ElliotNB helped me a lot debugging this by constantly testing the
patches. It is only fair that he is mentioned too, as I couldn't have
solved it without his help.
The TSRM keeps a hashtable mapping the thread IDs to the thread resource pointers.
It's possible that the thread disappears without us knowing, and then another thread
gets spawned some time later with the same ID as the disappeared thread.
Note that since it's a new thread the TSRM key pointer and cached pointer will be NULL.
The Apache request handler `php_handler()` will try to fetch some fields from the SAPI globals.
It uses a lazy thread resource allocation by calling `ts_resource(0);`.
This allocates a thread resource and sets up the TSRM pointers if they haven't been set up yet.
At least, that's what's supposed to happen. But since we are in a situation where the thread ID
still has the resources of the *old* thread associated in the hashtable,
the loop in `ts_resource_ex` will find that thread resource and assume the thread has been setup
already. But this is not the case since this thread is actually a new thread, just reusing the ID
of the old one, without any relation whatsoever to the old thread.
Because of this assumption, the TSRM pointers will not be setup, leading to a
NULL pointer dereference when trying to access the SAPI globals.
We can easily detect this scenario: if we're in the fallback path, and the pointer is NULL,
and we're looking for our own thread resource, we know we're actually reusing a thread ID.
In that case, we'll free up the old thread resources gracefully (gracefully because
there might still be resources open like database connection which need to be
shut down cleanly). After freeing the resources, we'll create the new resources for
this thread as if the stale resources never existed in the first place.
From that point forward, it is as if that situation never occurred.
The fact that this situation happens isn't that bad because a child process containing
threads will eventually be respawned anyway by the SAPI, so the stale thread resources
won't remain forever.
Note that we can't simply assign our own TSRM pointers to the existing
thread resource for our ID, since it was actually from a different thread
(just with the same ID!). Furthermore, the dynamically loaded extensions
have their own pointer, which is only set when their constructor is
called, so we'd have to call their constructor anyway...
I also tried to call the dtor and then the ctor again for those resources
on the pre-existing thread resource to reuse storage, but that didn't work properly
because other code doesn't expect something like that to happen, which breaks assumptions,
and this in turn caused Valgrind to (rightfully) complain about memory bugs.
Note 2: I also had to fix a bug in the core globals destruction because it
always assumed that the thread destroying them was the owning thread,
which on TSRM shutdown isn't always the case. A similar bug was fixed
recently with the JIT globals.
Closes GH-10863.
Don't misinterpret DJI info maker note as DJI maker note.
The DJI and DJI info maker note both share the "DJI" make string.
This caused the current code to try to interpret the DJI info maker note
as a DJI maker note. However, the DJI info maker note requires custom
parsing. Therefore, the misinterpretation actually caused the current
code to believe that there was an unrecoverable error in the IFD for the
maker note by returning false in the maker note parser. This in turn
caused the inability to parse other EXIF metadata.
This patch adds the identification of the DJI info maker note so that it
cannot be misinterpreted. Since we don't implement custom parsing, it
achieves this by setting the tag list to a special marker value (in this
case the NULL pointer). When this marker value is detected, the function
will just skip parsing the maker note and return true. Therefore, the
other code will believe that the IFD is not corrupt.
This approach is similar to handing an unrecognised maker note type
(see the loop on top of exif_process_IFD_in_MAKERNOTE() which also
returns true and treats it as a string). The end result of this patch
is that the DJI info maker note is considered as unknown to the caller of
exif_process_IFD_in_MAKERNOTE(), and therefore that the other EXIF
metadata can be parsed successfully.
Also fix debug output typos in exif.
Closes GH-10470.
Discovered this pre-existing problem while testing GH-10682.
Note: this problem existed *before* that PR.
* Not all paths throw a hierarchy request error
* xmlFreeNode must be used instead of xmlFree for the fragment to also
free its children.
* Free up nodes that couldn't be added when xmlAddChild fails.
I unified the error handling code that's exactly the same with a goto to
prevent at least some of such problems in the future.
Closes GH-10981.
It's actually not php-cli specific, nor SAPI specific.
We should delay the registration of the function into the function table
until after the compilation was successful, otherwise the function is
mistakingly registered and a NULL dereference will happen when trying to
call it.
I based my test of Nikita's test, so credits to him for the test:
https://github.com/php/php-src/pull/8933#issuecomment-1259881008
Closes GH-10989.
The ZVAL_ARR macro always set the zval type_info to IS_ARRAY_EX, even if the
hash table is immutable. Since in preg_replace_callback_array() we can return
the passed array directly, and that passed array can be immutable, we need to
reset the type_flags to keep the VM from performing ref-counting on the array.
Fixes GH-10968
Closes GH-10970
atoi()'s return value is actually undefined when an underflow or
overflow occurs. For example on 32-bit on my system the overflow test
which inputs "h2147483648" results in repetitions==2147483647 and on
64-bit this gives repetitions==-2147483648. The reason the test works on
32-bit is because there's a second undefined behaviour problem:
in case 'h' when repetitions==2147483647, we add 1 and divide by 2.
