Make sure we always return CURLM_CALL_MULTI_PERFORM when we reach
CURLM_STATE_DONE since the state is transient and it can very well
continue executing as there is nothing to wait for.
Bug: http://curl.haxx.se/mail/lib-2013-08/0211.html
Reported-by: Yi Huang
... this also makes sure that the progess callback gets called more
often during TFTP transfers.
Added test 1238 to verify.
Bug: http://curl.haxx.se/bug/view.cgi?id=1269
Reported-by: Jo3
I build curl.exe (using MingW) with '-DCURLDEBUG' and by importing from
libcurl.dll. Which means the new curl_easy_perform_ev() must be
exported from libcurl.dll.
Doing curl_multi_add_handle() on an easy handle that is already added to
a multi handle now returns this error code. It previously returned
CURLM_BAD_EASY_HANDLE for this condition.
The closure_handle is "owned" by the multi handle and it is
unconditional so the setting up of it should be in the Curl_multi_handle
function rather than curl_multi_add_handle.
This function is meant to work *exactly* as curl_easy_perform() but will
use the event-based libcurl API internally instead of
curl_multi_perform(). To avoid relying on an actual event-based library
and to not use non-portable functions (like epoll or similar), there's a
rather inefficient emulation layer implemented on top of Curl_poll()
instead.
There's currently some convenience logging done in curl_easy_perform_ev
which helps when tracking down problems. They may be suitable to remove
or change once things seem to be fine enough.
curl has a new --test-event option when built with debug enabled that
then uses curl_easy_perform_ev() instead of curl_easy_perform(). If
built without debug, using --test-event will only output a warning
message.
NOTE: curl_easy_perform_ev() is not part if the public API on purpose.
It is only present in debug builds of libcurl and MUST NOT be considered
stable even then. Use it for libcurl-testing purposes only.
runtests.pl now features an -e command line option that makes it use
--test-event for all curl command line tests. The man page is updated.
libcurl quietly truncates usernames, passwords, and options from
before an '@' sign in a URL to 255 (= MAX_CURL_PASSWORD_LENGTH - 1)
characters to fit in fixed-size buffers on the stack. Allocate a
buffer large enough to fit the parsed fields on the fly instead to
support longer passwords.
After this change, there are no more uses of MAX_CURL_OPTIONS_LENGTH
left, so stop defining that constant while at it. The hardcoded max
username and password length constants, on the other hand, are still
used in HTTP proxy credential handling (which this patch doesn't
touch).
Reported-by: Colby Ranger
Instead of nesting "if(success)" blocks and leaving the reader in
suspense about what happens in the !success case, deal with failure
cases early, usually with a simple goto to clean up and return from
the function.
No functional change intended. The main effect is to decrease the
indentation of this function slightly.
libcurl truncates usernames, passwords, and options set with
curl_easy_setopt to 255 (= MAX_CURL_PASSWORD_LENGTH - 1) characters.
This doesn't affect the return value from curl_easy_setopt(), so from
the caller's point of view, there is no sign anything strange has
happened, except that authentication fails.
For example:
# Prepare a long (300-char) password.
s=0123456789; s=$s$s$s$s$s$s$s$s$s$s; s=$s$s$s;
# Start a server.
nc -l -p 8888 | tee out & pid=$!
# Tell curl to pass the password to the server.
curl --user me:$s http://localhost:8888 & sleep 1; kill $pid
# Extract the password.
userpass=$(
awk '/Authorization: Basic/ {print $3}' <out |
tr -d '\r' |
base64 -d
)
password=${userpass#me:}
echo ${#password}
Expected result: 300
Actual result: 255
The fix is simple: allocate appropriately sized buffers on the heap
instead of trying to squeeze the provided values into fixed-size
on-stack buffers.
Bug: http://bugs.debian.org/719856
Reported-by: Colby Ranger
libcurl truncates usernames and passwords it reads from .netrc to
LOGINSIZE and PASSWORDSIZE (64) characters without any indication to
the user, to ensure the values returned from Curl_parsenetrc fit in a
caller-provided buffer.
Fix the interface by passing back dynamically allocated buffers
allocated to fit the user's input. The parser still relies on a
256-character buffer to read each line, though.
So now you can include an ~246-character password in your .netrc,
instead of the previous limit of 63 characters.
Reported-by: Colby Ranger
Instead of remembering before each "return" statement which temporary
allocations, if any, need to be freed, take care to set pointers to
NULL when no longer needed and use a goto to a common block to exit
the function and free all temporaries.
No functional change intended. Currently the only temporary buffer in
this function is "proxy" which is already correctly freed when
appropriate, but there will be more soon.
Use appropriately sized buffers on the heap instead of fixed-size
buffers on the stack, to allow for longer usernames and passwords.
