platform/upstream/libusb.git
12 years agoLinux: Add __read_sysfs_attr() helper function
Hans de Goede [Thu, 17 Feb 2011 11:45:10 +0000 (12:45 +0100)]
Linux: Add __read_sysfs_attr() helper function

On Linux we often need to read (postive) integers from sysfs. This patch
adds a helper function for this. This is a preparation patch for adding
a libusb_get_device_speed() function to libusb.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
12 years agoLinux: Stop kernel from re-attaching in-kernel driver after reset
Hans de Goede [Tue, 8 Feb 2011 15:37:18 +0000 (16:37 +0100)]
Linux: Stop kernel from re-attaching in-kernel driver after reset

When an interface is bound to the usbfs driver (iow claimed), the
kernel will unbind it, and then after the reset do a device_attach
on the interface, which will bind the default in kernel driver to
the interface.

So if an app has detached the in kernel driver, and claimed the
interface and then does a libusb_reset_device. Things end up with
the interface no longer being bound to the usbfs driver (so no longer
claimed) and instead it is bound to the in kernel driver (iow the in
kernel driver is re-attached).

We can stop this from happening by releasing all claimed interfaces
before the reset, as the kernel will not do the device attach after
reset, if no driver was bound to the interface before the reset.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
12 years agoCore: Use USB_MAXINTERFACES rather then sizeof(claimed_interfaces)
Hans de Goede [Thu, 17 Feb 2011 11:35:46 +0000 (12:35 +0100)]
Core: Use USB_MAXINTERFACES rather then sizeof(claimed_interfaces)

claimed_interfaces is a long and thus its size differs on 32
versus 64 bits, this patch uses USB_MAXINTERFACES instead of
sizeof(claimed_interfaces) on interface number range checks
for consistent behavior independent of the arch.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
12 years agoLinux: Fix warnings seen when compiling with gcc-4.6
Hans de Goede [Wed, 9 Feb 2011 11:40:35 +0000 (12:40 +0100)]
Linux: Fix warnings seen when compiling with gcc-4.6

Compiling with gcc-4.6 results in the following warnings:
os/linux_usbfs.c: In function 'op_get_configuration':
os/linux_usbfs.c:1144:6: warning: variable 'r' set but not used ...
os/linux_usbfs.c: In function 'op_handle_events':
os/linux_usbfs.c:2091:40: warning: 'status' may be used uninitialized ...
os/linux_usbfs.c:2044:6: note: 'status' was declared here
dpfp.c: In function 'save_to_file':
dpfp.c:228:9: warning: variable 'ignore' set but not used ...
dpfp_threaded.c: In function 'save_to_file':
dpfp_threaded.c:257:9: warning: variable 'ignore' set but not used ...

This patch fixes these. (The second comes from handle_control_completion()
which gcc seems to inline into op_handle_events().)

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
[stuge: Add fixes for dpfp examples and update source references]

12 years agoDarwin: Fix #108 crash when a device is unplugged
Vitali Lovich [Tue, 14 Jun 2011 06:51:57 +0000 (08:51 +0200)]
Darwin: Fix #108 crash when a device is unplugged

12 years agolibusb.h: Add missing LIBUSB_CLASS_ codes currently defined by usb.org
Peter Stuge [Mon, 13 Jun 2011 20:58:23 +0000 (22:58 +0200)]
libusb.h: Add missing LIBUSB_CLASS_ codes currently defined by usb.org

LIBUSB_CLASS_PHYSICAL
LIBUSB_CLASS_SMART_CARD
LIBUSB_CLASS_CONTENT_SECURITY
LIBUSB_CLASS_VIDEO
LIBUSB_CLASS_PERSONAL_HEALTHCARE
LIBUSB_CLASS_DIAGNOSTIC_DEVICE

Per http://www.usb.org/developers/defined_class

12 years agolibusb.h: Add LIBUSB_CLASS_IMAGE in addition to LIBUSB_CLASS_PTP
Peter Stuge [Mon, 13 Jun 2011 20:53:40 +0000 (22:53 +0200)]
libusb.h: Add LIBUSB_CLASS_IMAGE in addition to LIBUSB_CLASS_PTP

The old LIBUSB_CLASS_PTP name is kept for compatibility. Fixes #106.

12 years agolibusb.h: Fix LIBUSB_CLASS_PRINTER description typo
Peter Stuge [Mon, 13 Jun 2011 20:52:48 +0000 (22:52 +0200)]
libusb.h: Fix LIBUSB_CLASS_PRINTER description typo

References #106.

12 years agoDarwin: Fix #100 warning cast to pointer from integer of different size
Sean McBride [Fri, 11 Feb 2011 16:28:04 +0000 (11:28 -0500)]
Darwin: Fix #100 warning cast to pointer from integer of different size

12 years agoconfigure.ac: Fix #97 clang warning about -fgnu89-inline
Sean McBride [Mon, 13 Jun 2011 20:26:43 +0000 (22:26 +0200)]
configure.ac: Fix #97 clang warning about -fgnu89-inline

Remove the test that sets the flag since GNU89 inline semantics are not
required by libusb.