This is signed-wrap undefined behaviour and accidentally triggers the
overflow check like we wanted to.
Avoid all this trouble and use strtol with explicit error checking.
This also fixes a semantic bug where repetitions==INT_MAX would result
in the overflow check to trigger, even though there is no overflow.
Closes GH-10943.
The recent clang-16 throws errors for implicitly defined functions by
default. In many ./configure tests, an undefined function (which is
"implicitly defined" when you try to call it) is undefined because it
really does not exist. But in one case, utf8_to_mutf7() is undefined
because we forgot to include the header that defines it.
This commit updates the test for utf8_to_mutf7:
* We now include the header (c-client.h) that defines it.
* A "checking... yes/no" message was added to the test.
* The test was switched from PHP_IMAP_TEST_BUILD to AC_COMPILE_IFELSE.
This was the easiest way to avoid a return-type mismatch that runs
afoul of -Werror=implicit-int.
* CPPFLAGS is temporarily amended with the -I flag needed to find
c-client.h.
Fixes GH-10947.
Closes GH-10948
Signed-off-by: George Peter Banyard <girgias@php.net>
The alignment of sqldata is in most cases only the basic alignment,
so the code type-puns it to a larger type, it *can* crash due to the
misaligned access. This is only an issue for types > 4 bytes because
every sensible system requires an alignment of at least 4 bytes for
allocated data.
Even though this patch uses memcpy, the compiler is smart enough to
optimise it to something more efficient, especially on x86.
This is just the usual approach to solve these alignment problems.
Actually, unaligned memory access is undefined behaviour, so even on x86
platforms, where the bug doesn't cause a crash, this can be problematic.
Furthermore, even though the issue talks about a 64-bit kernel and
32-bit userspace, this doesn't necessarily need to be the case to
trigger this crash.
Test was Co-authored-by: rvk01
Closes GH-10920.
The stream context inside `mysqlnd_vio::enable_ssl()` is leaking.
In particular: when `php_stream_context_set()` get called the refcount
of `context` is increased by 1, which means that `context` will now
have a refcount of 2. Later on we remove the context from the stream
by calling `php_stream_context_set(stream, NULL)` but that leaves our
`context` with a refcount of 1, and therefore it's never destroyed.
In my test case this yielded a leak of 1456 bytes per connection
(but could be more depending on your settings ofc).
Annoyingly, Valgrind doesn't find it because the context is still
in the `EG(regular_list)` and will thus be destroyed at the end of
the request. However, I still think this bug needs to be fixed because
as the users in the issue report already mentioned:
there can be long-running PHP scripts.
Fix it by decreasing the refcount to transfer the ownership.
Closes GH-10909.
The char arrays were too small for a long on 64-bit systems, which
resulted in cutting off the string at the end with a NUL byte. Use a
size of MAX_LENGTH_OF_LONG to fix this issue instead of a fixed size
of 11 chars.
Closes GH-10525.
get_browser() implements a lazy parse system for the browscap
INI configuration. There are two possible moments when a browscap
configuration can be loaded: during module startup or during request.
In case of module startup, the strings are persistent strings, while for
the request they are not.
The INI parser must therefore know whether to create persistent or
non-persistent strings. It does this by looking at
CG(ini_parser_unbuffered_errors). If that value is 1 it's persistent,
otherwise non-persistent. Note that this also controls how the errors
are reported: if it's 1 then the errors are sent to stderr, otherwise we
get E_WARNINGs.
Currently, a hardcoded value of 1 is always used for that CG value in
browscap_read_file(). This means we'll always create persistent strings
*and* we'll not report parse errors correctly as E_WARNINGs.
We fix both the crash and the lack of warnings by passing the value of
persistent instead of a hardcoded 1.
This is also in line with how other INI parsing code is called in
ext/standard: they also make sure that during request a value of 0 is
passed.
Closes GH-10883.
This happens when there are spaces are in the path info. The reason is
that Apache decodes the path info part in the SCRIPT_NAME as per CGI
RFC. FPM tries to strip path info from the SCRIPT_NAME but the
comparison is done against SCRIPT_FILENAME which is not decoded. For
that to work we have to decode it before comparison if there is any
encoded character.
Closes GH-10869
Fixes GH-8789.
Fixes GH-10015.
This is one small part of the underlying bug for GH-10737, as in my
attempts to reproduce the issue I constantly hit this crash easily.
(The fix for the other underlying issue for that bug will follow soon.)
It's possible that a signal arrives at a thread that never handled a PHP
request before. This causes the signal globals to dereference a NULL
pointer because the TSRM pointers for the thread aren't set up to point
to the thread resources yet.
PR GH-9766 previously fixed this for master by ignoring the signal if
the thread didn't handle a PHP request yet. While this fixes the crash
bug, I think the solution is suboptimal for 3 reasons:
1) The signal is ignored and a message is printed saying there is a bug.
However, this is not a bug at all. For example in Apache, the signal
set up happens on child process creation, and the thread resource
creation happens lazily when the first request is handled by the
thread. Hence, the fact that the thread resources aren't set up yet
is not actually buggy behaviour.
