zlog_buf_prefix() can return a larger length than what actually was
written due to its use of snprintf(). The code in
zlog_stream_prefix_ex() does not take this into account, other callers
do. What ends up happening then is that stream->length is set to the
length as if snprintf() was able to write all bytes, causing
stream->length to become larger than stream->buf.size, causing a
segfault.
In case the buffer was too small we try with a larger buffer up to a
limit of zlog_limit. This makes sure that the stream length will remain
bounded by the buffer size.
This also adds assertions to make the programmer intent clear and catch
this more easily in debug builds.
Closes GH-16680.
In the test, I have an internal `__call` function for `_ZendTestMagicCallForward` that calls the global function with name `$name` via `call_user_function`.
Note that observer writes the pointer to the previously observed frame in the last temporary of the new call frame (`*prev_observed_frame`).
The following happens:
First, we call `$test->callee`, this will be handled via a trampoline with T=2 for the two arguments. The call frame is allocated at this point. This call frame is not observed because it has `ZEND_ACC_CALL_VIA_TRAMPOLINE` set. Next we use `ZEND_CALL_TRAMPOLINE` to call the trampoline, this reuses the stack frame allocated earlier with T=2, but this time it is observed. The pointer to the previous frame is written outside of the call frame because `T` is too small (should be 3). We are now in the internal function `_ZendTestMagicCallForward::__call` where we call the global function `callee`. This will push a new call frame which will overlap `*prev_observed_frame`. This value gets overwritten by `zend_init_func_execute_data` when `EX(opline)` is set because `*prev_observed_frame` overlaps with `EX(opline)`. From now on, `*prev_observed_frame` is corrupted. When `zend_observer_fcall_end` is called this will result in reading wrong value `*prev_observed_frame` into `current_observed_frame`. This causes issues in `zend_observer_fcall_end_all` leading to the segfault we observe.
Despite function with `ZEND_ACC_CALL_VIA_TRAMPOLINE` not being observed, the reuse of call frames makes problems when `T` is not large enough.
To fix this, we make sure to add 1 to `T` if `ZEND_OBSERVER_ENABLED` is true.
Closes GH-16252.
When functions' or class methods' availability is based on some preprocessor
condition, the generated arginfo header files wrap the declarations in the
preprocessor `#if` conditional blocks, one per declaration, even if they are in
the same conditional block based on comments in the stub file. Instead of
having multiple conditional blocks one after the other with the same condition,
combine them into a single conditional block.
When a class (or enum) has no methods, rather than using an array that only
contains `ZEND_FE_END`, use `NULL` for the functions. The implementation of
class registration for internal classes, `do_register_internal_class()` in
zend_API.c, already skips classes where the functions are `NULL`. By removing
these unneeded arrays, we can reduce the size of the header files, while also
removing an unneeded call to zend_register_functions() for each internal class
with no extra methods.
Currently, internal classes are registered with the following code:
INIT_CLASS_ENTRY(ce, "InternalClass", class_InternalClass_methods);
class_entry = zend_register_internal_class_ex(&ce, NULL);
class_entry->ce_flags |= ...;
This has worked well so far, except if InternalClass is readonly. It is because some inheritance checks are run by zend_register_internal_class_ex before ZEND_ACC_READONLY_CLASS is added to ce_flags.
The issue is fixed by adding a zend_register_internal_class_with_flags() zend API function that stubs can use from now on. This function makes sure to add the flags before running any checks. Since the new API is not available in lower PHP versions, gen_stub.php has to keep support for the existing API for PHP 8.3 and below.
is_zend_ptr() expected zend_mm_heap.huge_list to be circular, but it's in fact NULL-terminated. It could crash when at least one huge block exists and the ptr did not belong to any block.
This is necessary because `zend_get_attribute_object()` will use the persistent
string with the parameter name as the index for a newly created non-persistent
HashTable, which is not legal.
As parameter names are expected to be short-ish, reasonably common terms and
need to sit around in memory anyways, we might as well make them an interned
string, circumstepping the issue without needing to duplicate the parameter
name into a non-persistent string.