Callers never pass anything longer than MAX_CURL_USER_LENGTH (resp.
MAX_CURL_PASSWORD_LENGTH), so no functional change inteded yet.
This makes the socket callback get called with the proper bitmask as
otherwise the application could be left hanging waiting for reading on
an upload connection!
Bug: http://curl.haxx.se/mail/lib-2013-08/0043.html
Reported-by: Bill Doyle
With everything being struct SessionHandle pointers now, this rename
makes multi.c use the library-wide practise of calling that pointer
'data' instead of the previously used 'easy'.
Moved Curl_easy_addmulti() from easy.c to multi.c, renamed it to
easy_addmulti and made it static.
Removed Curl_easy_initHandleData() and uses of it since it was emptied
in commit cdda92ab67b47d74a.
All protocol handler structs are now opaque (void *) in the
SessionHandle struct and moved in the request-specific sub-struct
'SingleRequest'. The intension is to keep the protocol specific
knowledge in their own dedicated source files [protocol].c etc.
There's some "leakage" where this policy is violated, to be addressed at
a later point in time.
1 - always allocate the struct in protocol->setup_connection. Some
protocol handlers had to get this function added.
2 - always free at the end of a request. This is also an attempt to keep
less memory in the handle after it is completed.
CURLOPT_DNS_USE_GLOBAL_CACHE broke in commit c43127414d (been
broken since the libcurl 7.29.0 release). While this option has been
documented as deprecated for almost a decade and nobody even reported
this bug, it should remain functional.
Added test case 1512 to verify
The previous naming scheme ftp_state_post_XXXX() wasn't really helpful
as it wasn't always immediately after 'xxxx' and it wasn't easy to
understand what it does based on such a name.
This new one is instead ftp_state_yyyy() where yyyy describes what it
does or sends.
This is a regression as this logic used to work. It isn't clear when it
broke, but I'm assuming in 7.28.0 when we went all-multi internally.
This likely never worked with the multi interface. As the failed
connection is detected once the multi state has reached DO_MORE, the
Curl_do_more() function was now expanded somewhat so that the
ftp_do_more() function can request to go "back" to the previous state
when it makes another attempt - using PASV.
Added test case 1233 to verify this fix. It has the little issue that it
assumes no service is listening/accepting connections on port 1...
Reported-by: byte_bucket in the #curl IRC channel
For some reason, OS X 10.5's GCC suddenly stopped working correctly with
macros that change MD5_Init etc. in the code to CC_MD5_Init etc., so I
worked around this by removing use of the macros and inserting static
functions that just call CommonCrypto's implementations of the functions
instead.
This changes the previous check for untrusted certs to a check for
certs explicitely marked as trusted.
The change is backward-compatible (tested with certdata.txt v1.80).
The internal function that's used to detect known file extensions for
the default Content-Type got the the wrong pointer passed in when
CURLFORM_BUFFER + CURLFORM_BUFFERPTR were used. This had the effect that
strlen() would be used which could lead to an out-of-bounds read (and
thus segfault). In most cases it would only lead to it not finding or
using the correct default content-type.
It also showed that test 554 and test 587 were testing for the
previous/wrong behavior and now they're updated as well.
Bug: http://curl.haxx.se/bug/view.cgi?id=1262
Reported-by: Konstantin Isakov
Christian Heimes brought to our attention that the certdata.txt
format has recently changed [1], causing ca-bundle.crt created
with mk-ca-bundle.[pl|vbs] to include untrusted certs.
[1] http://lists.debian.org/debian-release/2012/11/msg00411.html
The motivation for having a separate struct that keep track of an easy
handle when using the multi handle was removed when we switched to
always using the multi interface internally. Now they were just two
separate struct that was always allocated for each easy handle.
This first step just moves the Curl_one_easy struct members into the
SessionHandle struct and hides this somehow (== keeps the source code
changes to a minimum) by defining Curl_one_easy to SessionHandle
The biggest changes in this commit are:
1 - the linked list of easy handles had to be changed somewhat due
to the new struct layout. This made the main linked list pointer
get renamed to 'easyp' and there's also a new pointer to the last
node, called easylp. It is no longer circular but ends with ->next
pointing to NULL. New nodes are still added last.
2 - easy->state is now called easy->mstate to avoid name collision
(This doesn't need to appear in the release notes.) I noticed a few places
where infof() was called, and there should've been an LF at the end of the
string, but there wasn't.
It turns out Snow Leopard not only has SecItemCopyMatching() defined in
a header not included by the omnibus header, but it won't work for our
purposes, because searching for SecIdentityRef objects wasn't added
to that API until Lion. So we now use the old SecKeychainSearch API
instead if the user is building under, or running under, Snow Leopard.
