diff options
| author | Robin Watts <robin.watts@artifex.com> | 2012-03-06 19:34:06 +0000 |
|---|---|---|
| committer | Robin Watts <robin@ghostscript.com> | 2012-03-06 19:49:48 +0000 |
| commit | dae93920d513842f6c9a96e833043f3f6f8e0681 (patch) | |
| tree | a54fb901c5cab53fbbf97deced91d1db75984654 /fitz | |
| parent | a958c1e2effd97d8f17779ab975d72ee67b1ab09 (diff) | |
| download | mupdf-dae93920d513842f6c9a96e833043f3f6f8e0681.tar.xz | |
Fix ref counting bugs in race condition correction code.
When we attempt to insert a key/value pair into the store, we have
to allow for the possibility that a racing thread may have already
inserted an equivalent key/value. We have special code in place to
handle this eventuality; if we spot an existing entry, we take the
existing one in preference to our new key/value pair.
This means that fz_store_item needs to take a new reference to any
existing thing it finds before returning it.
Currently the only store user that is exposed to this possibility
is pdf_image; it spots an existing tile being returned, and was
inadvertently double freeing the key.
Diffstat (limited to 'fitz')
| -rw-r--r-- | fitz/res_store.c | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/fitz/res_store.c b/fitz/res_store.c index 2a7680e3..cc635052 100644 --- a/fitz/res_store.c +++ b/fitz/res_store.c @@ -266,14 +266,15 @@ fz_store_item(fz_context *ctx, void *key, void *val_, unsigned int itemsize, fz_ } if (existing) { + /* Take a new reference */ + existing->val->refs++; fz_unlock(ctx, FZ_LOCK_ALLOC); fz_free(ctx, item); return existing->val; } } /* Now we can never fail, bump the ref */ - if (val->refs > 0) - val->refs++; + val->refs++; /* Regardless of whether it's indexed, it goes into the linked list */ item->next = store->head; if (item->next) |