Although the issue was demonstrated using Curl, the issue is purely in
the streams layer of PHP.
Full analysis is written in GH-11078 [1], but here is the brief version:
Here's what actually happens:
1) We're creating a FILE handle from a stream using the casting mechanism.
This will create a cookie-based FILE handle using funopen.
2) We're reading stream data using fread from the userspace stream. This will
temporarily set a buffer into a field _bf.base [2]. This buffer is now equal
to the upload buffer that Curl allocated and note that that buffer is owned
by Curl.
3) The fatal error occurs and we bail out from the fread function, notice how
the reset code is never executed and so the buffer will still point to
Curl's upload buffer instead of FILE's own buffer [3].
4) The resources are destroyed, this includes our opened stream and because the
FILE handle is cached, it gets destroyed as well.
In fact, the stream code calls through fclose on purpose in this case.
5) The fclose code frees the _bs.base buffer [4].
However, this is not the buffer that FILE owns but the one that Curl owns
because it isn't reset properly due to the bailout!
6) The objects are getting destroyed, and so the curl free logic is invoked.
When Curl tries to gracefully clean up, it tries to free the buffer.
But that buffer is actually already freed mistakingly by the C library!
This also explains why we can't reproduce it on Linux: this bizarre buffer
swapping only happens on macOS and BSD, not on Linux.
To solve this, we switch to an unbuffered mode for cookie-based FILEs.
This avoids any stateful problems related to buffers especially when the
bailout mechanism triggers. As streams have their own buffering
mechanism, I don't expect this to impact performance.
[1] https://github.com/php/php-src/issues/11078#issuecomment-2155616843
[2] 5e566be7a7/stdio/FreeBSD/fread.c (L102-L103)
[3] 5e566be7a7/stdio/FreeBSD/fread.c (L117)
[4] 5e566be7a7/stdio/FreeBSD/fclose.c (L66-L67)
Closes GH-14524.
Some modules may reset _fmode, which causes mangling of line endings.
Always be explicit like we do in other places where the native open call
is used.
Closes GH-14218.
- Declared compatibility expectations of stub files are now enforced by a ZEND_STATIC_ASSERT call at the top of arginfo files
- Property registration for PHP 7 is fixed: function zend_declare_property_ex() is used again instead of zend_declare_typed_property(). This has been a regression since I added support for exposing doc comments.
- As a defensive measure, deep cloning is performed before newer features (type declarations, attributes etc.) are discarded before generating legacy arginfo files. Until now, some of the objects were forgotten to be taken care of. These omissions may have resulted in some weird bugs in theory (but probably they didn't have much impact in practice).
- PHP version related conditions inside *non-legacy arginfo files* used to possibly check for the 70000 version iD until now if compatibility with PHP 7.0 was declared in a stub. This was not 100% correct, since non-legacy arginfo files are only for PHP 8.0+. Now, I made sure that at least PHP version ID 80000 is used in the preprocessor conditions. The solution was a bit tricky though...
- GH-11958: DNF types in trait properties do not get bound properly
- GH-11883: Memory leak in zend_type_release() for non-arena allocated DNF types
- Internal trait bound to userland class would not be arena allocated
- Property DNF types were not properly deep copied during lazy loading
Co-authored-by: Ilija Tovilo <ilija.tovilo@me.com>
Co-authored-by: ju1ius <jules.bernable@gmail.com>
It looks like the config.w32 uses CHECK_HEADER_ADD_INCLUDE to add the include
path to libxml into the search path.
That doesn't happen in zend-test.
To add to the Windows trouble, libxml is statically linked in, ext/libxml can
only be built statically but ext/zend-test can be built both statically and
dynamically.
So the regression tests won't work in all possible configurations anyway on Windows.
All of this is no problem on Linux because it just uses dynamic linking
and pkg-config, without any magic.
Signed-off-by: Ben Ramsey <ramsey@php.net>
