2 * I/O functions for libusbx
3 * Copyright © 2007-2009 Daniel Drake <dsd@gentoo.org>
4 * Copyright © 2001 Johannes Erdfelt <johannes@erdfelt.com>
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
33 #ifdef HAVE_SYS_TIME_H
36 #ifdef USBI_TIMERFD_AVAILABLE
37 #include <sys/timerfd.h>
44 * \page io Synchronous and asynchronous device I/O
46 * \section intro Introduction
48 * If you're using libusbx in your application, you're probably wanting to
49 * perform I/O with devices - you want to perform USB data transfers.
51 * libusbx offers two separate interfaces for device I/O. This page aims to
52 * introduce the two in order to help you decide which one is more suitable
53 * for your application. You can also choose to use both interfaces in your
54 * application by considering each transfer on a case-by-case basis.
56 * Once you have read through the following discussion, you should consult the
57 * detailed API documentation pages for the details:
61 * \section theory Transfers at a logical level
63 * At a logical level, USB transfers typically happen in two parts. For
64 * example, when reading data from a endpoint:
65 * -# A request for data is sent to the device
66 * -# Some time later, the incoming data is received by the host
68 * or when writing data to an endpoint:
70 * -# The data is sent to the device
71 * -# Some time later, the host receives acknowledgement from the device that
72 * the data has been transferred.
74 * There may be an indefinite delay between the two steps. Consider a
75 * fictional USB input device with a button that the user can press. In order
76 * to determine when the button is pressed, you would likely submit a request
77 * to read data on a bulk or interrupt endpoint and wait for data to arrive.
78 * Data will arrive when the button is pressed by the user, which is
79 * potentially hours later.
81 * libusbx offers both a synchronous and an asynchronous interface to performing
82 * USB transfers. The main difference is that the synchronous interface
83 * combines both steps indicated above into a single function call, whereas
84 * the asynchronous interface separates them.
86 * \section sync The synchronous interface
88 * The synchronous I/O interface allows you to perform a USB transfer with
89 * a single function call. When the function call returns, the transfer has
90 * completed and you can parse the results.
92 * If you have used the libusb-0.1 before, this I/O style will seem familar to
93 * you. libusb-0.1 only offered a synchronous interface.
95 * In our input device example, to read button presses you might write code
96 * in the following style:
98 unsigned char data[4];
100 int r = libusb_bulk_transfer(handle, LIBUSB_ENDPOINT_IN, data, sizeof(data), &actual_length, 0);
101 if (r == 0 && actual_length == sizeof(data)) {
102 // results of the transaction can now be found in the data buffer
103 // parse them here and report button press
109 * The main advantage of this model is simplicity: you did everything with
110 * a single simple function call.
112 * However, this interface has its limitations. Your application will sleep
113 * inside libusb_bulk_transfer() until the transaction has completed. If it
114 * takes the user 3 hours to press the button, your application will be
115 * sleeping for that long. Execution will be tied up inside the library -
116 * the entire thread will be useless for that duration.
118 * Another issue is that by tieing up the thread with that single transaction
119 * there is no possibility of performing I/O with multiple endpoints and/or
120 * multiple devices simultaneously, unless you resort to creating one thread
123 * Additionally, there is no opportunity to cancel the transfer after the
124 * request has been submitted.
126 * For details on how to use the synchronous API, see the
127 * \ref syncio "synchronous I/O API documentation" pages.
129 * \section async The asynchronous interface
131 * Asynchronous I/O is the most significant new feature in libusb-1.0.
132 * Although it is a more complex interface, it solves all the issues detailed
135 * Instead of providing which functions that block until the I/O has complete,
136 * libusbx's asynchronous interface presents non-blocking functions which
137 * begin a transfer and then return immediately. Your application passes a
138 * callback function pointer to this non-blocking function, which libusbx will
139 * call with the results of the transaction when it has completed.
141 * Transfers which have been submitted through the non-blocking functions
142 * can be cancelled with a separate function call.
144 * The non-blocking nature of this interface allows you to be simultaneously
145 * performing I/O to multiple endpoints on multiple devices, without having
148 * This added flexibility does come with some complications though:
149 * - In the interest of being a lightweight library, libusbx does not create
150 * threads and can only operate when your application is calling into it. Your
151 * application must call into libusbx from it's main loop when events are ready
152 * to be handled, or you must use some other scheme to allow libusbx to
153 * undertake whatever work needs to be done.
154 * - libusbx also needs to be called into at certain fixed points in time in
155 * order to accurately handle transfer timeouts.
156 * - Memory handling becomes more complex. You cannot use stack memory unless
157 * the function with that stack is guaranteed not to return until the transfer
158 * callback has finished executing.
159 * - You generally lose some linearity from your code flow because submitting
160 * the transfer request is done in a separate function from where the transfer
161 * results are handled. This becomes particularly obvious when you want to
162 * submit a second transfer based on the results of an earlier transfer.
164 * Internally, libusbx's synchronous interface is expressed in terms of function
165 * calls to the asynchronous interface.
167 * For details on how to use the asynchronous API, see the
168 * \ref asyncio "asynchronous I/O API" documentation pages.
173 * \page packetoverflow Packets and overflows
175 * \section packets Packet abstraction
177 * The USB specifications describe how data is transmitted in packets, with
178 * constraints on packet size defined by endpoint descriptors. The host must
179 * not send data payloads larger than the endpoint's maximum packet size.
181 * libusbx and the underlying OS abstract out the packet concept, allowing you
182 * to request transfers of any size. Internally, the request will be divided
183 * up into correctly-sized packets. You do not have to be concerned with
184 * packet sizes, but there is one exception when considering overflows.
186 * \section overflow Bulk/interrupt transfer overflows
188 * When requesting data on a bulk endpoint, libusbx requires you to supply a
189 * buffer and the maximum number of bytes of data that libusbx can put in that
190 * buffer. However, the size of the buffer is not communicated to the device -
191 * the device is just asked to send any amount of data.
193 * There is no problem if the device sends an amount of data that is less than
194 * or equal to the buffer size. libusbx reports this condition to you through
195 * the \ref libusb_transfer::actual_length "libusb_transfer.actual_length"
198 * Problems may occur if the device attempts to send more data than can fit in
199 * the buffer. libusbx reports LIBUSB_TRANSFER_OVERFLOW for this condition but
200 * other behaviour is largely undefined: actual_length may or may not be
201 * accurate, the chunk of data that can fit in the buffer (before overflow)
202 * may or may not have been transferred.
204 * Overflows are nasty, but can be avoided. Even though you were told to
205 * ignore packets above, think about the lower level details: each transfer is
206 * split into packets (typically small, with a maximum size of 512 bytes).
207 * Overflows can only happen if the final packet in an incoming data transfer
208 * is smaller than the actual packet that the device wants to transfer.
209 * Therefore, you will never see an overflow if your transfer buffer size is a
210 * multiple of the endpoint's packet size: the final packet will either
211 * fill up completely or will be only partially filled.
215 * @defgroup asyncio Asynchronous device I/O
217 * This page details libusbx's asynchronous (non-blocking) API for USB device
218 * I/O. This interface is very powerful but is also quite complex - you will
219 * need to read this page carefully to understand the necessary considerations
220 * and issues surrounding use of this interface. Simplistic applications
221 * may wish to consider the \ref syncio "synchronous I/O API" instead.
223 * The asynchronous interface is built around the idea of separating transfer
224 * submission and handling of transfer completion (the synchronous model
225 * combines both of these into one). There may be a long delay between
226 * submission and completion, however the asynchronous submission function
227 * is non-blocking so will return control to your application during that
228 * potentially long delay.