[stuge: Also remove reference to the test result]

12 years agoWindows: Make cancel_io() in poll_windows.c static
Pete Batard [Tue, 14 Jun 2011 23:38:58 +0000 (01:38 +0200)]
Windows: Make cancel_io() in poll_windows.c static

References #97.

12 years agoWindows: Make upperize() in windows_usb.h static
Peter Stuge [Mon, 13 Jun 2011 20:21:02 +0000 (22:21 +0200)]
Windows: Make upperize() in windows_usb.h static

References #97.

12 years agodescriptor.c: Fix buffer read overflow caught by valgrind
Vitali Lovich [Thu, 17 Mar 2011 00:33:57 +0000 (17:33 -0700)]
descriptor.c: Fix buffer read overflow caught by valgrind

In parse_interface() an unexpected descriptor would be parsed without
validating the descriptor's length.  It is possible for size to be 0 at
this point, which means that the parsing would read past the end of the
source buffer.  Fix #83 by checking the length of the remaining buffer
before parsing.

12 years agoClean up in-flight transfers and device handle when closing a device
Vitali Lovich [Thu, 17 Mar 2011 02:51:40 +0000 (19:51 -0700)]
Clean up in-flight transfers and device handle when closing a device

Any in-flight transfers should properly invalidate their references
to device handles that are being closed.  Additionally, they should be
removed from the transfer-in-flight list.  This is done with the events
lock held to protect against another thread processing the same transfer.
The events lock is initialized as a recursive mutex, because the device
close code might itself be called while an event is being handled.
Fixes #82.

[stuge: Trivial rework to reduce indenting]

12 years agoAdd recursive mutexes to threading abstraction
Vitali Lovich [Thu, 17 Mar 2011 02:43:30 +0000 (19:43 -0700)]
Add recursive mutexes to threading abstraction

This is necessary for the device close path which needs to attain the
events lock, but which might itself be called while handling an event.
The events lock is necessary to properly clean up transfers which might
still be pointing to the device. References #82.

[stuge: Move usbi_mutex_init_recursive() into threads_posix.c]
[stuge: Must also #define _XOPEN_SOURCE 500 to be able to build]
[pbatard: Un-inline usbi_mutex_init_recursive() to make Cygwin happy]

12 years agoAdd USBI_TRANSFER_CANCELLING and _DEVICE_DISAPPEARED status flags
Vitali Lovich [Thu, 17 Mar 2011 02:48:35 +0000 (19:48 -0700)]
Add USBI_TRANSFER_CANCELLING and _DEVICE_DISAPPEARED status flags

The flags are used to indicate if a cancellation has started, and if
a cancellation has failed because the device is no longer available.
References #82.

12 years agoLinux: Detect and report that device has gone away while discarding URBs
Vitali Lovich [Thu, 17 Mar 2011 02:50:04 +0000 (19:50 -0700)]
Linux: Detect and report that device has gone away while discarding URBs

Return LIBUSB_ERROR_NO_DEVICE if the device is no longer available
(ENODEV) to discard the urbs. References #82.

12 years agolibusbi.h: Add comma after last value in enum usbi_transfer_flags
Peter Stuge [Mon, 6 Jun 2011 00:46:46 +0000 (02:46 +0200)]
libusbi.h: Add comma after last value in enum usbi_transfer_flags

12 years agothreads_posix.h: Whitespace fix
Peter Stuge [Sun, 5 Jun 2011 21:28:07 +0000 (23:28 +0200)]
threads_posix.h: Whitespace fix

12 years agolibusb/Makefile.am: Correct threading files in libusb_1_0_la_SOURCES
Peter Stuge [Sun, 5 Jun 2011 21:21:08 +0000 (23:21 +0200)]
libusb/Makefile.am: Correct threading files in libusb_1_0_la_SOURCES

12 years agoLinux: Fix #70 race condition in sysfs_get_device_list()
Alan Ott [Thu, 21 Jul 2011 14:37:48 +0000 (16:37 +0200)]
Linux: Fix #70 race condition in sysfs_get_device_list()

Change the way libusb chooses between using sysfs and usbfs for information
about the attached devies.  Using the old method, a race condition could
occur if a device was unplugged just before (or during) the call to
libusb_get_device_list(), corrupting the internal sysfs_can_relate_devices
and sysfs_has_descriptors variables and preventing libusb_get_device_list()
from working in future calls.

The old method was based on the assumption that if certain sysfs files
(eg: busnum) could not be opened, that indicated an inadequacy of sysfs
(ie: the running kernel's sysfs version did not contain those files),
when in reality those files couldn't be opened because the device had
been unplugged.

