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

[PATCH 6/7] verify_path(): disallow symlinks in .gitattributes and .gitignore

From
Jeff King <peff@peff.net>
Date
Oct 5, 2020, 07:24 UTC
Message-ID
<20201005072406.GF2291074@coredump.intra.peff.net>
In-Reply-To
<20201005071751.GA2290770@coredump.intra.peff.net>

In commit 10ecfa7649 (verify_path: disallow symlinks in .gitmodules, 2018-05-04) we made it impossible to load a .gitmodules file that's a symlink into the index. The security reasons for doing so are described there. We also discussed forbidding symlinks of other .git files as part of that fix, but the tradeoff was less compelling:

  1. Unlike .gitmodules, the other files don't have content-level fsck
     checks. So an attacker using symlinks to evade those checks isn't a
     problem.
  2. Unlike .gitmodules, Git will never write .gitignore or
     .gitattributes itself, making it much less likely to use them to
     write outside the repo. They could be used for out-of-repo reads,
     however.
  3. The .gitmodules change was part of a critical bug-fix that was
     not publicly disclosed until it was released. Changing the other
     files was not needed for the minimal fix.

However, it's still a reasonable idea to forbid symlinks for these files:

  - As noted, they can still be used to read out-of-repo files (which is
    fairly restricted, but in some circumstances you can probe file
    content by speculatively creating files and seeing if they get
    ignored)
  - They don't currently behave well in all cases. We sometimes read
    these files from the index, where we _don't_ follow symlinks (we'd
    just treat the symlink target as the .gitignore or .gitattributes
    content, which is actively wrong).

This patch forbids symlinked versions of these files from entering the index. We already have helpers for obscured forms of the names from e7cb0b4455 (is_ntfs_dotgit: match other .git files, 2018-05-11) and 0fc333ba20 (is_hfs_dotgit: match other .git files, 2018-05-02), which were done as part of the series touching .gitmodules.

Signed-off-by: Jeff King <peff@peff.net>
---
I note that neither these new tests nor the existing .gitmodules ones
confirm that we catch the obscured ntfs/hfs forms in the actual code
paths (instead, we feed them to a synthetic test-tool helper in t0060).
I think that's OK, but if we wanted to be super-paranoid we could beef
up these tests with trickier names.
 read-cache.c              | 12 +++++++++---
 t/t7450-bad-meta-files.sh | 29 +++++++++++++++++++++++++++++
 2 files changed, 38 insertions(+), 3 deletions(-)
diff --git a/read-cache.c b/read-cache.c
index ecf6f68994..63aec6c35d 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -947,7 +947,9 @@ static int verify_dotfile(const char *rest, unsigned mode)
 			return 0;
 		if (S_ISLNK(mode)) {
 			rest += 3;
-			if (skip_iprefix(rest, "modules", &rest) &&
+			if ((skip_iprefix(rest, "modules", &rest) ||
+			     skip_iprefix(rest, "ignore", &rest) ||
+			     skip_iprefix(rest, "attributes", &rest)) &&
 			    (*rest == '\0' || is_dir_sep(*rest)))
 				return 0;
 		}
@@ -980,7 +982,9 @@ int verify_path(const char *path, unsigned mode)
 				if (is_hfs_dotgit(path))
 					return 0;
 				if (S_ISLNK(mode)) {
-					if (is_hfs_dotgitmodules(path))
+					if (is_hfs_dotgitmodules(path) ||
+					    is_hfs_dotgitignore(path) ||
+					    is_hfs_dotgitattributes(path))
 						return 0;
 				}
 			}
@@ -992,7 +996,9 @@ int verify_path(const char *path, unsigned mode)
 				if (is_ntfs_dotgit(path))
 					return 0;
 				if (S_ISLNK(mode)) {
-					if (is_ntfs_dotgitmodules(path))
+					if (is_ntfs_dotgitmodules(path) ||
+					    is_ntfs_dotgitignore(path) ||
+					    is_ntfs_dotgitattributes(path))
 						return 0;
 				}
 			}
diff --git a/t/t7450-bad-meta-files.sh b/t/t7450-bad-meta-files.sh
index b73985157f..6a038ed55b 100755
--- a/t/t7450-bad-meta-files.sh
+++ b/t/t7450-bad-meta-files.sh
@@ -267,4 +267,33 @@ test_expect_success 'git dirs of sibling submodules must not be nested' '
 	test_i18ngrep "is inside git dir" err
 '
 