230 * \section asyncabstraction Transfer abstraction
232 * For the asynchronous I/O, libusbx implements the concept of a generic
233 * transfer entity for all types of I/O (control, bulk, interrupt,
234 * isochronous). The generic transfer object must be treated slightly
235 * differently depending on which type of I/O you are performing with it.
237 * This is represented by the public libusb_transfer structure type.
239 * \section asynctrf Asynchronous transfers
241 * We can view asynchronous I/O as a 5 step process:
242 * -# <b>Allocation</b>: allocate a libusb_transfer
243 * -# <b>Filling</b>: populate the libusb_transfer instance with information
244 * about the transfer you wish to perform
245 * -# <b>Submission</b>: ask libusbx to submit the transfer
246 * -# <b>Completion handling</b>: examine transfer results in the
247 * libusb_transfer structure
248 * -# <b>Deallocation</b>: clean up resources
251 * \subsection asyncalloc Allocation
253 * This step involves allocating memory for a USB transfer. This is the
254 * generic transfer object mentioned above. At this stage, the transfer
255 * is "blank" with no details about what type of I/O it will be used for.
257 * Allocation is done with the libusb_alloc_transfer() function. You must use
258 * this function rather than allocating your own transfers.
260 * \subsection asyncfill Filling
262 * This step is where you take a previously allocated transfer and fill it
263 * with information to determine the message type and direction, data buffer,
264 * callback function, etc.
266 * You can either fill the required fields yourself or you can use the
267 * helper functions: libusb_fill_control_transfer(), libusb_fill_bulk_transfer()
268 * and libusb_fill_interrupt_transfer().
270 * \subsection asyncsubmit Submission
272 * When you have allocated a transfer and filled it, you can submit it using
273 * libusb_submit_transfer(). This function returns immediately but can be
274 * regarded as firing off the I/O request in the background.
276 * \subsection asynccomplete Completion handling
278 * After a transfer has been submitted, one of four things can happen to it:
280 * - The transfer completes (i.e. some data was transferred)
281 * - The transfer has a timeout and the timeout expires before all data is
283 * - The transfer fails due to an error
284 * - The transfer is cancelled
286 * Each of these will cause the user-specified transfer callback function to
287 * be invoked. It is up to the callback function to determine which of the
288 * above actually happened and to act accordingly.
290 * The user-specified callback is passed a pointer to the libusb_transfer
291 * structure which was used to setup and submit the transfer. At completion
292 * time, libusbx has populated this structure with results of the transfer:
293 * success or failure reason, number of bytes of data transferred, etc. See
294 * the libusb_transfer structure documentation for more information.
296 * \subsection Deallocation
298 * When a transfer has completed (i.e. the callback function has been invoked),
299 * you are advised to free the transfer (unless you wish to resubmit it, see
300 * below). Transfers are deallocated with libusb_free_transfer().
302 * It is undefined behaviour to free a transfer which has not completed.
304 * \section asyncresubmit Resubmission
306 * You may be wondering why allocation, filling, and submission are all
307 * separated above where they could reasonably be combined into a single
310 * The reason for separation is to allow you to resubmit transfers without
311 * having to allocate new ones every time. This is especially useful for
312 * common situations dealing with interrupt endpoints - you allocate one
313 * transfer, fill and submit it, and when it returns with results you just
314 * resubmit it for the next interrupt.
316 * \section asynccancel Cancellation
318 * Another advantage of using the asynchronous interface is that you have
319 * the ability to cancel transfers which have not yet completed. This is
320 * done by calling the libusb_cancel_transfer() function.
322 * libusb_cancel_transfer() is asynchronous/non-blocking in itself. When the
323 * cancellation actually completes, the transfer's callback function will
324 * be invoked, and the callback function should check the transfer status to
325 * determine that it was cancelled.
327 * Freeing the transfer after it has been cancelled but before cancellation
328 * has completed will result in undefined behaviour.
330 * When a transfer is cancelled, some of the data may have been transferred.
331 * libusbx will communicate this to you in the transfer callback. Do not assume
332 * that no data was transferred.
334 * \section bulk_overflows Overflows on device-to-host bulk/interrupt endpoints
336 * If your device does not have predictable transfer sizes (or it misbehaves),
337 * your application may submit a request for data on an IN endpoint which is
338 * smaller than the data that the device wishes to send. In some circumstances
339 * this will cause an overflow, which is a nasty condition to deal with. See
340 * the \ref packetoverflow page for discussion.
342 * \section asyncctrl Considerations for control transfers
344 * The <tt>libusb_transfer</tt> structure is generic and hence does not
345 * include specific fields for the control-specific setup packet structure.
347 * In order to perform a control transfer, you must place the 8-byte setup
348 * packet at the start of the data buffer. To simplify this, you could
349 * cast the buffer pointer to type struct libusb_control_setup, or you can
350 * use the helper function libusb_fill_control_setup().
352 * The wLength field placed in the setup packet must be the length you would
353 * expect to be sent in the setup packet: the length of the payload that
354 * follows (or the expected maximum number of bytes to receive). However,
355 * the length field of the libusb_transfer object must be the length of
356 * the data buffer - i.e. it should be wLength <em>plus</em> the size of
357 * the setup packet (LIBUSB_CONTROL_SETUP_SIZE).
359 * If you use the helper functions, this is simplified for you:
360 * -# Allocate a buffer of size LIBUSB_CONTROL_SETUP_SIZE plus the size of the
361 * data you are sending/requesting.
362 * -# Call libusb_fill_control_setup() on the data buffer, using the transfer
363 * request size as the wLength value (i.e. do not include the extra space you
364 * allocated for the control setup).
365 * -# If this is a host-to-device transfer, place the data to be transferred
366 * in the data buffer, starting at offset LIBUSB_CONTROL_SETUP_SIZE.
367 * -# Call libusb_fill_control_transfer() to associate the data buffer with
368 * the transfer (and to set the remaining details such as callback and timeout).
369 * - Note that there is no parameter to set the length field of the transfer.
370 * The length is automatically inferred from the wLength field of the setup
372 * -# Submit the transfer.
374 * The multi-byte control setup fields (wValue, wIndex and wLength) must
375 * be given in little-endian byte order (the endianness of the USB bus).
376 * Endianness conversion is transparently handled by
377 * libusb_fill_control_setup() which is documented to accept host-endian
380 * Further considerations are needed when handling transfer completion in
381 * your callback function:
382 * - As you might expect, the setup packet will still be sitting at the start
383 * of the data buffer.
384 * - If this was a device-to-host transfer, the received data will be sitting
385 * at offset LIBUSB_CONTROL_SETUP_SIZE into the buffer.
386 * - The actual_length field of the transfer structure is relative to the
387 * wLength of the setup packet, rather than the size of the data buffer. So,
388 * if your wLength was 4, your transfer's <tt>length</tt> was 12, then you
389 * should expect an <tt>actual_length</tt> of 4 to indicate that the data was
390 * transferred in entirity.
392 * To simplify parsing of setup packets and obtaining the data from the
393 * correct offset, you may wish to use the libusb_control_transfer_get_data()
394 * and libusb_control_transfer_get_setup() functions within your transfer
397 * Even though control endpoints do not halt, a completed control transfer
398 * may have a LIBUSB_TRANSFER_STALL status code. This indicates the control
399 * request was not supported.
401 * \section asyncintr Considerations for interrupt transfers
403 * All interrupt transfers are performed using the polling interval presented
404 * by the bInterval value of the endpoint descriptor.