The new method checks the adequacy of sysfs during libusb_init()
(op_init()) and if a sysfs file cannot be opened, it is now assumed that
it is because the device has been unplugged, not because sysfs is
inadequate.

Signed-off-by: Alan Ott <alan@signal11.us>
[stuge: Include closedir() bugfix posted in ticket by Arne Laansoo]
[stuge: Remove dead code in sysfs_scan_device() found by Hans de Goede]

13 years agoDarwin: Fix #63 error when apps use Objective-C garbage collection
Sean McBride [Thu, 10 Feb 2011 22:11:29 +0000 (17:11 -0500)]
Darwin: Fix #63 error when apps use Objective-C garbage collection

13 years agoDarwin: Fix #28 clang analyzer warning about unbalanced retain/release
Sean McBride [Fri, 5 Feb 2010 16:50:34 +0000 (11:50 -0500)]
Darwin: Fix #28 clang analyzer warning about unbalanced retain/release

13 years agoio.c: Fix clang analyzer warning about unused variable
Sean McBride [Fri, 5 Feb 2010 16:50:34 +0000 (11:50 -0500)]
io.c: Fix clang analyzer warning about unused variable

References #28.

13 years ago.gitignore: Add m4 subdirectory created by libtool
Peter Stuge [Tue, 10 May 2011 08:44:18 +0000 (10:44 +0200)]
.gitignore: Add m4 subdirectory created by libtool

13 years agoLinux: Do not include useless pthread.h
Ludovic Rousseau [Mon, 15 Nov 2010 13:45:50 +0000 (14:45 +0100)]
Linux: Do not include useless pthread.h

The code does not use threads so no need to #include <pthread.h>
Fixes #74.

13 years agoWindows: minor code improvements
Pete Batard [Tue, 8 Mar 2011 17:41:05 +0000 (17:41 +0000)]
Windows: minor code improvements

* mostly warnings from VS2010's Intellisense

13 years agoWindows: prevent set_configuration request from being sent using WinUSB
Pete Batard [Tue, 8 Mar 2011 17:40:39 +0000 (17:40 +0000)]
Windows: prevent set_configuration request from being sent using WinUSB

* multiple WinUSB set configuration requests can stall a device and
  WinUSB does not support setting of alt configuration in the first place
* issue reported and investigated by Xiaofan Chen

13 years agoWindows: fixed default WinUSB timeout and ineffective policy settings
Graeme Gill [Tue, 8 Mar 2011 17:39:16 +0000 (17:39 +0000)]
Windows: fixed default WinUSB timeout and ineffective policy settings

* default WinUSB timeout was 5 seconds, which may be to small.
  PIPE_TRANSFER_TIMEOUT policy is now set to infinite for all EPs.
* other WinUSB policy settings were called before the endpoints
  had been set. This is now fixed.

13 years agoWindows: Don't leak pipe fds
Pete Batard [Tue, 8 Mar 2011 17:37:40 +0000 (17:37 +0000)]
Windows: Don't leak pipe fds

use _open() and _close() rather than _open_osfhandle() and CloseHandle()

* use of CloseHandle() prevented the pipe fds from being
  relinquished on libusb_exit()
* leaked fds could lead to the OS running out of new fds
  and LIBUSB_ERROR_NO_MEM being returned as a result
* using _open() avoids _open_osfhandle() redef for cygwin
* issue reported by Stephano Antonelli

13 years agoWindows: enumeration overhaul
Peter Stuge [Tue, 10 May 2011 08:05:29 +0000 (10:05 +0200)]
Windows: enumeration overhaul

* uses multiple passes in a single call
* uses a hash table, in anticipation for hotplug
* adds a (dummy) HUB interface for harmonization
* adds calloc on device struct creation in core
  (to detect and avoid unnecessary double initialization)

13 years agoWindows: MS project files part 4 - MSVC6 project files
Pete Batard [Mon, 9 May 2011 23:56:18 +0000 (00:56 +0100)]
Windows: MS project files part 4 - MSVC6 project files

Fixes #57 and #1.

13 years agoWindows: MS project files part 3 - VS2005 project files
Pete Batard [Mon, 9 May 2011 23:53:09 +0000 (00:53 +0100)]
Windows: MS project files part 3 - VS2005 project files

References #57.

13 years agoWindows: MS project files part 2 - WDK project files
Pete Batard [Mon, 9 May 2011 23:38:49 +0000 (00:38 +0100)]
Windows: MS project files part 2 - WDK project files

References #57.

13 years agoWindows: MS project files part 1 - support files & preliminaries
Pete Batard [Mon, 9 May 2011 23:38:12 +0000 (00:38 +0100)]
Windows: MS project files part 1 - support files & preliminaries

* adds the MSVC support files used by MSVC6, VS2005 and WDK
* also adds missing defines required by MS compilers in libusb.h
* also anticipatory sets .gitattributes for CRLF handling
* also adds manually maintained .def file for import lib generation

References #57.