Bug: http://sourceforge.net/p/curl/bugs/1255/
Reported by: Edward Rudd
Previously we used __MAC_10_X and __IPHONE_X to mark digest-generating
code that was specific to OS X and iOS. Now we use
__MAC_OS_X_VERSION_MAX_ALLOWED and __IPHONE_OS_VERSION_MAX_ALLOWED
instead of those macros.
Bug: http://sourceforge.net/p/curl/bugs/1255/
Reported by: Edward Rudd
For the standard VMS text file formats, VMS needs to read the file to
get the actual file size.
For the standard VMS binary file formats, VMS needs a special format of
fopen() call so that it stops reading at the logical end of file instead
of at the end of the blocks allocated to the file.
I structured the patch this way as I was not sure about changing the
structures or parameters to the routines, but would prefer to only call
the stat() function once and pass the information to where the fopen()
call is made.
Bug: https://sourceforge.net/p/curl/bugs/758/
The code for CURLFORM_FILECONTENT had its check for duplicate options
wrong so that it would reject CURLFORM_PTRNAME if used in combination
with it (but not CURLFORM_COPYNAME)! The flags field used for this
purpose cannot be interpreted that broadly.
Bug: http://curl.haxx.se/mail/lib-2013-07/0258.html
Reported-by: Byrial Jensen
Linking on Solaris 10 x86 with Sun Studio 12 failed when we upgraded
automake for the release builds.
Bug: http://curl.haxx.se/bug/view.cgi?id=1217
Reported-by: Dagobert Michelsen
Commit 6d30f8ebed didn't work properly. First, it used the wrong
array index, but this fix also:
1 - only does the copying if indeed there was any activity
2 - makes sure to properly translate between internal and external
bitfields, which are not guaranteed to match
Reported-by: Evgeny Turnaev
Instead of going 50,100,150 etc millisecond delay time when nothing has
been found to do or wait for, we now start lower and double each loop as
in 4,8,16,32 etc.
This lowers the minimum wait without sacrifizing the longer wait too
much with unnecessary CPU cycles burnt.
Bug: http://curl.haxx.se/mail/lib-2013-07/0103.html
Reported-by: Andreas Malzahn
In the case of an active connection when ftp_do_more() detects that the
server has connected back, it must make sure to mark it as complete so
that the multi_runsingle() function will detect this and move on to the
next state.
Bug: http://curl.haxx.se/mail/lib-2013-07/0115.html
Reported-by: Clemens Gruber
CURLOPT_XFERINFOFUNCTION is now the preferred progress callback function
and CURLOPT_PROGRESSFUNCTION is considered deprecated.
This new callback uses pure 'curl_off_t' arguments to pass on full
resolution sizes. It otherwise retains the same characteristics: the
same call rate, the same meanings for the arguments and the return code
is used the same way.
The progressfunc.c example is updated to show how to use the new
callback for newer libcurls while supporting the older one if built with
an older libcurl or even built with a newer libcurl while running with
an older.
This reverts commit 7ed25cc, reinstating commit 8ec2cb5.
As of 18-jul-2013 we still do have code in libcurl that makes use of these
memory functions. Commit 8ec2cb5 comment still applies and is yet valid.
These memory functions are solely used in Windows builds, so all related
code is protected with '#ifdef WIN32' preprocessor conditional compilation
directives.
Specifically, wcsdup() _wcsdup() are used when building a Windows target with
UNICODE and USE_WINDOWS_SSPI preprocessor symbols defined. This is the case
when building a Windows UNICODE target with Windows native SSL/TLS support
enabled.
Realizing that wcsdup() _wcsdup() are used is a bit tricky given that usage
of these is hidden behind _tcsdup() which is MS way of dealing with code
that must tolerate UNICODE and non-UNICODE compilation. Additionally, MS
header files and those compatible from other compilers use this preprocessor
conditional compilation directive in order to select at compilation time
whether 'wide' or 'ansi' MS API functions are used.
Without this code, Windows build targets with Windows native SSL/TLS support
enabled and MemoryTracking support enabled misbehave in tracking memory usage,
regardless of being a UNICODE enabled build or not.
Fixed issue with static build for MSVC2010.
After some investigation I've discovered known issue
http://public.kitware.com/Bug/view.php?id=11240 When .rc file is linked
to static lib it fails with following linker error
LINK : warning LNK4068: /MACHINE not specified; defaulting to X86
file.obj : fatal error LNK1112: module machine type 'x64' conflicts with
target machine type 'X86'
Fix add target property /MACHINE: for MSVC generation.
Also removed old workarounds - it caused errors during msvc build.
