git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH 7/5] fsck: use streaming interface for large blobs in pack

From
Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Date
Jul 10, 2016, 10:45 UTC
Message-ID
<20160710104555.27478-1-pclouds@gmail.com>
In-Reply-To
<20160705170558.10906-1-pclouds@gmail.com>

For blobs, we want to make sure the on-disk data is not corrupted (i.e. can be inflated and produce the expected SHA-1). Blob content is opaque, there's nothing else inside to check for.

For really large blobs, we may want to avoid unpacking the entire blob in memory, just to check whether it produces the same SHA-1. On 32-bit systems, we may not have enough virtual address space for such memory allocation. And even on 64-bit where it's not a problem, allocating a lot more memory could result in kicking other parts of systems to swap file, generating lots of I/O and slowing everything down.

For this particular operation, not unpacking the blob and letting check_sha1_signature, which supports streaming interface, do the job is sufficient. check_sha1_signature() is not shown in the diff, unfortunately. But if will be called when "data_valid && !data" is false.

We will call the callback function "fn" with NULL as "data". The only callback of this function is fsck_obj_buffer(), which does not touch "data" at all if it's a blob.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 pack-check.c | 15 +++++++++++++--
 pack.h       |  1 +
 2 files changed, 14 insertions(+), 2 deletions(-)
diff --git a/pack-check.c b/pack-check.c
index 1da89a4..0777766 100644
--- a/pack-check.c
+++ b/pack-check.c
@@ -105,6 +105,8 @@ static int verify_packfile(struct packed_git *p,
 		void *data;
 		enum object_type type;
 		unsigned long size;
+		off_t curpos;
+		int data_valid = 0;
 
 		if (p->index_version > 1) {
 			off_t offset = entries[i].offset;
@@ -116,8 +118,17 @@ static int verify_packfile(struct packed_git *p,
 					    sha1_to_hex(entries[i].sha1),
 					    p->pack_name, (uintmax_t)offset);
 		}
-		data = unpack_entry(p, entries[i].offset, &type, &size);
-		if (!data)
+
+		curpos = entries[i].offset;
+		type = unpack_object_header(p, w_curs, &curpos, &size);
+		unuse_pack(w_curs);
+
+		if (type != OBJ_BLOB || size < big_file_threshold) {
+			data = unpack_entry(p, entries[i].offset, &type, &size);
+			data_valid = 1;
+		}
+
+		if (data_valid && !data)
 			err = error("cannot unpack %s from %s at offset %"PRIuMAX"",
 				    sha1_to_hex(entries[i].sha1), p->pack_name,
 				    (uintmax_t)entries[i].offset);
diff --git a/pack.h b/pack.h
index 3223f5a..0e77429 100644
--- a/pack.h
+++ b/pack.h
@@ -74,6 +74,7 @@ struct pack_idx_entry {
 
 
 struct progress;
+/* Note, the data argument could be NULL if object type is blob */
 typedef int (*verify_fn)(const unsigned char*, enum object_type, unsigned long, void*, int*);
 
 extern const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects, int nr_objects, const struct pack_idx_option *, const unsigned char *sha1);
-- 
2.8.2.537.g0965dd9
Previous: Nguyễn Thái Ngọc DuyNext: Junio C Hamano
Message 22 of 39 in “[bug] Reliably Reproducible Bad Packing of Objects”
  1. Christoph MichelbachJun 24, 2016
  2. Duy NguyenJul 2, 2016
  3. Duy NguyenJul 2, 2016
  4. 0/5 Number truncation with 4+ GB files on 32-bit systemsNguyễn Thái Ngọc Duy, Jul 5, 2016
  5. 1/5 pack-objects: pass length to check_pack_crc() without truncationNguyễn Thái Ngọc Duy, Jul 5, 2016
  6. Junio C HamanoJul 12, 2016
  7. 5/5 index-pack: correct "offset" type in unpack_entry_data()Nguyễn Thái Ngọc Duy, Jul 5, 2016
  8. 4/5 index-pack: report correct bad object offsets even if they are largeNguyễn Thái Ngọc Duy, Jul 5, 2016
  9. Junio C HamanoJul 12, 2016
  10. Junio C HamanoJul 12, 2016
  11. 2/5 sha1_file.c: use type off_t* for object_info->disk_sizepNguyễn Thái Ngọc Duy, Jul 5, 2016
  12. Junio C HamanoJul 12, 2016
  13. Junio C HamanoJul 12, 2016
  14. 3/5 index-pack: correct "len" type in unpack_data()Nguyễn Thái Ngọc Duy, Jul 5, 2016
  15. Johannes SixtJul 5, 2016
  16. Duy NguyenJul 6, 2016
  17. Junio C HamanoJul 6, 2016
  18. Junio C HamanoJul 6, 2016
  19. Christoph MichelbachJul 5, 2016
  20. Duy NguyenJul 10, 2016
  21. 6/5 pack-objects: do not truncate result in-pack object size on 32-bit systemsNguyễn Thái Ngọc Duy, Jul 10, 2016
  22. 7/5 fsck: use streaming interface for large blobs in packNguyễn Thái Ngọc Duy, Jul 10, 2016
  23. Junio C HamanoJul 12, 2016
  24. Junio C HamanoJul 12, 2016
  25. Junio C HamanoJul 12, 2016
  26. Junio C HamanoJul 12, 2016
  27. Junio C HamanoJul 12, 2016
  28. Duy NguyenJul 13, 2016
  29. 0/7 Number truncation with 4+ GB files on 32-bit systemsNguyễn Thái Ngọc Duy, Jul 13, 2016
  30. 4/7 index-pack: report correct bad object offsets even if they are largeNguyễn Thái Ngọc Duy, Jul 13, 2016
  31. 3/7 index-pack: correct "len" type in unpack_data()Nguyễn Thái Ngọc Duy, Jul 13, 2016
  32. 2/7 sha1_file.c: use type off_t* for object_info->disk_sizepNguyễn Thái Ngọc Duy, Jul 13, 2016
  33. 5/7 index-pack: correct "offset" type in unpack_entry_data()Nguyễn Thái Ngọc Duy, Jul 13, 2016
  34. 7/7 fsck: use streaming interface for large blobs in packNguyễn Thái Ngọc Duy, Jul 13, 2016
  35. 6/7 pack-objects: do not truncate result in-pack object size on 32-bit systemsNguyễn Thái Ngọc Duy, Jul 13, 2016
  36. 1/7 pack-objects: pass length to check_pack_crc() without truncationNguyễn Thái Ngọc Duy, Jul 13, 2016
  37. Junio C HamanoJul 13, 2016
  38. Christoph MichelbachJul 6, 2016
  39. Duy NguyenJul 6, 2016

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.