406 * \section asynciso Considerations for isochronous transfers
408 * Isochronous transfers are more complicated than transfers to
409 * non-isochronous endpoints.
411 * To perform I/O to an isochronous endpoint, allocate the transfer by calling
412 * libusb_alloc_transfer() with an appropriate number of isochronous packets.
414 * During filling, set \ref libusb_transfer::type "type" to
415 * \ref libusb_transfer_type::LIBUSB_TRANSFER_TYPE_ISOCHRONOUS
416 * "LIBUSB_TRANSFER_TYPE_ISOCHRONOUS", and set
417 * \ref libusb_transfer::num_iso_packets "num_iso_packets" to a value less than
418 * or equal to the number of packets you requested during allocation.
419 * libusb_alloc_transfer() does not set either of these fields for you, given
420 * that you might not even use the transfer on an isochronous endpoint.
422 * Next, populate the length field for the first num_iso_packets entries in
423 * the \ref libusb_transfer::iso_packet_desc "iso_packet_desc" array. Section
424 * 5.6.3 of the USB2 specifications describe how the maximum isochronous
425 * packet length is determined by the wMaxPacketSize field in the endpoint
427 * Two functions can help you here:
429 * - libusb_get_max_iso_packet_size() is an easy way to determine the max
430 * packet size for an isochronous endpoint. Note that the maximum packet
431 * size is actually the maximum number of bytes that can be transmitted in
432 * a single microframe, therefore this function multiplies the maximum number
433 * of bytes per transaction by the number of transaction opportunities per
435 * - libusb_set_iso_packet_lengths() assigns the same length to all packets
436 * within a transfer, which is usually what you want.
438 * For outgoing transfers, you'll obviously fill the buffer and populate the
439 * packet descriptors in hope that all the data gets transferred. For incoming
440 * transfers, you must ensure the buffer has sufficient capacity for
441 * the situation where all packets transfer the full amount of requested data.
443 * Completion handling requires some extra consideration. The
444 * \ref libusb_transfer::actual_length "actual_length" field of the transfer
445 * is meaningless and should not be examined; instead you must refer to the
446 * \ref libusb_iso_packet_descriptor::actual_length "actual_length" field of
447 * each individual packet.
449 * The \ref libusb_transfer::status "status" field of the transfer is also a
451 * - If the packets were submitted and the isochronous data microframes
452 * completed normally, status will have value
453 * \ref libusb_transfer_status::LIBUSB_TRANSFER_COMPLETED
454 * "LIBUSB_TRANSFER_COMPLETED". Note that bus errors and software-incurred
455 * delays are not counted as transfer errors; the transfer.status field may
456 * indicate COMPLETED even if some or all of the packets failed. Refer to
457 * the \ref libusb_iso_packet_descriptor::status "status" field of each
458 * individual packet to determine packet failures.
459 * - The status field will have value
460 * \ref libusb_transfer_status::LIBUSB_TRANSFER_ERROR
461 * "LIBUSB_TRANSFER_ERROR" only when serious errors were encountered.
462 * - Other transfer status codes occur with normal behaviour.
464 * The data for each packet will be found at an offset into the buffer that
465 * can be calculated as if each prior packet completed in full. The
466 * libusb_get_iso_packet_buffer() and libusb_get_iso_packet_buffer_simple()
467 * functions may help you here.
469 * \section asyncmem Memory caveats
471 * In most circumstances, it is not safe to use stack memory for transfer
472 * buffers. This is because the function that fired off the asynchronous
473 * transfer may return before libusbx has finished using the buffer, and when
474 * the function returns it's stack gets destroyed. This is true for both
475 * host-to-device and device-to-host transfers.
477 * The only case in which it is safe to use stack memory is where you can
478 * guarantee that the function owning the stack space for the buffer does not
479 * return until after the transfer's callback function has completed. In every
480 * other case, you need to use heap memory instead.
482 * \section asyncflags Fine control
484 * Through using this asynchronous interface, you may find yourself repeating
485 * a few simple operations many times. You can apply a bitwise OR of certain
486 * flags to a transfer to simplify certain things:
487 * - \ref libusb_transfer_flags::LIBUSB_TRANSFER_SHORT_NOT_OK
488 * "LIBUSB_TRANSFER_SHORT_NOT_OK" results in transfers which transferred
489 * less than the requested amount of data being marked with status
490 * \ref libusb_transfer_status::LIBUSB_TRANSFER_ERROR "LIBUSB_TRANSFER_ERROR"
491 * (they would normally be regarded as COMPLETED)
492 * - \ref libusb_transfer_flags::LIBUSB_TRANSFER_FREE_BUFFER
493 * "LIBUSB_TRANSFER_FREE_BUFFER" allows you to ask libusbx to free the transfer
494 * buffer when freeing the transfer.
495 * - \ref libusb_transfer_flags::LIBUSB_TRANSFER_FREE_TRANSFER
496 * "LIBUSB_TRANSFER_FREE_TRANSFER" causes libusbx to automatically free the
497 * transfer after the transfer callback returns.
499 * \section asyncevent Event handling
501 * In accordance of the aim of being a lightweight library, libusbx does not
502 * create threads internally. This means that libusbx code does not execute
503 * at any time other than when your application is calling a libusbx function.
504 * However, an asynchronous model requires that libusbx perform work at various
505 * points in time - namely processing the results of previously-submitted
506 * transfers and invoking the user-supplied callback function.
508 * This gives rise to the libusb_handle_events() function which your
509 * application must call into when libusbx has work do to. This gives libusbx
510 * the opportunity to reap pending transfers, invoke callbacks, etc.
512 * The first issue to discuss here is how your application can figure out
513 * when libusbx has work to do. In fact, there are two naive options which
514 * do not actually require your application to know this:
515 * -# Periodically call libusb_handle_events() in non-blocking mode at fixed
516 * short intervals from your main loop
517 * -# Repeatedly call libusb_handle_events() in blocking mode from a dedicated
520 * The first option is plainly not very nice, and will cause unnecessary
521 * CPU wakeups leading to increased power usage and decreased battery life.
522 * The second option is not very nice either, but may be the nicest option
523 * available to you if the "proper" approach can not be applied to your
524 * application (read on...).
526 * The recommended option is to integrate libusbx with your application main
527 * event loop. libusbx exposes a set of file descriptors which allow you to do
528 * this. Your main loop is probably already calling poll() or select() or a
529 * variant on a set of file descriptors for other event sources (e.g. keyboard
530 * button presses, mouse movements, network sockets, etc). You then add
531 * libusbx's file descriptors to your poll()/select() calls, and when activity
532 * is detected on such descriptors you know it is time to call
533 * libusb_handle_events().
535 * There is one final event handling complication. libusbx supports
536 * asynchronous transfers which time out after a specified time period, and
537 * this requires that libusbx is called into at or after the timeout so that
538 * the timeout can be handled. So, in addition to considering libusbx's file
539 * descriptors in your main event loop, you must also consider that libusbx
540 * sometimes needs to be called into at fixed points in time even when there
541 * is no file descriptor activity.
543 * For the details on retrieving the set of file descriptors and determining
544 * the next timeout, see the \ref poll "polling and timing" API documentation.
548 * @defgroup poll Polling and timing
550 * This page documents libusbx's functions for polling events and timing.
551 * These functions are only necessary for users of the
552 * \ref asyncio "asynchronous API". If you are only using the simpler
553 * \ref syncio "synchronous API" then you do not need to ever call these
556 * The justification for the functionality described here has already been
557 * discussed in the \ref asyncevent "event handling" section of the
558 * asynchronous API documentation. In summary, libusbx does not create internal
559 * threads for event processing and hence relies on your application calling
560 * into libusbx at certain points in time so that pending events can be handled.