Bug: http://curl.haxx.se/mail/lib-2013-07/0046.html
I just noticed that OS X no longer supports SSLv2. Other TLS engines return
an error if the requested protocol isn't supported by the underlying
engine, so we do that now for SSLv2 if the framework returns an error
when trying to turn on SSLv2 support. (Note: As always, SSLv2 support is
only enabled in curl when starting the app with the -2 argument; it's off
by default. SSLv2 is really old and insecure.)
When doing multi-part formposts, libcurl used a pseudo-random value that
was seeded with time(). This turns out to be bad for users who formpost
data that is provided with users who then can guess how the boundary
string will look like and then they can forge a different formpost part
and trick the receiver.
My advice to such implementors is (still even after this change) to not
rely on the boundary strings being cryptographically strong. Fix your
code and logic to not depend on them that much!
I moved the Curl_rand() function into the sslgen.c source file now to be
able to take advantage of the SSL library's random function if it
provides one. If not, try to use the RANDOM_FILE for seeding and as a
last resort keep the old logic, just modified to also add microseconds
which makes it harder to properly guess the exact seed.
The formboundary() function in formdata.c is now using 64 bit entropy
for the boundary and therefore the string of dashes was reduced by 4
letters and there are 16 hex digits following it. The total length is
thus still the same.
Bug: http://curl.haxx.se/bug/view.cgi?id=1251
Reported-by: "Floris"
When using %x, the number must be treated as unsigned as otherwise it
would get sign-extended on for example 64bit machines and do wrong
output. This problem showed when doing printf("%08x", 0xffeeddcc) on a
64bit host.
Follow-up fix from 7d80ed64e4.
The SessionHandle may not be around to use when we restore the sigpipe
sighandler so we store the no_signal boolean in the local struct to know
if/how to restore.
When the c-ares based resolver backend failed to resolve a name, it
tried to show the name that failed from existing structs. This caused
the wrong output and shown hostname when for example --interface
[hostname] was used and that name resolving failed.
Now we use the hostname used in the actual resolve attempt in the error
message as well.
Bug: http://curl.haxx.se/bug/view.cgi?id=1191
Reported-by: Kim Vandry
When we recently started to treat a zero return code from SSL_read() as
an error we also got false positives - which primarily looks to be
because the OpenSSL documentation is wrong and a zero return code is not
at all an error case in many situations.
Now ossl_recv() will check with ERR_get_error() to see if there is a
stored error and only then consider it to be a true error if SSL_read()
returned zero.
Bug: http://curl.haxx.se/bug/view.cgi?id=1249
Reported-by: Nach M. S.
Patch-by: Nach M. S.
Something (a recent security update maybe?) changed in Lion, and now it
has changed SSLCopyPeerTrust such that it may return noErr but also give
us a null trust, which caught us off guard and caused an eventual crash.
... and restore the ordinary handling again when it returns. This is
done for curl_easy_perform() and curl_easy_cleanup() only for now - and
only when built to use OpenSSL as backend as this is the known culprit
for the spurious SIGPIPEs people have received.
Bug: http://curl.haxx.se/bug/view.cgi?id=1180
Reported by: Lluís Batlle i Rossell
This doesn't need to be in the release notes. I cleaned up a lot of the #if
lines in the code to use MAC_OS_X_VERSION_MIN_REQUIRED and
MAC_OS_X_VERSION_MAX_ALLOWED instead of checking for whether things like
__MAC_10_6 or whatever were defined, because for some SDKs Apple has released
they were defined out of place.
RFC3986 details how a path part passed in as part of a URI should be
"cleaned" from dot sequences before getting used. The described
algorithm is now implemented in lib/dotdot.c with the accompanied test
case in test 1395.
Bug: http://curl.haxx.se/bug/view.cgi?id=1200
Reported-by: Alex Vinnik
Security problem: CVE-2013-2174
If a program would give a string like "%FF" to curl_easy_unescape() but
ask for it to decode only the first byte, it would still parse and
decode the full hex sequence. The function then not only read beyond the
allowed buffer but it would also deduct the *unsigned* counter variable
for how many more bytes there's left to read in the buffer by two,
making the counter wrap. Continuing this, the function would go on
reading beyond the buffer and soon writing beyond the allocated target
buffer...
Bug: http://curl.haxx.se/docs/adv_20130622.html
Reported-by: Timo Sirainen
As a remedy to the problem when a socket gets closed and a new one is
opened with the same file descriptor number and as a result
multi.c:singlesocket() doesn't detect the difference, the new function
Curl_multi_closed() gets told when a socket is closed so that it can be
removed from the socket hash. When the old one has been removed, a new
socket should be detected fine by the singlesocket() on next invoke.
Bug: http://curl.haxx.se/bug/view.cgi?id=1248
Reported-by: Erik Johansson
