summaryrefslogtreecommitdiff
path: root/core/fxcrt/cfx_retain_ptr.h
diff options
context:
space:
mode:
authortsepez <tsepez@chromium.org>2017-01-17 11:05:57 -0800
committerCommit bot <commit-bot@chromium.org>2017-01-17 11:05:57 -0800
commit783a7e048c677d26aaf3884304627bbe27cff546 (patch)
treeb03feaa32114472f54855c1b702e5bfda28d2be1 /core/fxcrt/cfx_retain_ptr.h
parentb9fbe6e9af590a91ab030d2523a147e972816b32 (diff)
downloadpdfium-783a7e048c677d26aaf3884304627bbe27cff546.tar.xz
Avoid endless loop deleting CFGAS_GEFont.
It's a ref-counted class, so if we're in the destructor, the ref count has hit zero. We can't make a new ref pointer to itself here, as it will re-invoke the destructor when it goes out of scope. This should have been an obvious anti-pattern in hindsight. The object in question can't be in the m_pFontManager, since the font manager retains a reference, and we wouldn't get to this destructor while that is present. So the cleanup isn't required. Fixing this revealed a free-delete mismatch in cxfa_textlayout.cpp. I also converted to use unique_ptrs in a few places near this issue. Fixing this revealed a UAF in CFGAS_GEFont, memcpy'ing a RetainPtr is not a good idea as it doesn't bump the ref count. Also protect and friend the CFGAS_GEFont destructor, to make sure random deletes don't happen. Also kill off a const cast, and remove unnecessary conversion to retain_ptr when we already have one. TEST=look for absence of -11 in XFA corpus test logs, bots not currently noticing the segv. Argh. Review-Url: https://codereview.chromium.org/2631703003
Diffstat (limited to 'core/fxcrt/cfx_retain_ptr.h')
-rw-r--r--core/fxcrt/cfx_retain_ptr.h1
1 files changed, 1 insertions, 0 deletions
diff --git a/core/fxcrt/cfx_retain_ptr.h b/core/fxcrt/cfx_retain_ptr.h
index f70faf1464..62b26942ba 100644
--- a/core/fxcrt/cfx_retain_ptr.h
+++ b/core/fxcrt/cfx_retain_ptr.h
@@ -83,6 +83,7 @@ class CFX_Retainable {
void Retain() { ++m_nRefCount; }
void Release() {
+ ASSERT(m_nRefCount > 0);
if (--m_nRefCount == 0)
delete this;
}