561 * In order to know precisely when libusbx needs to be called into, libusbx
562 * offers you a set of pollable file descriptors and information about when
563 * the next timeout expires.
565 * If you are using the asynchronous I/O API, you must take one of the two
566 * following options, otherwise your I/O will not complete.
568 * \section pollsimple The simple option
570 * If your application revolves solely around libusbx and does not need to
571 * handle other event sources, you can have a program structure as follows:
573 // initialize libusbx
574 // find and open device
575 // maybe fire off some initial async I/O
577 while (user_has_not_requested_exit)
578 libusb_handle_events(ctx);
583 * With such a simple main loop, you do not have to worry about managing
584 * sets of file descriptors or handling timeouts. libusb_handle_events() will
585 * handle those details internally.
587 * \section pollmain The more advanced option
589 * \note This functionality is currently only available on Unix-like platforms.
590 * On Windows, libusb_get_pollfds() simply returns NULL. Exposing event sources
591 * on Windows will require some further thought and design.
593 * In more advanced applications, you will already have a main loop which
594 * is monitoring other event sources: network sockets, X11 events, mouse
595 * movements, etc. Through exposing a set of file descriptors, libusbx is
596 * designed to cleanly integrate into such main loops.
598 * In addition to polling file descriptors for the other event sources, you
599 * take a set of file descriptors from libusbx and monitor those too. When you
600 * detect activity on libusbx's file descriptors, you call
601 * libusb_handle_events_timeout() in non-blocking mode.
603 * What's more, libusbx may also need to handle events at specific moments in
604 * time. No file descriptor activity is generated at these times, so your
605 * own application needs to be continually aware of when the next one of these
606 * moments occurs (through calling libusb_get_next_timeout()), and then it
607 * needs to call libusb_handle_events_timeout() in non-blocking mode when
608 * these moments occur. This means that you need to adjust your
609 * poll()/select() timeout accordingly.
611 * libusbx provides you with a set of file descriptors to poll and expects you
612 * to poll all of them, treating them as a single entity. The meaning of each
613 * file descriptor in the set is an internal implementation detail,
614 * platform-dependent and may vary from release to release. Don't try and
615 * interpret the meaning of the file descriptors, just do as libusbx indicates,
616 * polling all of them at once.
618 * In pseudo-code, you want something that looks like:
620 // initialise libusbx
622 libusb_get_pollfds(ctx)
623 while (user has not requested application exit) {
624 libusb_get_next_timeout(ctx);
625 poll(on libusbx file descriptors plus any other event sources of interest,
626 using a timeout no larger than the value libusbx just suggested)
627 if (poll() indicated activity on libusbx file descriptors)
628 libusb_handle_events_timeout(ctx, &zero_tv);
629 if (time has elapsed to or beyond the libusbx timeout)
630 libusb_handle_events_timeout(ctx, &zero_tv);
631 // handle events from other sources here
637 * \subsection polltime Notes on time-based events
639 * The above complication with having to track time and call into libusbx at
640 * specific moments is a bit of a headache. For maximum compatibility, you do
641 * need to write your main loop as above, but you may decide that you can
642 * restrict the supported platforms of your application and get away with
643 * a more simplistic scheme.
645 * These time-based event complications are \b not required on the following
648 * - Linux, provided that the following version requirements are satisfied:
649 * - Linux v2.6.27 or newer, compiled with timerfd support
650 * - glibc v2.9 or newer
651 * - libusbx v1.0.5 or newer
653 * Under these configurations, libusb_get_next_timeout() will \em always return
654 * 0, so your main loop can be simplified to:
656 // initialise libusbx
658 libusb_get_pollfds(ctx)
659 while (user has not requested application exit) {
660 poll(on libusbx file descriptors plus any other event sources of interest,
661 using any timeout that you like)
662 if (poll() indicated activity on libusbx file descriptors)
663 libusb_handle_events_timeout(ctx, &zero_tv);
664 // handle events from other sources here
670 * Do remember that if you simplify your main loop to the above, you will
671 * lose compatibility with some platforms (including legacy Linux platforms,
672 * and <em>any future platforms supported by libusbx which may have time-based
673 * event requirements</em>). The resultant problems will likely appear as
674 * strange bugs in your application.
676 * You can use the libusb_pollfds_handle_timeouts() function to do a runtime
677 * check to see if it is safe to ignore the time-based event complications.
678 * If your application has taken the shortcut of ignoring libusbx's next timeout
679 * in your main loop, then you are advised to check the return value of
680 * libusb_pollfds_handle_timeouts() during application startup, and to abort
681 * if the platform does suffer from these timing complications.
683 * \subsection fdsetchange Changes in the file descriptor set
685 * The set of file descriptors that libusbx uses as event sources may change
686 * during the life of your application. Rather than having to repeatedly
687 * call libusb_get_pollfds(), you can set up notification functions for when
688 * the file descriptor set changes using libusb_set_pollfd_notifiers().
690 * \subsection mtissues Multi-threaded considerations
692 * Unfortunately, the situation is complicated further when multiple threads
693 * come into play. If two threads are monitoring the same file descriptors,
694 * the fact that only one thread will be woken up when an event occurs causes
697 * The events lock, event waiters lock, and libusb_handle_events_locked()
698 * entities are added to solve these problems. You do not need to be concerned
699 * with these entities otherwise.
701 * See the extra documentation: \ref mtasync
704 /** \page mtasync Multi-threaded applications and asynchronous I/O
706 * libusbx is a thread-safe library, but extra considerations must be applied
707 * to applications which interact with libusbx from multiple threads.
709 * The underlying issue that must be addressed is that all libusbx I/O
710 * revolves around monitoring file descriptors through the poll()/select()
711 * system calls. This is directly exposed at the
712 * \ref asyncio "asynchronous interface" but it is important to note that the
713 * \ref syncio "synchronous interface" is implemented on top of the
714 * asynchonrous interface, therefore the same considerations apply.
716 * The issue is that if two or more threads are concurrently calling poll()
717 * or select() on libusbx's file descriptors then only one of those threads
718 * will be woken up when an event arrives. The others will be completely
719 * oblivious that anything has happened.
721 * Consider the following pseudo-code, which submits an asynchronous transfer
722 * then waits for its completion. This style is one way you could implement a
723 * synchronous interface on top of the asynchronous interface (and libusbx
724 * does something similar, albeit more advanced due to the complications
725 * explained on this page).
728 void cb(struct libusb_transfer *transfer)
730 int *completed = transfer->user_data;
735 struct libusb_transfer *transfer;
736 unsigned char buffer[LIBUSB_CONTROL_SETUP_SIZE];
739 transfer = libusb_alloc_transfer(0);
740 libusb_fill_control_setup(buffer,
741 LIBUSB_REQUEST_TYPE_VENDOR | LIBUSB_ENDPOINT_OUT, 0x04, 0x01, 0, 0);
742 libusb_fill_control_transfer(transfer, dev, buffer, cb, &completed, 1000);
743 libusb_submit_transfer(transfer);
746 poll(libusbx file descriptors, 120*1000);
747 if (poll indicates activity)
748 libusb_handle_events_timeout(ctx, &zero_tv);
750 printf("completed!");
755 * Here we are <em>serializing</em> completion of an asynchronous event
756 * against a condition - the condition being completion of a specific transfer.
757 * The poll() loop has a long timeout to minimize CPU usage during situations
758 * when nothing is happening (it could reasonably be unlimited).
