For some reason, qpdf from the beginning was replacing indirect
references to null with literal null in arrays even after removing the
old behavior of flattening scalar references. This seems like a bad
idea.
This change works around STL problems with Embarcadero C++ Builder
version 10.2, but std::vector is more common than std::list in qpdf,
and this is a relatively new API, so an API change is tolerable.
Thanks to Thorsten Schöning <6223655+ams-tschoening@users.noreply.github.com>
for the fix.
Use PointerHolder in several places where manually memory allocation
and deallocation were being used. This helps to protect against memory
leaks when exceptions are thrown in surprising places.
This makes all integer type conversions that have potential data loss
explicit with calls that do range checks and raise an exception. After
this commit, qpdf builds with no warnings when -Wsign-conversion
-Wconversion is used with gcc or clang or when -W3 -Wd4800 is used
with MSVC. This significantly reduces the likelihood of potential
crashes from bogus integer values.
There are some parts of the code that take int when they should take
size_t or an offset. Such places would make qpdf not support files
with more than 2^31 of something that usually wouldn't be so large. In
the event that such a file shows up and is valid, at least qpdf would
raise an error in the right spot so the issue could be legitimately
addressed rather than failing in some weird way because of a silent
overflow condition.
On read, ignore /DecodeParms when empty list; on write, delete it.
Some files have been found that include an empty list for
/DecodeParms, but this is not technically compliant with the spec, and
the only sensible interpretation is to treat it as if there are no
decode parameters.
The preservation of outlines didn't provide very useful behavior
anyway as it copied all outlines but most didn't work. This
implementation also caused a very significant performance hit and so
is being reverted until a proper solution can be coded. The eventual
solution will not be compatible with the reverted solution anyway, so
it's best not to leave this in.
We've actually seen a PDF file in the wild that contained EI
surrounded by delimiters inside the image data, which confused qpdf's
naive code. This significantly improves EI detection.
When linearizing a file or getting the list of all pages in a file,
detect if the pages tree contains a duplicated page object and, if so,
shallow copy it. This makes it possible to have a one to one mapping
of page positions to page objects.
When generating appearance streams for variable text annotations,
properly handle the cases of there being no appearance dictionary, no
appearance stream, or an appearance stream with no BMC..EMC marker.
With the exception of form field annotations when /NeedAppearances is
true, remove annotations that don't have appearance streams when
flattening. There is no reason to keep these when flattening since
they are invisible. This may include unchecked checkboxes, unshown
popup windows, etc.
Allow fine control over how passwords are encoded for writing, and
allow password for reading to be given as a hexademical encoded
string. Allow suppression of password recovery as a means to ensure
that the password you specify is actually the right one.
Setting encryption permissions for R >= 3 set permission bits in
groups corresponding to menu options in Acrobat 5. The new API allows
the bits to be set individually.
If set, we avoid using Windows I/O HANDLE, which is disallowed in some
versions of the Windows SDK, such as for Windows phones.
QUtil::same_file will always return false in this case. Only applies
to Windows builds.
The original QPDF is only required now when the source
QPDFObjectHandle is a stream that gets its stream data from a
QPDFObjectHandle::StreamDataProvider.
Instead of calling assert for problems found during checking
linearization data, throw an exception which is later caught and
issued as an error. Ideally we would handle errors more robustly, but
this is still a significant improvement.
On certain operations, such as iterating through all objects and
adding new indirect objects, walk through the entire object structure
and explicitly resolve any indirect references to non-existent
objects. That prevents new objects from springing into existence and
causing the previously dangling references to point to them.
Instead of directly putting the contents of the annotation appearance
streams into the page's content stream, add commands to render the
form xobjects directly. This is a more robust way to do it than the
original solution as it works properly with patterns and avoids
problems with resource name clashes between the pages and the form
xobjects.
Flatten annotations by integrating their appearance streams into the
content stream of the containing page. In the case of form fields,
only flatten if /NeedAppearance is false (or equivalently absent). If
flattening form fields, also remove /AcroForm from the document
catalog.
Rather than keeping a list of buffers for every write, accumulate
bytes in a single buffer, doubling the size of the buffer when needed
to accommodate new data.
This is not the best possible implementation, but the change was
implemented in this way to avoid changing the shape of Pl_Buffer and
thus breaking backward compatibility.