13 years agoMove library version number from configure.ac to libusb/version.h
Peter Stuge [Mon, 9 May 2011 06:12:24 +0000 (08:12 +0200)]
Move library version number from configure.ac to libusb/version.h

This is neccessary to support native MS builds. The Windows resource
file libusb/libusb-1.0.rc must include the release version, which was
previously only available after configure had run and had substituted
the numbers into a generated libusb/libusb-1.0.rc file.

The version atoms are now stored as CPP style #defines in libusb/version.h
so that the .rc no longer needs to be generated but can simply include the
header file and access the version information directly.

The m4 macro LU_DEFINE_VERSION_ATOM() was added to configure.ac to get
version atoms from libusb/version.h for use in AC_INIT(). The macro
handles C and C++ style comments in version.h, but can easily be made
to fail by obscuring the file. Please don't do that.

Tested with MinGW using autoconf, and manual compile of libusb-1.0.rc
using RC.EXE Version 5.2.3690.0 from Visual C++ 2005 Express Edition.

13 years agoWindows: Debug builds create the same .dll file name as non-debug
Peter Stuge [Tue, 10 May 2011 06:36:32 +0000 (08:36 +0200)]
Windows: Debug builds create the same .dll file name as non-debug

Thanks to Pete for pointing this out!

13 years agoWindows: explicit use of ANSI or WideChar calls
Pete Batard [Tue, 8 Mar 2011 17:18:48 +0000 (17:18 +0000)]
Windows: explicit use of ANSI or WideChar calls

* This is in anticipation of adding MSVC project files
  that use Unicode rather than MBCS

13 years agoWindows: Remove SetupAPI, AdvAPI32 and OLE32 link-time dependencies
Pete Batard [Tue, 8 Mar 2011 17:16:23 +0000 (17:16 +0000)]
Windows: Remove SetupAPI, AdvAPI32 and OLE32 link-time dependencies

* this ensures that libusb dependent applications only need
  to explicitly link against libusb on Windows

The run-time dependency on Cfgmgr32.dll, OLE32.dll and SetupAPI.dll remains.

13 years agoAdd .gitattributes to avoid Windows line ending issues with autocrlf
Pete Batard [Tue, 8 Mar 2011 17:12:33 +0000 (17:12 +0000)]
Add .gitattributes to avoid Windows line ending issues with autocrlf

13 years agoWindows: Touch up FileDescription and ProductName resource strings
Peter Stuge [Sun, 27 Feb 2011 08:37:54 +0000 (09:37 +0100)]
Windows: Touch up FileDescription and ProductName resource strings

13 years agoWindows: Remove emulated direct device access via HID API
Peter Stuge [Sun, 27 Feb 2011 06:20:40 +0000 (07:20 +0100)]
Windows: Remove emulated direct device access via HID API

Either use a library such as HIDAPI to communicate with HID class
devices, or replace the kernel driver for the device with one that
offers support for the complete libusb API.

13 years agoRevert libusb_strerror() until we have i18n and l10n
Peter Stuge [Sun, 27 Feb 2011 05:59:15 +0000 (06:59 +0100)]
Revert libusb_strerror() until we have i18n and l10n

13 years agodpfp_threaded: return instead of pthread_exit() to avoid Cygwin warning
Pete Batard [Wed, 19 Jan 2011 14:00:12 +0000 (14:00 +0000)]
dpfp_threaded: return instead of pthread_exit() to avoid Cygwin warning

13 years agoWindows: Rename various variables named "index" to avoid shadow warnings
Pete Batard [Wed, 19 Jan 2011 13:52:08 +0000 (13:52 +0000)]
Windows: Rename various variables named "index" to avoid shadow warnings

13 years agoWindows: Fix logic in enumeration of driver name and port number
Pete Batard [Tue, 5 Oct 2010 11:50:53 +0000 (12:50 +0100)]
Windows: Fix logic in enumeration of driver name and port number

13 years agoWindows: Allow claiming any interface in composite device using WinUSB
Pete Batard [Tue, 5 Oct 2010 11:50:13 +0000 (12:50 +0100)]
Windows: Allow claiming any interface in composite device using WinUSB

Issue reported by Benjamin Dobell, see #48. The call to
WinUsb_QueryInterfaceSettings() is removed since it doesn't work
on any other interface but the first (a WinUSB limitation).

This allows the use of WinUSB.sys in lieu of the usbccgp.sys
"USB Generic Parent" default driver for composite devices.

13 years agoWindows: Allow arbitrary bConfigurationValue in config descriptors
Pete Batard [Tue, 5 Oct 2010 11:47:15 +0000 (12:47 +0100)]
Windows: Allow arbitrary bConfigurationValue in config descriptors

Unlike bInterfaceNumber, bConfigurationValue does not have to be
either sequential or contiguous. The bConfigurationValue is now
included in debug output. Fixes #48 reported by Benjamin Dobell.