760 * If this is the only thread that is polling libusbx's file descriptors, there
761 * is no problem: there is no danger that another thread will swallow up the
762 * event that we are interested in. On the other hand, if there is another
763 * thread polling the same descriptors, there is a chance that it will receive
764 * the event that we were interested in. In this situation, <tt>myfunc()</tt>
765 * will only realise that the transfer has completed on the next iteration of
766 * the loop, <em>up to 120 seconds later.</em> Clearly a two-minute delay is
767 * undesirable, and don't even think about using short timeouts to circumvent
770 * The solution here is to ensure that no two threads are ever polling the
771 * file descriptors at the same time. A naive implementation of this would
772 * impact the capabilities of the library, so libusbx offers the scheme
773 * documented below to ensure no loss of functionality.
775 * Before we go any further, it is worth mentioning that all libusb-wrapped
776 * event handling procedures fully adhere to the scheme documented below.
777 * This includes libusb_handle_events() and its variants, and all the
778 * synchronous I/O functions - libusbx hides this headache from you.
780 * \section Using libusb_handle_events() from multiple threads
782 * Even when only using libusb_handle_events() and synchronous I/O functions,
783 * you can still have a race condition. You might be tempted to solve the
784 * above with libusb_handle_events() like so:
787 libusb_submit_transfer(transfer);
790 libusb_handle_events(ctx);
792 printf("completed!");
795 * This however has a race between the checking of completed and
796 * libusb_handle_events() acquiring the events lock, so another thread
797 * could have completed the transfer, resulting in this thread hanging
798 * until either a timeout or another event occurs. See also commit
799 * 6696512aade99bb15d6792af90ae329af270eba6 which fixes this in the
800 * synchronous API implementation of libusb.
802 * Fixing this race requires checking the variable completed only after
803 * taking the event lock, which defeats the concept of just calling
804 * libusb_handle_events() without worrying about locking. This is why
805 * libusb-1.0.9 introduces the new libusb_handle_events_timeout_completed()
806 * and libusb_handle_events_completed() functions, which handles doing the
807 * completion check for you after they have acquired the lock:
810 libusb_submit_transfer(transfer);
813 libusb_handle_events_completed(ctx, &completed);
815 printf("completed!");
818 * This nicely fixes the race in our example. Note that if all you want to
819 * do is submit a single transfer and wait for its completion, then using
820 * one of the synchronous I/O functions is much easier.
822 * \section eventlock The events lock
824 * The problem is when we consider the fact that libusbx exposes file
825 * descriptors to allow for you to integrate asynchronous USB I/O into
826 * existing main loops, effectively allowing you to do some work behind
827 * libusbx's back. If you do take libusbx's file descriptors and pass them to
828 * poll()/select() yourself, you need to be aware of the associated issues.
830 * The first concept to be introduced is the events lock. The events lock
831 * is used to serialize threads that want to handle events, such that only
832 * one thread is handling events at any one time.
834 * You must take the events lock before polling libusbx file descriptors,
835 * using libusb_lock_events(). You must release the lock as soon as you have
836 * aborted your poll()/select() loop, using libusb_unlock_events().
838 * \section threadwait Letting other threads do the work for you
840 * Although the events lock is a critical part of the solution, it is not
841 * enough on it's own. You might wonder if the following is sufficient...
843 libusb_lock_events(ctx);
845 poll(libusbx file descriptors, 120*1000);
846 if (poll indicates activity)
847 libusb_handle_events_timeout(ctx, &zero_tv);
849 libusb_unlock_events(ctx);
851 * ...and the answer is that it is not. This is because the transfer in the
852 * code shown above may take a long time (say 30 seconds) to complete, and
853 * the lock is not released until the transfer is completed.
855 * Another thread with similar code that wants to do event handling may be
856 * working with a transfer that completes after a few milliseconds. Despite
857 * having such a quick completion time, the other thread cannot check that
858 * status of its transfer until the code above has finished (30 seconds later)
859 * due to contention on the lock.
861 * To solve this, libusbx offers you a mechanism to determine when another
862 * thread is handling events. It also offers a mechanism to block your thread
863 * until the event handling thread has completed an event (and this mechanism
864 * does not involve polling of file descriptors).
866 * After determining that another thread is currently handling events, you
867 * obtain the <em>event waiters</em> lock using libusb_lock_event_waiters().
868 * You then re-check that some other thread is still handling events, and if
869 * so, you call libusb_wait_for_event().
871 * libusb_wait_for_event() puts your application to sleep until an event
872 * occurs, or until a thread releases the events lock. When either of these
873 * things happen, your thread is woken up, and should re-check the condition
874 * it was waiting on. It should also re-check that another thread is handling
875 * events, and if not, it should start handling events itself.
877 * This looks like the following, as pseudo-code:
880 if (libusb_try_lock_events(ctx) == 0) {
881 // we obtained the event lock: do our own event handling
883 if (!libusb_event_handling_ok(ctx)) {
884 libusb_unlock_events(ctx);
887 poll(libusbx file descriptors, 120*1000);
888 if (poll indicates activity)
889 libusb_handle_events_locked(ctx, 0);
891 libusb_unlock_events(ctx);
893 // another thread is doing event handling. wait for it to signal us that
894 // an event has completed
895 libusb_lock_event_waiters(ctx);
898 // now that we have the event waiters lock, double check that another
899 // thread is still handling events for us. (it may have ceased handling
900 // events in the time it took us to reach this point)
901 if (!libusb_event_handler_active(ctx)) {
902 // whoever was handling events is no longer doing so, try again
903 libusb_unlock_event_waiters(ctx);
907 libusb_wait_for_event(ctx, NULL);
909 libusb_unlock_event_waiters(ctx);
911 printf("completed!\n");
914 * A naive look at the above code may suggest that this can only support
915 * one event waiter (hence a total of 2 competing threads, the other doing
916 * event handling), because the event waiter seems to have taken the event
917 * waiters lock while waiting for an event. However, the system does support
918 * multiple event waiters, because libusb_wait_for_event() actually drops
919 * the lock while waiting, and reaquires it before continuing.
921 * We have now implemented code which can dynamically handle situations where
922 * nobody is handling events (so we should do it ourselves), and it can also
923 * handle situations where another thread is doing event handling (so we can
924 * piggyback onto them). It is also equipped to handle a combination of
925 * the two, for example, another thread is doing event handling, but for
926 * whatever reason it stops doing so before our condition is met, so we take
927 * over the event handling.
929 * Four functions were introduced in the above pseudo-code. Their importance
930 * should be apparent from the code shown above.
931 * -# libusb_try_lock_events() is a non-blocking function which attempts
932 * to acquire the events lock but returns a failure code if it is contended.
933 * -# libusb_event_handling_ok() checks that libusbx is still happy for your
934 * thread to be performing event handling. Sometimes, libusbx needs to
935 * interrupt the event handler, and this is how you can check if you have
936 * been interrupted. If this function returns 0, the correct behaviour is
937 * for you to give up the event handling lock, and then to repeat the cycle.
938 * The following libusb_try_lock_events() will fail, so you will become an
939 * events waiter. For more information on this, read \ref fullstory below.
940 * -# libusb_handle_events_locked() is a variant of
941 * libusb_handle_events_timeout() that you can call while holding the
942 * events lock. libusb_handle_events_timeout() itself implements similar
943 * logic to the above, so be sure not to call it when you are
944 * "working behind libusbx's back", as is the case here.