If we are unable to filter a page's content streams, don't attempt to
remove objects from the page's resource dictionary. Also provide a
command line option to suppress resource removal in case we ever need
this as a workaround for some bug or broken PDF files.
This is the beginning of higher-level API support using helper
classes. The goal is to be able to add more helpers without continuing
to pollute QPDF's and QPDFObjectHandle's public interfaces.
Remove calls to assertPageObject(). All cases in the library that
called assertPageObject() work fine if you don't call
assertPageObject() because nothing assumes anything that was being
checked by that call. Removing the calls enables more files to be
successfully processed.
Prior to this fix, if there was a loop detected in following /Prev
pointers in xref streams/tables, it would cause qpdf to lose data.
Note that this condition causes many PDF readers to hang or fail.
The QPDF_String::getUTF8Val() method was not treating strings that
weren't explicitly Unicode as PDF Doc Encoded. This only affects
characters in the range 0x80 through 0xa0.
Implement a TokenFilter class and refactor Pl_QPDFTokenizer to use a
TokenFilter class called ContentNormalizer. Pl_QPDFTokenizer is now a
general filter that passes data through a TokenFilter.
Adding a trailing newline in content normalization damages files whose
contents are split across streams in the middle of tokens. Let
QPDFWriter add the newline with the indicator to ignore the newline,
which it already does. This changes the way some qdf files look.
Significant enhancements to the lexer to improve EOF handling and to
support comments and spaces as tokens. Various other minor issues were
fixed as well.
Add options to enable the raw encryption key to be directly shown or
specified. Thanks to Didier Stevens <didier.stevens@gmail.com> for the
idea and contribution of one implementation of this idea.
Make sure to link from the source tree before linking from the system.
In many environments, this is necessary to allow a newly built qpdf to
link properly instead of trying to link or resolve libraries from an
older installed version.
This commit adds several API methods that enable control over which
types of filters QPDF will attempt to decode. It also adds support for
/RunLengthDecode and /DCTDecode filters for both encoding and
decoding.
This reverts commit 8ee83ca722.
This is being removed because qpdf now has its own page rotation. The
example was an excellent contribution to qpdf, but now it illustrates
rotating pages "by hand", which is no longer needed because of
QPDFObjectHandle::rotatePage.
This is added to contrib rather than examples because it requires
c++-11 and lacks a test suite, but it is still useful enough to
include with the distribution.
When requested, QPDFWriter will do more aggress prechecking of streams
to make sure it can actually succeed in decoding them before
attempting to do so. This will allow preservation of raw data even
when the raw data is corrupted relative to the specified filters.
QPDFObjectHandle::parseInternal now issues warnings instead of
throwing exceptions for all error conditions that it finds (except
internal logic errors) and has stronger recovery for things like
invalid tokens and malformed dictionaries. This should improve qpdf's
ability to recover from a wide range of broken files that currently
cause it to fail.
fixes #117
fixes #118
fixes #119
fixes #120
Several other infinite loop bugs were fixed by previous changes.
Include their test files in the test suite.
During parsing of an object, sometimes parts of the object have to be
resolved. An example is stream lengths. If such an object directly or
indirectly points to the object being parsed, it can cause an infinite
loop. Guard against all cases of re-entrant resolution of objects.
This is CVE-2017-9208.
The QPDF library uses object ID 0 internally as a sentinel to
represent a direct object, but prior to this fix, was not blocking
handling of 0 0 obj or 0 0 R as a special case. Creating an object in
the file with 0 0 obj could cause various infinite loops. The PDF spec
doesn't allow for object 0. Having qpdf handle object 0 might be a
better fix, but changing all the places in the code that assumes objid
== 0 means direct would be risky.
This is CVE-2017-9210.
The description string for an error message included unparsing an
object, which is too complex of a thing to try to do while throwing an
exception. There was only one example of this in the entire codebase,
so it is not a pervasive problem. Fixing this eliminated one class of
infinite loop errors.
For non-encrypted files, determinstic ID generation uses file contents
instead of timestamp and file name. At a small runtime cost, this
enables generation of the same /ID if the same inputs are converted in
the same way multiple times.
fix-qdf was previously hard-coding the number of bytes for the f2
field of the xref stream entry. This addresses issue #37. Thanks
aluebcke for reporting.
As reported in issue #40, a call to CryptAcquireContext in
SecureRandomDataProvider fails in a fresh windows install prior to any
user keys being created in AppData\Roaming\Microsoft\Crypto\RSA.
