aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorNicolas Pitre <nico@cam.org>2007-04-09 01:06:31 -0400
committerJunio C Hamano <junkio@cox.net>2007-04-10 12:48:14 -0700
commit78d1e84fe5586ed45740b915b52e6856bb3f1c44 (patch)
treedd19f075bfeaaa0c93d6d462f67c589ae73fa16d
parentd7dd02231f75604e388afb905f7bf8afd1bf4b24 (diff)
downloadgit-78d1e84fe5586ed45740b915b52e6856bb3f1c44.tar.gz
git-78d1e84fe5586ed45740b915b52e6856bb3f1c44.tar.xz
compute a CRC32 for each object as stored in a pack
The most important optimization for performance when repacking is the ability to reuse data from a previous pack as is and bypass any delta or even SHA1 computation by simply copying the raw data from one pack to another directly. The problem with this is that any data corruption within a copied object would go unnoticed and the new (repacked) pack would be self-consistent with its own checksum despite containing a corrupted object. This is a real issue that already happened at least once in the past. In some attempt to prevent this, we validate the copied data by inflating it and making sure no error is signaled by zlib. But this is still not perfect as a significant portion of a pack content is made of object headers and references to delta base objects which are not deflated and therefore not validated when repacking actually making the pack data reuse still not as safe as it could be. Of course a full SHA1 validation could be performed, but that implies full data inflating and delta replaying which is extremely costly, which cost the data reuse optimization was designed to avoid in the first place. So the best solution to this is simply to store a CRC32 of the raw pack data for each object in the pack index. This way any object in a pack can be validated before being copied as is in another pack, including header and any other non deflated data. Why CRC32 instead of a faster checksum like Adler32? Quoting Wikipedia: Jonathan Stone discovered in 2001 that Adler-32 has a weakness for very short messages. He wrote "Briefly, the problem is that, for very short packets, Adler32 is guaranteed to give poor coverage of the available bits. Don't take my word for it, ask Mark Adler. :-)" The problem is that sum A does not wrap for short messages. The maximum value of A for a 128-byte message is 32640, which is below the value 65521 used by the modulo operation. An extended explanation can be found in RFC 3309, which mandates the use of CRC32 instead of Adler-32 for SCTP, the Stream Control Transmission Protocol. In the context of a GIT pack, we have lots of small objects, especially deltas, which are likely to be quite small and in a size range for which Adler32 is dimed not to be sufficient. Another advantage of CRC32 is the possibility for recovery from certain types of small corruptions like single bit errors which are the most probable type of corruptions. OK what this patch does is to compute the CRC32 of each object written to a pack within pack-objects. It is not written to the index yet and it is obviously not validated when reusing pack data yet either. Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
-rw-r--r--builtin-pack-objects.c6
-rw-r--r--csum-file.c14
-rw-r--r--csum-file.h4
3 files changed, 24 insertions, 0 deletions
diff --git a/builtin-pack-objects.c b/builtin-pack-objects.c
index d0be87944..03e36f018 100644
--- a/builtin-pack-objects.c
+++ b/builtin-pack-objects.c
@@ -44,6 +44,7 @@ struct object_entry {
* be used as the base objectto delta huge
* objects against.
*/
+ uint32_t crc32; /* crc of raw pack data for this object */
};
/*
@@ -381,6 +382,9 @@ static unsigned long write_object(struct sha1file *f,
enum object_type obj_type;
int to_reuse = 0;
+ if (!pack_to_stdout)
+ crc32_begin(f);
+
obj_type = entry->type;
if (! entry->in_pack)
to_reuse = 0; /* can't reuse what we don't have */
@@ -496,6 +500,8 @@ static unsigned long write_object(struct sha1file *f,
if (entry->delta)
written_delta++;
written++;
+ if (!pack_to_stdout)
+ entry->crc32 = crc32_end(f);
return hdrlen + datalen;
}
diff --git a/csum-file.c b/csum-file.c
index b7174c6c0..7c806ada4 100644
--- a/csum-file.c
+++ b/csum-file.c
@@ -49,6 +49,8 @@ int sha1close(struct sha1file *f, unsigned char *result, int update)
int sha1write(struct sha1file *f, void *buf, unsigned int count)
{
+ if (f->do_crc)
+ f->crc32 = crc32(f->crc32, buf, count);
while (count) {
unsigned offset = f->offset;
unsigned left = sizeof(f->buffer) - offset;
@@ -91,6 +93,7 @@ struct sha1file *sha1create(const char *fmt, ...)
f->fd = fd;
f->error = 0;
f->offset = 0;
+ f->do_crc = 0;
SHA1_Init(&f->ctx);
return f;
}
@@ -111,6 +114,7 @@ struct sha1file *sha1fd(int fd, const char *name)
f->fd = fd;
f->error = 0;
f->offset = 0;
+ f->do_crc = 0;
SHA1_Init(&f->ctx);
return f;
}
@@ -143,4 +147,14 @@ int sha1write_compressed(struct sha1file *f, void *in, unsigned int size)
return size;
}
+void crc32_begin(struct sha1file *f)
+{
+ f->crc32 = crc32(0, Z_NULL, 0);
+ f->do_crc = 1;
+}
+uint32_t crc32_end(struct sha1file *f)
+{
+ f->do_crc = 0;
+ return f->crc32;
+}
diff --git a/csum-file.h b/csum-file.h
index 3ad1a992a..7e1339189 100644
--- a/csum-file.h
+++ b/csum-file.h
@@ -7,6 +7,8 @@ struct sha1file {
unsigned int offset, namelen;
SHA_CTX ctx;
char name[PATH_MAX];
+ int do_crc;
+ uint32_t crc32;
unsigned char buffer[8192];
};
@@ -15,5 +17,7 @@ extern struct sha1file *sha1create(const char *fmt, ...) __attribute__((format (
extern int sha1close(struct sha1file *, unsigned char *, int);
extern int sha1write(struct sha1file *, void *, unsigned int);
extern int sha1write_compressed(struct sha1file *, void *, unsigned int);
+extern void crc32_begin(struct sha1file *);
+extern uint32_t crc32_end(struct sha1file *);
#endif