945 * -# libusb_event_handler_active() determines if someone is currently
946 * holding the events lock
948 * You might be wondering why there is no function to wake up all threads
949 * blocked on libusb_wait_for_event(). This is because libusbx can do this
950 * internally: it will wake up all such threads when someone calls
951 * libusb_unlock_events() or when a transfer completes (at the point after its
952 * callback has returned).
954 * \subsection fullstory The full story
956 * The above explanation should be enough to get you going, but if you're
957 * really thinking through the issues then you may be left with some more
958 * questions regarding libusbx's internals. If you're curious, read on, and if
959 * not, skip to the next section to avoid confusing yourself!
961 * The immediate question that may spring to mind is: what if one thread
962 * modifies the set of file descriptors that need to be polled while another
963 * thread is doing event handling?
965 * There are 2 situations in which this may happen.
966 * -# libusb_open() will add another file descriptor to the poll set,
967 * therefore it is desirable to interrupt the event handler so that it
968 * restarts, picking up the new descriptor.
969 * -# libusb_close() will remove a file descriptor from the poll set. There
970 * are all kinds of race conditions that could arise here, so it is
971 * important that nobody is doing event handling at this time.
973 * libusbx handles these issues internally, so application developers do not
974 * have to stop their event handlers while opening/closing devices. Here's how
975 * it works, focusing on the libusb_close() situation first:
977 * -# During initialization, libusbx opens an internal pipe, and it adds the read
978 * end of this pipe to the set of file descriptors to be polled.
979 * -# During libusb_close(), libusbx writes some dummy data on this control pipe.
980 * This immediately interrupts the event handler. libusbx also records
981 * internally that it is trying to interrupt event handlers for this
982 * high-priority event.
983 * -# At this point, some of the functions described above start behaving
985 * - libusb_event_handling_ok() starts returning 1, indicating that it is NOT
986 * OK for event handling to continue.
987 * - libusb_try_lock_events() starts returning 1, indicating that another
988 * thread holds the event handling lock, even if the lock is uncontended.
989 * - libusb_event_handler_active() starts returning 1, indicating that
990 * another thread is doing event handling, even if that is not true.
991 * -# The above changes in behaviour result in the event handler stopping and
992 * giving up the events lock very quickly, giving the high-priority
993 * libusb_close() operation a "free ride" to acquire the events lock. All
994 * threads that are competing to do event handling become event waiters.
995 * -# With the events lock held inside libusb_close(), libusbx can safely remove
996 * a file descriptor from the poll set, in the safety of knowledge that
997 * nobody is polling those descriptors or trying to access the poll set.
998 * -# After obtaining the events lock, the close operation completes very
999 * quickly (usually a matter of milliseconds) and then immediately releases
1001 * -# At the same time, the behaviour of libusb_event_handling_ok() and friends
1002 * reverts to the original, documented behaviour.
1003 * -# The release of the events lock causes the threads that are waiting for
1004 * events to be woken up and to start competing to become event handlers
1005 * again. One of them will succeed; it will then re-obtain the list of poll
1006 * descriptors, and USB I/O will then continue as normal.
1008 * libusb_open() is similar, and is actually a more simplistic case. Upon a
1009 * call to libusb_open():
1011 * -# The device is opened and a file descriptor is added to the poll set.
1012 * -# libusbx sends some dummy data on the control pipe, and records that it
1013 * is trying to modify the poll descriptor set.
1014 * -# The event handler is interrupted, and the same behaviour change as for
1015 * libusb_close() takes effect, causing all event handling threads to become
1017 * -# The libusb_open() implementation takes its free ride to the events lock.
1018 * -# Happy that it has successfully paused the events handler, libusb_open()
1019 * releases the events lock.
1020 * -# The event waiter threads are all woken up and compete to become event
1021 * handlers again. The one that succeeds will obtain the list of poll
1022 * descriptors again, which will include the addition of the new device.
1024 * \subsection concl Closing remarks
1026 * The above may seem a little complicated, but hopefully I have made it clear
1027 * why such complications are necessary. Also, do not forget that this only
1028 * applies to applications that take libusbx's file descriptors and integrate
1029 * them into their own polling loops.
1031 * You may decide that it is OK for your multi-threaded application to ignore
1032 * some of the rules and locks detailed above, because you don't think that
1033 * two threads can ever be polling the descriptors at the same time. If that
1034 * is the case, then that's good news for you because you don't have to worry.
1035 * But be careful here; remember that the synchronous I/O functions do event
1036 * handling internally. If you have one thread doing event handling in a loop
1037 * (without implementing the rules and locking semantics documented above)
1038 * and another trying to send a synchronous USB transfer, you will end up with
1039 * two threads monitoring the same descriptors, and the above-described
1040 * undesirable behaviour occuring. The solution is for your polling thread to
1041 * play by the rules; the synchronous I/O functions do so, and this will result
1042 * in them getting along in perfect harmony.
1044 * If you do have a dedicated thread doing event handling, it is perfectly
1045 * legal for it to take the event handling lock for long periods of time. Any
1046 * synchronous I/O functions you call from other threads will transparently
1047 * fall back to the "event waiters" mechanism detailed above. The only
1048 * consideration that your event handling thread must apply is the one related
1049 * to libusb_event_handling_ok(): you must call this before every poll(), and
1050 * give up the events lock if instructed.
1053 int usbi_io_init(struct libusb_context *ctx)
1057 usbi_mutex_init(&ctx->flying_transfers_lock, NULL);
1058 usbi_mutex_init(&ctx->pollfds_lock, NULL);
1059 usbi_mutex_init(&ctx->pollfd_modify_lock, NULL);
1060 usbi_mutex_init_recursive(&ctx->events_lock, NULL);
1061 usbi_mutex_init(&ctx->event_waiters_lock, NULL);
1062 usbi_cond_init(&ctx->event_waiters_cond, NULL);
1063 list_init(&ctx->flying_transfers);
1064 list_init(&ctx->pollfds);
1066 /* FIXME should use an eventfd on kernels that support it */
1067 r = usbi_pipe(ctx->ctrl_pipe);
1069 r = LIBUSB_ERROR_OTHER;
1073 r = usbi_add_pollfd(ctx, ctx->ctrl_pipe[0], POLLIN);