13 years agoDarwin: Schedule isochronous transfers further in the future
Thomas Röfer [Fri, 19 Nov 2010 21:03:11 +0000 (22:03 +0100)]
Darwin: Schedule isochronous transfers further in the future

This is required to make OpenKinect transfer data at 30 fps. This
change was originally made by Theo Watson, and Thomas improved on it.
The commit came via OpenKinect into libusb thanks to Kyle Machulis and
Hector Martin. Fixes #66.

13 years agoDarwin: Fix #65 memory leak in submit_iso_transfer()
Thomas Röfer [Fri, 19 Nov 2010 21:03:11 +0000 (22:03 +0100)]
Darwin: Fix #65 memory leak in submit_iso_transfer()

This was also independently fixed by Hoi-Ho Chan, but Thomas caught
an additional problem with his fix.

This commit came via OpenKinect into libusb thanks to Kyle Machulis
and Hector Martin.

13 years agoconfigure.ac: Refactor Windows backend settings into one occurence
Peter Stuge [Sat, 26 Feb 2011 03:21:05 +0000 (04:21 +0100)]
configure.ac: Refactor Windows backend settings into one occurence

The Windows backend is used both with MinGW and Cygwin, and since most
settings are identical they can be set in one place.

13 years agoconfigure.ac: Rename AM_LDFLAGS to LTLDFLAGS and actually use them
Peter Stuge [Mon, 15 Nov 2010 18:58:51 +0000 (19:58 +0100)]
configure.ac: Rename AM_LDFLAGS to LTLDFLAGS and actually use them

The new variable name tries to clarify that libtool is being used.
Linker flags must thus always be specified with -Wl.

Factor out the libtool flag -no-undefined from host specific cases.
The flag is required to build a Windows DLL, but is correct also for
the other supported systems.

Also, start actually using LTLDFLAGS in libusb/Makefile.am, so that
libtool will see the options set by configure.

13 years agoconfigure.ac: Clean up PC_LIBS_PRIVATE and AM_LDFLAGS
Peter Stuge [Sun, 21 Nov 2010 22:38:46 +0000 (23:38 +0100)]
configure.ac: Clean up PC_LIBS_PRIVATE and AM_LDFLAGS

Move linker options for when linking the library itself from
PC_LIBS_PRIVATE into AM_LDFLAGS. PC_LIBS_PRIVATE should only
contain flags needed to link applications statically against libusb.
Fixes #72.

13 years agoconfigure.ac: Call AC_CONFIG_FILES() for each output file
Peter Stuge [Fri, 25 Feb 2011 01:33:29 +0000 (02:33 +0100)]
configure.ac: Call AC_CONFIG_FILES() for each output file

This avoids a problem with Windows line endings.

13 years agoconfigure.ac: Whitespace changes and trivial reordering
Peter Stuge [Sun, 21 Nov 2010 22:37:27 +0000 (23:37 +0100)]
configure.ac: Whitespace changes and trivial reordering

13 years agoconfigure.ac: Quote AC_COMPILE_IFELSE() input
Peter Stuge [Sun, 21 Nov 2010 22:10:35 +0000 (23:10 +0100)]
configure.ac: Quote AC_COMPILE_IFELSE() input

13 years agoconfigure.ac: Define booleans to 1 when set, instead of an empty string
Peter Stuge [Sun, 21 Nov 2010 22:09:02 +0000 (23:09 +0100)]
configure.ac: Define booleans to 1 when set, instead of an empty string

This makes the generated config.h look a lot nicer.

13 years agoconfigure.ac: Clean up redundancy and fix LIBS on Linux
Peter Stuge [Sat, 30 Oct 2010 20:28:59 +0000 (22:28 +0200)]
configure.ac: Clean up redundancy and fix LIBS on Linux

LIBS, AM_LDFLAGS, OS_ conditionals and THREADS_POSIX had a bit of
redundancy throughout the file, and on Linux LIBS ended up missing
the -pthread flag.

13 years agoconfigure.ac: Touch up Darwin and Cygwin OS messages
Peter Stuge [Sat, 26 Feb 2011 03:08:38 +0000 (04:08 +0100)]
configure.ac: Touch up Darwin and Cygwin OS messages

13 years agoconfigure.ac: Do not use -pthread on Darwin
Peter Stuge [Sat, 30 Oct 2010 20:13:05 +0000 (22:13 +0200)]
configure.ac: Do not use -pthread on Darwin

It's not needed because Darwin has POSIX Threads in libc. Fixes #96.

13 years agoconfigure.ac: Check for poll.h, and for nfds_t on Darwin
Peter Stuge [Sat, 30 Oct 2010 20:07:37 +0000 (22:07 +0200)]
configure.ac: Check for poll.h, and for nfds_t on Darwin

On Linux, assume nfds_t is always available.