2) I believe since it was believed to be buggy behaviour, that fix was
only applied to master, so 8.1 & 8.2 keep on crashing.
3) We can do better than ignoring the signal. By just acting in the
same way as if the signals aren't active. This means we need to
take the same path as if the TSRM had already shut down.
Closes GH-10861.
* Missing check: SQLAllocHandle() for the environment wasn't checked in
pdo_odbc_handle_factory(). Add a check similar to the other ones for
SQLAllocHandle().
* Inconsistent check: one of the SQLAllocHandle() calls wasn't checked
for SQL_SUCCESS_WITH_INFO. However, looking at the other uses and the
documentation we should probably check this as well.
Furthermore, since there was a mix of "SQLAllocHandle: reason" and
"SQLAllocHandle (reason)" in the error reporting, I made them
consistently use the first option as that seems to be the most used for
error reporting in this file.
Closes GH-10740.
Fixes GH-10801
Named arguments are not supported by the constant evaluation routine, in
the sense that they are ignored. This causes two issues:
- It causes a crash because not all oplines belonging to the call are
removed, which results in SEND_VA{L,R} which should've been removed.
- It causes semantic issues (demonstrated in the test case).
This case never worked anyway, leading to crashes or incorrect behaviour,
so just prevent CTE of calls with named parameters for now.
We can choose to support it later, but introducing support for this in
a stable branch seems too dangerous.
This patch does not change the removal of SEND_* opcodes in remove_call
because the crash bug can't be triggered anymore with this patch as
there are no named parameters anymore and no variadic CTE functions
exist.
Closes GH-10811.
We need to carry around a reference to the underlying Bucket to be able to modify it by reference.
Closes GH-10749
Signed-off-by: George Peter Banyard <girgias@php.net>
Disable opcache.consistency_checks.
This feature does not work right now and leads to memory leaks and other
problems. For analysis and discussion see GH-8065. In GH-10624 it was
decided to disable the feature to prevent problems for end users.
If end users which to get some consistency guarantees, they can rely on
opcache.protect_memory.
Closes GH-10798.
Fixes GH-8646
See https://github.com/php/php-src/issues/8646 for thorough discussion.
Interned strings that hold class entries can get a corresponding slot in map_ptr for the CE cache.
map_ptr works like a bump allocator: there is a counter which increases to allocate the next slot in the map.
For class name strings in non-opcache we have:
- on startup: permanent + interned
- on request: interned
For class name strings in opcache we have:
- on startup: permanent + interned
- on request: either not interned at all, which we can ignore because they won't get a CE cache entry
or they were already permanent + interned
or we get a new permanent + interned string in the opcache persistence code
Notice that the map_ptr layout always has the permanent strings first, and the request strings after.
In non-opcache, a request string may get a slot in map_ptr, and that interned request string
gets destroyed at the end of the request. The corresponding map_ptr slot can thereafter never be used again.
This causes map_ptr to keep reallocating to larger and larger sizes.
We solve it as follows:
We can check whether we had any interned request strings, which only happens in non-opcache.
If we have any, we reset map_ptr to the last permanent string.
We can't lose any permanent strings because of map_ptr's layout.
Closes GH-10783.
Due to an incorrect check, the datetime was never actually set.
To test this we need to write the file using phar, but read the file
using a different method to not get a cached, or a value that's been
transformed twice and is therefore accidentally correct.
Closes GH-10769
The docs say that this function returns true on success, and false on
error. This function always returns true in the current implementation
because the success return value from ftp_close() is never propagated to
userland. This affects one test: since the test server exits after an
invalid login, the ftp close correctly fails (because the server has
gone away).
Remove capstone include folder.
For most of the supported systems it worked fine somehow despite
the pkg-config --cflags, but is always include it even on Linux.
Closes GH-10732.
Fixes GH-10715
When a string starting with a NUL character is passed to
phpdbg_vprint(), the vasprintf() will return that 0 characters have been
printed. This causes msglen == 0. When phpdbg_process_print() is called
with a message of length 0, the -1 to check for '\n' will perform an out
of bounds read. Since nothing is printed anyway for msglen == 0, it
seems best to just skip the printing routine for this case.
Closes GH-10720.
SSL_CTX_set_tmp_dh() and SSL_CTX_set0_tmp_dh_pkey() return 1 on success
and 0 on error. But only < 0 was checked which means that errors were
never caught.
Closes GH-10705.
Fixes GH-10692
php_fopen_primary_script() does not initialize all fields of
zend_file_handle. So when it fails and when fastcgi is true, the
zend_destroy_file_handle() function will try to free uninitialized
pointers, causing a segmentation fault. Fix it by zero-initializing file
handles just like the zend_stream_init_fp() counterpart does.
Closes GH-10697.
zend_update_static_property_ex() returns a zend_result, but the return
value is stored here in a bool. A bool is unsigned on my system, so in
case zend_update_static_property_ex() returns FAILURE (== -1) this gets
converted to 1 instead. This is not a valid zend_result value. This
means that (transitive) callers could mistakingly think the function
succeeded while it did in fact not succeed. Fix it by changing the type
to zend_result.
Closes GH-10691.