1075 goto err_close_pipe;
1077 /* create hotplug pipe */
1078 r = usbi_pipe(ctx->hotplug_pipe);
1080 r = LIBUSB_ERROR_OTHER;
1085 fcntl(ctx->hotplug_pipe[1], F_SETFD, O_NONBLOCK);
1087 r = usbi_add_pollfd(ctx, ctx->hotplug_pipe[0], POLLIN);
1089 goto err_close_hp_pipe;
1091 #ifdef USBI_TIMERFD_AVAILABLE
1092 ctx->timerfd = timerfd_create(usbi_backend->get_timerfd_clockid(),
1094 if (ctx->timerfd >= 0) {
1095 usbi_dbg("using timerfd for timeouts");
1096 r = usbi_add_pollfd(ctx, ctx->timerfd, POLLIN);
1098 usbi_remove_pollfd(ctx, ctx->ctrl_pipe[0]);
1099 close(ctx->timerfd);
1100 goto err_close_hp_pipe;
1103 usbi_dbg("timerfd not available (code %d error %d)", ctx->timerfd, errno);
1111 usbi_close(ctx->hotplug_pipe[0]);
1112 usbi_close(ctx->hotplug_pipe[1]);
1114 usbi_close(ctx->ctrl_pipe[0]);
1115 usbi_close(ctx->ctrl_pipe[1]);
1117 usbi_mutex_destroy(&ctx->flying_transfers_lock);
1118 usbi_mutex_destroy(&ctx->pollfds_lock);
1119 usbi_mutex_destroy(&ctx->pollfd_modify_lock);
1120 usbi_mutex_destroy(&ctx->events_lock);
1121 usbi_mutex_destroy(&ctx->event_waiters_lock);
1122 usbi_cond_destroy(&ctx->event_waiters_cond);
1126 void usbi_io_exit(struct libusb_context *ctx)
1128 usbi_remove_pollfd(ctx, ctx->ctrl_pipe[0]);
1129 usbi_close(ctx->ctrl_pipe[0]);
1130 usbi_close(ctx->ctrl_pipe[1]);
1131 usbi_remove_pollfd(ctx, ctx->hotplug_pipe[0]);
1132 usbi_close(ctx->hotplug_pipe[0]);
1133 usbi_close(ctx->hotplug_pipe[1]);
1134 #ifdef USBI_TIMERFD_AVAILABLE
1135 if (usbi_using_timerfd(ctx)) {
1136 usbi_remove_pollfd(ctx, ctx->timerfd);
1137 close(ctx->timerfd);
1140 usbi_mutex_destroy(&ctx->flying_transfers_lock);
1141 usbi_mutex_destroy(&ctx->pollfds_lock);
1142 usbi_mutex_destroy(&ctx->pollfd_modify_lock);
1143 usbi_mutex_destroy(&ctx->events_lock);
1144 usbi_mutex_destroy(&ctx->event_waiters_lock);
1145 usbi_cond_destroy(&ctx->event_waiters_cond);
1148 static int calculate_timeout(struct usbi_transfer *transfer)
1151 struct timespec current_time;
1152 unsigned int timeout =
1153 USBI_TRANSFER_TO_LIBUSB_TRANSFER(transfer)->timeout;
1158 r = usbi_backend->clock_gettime(USBI_CLOCK_MONOTONIC, ¤t_time);
1160 usbi_err(ITRANSFER_CTX(transfer),
1161 "failed to read monotonic clock, errno=%d", errno);
1165 current_time.tv_sec += timeout / 1000;
1166 current_time.tv_nsec += (timeout % 1000) * 1000000;
1168 while (current_time.tv_nsec >= 1000000000) {
1169 current_time.tv_nsec -= 1000000000;
1170 current_time.tv_sec++;
1173 TIMESPEC_TO_TIMEVAL(&transfer->timeout, ¤t_time);
1177 /* add a transfer to the (timeout-sorted) active transfers list.
1178 * returns 1 if the transfer has a timeout and it is the timeout next to
1180 static int add_to_flying_list(struct usbi_transfer *transfer)
1182 struct usbi_transfer *cur;
1183 struct timeval *timeout = &transfer->timeout;
1184 struct libusb_context *ctx = ITRANSFER_CTX(transfer);
1188 usbi_mutex_lock(&ctx->flying_transfers_lock);
1190 /* if we have no other flying transfers, start the list with this one */
1191 if (list_empty(&ctx->flying_transfers)) {
1192 list_add(&transfer->list, &ctx->flying_transfers);
1196 /* if we have infinite timeout, append to end of list */
1197 if (!timerisset(timeout)) {
1198 list_add_tail(&transfer->list, &ctx->flying_transfers);
1199 /* first is irrelevant in this case */
1203 /* otherwise, find appropriate place in list */
1204 list_for_each_entry(cur, &ctx->flying_transfers, list, struct usbi_transfer) {
1205 /* find first timeout that occurs after the transfer in question */
1206 struct timeval *cur_tv = &cur->timeout;
1208 if (!timerisset(cur_tv) || (cur_tv->tv_sec > timeout->tv_sec) ||
1209 (cur_tv->tv_sec == timeout->tv_sec &&
1210 cur_tv->tv_usec > timeout->tv_usec)) {
1211 list_add_tail(&transfer->list, &cur->list);
1216 /* first is 0 at this stage (list not empty) */
1218 /* otherwise we need to be inserted at the end */
1219 list_add_tail(&transfer->list, &ctx->flying_transfers);
1221 #ifdef USBI_TIMERFD_AVAILABLE
1222 if (first && usbi_using_timerfd(ctx) && timerisset(timeout)) {
1223 /* if this transfer has the lowest timeout of all active transfers,
1224 * rearm the timerfd with this transfer's timeout */
1225 const struct itimerspec it = { {0, 0},
1226 { timeout->tv_sec, timeout->tv_usec * 1000 } };
1227 usbi_dbg("arm timerfd for timeout in %dms (first in line)",
1228 USBI_TRANSFER_TO_LIBUSB_TRANSFER(transfer)->timeout);
1229 r = timerfd_settime(ctx->timerfd, TFD_TIMER_ABSTIME, &it, NULL);
1231 usbi_warn(ctx, "failed to arm first timerfd (errno %d)", errno);
1232 r = LIBUSB_ERROR_OTHER;
1239 usbi_mutex_unlock(&ctx->flying_transfers_lock);
1243 /** \ingroup asyncio
1244 * Allocate a libusbx transfer with a specified number of isochronous packet
1245 * descriptors. The returned transfer is pre-initialized for you. When the new
1246 * transfer is no longer needed, it should be freed with
1247 * libusb_free_transfer().
1249 * Transfers intended for non-isochronous endpoints (e.g. control, bulk,
1250 * interrupt) should specify an iso_packets count of zero.
1252 * For transfers intended for isochronous endpoints, specify an appropriate
1253 * number of packet descriptors to be allocated as part of the transfer.
1254 * The returned transfer is not specially initialized for isochronous I/O;
1255 * you are still required to set the
1256 * \ref libusb_transfer::num_iso_packets "num_iso_packets" and
1257 * \ref libusb_transfer::type "type" fields accordingly.
1259 * It is safe to allocate a transfer with some isochronous packets and then
1260 * use it on a non-isochronous endpoint. If you do this, ensure that at time
1261 * of submission, num_iso_packets is 0 and that type is set appropriately.
1263 * \param iso_packets number of isochronous packet descriptors to allocate
1264 * \returns a newly allocated transfer, or NULL on error
1267 struct libusb_transfer * LIBUSB_CALL libusb_alloc_transfer(
1270 size_t os_alloc_size = usbi_backend->transfer_priv_size
1271 + (usbi_backend->add_iso_packet_size * iso_packets);
1272 size_t alloc_size = sizeof(struct usbi_transfer)
1273 + sizeof(struct libusb_transfer)
1274 + (sizeof(struct libusb_iso_packet_descriptor) * iso_packets)
1276 struct usbi_transfer *itransfer = calloc(1, alloc_size);
1280 itransfer->num_iso_packets = iso_packets;
1281 usbi_mutex_init(&itransfer->lock, NULL);
1282 return USBI_TRANSFER_TO_LIBUSB_TRANSFER(itransfer);
1285 /** \ingroup asyncio
1286 * Free a transfer structure. This should be called for all transfers
1287 * allocated with libusb_alloc_transfer().
1289 * If the \ref libusb_transfer_flags::LIBUSB_TRANSFER_FREE_BUFFER
1290 * "LIBUSB_TRANSFER_FREE_BUFFER" flag is set and the transfer buffer is
1291 * non-NULL, this function will also free the transfer buffer using the
1292 * standard system memory allocator (e.g. free()).
1294 * It is legal to call this function with a NULL transfer. In this case,
1295 * the function will simply return safely.
1297 * It is not legal to free an active transfer (one which has been submitted
1298 * and has not yet completed).