Thanks michalrames.
Pushing inherited objects to pages and getting all pages were both
prone to stack overflow infinite loops if there were loops in the
Pages dictionary. There is a general weakness in the code in that any
part of the code that traverses the Pages structure would be prone to
this and would have to implement its own loop detection. A more robust
fix may provide some general method for handling the Pages structure,
but it's probably not worth doing.
Note: addition of *Internal2 private functions was done rather than
changing signatures of existing methods to avoid breaking
compatibility.
Converting a password to an encryption key is supposed to copy up to a
certain number of bytes from a digest. Make sure never to copy more
than the size of the digest.
When checking two objects preceding R while parsing, ensure that the
objects are direct. This avoids stuff like 1 0 obj containing 1 0 R 0 R
from causing an infinite loop in object resolution.
Original reported here:
https://bugs.launchpad.net/ubuntu/+source/qpdf/+bug/1397413
The PDF specification says that the /Type key for nodes in the pages
dictionary (both /Page and /Pages) is required, but some PDF files
omit them. Use the presence of other keys to determine the type of
pages tree node this is if the type key is not found.
QPDFWriter was trying to make /Filter and /DecodeParms direct in all
cases, but there are some cases where /DecodeParms may refer to a
stream, which can't be direct. QPDFWriter doesn't actually need
/DecodeParms to be direct in that case because it won't be able to
filter the stream. Until we can handle this type of stream, just don't
make /Filter and /DecodeParms direct if we can't filter the stream
anyway.
Fixes #34
Fix problem: if the last object in the first part of a linearized file
had an offset that was below 65536 by less than the size of the hint
stream, the xref stream was invalid and the resulting file is not
usable.
Add new RandomDataProvider object and implement existing random number
generation in terms of that. This enables end users to supply their
own random data providers.
If NO_GET_ENVIRONMENT is #defined at compile time on Windows, do not
call GetEnvironmentVariable. QUtil::get_env will always return
false. This option is not available through configure. This was
added to support a specific user's requirements to avoid calling
GetEnvironmentVariable from the Windows API. Nothing in qpdf outside
the test coverage system in qtest relies on QUtil::get_env.
In compare image tests, use the gs device tiff24nc instead of tiff12nc
since the 4 bit per sample images created by tiff12nc could sometimes
trigger a bug in tiffcmp. Fixes #20.
Ideally, the library should never call assert outside of test code,
but it does in several places. For some cases where the assertion
might conceivably fail because of a problem with the input data,
replace assertions with exceptions so that they can be trapped by the
calling application. This commit surely misses some cases and
replaced some cases unnecessarily, but it should still be an
improvement.
In places where std::vector<T>(size_t) was used, either validate that
the size parameter is sane or refactor code to avoid the need to
pre-allocate the vector.
The /W array was not sanitized, possibly causing an integer overflow
in a multiplication. An analysis of the code suggests that there were
no possible exploits based on this since the problems were in checking
expected values but bounds checks were performed on actual values.
4.2.0 was binary incompatible in spite of there being no deletions or
changes to any public methods. As such, we have to bump the ABI and
are fixing some API breakage while we're at it.
Previous 4.3.0 target is now 5.1.0.
Space rather than newline after xref, missing /ID in trailer for
encrypted file. This enables qpdf to handle some files that xpdf can
handle. Adobe reader can't necessarily handle them.
Rework QPDFWriter to always track old object IDs and QPDFObjGen
instead of int, thus not discarding the generation number. Switch to
QPDF::getCompressibleObjGen() to properly handle the case of an old
object eligible for compression that has a generation of other than
zero.
Make --enable-werror work properly on msvc, handle extra warnings
flags for msvc in configure.ac instead of hardcoding into
make/msvc.mk, separate warnings flags into WFLAGS in autoconf.mk to
avoid duplication and to make it easier to override.
Some distributions (like debian) don't want .la files to be installed,
but the responsibility for doing this should like in the packaging,
not in qpdf itself.
Fix exit status for case of errors without warnings, continue after
errors when possible, add test case for parsing a file with content
stream errors on some but not all pages.
Change object type Keyword to Operator, and place the order of the
object types in object_type_e in the same order as they are mentioned
in the PDF specification.
Note that this change only breaks backward compatibility with code
that has not yet been released.