On Darwin, fall back to unsigned int when poll() exists but there
is no nfds_t, such as on Mac OS X before 10.4.

On Windows (both MinGW and Cygwin), always use unsigned int instead
of nfds_t, and don't check for poll.h because we use our own poll()
implementation.

13 years agoDarwin: Add support for control requests on endpoints other than 0
Nathan Hjelm [Wed, 6 Oct 2010 02:08:08 +0000 (20:08 -0600)]
Darwin: Add support for control requests on endpoints other than 0

Signed-off-by: Nathan Hjelm <hjelmn@me.com>
13 years agoDarwin: Add more error checking for libusb_open()
Nathan Hjelm [Wed, 6 Oct 2010 05:11:50 +0000 (23:11 -0600)]
Darwin: Add more error checking for libusb_open()

Now also check the CreateDeviceAsyncEventSource() return code for errors.

Signed-off-by: Nathan Hjelm <hjelmn@me.com>
13 years agoCore: libusb_get_next_timeout() must consider all flying transfers
Peter Stuge [Wed, 3 Nov 2010 20:46:53 +0000 (21:46 +0100)]
Core: libusb_get_next_timeout() must consider all flying transfers

If transfer->flags indicated that a transfer had a timeout, but no
timeout was actually set, then libusb_get_next_timeout() would look
no further for a timeout, ignoring any transfers later in the list
which had a timeout set.

Since libusb has an internal 60 second timeout this bug could not
cause complete lockup, but it could cause a 60 second timeout even
when a transfer was submitted with a shorter timeout.

13 years agoLinux: Correctly catch read() errors for sysfs config descriptors
Ludovic Rousseau [Mon, 15 Nov 2010 13:55:51 +0000 (14:55 +0100)]
Linux: Correctly catch read() errors for sysfs config descriptors

read(2) returns ssize_t (signed) and not size_t (unsigned) or int.

Silence "warning: comparison of unsigned expression < 0 is always false"
and fix #75.

13 years agoLinux: Refactor cancellation into one function for all transfer types
Peter Stuge [Sun, 23 Jan 2011 03:01:25 +0000 (04:01 +0100)]
Linux: Refactor cancellation into one function for all transfer types

13 years agoLinux: Refactor discarding of URBs into a function and return all errors
Peter Stuge [Sun, 17 Oct 2010 06:19:56 +0000 (08:19 +0200)]
Linux: Refactor discarding of URBs into a function and return all errors

The Linux backend did not always return errors according to the
libusb_cancel_transfer() documentation.

13 years agoLinux: Set private number of URBs also for control transfers
Peter Stuge [Sun, 23 Jan 2011 03:01:01 +0000 (04:01 +0100)]
Linux: Set private number of URBs also for control transfers

13 years agoLinux: cancel URBs in reverse order
Alan Stern [Sun, 17 Oct 2010 04:57:06 +0000 (06:57 +0200)]
Linux: cancel URBs in reverse order

In a multi-URB transfer, URBs should be cancelled in reverse order of
submission. This prevents races that might otherwise occur (after URB N
is cancelled, data may be transferred for URB N+1 before it too can be
cancelled). Fixes #8.

13 years agoLinux: report correct value for urb->actual_length
Alan Stern [Sun, 17 Oct 2010 02:17:32 +0000 (04:17 +0200)]
Linux: report correct value for urb->actual_length

Transferred bytes are returned correctly by the kernel for every URB,
even when an error occurs.  Hence they should always be included in
the transfer statistics.  The same is true for isochronous packet
length and status.

[stuge: set itransfer->transferred so data is stored at correct offset]

13 years agoLinux: improve reporting of URB error status codes
Alan Stern [Fri, 22 Oct 2010 09:13:34 +0000 (11:13 +0200)]
Linux: improve reporting of URB error status codes

Fix the way overflow errors are handled (they can occur in any
transaction, not just the last one in a transfer) and add tests
for other commonly-occuring error statuses.

13 years agoLinux: improve detection of BULK_CONTINUATION kernel support
Alan Stern [Tue, 5 Oct 2010 15:08:40 +0000 (11:08 -0400)]
Linux: improve detection of BULK_CONTINUATION kernel support

Clarify the comments describing the BULK_CONTINATION flag, and improve
the subroutine that checks whether the flag is supported by making it
"future-proof", i.e., by working properly with kernel version numbers
higher than 2.6.

13 years agoREADME: Update webpage URL and add Peter's email address
Peter Stuge [Mon, 22 Nov 2010 04:14:00 +0000 (05:14 +0100)]
README: Update webpage URL and add Peter's email address

13 years agolibusb is not just for Linux!
Peter Stuge [Wed, 4 Aug 2010 05:30:22 +0000 (07:30 +0200)]
libusb is not just for Linux!

