summaryrefslogtreecommitdiff
path: root/lib
diff options
context:
space:
mode:
authorDavid Howells <dhowells@redhat.com>2019-02-14 19:20:15 +0300
committerJames Morris <james.morris@microsoft.com>2019-02-16 01:12:08 +0300
commitbb2ba2d75a2d673e76ddaf13a9bd30d6a8b1bb08 (patch)
tree9f3a8e081a191743d29454281be65d32792ec974 /lib
parenta08bf91ce28ed3ae7b6fef35d843fef8dc8c2cd9 (diff)
downloadlinux-bb2ba2d75a2d673e76ddaf13a9bd30d6a8b1bb08.tar.xz
assoc_array: Fix shortcut creation
Fix the creation of shortcuts for which the length of the index key value is an exact multiple of the machine word size. The problem is that the code that blanks off the unused bits of the shortcut value malfunctions if the number of bits in the last word equals machine word size. This is due to the "<<" operator being given a shift of zero in this case, and so the mask that should be all zeros is all ones instead. This causes the subsequent masking operation to clear everything rather than clearing nothing. Ordinarily, the presence of the hash at the beginning of the tree index key makes the issue very hard to test for, but in this case, it was encountered due to a development mistake that caused the hash output to be either 0 (keyring) or 1 (non-keyring) only. This made it susceptible to the keyctl/unlink/valid test in the keyutils package. The fix is simply to skip the blanking if the shift would be 0. For example, an index key that is 64 bits long would produce a 0 shift and thus a 'blank' of all 1s. This would then be inverted and AND'd onto the index_key, incorrectly clearing the entire last word. Fixes: 3cb989501c26 ("Add a generic associative array implementation.") Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: James Morris <james.morris@microsoft.com>
Diffstat (limited to 'lib')
-rw-r--r--lib/assoc_array.c8
1 files changed, 5 insertions, 3 deletions
diff --git a/lib/assoc_array.c b/lib/assoc_array.c
index c6659cb37033..59875eb278ea 100644
--- a/lib/assoc_array.c
+++ b/lib/assoc_array.c
@@ -768,9 +768,11 @@ all_leaves_cluster_together:
new_s0->index_key[i] =
ops->get_key_chunk(index_key, i * ASSOC_ARRAY_KEY_CHUNK_SIZE);
- blank = ULONG_MAX << (level & ASSOC_ARRAY_KEY_CHUNK_MASK);
- pr_devel("blank off [%zu] %d: %lx\n", keylen - 1, level, blank);
- new_s0->index_key[keylen - 1] &= ~blank;
+ if (level & ASSOC_ARRAY_KEY_CHUNK_MASK) {
+ blank = ULONG_MAX << (level & ASSOC_ARRAY_KEY_CHUNK_MASK);
+ pr_devel("blank off [%zu] %d: %lx\n", keylen - 1, level, blank);
+ new_s0->index_key[keylen - 1] &= ~blank;
+ }
/* This now reduces to a node splitting exercise for which we'll need
* to regenerate the disparity table.