<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Jul 11, 2017 at 10:49 PM, Devin Coughlin <span dir="ltr"><<a href="mailto:devin.coughlin@gmail.com" target="_blank">devin.coughlin@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word"><br><div><span class=""><blockquote type="cite"><div>On Jul 11, 2017, at 5:02 AM, Malhar Thakkar via cfe-dev <<a href="mailto:cfe-dev@lists.llvm.org" target="_blank">cfe-dev@lists.llvm.org</a>> wrote:</div><br class="m_4204122371872823108Apple-interchange-newline"><div><div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Jul 11, 2017 at 5:03 PM, Sven Verdoolaege <span dir="ltr"><<a href="mailto:skimo-cfe@kotnet.org" target="_blank">skimo-cfe@kotnet.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><span class="m_4204122371872823108gmail-">On Tue, Jul 11, 2017 at 10:33:06AM +0530, Malhar Thakkar wrote:<br>
> Hence, although evalCall() works perfectly for ISL, we may not be able to<br>
> generalize it for other C codebases.<br>
<br>
</span>I think it's reasonable to assume that frameworks that shield off<br>
free for reference counting, would also shield off malloc<br>
in order to initialize the reference counting.<br>
Of course, this may just be a lack of imagination on my part.<br>
Do you have any examples of frameworks that could use your<br>
annotations where this is not the case?<br></blockquote><div>Well, I haven't come across/thought of any such codebase which doesn't shield off malloc which is why I created a hypothetical test case. Now that you mention it, it does seem reasonable to assume that frameworks would shield off malloc as well. </div></div></div></div></div></blockquote><div><br></div></span><div>There are some frameworks/idioms that allow a transfer of ownership from a raw malloc’d pointer to a referenced-counted opaque pointer container (this is common in C++ where you use new to allocate and initialize an object). However, I think the ‘trusted implementation’ annotation could be extended to handle this as well. It would require the function that transfers ownership to the reference-counted implementation to be annotated, but I don’t think that is a high burden. That said, I wouldn’t worry about raw malloc’d pointer case for now.</div><span class=""><div><br></div><br><blockquote type="cite"><div><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div>Also, keeping MallocChecker aside, there are a lot of other checkers which may create some issues if there are additional side-effects (as Dr. Artem mentioned) in such annotated functions. However, I guess it may be safe to assume that there are no such additional side-effects in such "trusted" functions.</div></div></div></div></div></blockquote><div><br></div></span><div>I suspect that many of these side effects would have high-level invariants that the analyzer on its own wouldn’t be able to discover (for example, that a set created with a single item in it is not empty) — so custom modeling would be needed in any event, even if the calls were inlined.</div><div><br></div><div>Devin</div><span class=""><div><br></div><br><blockquote type="cite"><div><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
skimo<br></blockquote><div><br></div><div>Now, I was experimenting a bit more with evalCall-ing based on annotations and although this works like a charm for functions of the type obj_free() and obj_cow(), it is unable to avoid the problems created by obj_copy(). This is probably because of the lack of a core-foundation annotation which is analogous to isl_keep.<br></div><div>Consider the following example.</div><div><br></div><div><div></div><br><div><div class="m_4204122371872823108gmail-cpp m_4204122371872823108gmail-hljs" style="display:block;overflow-x:auto;padding:0.5em;background:rgb(248,248,248)"><div><font color="#1f7199" face="monospace">#define __isl_give __attribute__((cf_returns_<wbr>retained))</font></div><div><font color="#1f7199" face="monospace">#define __isl_take __attribute__((cf_consumed))</font></div><div><font color="#1f7199" face="monospace">#define __isl_null</font></div><div><font color="#1f7199" face="monospace">#define __isl_keep</font></div><div><font color="#1f7199" face="monospace">#define NULL 0</font></div><div><font color="#1f7199" face="monospace"><br></font></div><div><font color="#1f7199" face="monospace">typedef struct</font></div><div><font color="#1f7199" face="monospace">{</font></div><div><font color="#1f7199" face="monospace"> int ref;</font></div><div><font color="#1f7199" face="monospace">} isl_basic_map;</font></div><div><font