13 years agoRemove USBI_OS_HANDLES_TIMEOUT and fix int/isoc timeouts on Darwin
Nathan Hjelm [Wed, 6 Oct 2010 01:48:39 +0000 (19:48 -0600)]
Remove USBI_OS_HANDLES_TIMEOUT and fix int/isoc timeouts on Darwin

Backends set USBI_TRANSFER_OS_HANDLES_TIMEOUT for transfers instead.
Darwin only handles timeouts for bulk and control transfers, so the
backend now sets that flag accordingly, making libusb core handle
timeouts for interrupt and isochronous transfers. Fixes #31.

Signed-off-by: Nathan Hjelm <hjelmn@me.com>
[stuge: rework libusb_get_next_timeout() and enum usbi_transfer_flags]
[stuge: fix typo; set USBI_TRANSFER_TIMED_OUT flag correctly]

13 years agoSilence "warning: suggest braces around empty body in an 'if' statement"
Peter Stuge [Sun, 14 Nov 2010 02:14:00 +0000 (03:14 +0100)]
Silence "warning: suggest braces around empty body in an 'if' statement"

13 years agoDisallow libusb_get_string_descriptor_ascii() with index 0
Konrad Rzepecki [Sat, 13 Nov 2010 13:09:22 +0000 (14:09 +0100)]
Disallow libusb_get_string_descriptor_ascii() with index 0

String descriptor 0 is the list of supported language IDs in the device,
which can't have an ASCII representation. Calling the function with
index 0 is thus not really useful. Fixes #43.

13 years agoUse const for the usbi_parse_descriptor() format string
Ludovic Rousseau [Fri, 19 Mar 2010 14:35:14 +0000 (15:35 +0100)]
Use const for the usbi_parse_descriptor() format string

Constant strings should be "const char *" instead of "char *". Silence
"discards qualifiers from pointer target type" warnings. Fixes #61.

13 years agoLinux: Fix usbfs/sysfs config descriptor handling on big-endian
Hector Martin [Wed, 25 Aug 2010 01:21:57 +0000 (03:21 +0200)]
Linux: Fix usbfs/sysfs config descriptor handling on big-endian

usbfs endian-swaps, sysfs doesn't, not the other way around. Fixes #51;
descriptor endian parsing is backwards using sysfs on big-endian hosts.
usbfs is untested.

Signed-off-by: Hector Martin <hector@marcansoft.com>
[stuge: checked against Documentation/usb/proc_usb_info.txt]

13 years agoexamples: #include <libusb.h> without directory name
Ludovic Rousseau [Tue, 17 Aug 2010 18:53:04 +0000 (20:53 +0200)]
examples: #include <libusb.h> without directory name

The libusb-1.0 directory name belongs in the compile command, e.g.
via pkg-config --cflags. Fixes #58.

13 years agoAdd missing argument to libusb_wait_for_event() documentation, fix #55
Ludovic Rousseau [Thu, 24 Jun 2010 08:14:24 +0000 (10:14 +0200)]
Add missing argument to libusb_wait_for_event() documentation, fix #55

In the libusb_wait_for_event() sample code in the section "Letting other
threads do the work for you" the call to libusb_wait_for_event() was not
updated by commit 1df713d622ab4f0b03aad72d903ac7beb8fb3b90, which added
the libusb_context *ctx parameter to the function, so the sample code
was broken until now.

13 years agoexamples: Silence warnings about return value of fwrite()
Peter Stuge [Sat, 16 Oct 2010 12:22:30 +0000 (14:22 +0200)]
examples: Silence warnings about return value of fwrite()

13 years agoLinux: Don't free() URBs prematurely on hardware error; fix #54
Peter Stuge [Sat, 16 Oct 2010 09:47:14 +0000 (11:47 +0200)]
Linux: Don't free() URBs prematurely on hardware error; fix #54

When an URB in a multi-URB transfer had an error, handle_bulk_completion()
would free all URBs and segfault once the next one completed, instead of
cancelling the remaining URBs and cleaning up as usual.

This is basically the patch from the ticket, plus some restructuring
for increased readability of the function.

Many thanks to Brian Shirley and National Instruments for finding and
fixing this!

13 years agoWindows: add support for filter drivers
Pete Batard [Tue, 5 Oct 2010 11:43:41 +0000 (12:43 +0100)]
Windows: add support for filter drivers

* precedence is driver > upper filter > lower filter, and defined
  in struct driver_lookup lookup from set_device_paths()

13 years agoWindows: fix errcode shadow warning, and string and DLL_DECLARE fixes
Pete Batard [Tue, 5 Oct 2010 11:36:49 +0000 (12:36 +0100)]
Windows: fix errcode shadow warning, and string and DLL_DECLARE fixes

* use FormatMessageA, and overall improvement of windows_error_str()
* use safe_strlen() always
* better typing/init default
* more explicit debug message for devices with no driver

13 years agoWindows: use IOCTLs for HID input, output and feature reports
Pete Batard [Tue, 5 Oct 2010 11:29:32 +0000 (12:29 +0100)]
Windows: use IOCTLs for HID input, output and feature reports

* fixes feature reports not providing actual read size
  (reported by Axel Rohde http://marc.info/?m=127033070021994)
* removes the USE_HIDD_FOR_REPORTS macro
* IOCTL usage inspired from HIDAPI by Alan Ott

13 years agoFix libtool version setting
Daniel Drake [Mon, 4 Oct 2010 19:03:58 +0000 (20:03 +0100)]
Fix libtool version setting

This was being set too early to take effect.

