad) {
if (detail)
fprintf(stderr, "%s: bad hash [%u-%u]\n",
name, badstart, badstart + badsize - 1);
reportbad = inbad = 0;
}
}
/*
* Finished on a sour note, report the final bad stretch.
*/
if (inbad && detail)
fprintf(stderr, "%s: bad hash [%u-%u]\n",
name, badstart, badstart + badsize - 1);
stopreader();
nhregions = hinfo->nregions;
printf("%s: %lu chunks, %lu hashregions, %qu data bytes\n",
name, nchunks, nhregions, ndatabytes);
if (badhashes)
printf("%s: %u regions (%d chunks) had bad hashes, "
"%qu bytes affected\n",
name, badhashes, badchunks, badhashdata);
dump_readbufs();
#ifdef TIMEIT
printf("%qu bytes: read cycles: %qu, hash cycles: %qu, cmp cycles: %qu\n",
ndatabytes, rcycles, hcycles, ccycles);
#endif
return 0;
}