color="#1f7199" face="monospace"><br></font></div><div><font color="#1f7199" face="monospace">__isl_give isl_basic_map *isl_basic_map_cow(__isl_take isl_basic_map *bmap);</font></div><div><font color="#1f7199" face="monospace">__isl_null isl_basic_map *isl_basic_map_free(__isl_take isl_basic_map *bmap);</font></div><div><font color="#1f7199" face="monospace"><br></font></div><div><font color="#1f7199" face="monospace">__attribute__((annotate("rc_<wbr>ownership_trusted_<wbr>implementation"))) __isl_give isl_basic_map *isl_basic_map_copy(__isl_keep isl_basic_map *bmap)</font></div><div><font color="#1f7199" face="monospace">{</font></div><div><font color="#1f7199" face="monospace"> if (!bmap)</font></div><div><font color="#1f7199" face="monospace"> return NULL;</font></div><div><font color="#1f7199" face="monospace"><br></font></div><div><font color="#1f7199" face="monospace"> bmap->ref++;</font></div><div><font color="#1f7199" face="monospace"> return bmap;</font></div><div><font color="#1f7199" face="monospace">}</font></div><div><font color="#1f7199" face="monospace"><br></font></div><div><font color="#1f7199" face="monospace">void test_use_after_release_with_<wbr>trusted_implementation_<wbr>annotate_attribute(__isl_take isl_basic_map *bmap) {</font></div><div><font color="#1f7199" face="monospace"> bmap = isl_basic_map_cow(bmap);</font></div><div><font color="#1f7199" face="monospace"> isl_basic_map *temp = isl_basic_map_cow(isl_basic_<wbr>map_copy(bmap)); <b>// Here, the analyzer states "Object released".</b></font></div><div><font color="#1f7199" face="monospace"> isl_basic_map *temp2 = isl_basic_map_cow(bmap); <b>// Use-after-release for 'bmap' raised here.</b></font></div><div><font color="#1f7199" face="monospace"> isl_basic_map_free(temp2);</font></div><div><font color="#1f7199" face="monospace"> isl_basic_map_free(temp);</font></div><div><font color="#1f7199" face="monospace">}</font></div></div></div></div></div></div></div></div></blockquote><div><br></div></span><div>I don’t understand what is going on in the above. Can you please simplify the example and explain what is going on and what you expected the analyzer to do? What call is the analyzer complaining about? Is it inlining the call? Please present the simplest example that illustrates the problem.</div></div></div></blockquote><div><br></div><div></div><br><div><div class="hljs cpp" style="display:block;overflow-x:auto;padding:0.5em;color:rgb(51,51,51);background:rgb(248,248,248);font-family:monospace"><div><span class="hljs-meta" style="color:rgb(31,113,153)">#<span class="hljs-meta-keyword" style="font-weight:bold">define</span> __isl_keep</span></div><div><span class="hljs-meta" style="color:rgb(31,113,153)">isl_basic_map *foo(__isl_keep isl_basic_map *bmap);</span></div></div></div><div class="gmail_quote"><br></div>Consider the above code snippet. Now, I feel that when a parameter has no annotation associated to it in some function (foo, in this case), the analyzer assumes that the object returned by that function (foo) points to the same memory location as that pointed to by the argument passed to it (bmap). Please correct me if I'm wrong.</div><div class="gmail_quote"><br></div><div class="gmail_quote">This is probably the reason why, in the example specified in my previous email, the analyzer decrements the reference count for 'bmap' when the object returned from 'isl_basic_map_copy' is passed as __isl_take (cf_consumed) to 'isl_basic_map_cow'. This in turn results in a 'use-after-release' warning on the next line.<br><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word"><div><span class=""><div><br></div><blockquote type="cite"><div><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><div></div></div><div>Hence, I need more clarity as to what assumptions the checker/analyzer makes when it doesn't see any annotation associated with an object.</div></div></div></div></div></blockquote><div><br></div></span><div>The default assumption (when no annotation is present) is the equivalent of __isl_keep.</div><span class="HOEnZb"><font color="#888888"><div><br></div></font></span></div><span class="HOEnZb"><font color="#888888">Devin</font></span></div></blockquote></div><br></div></div><div hspace="streak-pt-mark" style="max-height:1px"><img alt="" style="width:0px;max-height:0px;overflow:hidden" src="https://mailfoogae.appspot.com/t?sender=aY3MxM2IxMDMxQGlpdGguYWMuaW4%3D&type=zerocontent&guid=738c4a06-9f25-4cb9-b4c5-efeb7ca20cab"><font color="#ffffff" size="1">ᐧ</font></div>