+test_expect_success 'create repo with symlinked .gitattributes file' '
+	git init symlink-attr &&
+	target=$(echo target | git -C symlink-attr hash-object -w --stdin) &&
+	tree=$(
+		printf "120000 blob $target\t.gitattributes\n" |
+		git -C symlink-attr mktree
+	)
+'
+
+test_expect_success 'refuse to load symlinked .gitattributes into index' '
+	test_must_fail git -C symlink-attr read-tree $tree 2>err &&
+	test_i18ngrep "invalid path.*gitattributes" err
+'
+
+test_expect_success 'create repo with symlinked .gitignore file' '
+	git init symlink-ignore &&
+	target=$(echo target | git -C symlink-ignore hash-object -w --stdin) &&
+	tree=$(
+		printf "120000 blob $target\t.gitignore\n" |
+		git -C symlink-ignore mktree
+	)
+'
+
+test_expect_success 'refuse to load symlinked .gitignore into index' '
+	test_must_fail git -C symlink-ignore read-tree $tree 2>err &&
+	test_i18ngrep "invalid path.*gitignore" err
+'
+
+
 test_done
-- 
2.28.0.1295.g4824feede7
Previous: Jeff KingNext: Jonathan Nieder
Message 22 of 55 in “forbidding symlinked .gitattributes and .gitignore”
  1. 0/7 forbidding symlinked .gitattributes and .gitignoreJeff King, Oct 5, 2020
  2. 1/7 fsck_tree(): fix shadowed variableJeff King, Oct 5, 2020
  3. Jonathan NiederOct 5, 2020
  4. Jeff KingOct 5, 2020
  5. Jonathan NiederOct 5, 2020
  6. 2/7 fsck_tree(): wrap some long linesJeff King, Oct 5, 2020
  7. Jonathan NiederOct 5, 2020
  8. 3/7 t7415: rename to expand scopeJeff King, Oct 5, 2020
  9. Jonathan NiederOct 5, 2020
  10. Jeff KingOct 5, 2020
  11. Jonathan NiederOct 5, 2020
  12. Jeff KingOct 5, 2020
  13. 4/7 t7450: test verify_path() handling of gitmodulesJeff King, Oct 5, 2020
  14. Jonathan NiederOct 5, 2020
  15. Jeff KingOct 5, 2020
  16. Jonathan NiederOct 5, 2020
  17. 5/7 t0060: test obscured .gitattributes and .gitignore matchingJeff King, Oct 5, 2020
  18. Jonathan NiederOct 5, 2020
  19. Jeff KingOct 5, 2020
  20. Johannes SchindelinOct 5, 2020
  21. Jeff KingOct 6, 2020
  22. 6/7 verify_path(): disallow symlinks in .gitattributes and .gitignoreJeff King, Oct 5, 2020
  23. Jonathan NiederOct 5, 2020
  24. Jeff KingOct 5, 2020
  25. 7/7 fsck: complain when .gitattributes or .gitignore is a symlinkJeff King, Oct 5, 2020
  26. Jonathan NiederOct 5, 2020
  27. Jeff KingOct 5, 2020
  28. Jonathan NiederOct 5, 2020
  29. Jeff KingOct 5, 2020
  30. 0/8 forbidding symlinked .gitattributes and .gitignoreJeff King, Oct 5, 2020
  31. 1/8 fsck_tree(): fix shadowed variableJeff King, Oct 5, 2020
  32. 2/8 fsck_tree(): wrap some long linesJeff King, Oct 5, 2020
  33. 3/8 t7415: rename to expand scopeJeff King, Oct 5, 2020
  34. 4/8 t7450: test verify_path() handling of gitmodulesJeff King, Oct 5, 2020
  35. 5/8 t7450: test .gitmodules symlink matching against obscured namesJeff King, Oct 5, 2020
  36. 7/8 verify_path(): disallow symlinks in .gitattributes and .gitignoreJeff King, Oct 5, 2020
  37. Jonathan NiederOct 27, 2020
  38. Jeff KingOct 27, 2020
  39. Junio C HamanoOct 27, 2020
  40. Jeff KingOct 28, 2020
  41. Jonathan NiederOct 27, 2020
  42. Junio C HamanoOct 28, 2020
  43. 6/8 t0060: test obscured .gitattributes and .gitignore matchingJeff King, Oct 5, 2020
  44. 8/8 fsck: complain when .gitattributes or .gitignore is a symlinkJeff King, Oct 5, 2020
  45. Junio C HamanoOct 6, 2020
  46. Philip OakleyOct 20, 2020
  47. documentation symlink restrictions for .git* filesJeff King, Oct 23, 2020
  48. Jeff KingOct 23, 2020
  49. Philip OakleyOct 26, 2020
  50. Jeff KingOct 26, 2020
  51. Junio C HamanoOct 26, 2020
  52. Jeff KingOct 27, 2020
  53. Junio C HamanoOct 27, 2020
  54. Philip OakleyOct 27, 2020
  55. Junio C HamanoOct 28, 2020

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.