13 years agoPopulate the pkg-config Libs.private field
Mike Frysinger [Mon, 4 Oct 2010 19:01:07 +0000 (20:01 +0100)]
Populate the pkg-config Libs.private field

13 years agoLinux: fix build failure when CLOCK_MONOTONIC does not exist
Mike Frysinger [Fri, 20 Aug 2010 04:09:46 +0000 (00:09 -0400)]
Linux: fix build failure when CLOCK_MONOTONIC does not exist

This is not available on all platforms.

13 years agoUpdate libtool version info
Daniel Drake [Mon, 4 Oct 2010 17:45:18 +0000 (18:45 +0100)]
Update libtool version info

With input from various people on the mailing list, update the libtool
versioning info and start to update this on every release.

The next libusb release will not need a change here. All following ones
will.

13 years agoUpdated documentation for unsupported Darwin/Windows calls
Pete Batard [Thu, 5 Aug 2010 15:40:09 +0000 (16:40 +0100)]
Updated documentation for unsupported Darwin/Windows calls

13 years agoWindows: Make libusb_pollfd() return error
Pete Batard [Sun, 19 Sep 2010 13:24:25 +0000 (14:24 +0100)]
Windows: Make libusb_pollfd() return error

Windows does not have numerical file descriptors but does have the
concept of event sources. Exposing these event sources will require
some careful thought and design, to be completed later.

13 years agoMinGW: Use --add-stdcall-alias linker option
Daniel Drake [Sun, 19 Sep 2010 13:21:01 +0000 (14:21 +0100)]
MinGW: Use --add-stdcall-alias linker option

Naming of symbols inside the library is inconsistent on Windows:
http://wyw.dcweb.cn/stdcall.htm

Use this linker option to add aliases which add compatibility with
the "MSVC DLL" platform when a DEF file is used. It also better matches
the appearance of the Windows API itself.

13 years agoIntroduced calling convention (for Windows)
Pete Batard [Fri, 13 Aug 2010 10:59:49 +0000 (11:59 +0100)]
Introduced calling convention (for Windows)

Under Windows, a variety of compilers and configurations are available,
meaning that the manner of parameter passing (e.g. registers vs stack)
can vary.

Match the Windows API calling convention and document this appropriately.
This calling convention will be used regardless of the configuration of
the user's development platform.

The only user-level complication is that all functions used as libusb
callbacks must use the same calling convention as libusb. The
LIBUSB_CALL macro is provided to make this easy.

Signed-off-by: Michael Plante <michael.plante@gmail.com>
Signed-off-by: Pete Batard <pbatard@gmail.com>
[dsd: slight change of strategy, add documentation]

13 years agoWindows: fix string copy in sanitize_path
Pete Batard [Sat, 7 Aug 2010 03:56:04 +0000 (21:56 -0600)]
Windows: fix string copy in sanitize_path

The NULL byte was not being copied.

13 years agoWindows: DuplicateHandle failure is not a warning
Pete Batard [Thu, 29 Jul 2010 23:30:31 +0000 (00:30 +0100)]
Windows: DuplicateHandle failure is not a warning

This is because WinUSB handles are not duplicable, and generate a
lot of these warnings.

13 years agoremoved trailing whitespaces
Pete Batard [Thu, 29 Jul 2010 23:23:05 +0000 (00:23 +0100)]
removed trailing whitespaces

13 years agoWindows: Mark interfaces as unused when releasing
Pete Batard [Thu, 5 Aug 2010 00:18:32 +0000 (18:18 -0600)]
Windows: Mark interfaces as unused when releasing

Without this important cleanup, the autoclaim code could attempt to
reuse an unclaimed interface as if it were already claimed.

13 years agoWindows: fix device path string duplication
Pete Batard [Thu, 5 Aug 2010 00:15:59 +0000 (18:15 -0600)]
Windows: fix device path string duplication

13 years agoDarwin: Cache device configuration value
Nathan Hjelm [Tue, 3 Aug 2010 03:27:38 +0000 (21:27 -0600)]
Darwin: Cache device configuration value

Cache device configuration value to bring the Darwin backend more in
line with the libusb spec. To handle buggy devices GetConfiguration is
not called unless the device has more than one configuration.