1300 * \param transfer the transfer to free
1302 void API_EXPORTED libusb_free_transfer(struct libusb_transfer *transfer)
1304 struct usbi_transfer *itransfer;
1308 if (transfer->flags & LIBUSB_TRANSFER_FREE_BUFFER && transfer->buffer)
1309 free(transfer->buffer);
1311 itransfer = LIBUSB_TRANSFER_TO_USBI_TRANSFER(transfer);
1312 usbi_mutex_destroy(&itransfer->lock);
1316 #ifdef USBI_TIMERFD_AVAILABLE
1317 static int disarm_timerfd(struct libusb_context *ctx)
1319 const struct itimerspec disarm_timer = { { 0, 0 }, { 0, 0 } };
1323 r = timerfd_settime(ctx->timerfd, 0, &disarm_timer, NULL);
1325 return LIBUSB_ERROR_OTHER;
1330 /* iterates through the flying transfers, and rearms the timerfd based on the
1331 * next upcoming timeout.
1332 * must be called with flying_list locked.
1333 * returns 0 if there was no timeout to arm, 1 if the next timeout was armed,
1334 * or a LIBUSB_ERROR code on failure.
1336 static int arm_timerfd_for_next_timeout(struct libusb_context *ctx)
1338 struct usbi_transfer *transfer;
1340 list_for_each_entry(transfer, &ctx->flying_transfers, list, struct usbi_transfer) {
1341 struct timeval *cur_tv = &transfer->timeout;
1343 /* if we've reached transfers of infinite timeout, then we have no
1345 if (!timerisset(cur_tv))
1348 /* act on first transfer that is not already cancelled */
1349 if (!(transfer->flags & USBI_TRANSFER_TIMED_OUT)) {
1351 const struct itimerspec it = { {0, 0},
1352 { cur_tv->tv_sec, cur_tv->tv_usec * 1000 } };
1353 usbi_dbg("next timeout originally %dms", USBI_TRANSFER_TO_LIBUSB_TRANSFER(transfer)->timeout);
1354 r = timerfd_settime(ctx->timerfd, TFD_TIMER_ABSTIME, &it, NULL);
1356 return LIBUSB_ERROR_OTHER;
1362 return disarm_timerfd(ctx);
1365 static int arm_timerfd_for_next_timeout(struct libusb_context *ctx)
1372 /** \ingroup asyncio
1373 * Submit a transfer. This function will fire off the USB transfer and then
1374 * return immediately.
1376 * \param transfer the transfer to submit
1377 * \returns 0 on success
1378 * \returns LIBUSB_ERROR_NO_DEVICE if the device has been disconnected
1379 * \returns LIBUSB_ERROR_BUSY if the transfer has already been submitted.
1380 * \returns LIBUSB_ERROR_NOT_SUPPORTED if the transfer flags are not supported
1381 * by the operating system.
1382 * \returns another LIBUSB_ERROR code on other failure
1384 int API_EXPORTED libusb_submit_transfer(struct libusb_transfer *transfer)
1386 struct libusb_context *ctx = TRANSFER_CTX(transfer);
1387 struct usbi_transfer *itransfer =
1388 LIBUSB_TRANSFER_TO_USBI_TRANSFER(transfer);
1392 usbi_mutex_lock(&itransfer->lock);
1393 itransfer->transferred = 0;
1394 itransfer->flags = 0;
1395 r = calculate_timeout(itransfer);
1397 r = LIBUSB_ERROR_OTHER;
1401 r = add_to_flying_list(itransfer);
1404 r = usbi_backend->submit_transfer(itransfer);
1406 usbi_mutex_lock(&ctx->flying_transfers_lock);
1407 list_del(&itransfer->list);
1408 arm_timerfd_for_next_timeout(ctx);
1409 usbi_mutex_unlock(&ctx->flying_transfers_lock);
1413 updated_fds = (itransfer->flags & USBI_TRANSFER_UPDATED_FDS);
1414 usbi_mutex_unlock(&itransfer->lock);
1416 usbi_fd_notification(ctx);
1420 /** \ingroup asyncio
1421 * Asynchronously cancel a previously submitted transfer.
1422 * This function returns immediately, but this does not indicate cancellation
1423 * is complete. Your callback function will be invoked at some later time
1424 * with a transfer status of
1425 * \ref libusb_transfer_status::LIBUSB_TRANSFER_CANCELLED
1426 * "LIBUSB_TRANSFER_CANCELLED."
1428 * \param transfer the transfer to cancel
1429 * \returns 0 on success
1430 * \returns LIBUSB_ERROR_NOT_FOUND if the transfer is already complete or
1432 * \returns a LIBUSB_ERROR code on failure
1434 int API_EXPORTED libusb_cancel_transfer(struct libusb_transfer *transfer)
1436 struct usbi_transfer *itransfer =
1437 LIBUSB_TRANSFER_TO_USBI_TRANSFER(transfer);
1441 usbi_mutex_lock(&itransfer->lock);
1442 r = usbi_backend->cancel_transfer(itransfer);
1444 if (r != LIBUSB_ERROR_NOT_FOUND &&
1445 r != LIBUSB_ERROR_NO_DEVICE)
1446 usbi_err(TRANSFER_CTX(transfer),
1447 "cancel transfer failed error %d", r);
1449 usbi_dbg("cancel transfer failed error %d", r);
1451 if (r == LIBUSB_ERROR_NO_DEVICE)
1452 itransfer->flags |= USBI_TRANSFER_DEVICE_DISAPPEARED;
1455 itransfer->flags |= USBI_TRANSFER_CANCELLING;
1457 usbi_mutex_unlock(&itransfer->lock);
1461 /* Handle completion of a transfer (completion might be an error condition).
1462 * This will invoke the user-supplied callback function, which may end up
1463 * freeing the transfer. Therefore you cannot use the transfer structure
1464 * after calling this function, and you should free all backend-specific
1465 * data before calling it.
1466 * Do not call this function with the usbi_transfer lock held. User-specified
1467 * callback functions may attempt to directly resubmit the transfer, which
1468 * will attempt to take the lock. */
1469 int usbi_handle_transfer_completion(struct usbi_transfer *itransfer,
1470 enum libusb_transfer_status status)
1472 struct libusb_transfer *transfer =
1473 USBI_TRANSFER_TO_LIBUSB_TRANSFER(itransfer);
1474 struct libusb_context *ctx = TRANSFER_CTX(transfer);
1478 /* FIXME: could be more intelligent with the timerfd here. we don't need
1479 * to disarm the timerfd if there was no timer running, and we only need
1480 * to rearm the timerfd if the transfer that expired was the one with
1481 * the shortest timeout. */
1483 usbi_mutex_lock(&ctx->flying_transfers_lock);
1484 list_del(&itransfer->list);
1485 if (usbi_using_timerfd(ctx))
1486 r = arm_timerfd_for_next_timeout(ctx);
1487 usbi_mutex_unlock(&ctx->flying_transfers_lock);
1488 if (usbi_using_timerfd(ctx) && (r < 0))
1491 if (status == LIBUSB_TRANSFER_COMPLETED
1492 && transfer->flags & LIBUSB_TRANSFER_SHORT_NOT_OK) {
1493 int rqlen = transfer->length;
1494 if (transfer->type == LIBUSB_TRANSFER_TYPE_CONTROL)
1495 rqlen -= LIBUSB_CONTROL_SETUP_SIZE;
1496 if (rqlen != itransfer->transferred) {
1497 usbi_dbg("interpreting short transfer as error");
1498 status = LIBUSB_TRANSFER_ERROR;
1502 flags = transfer->flags;
1503 transfer->status = status;
1504 transfer->actual_length = itransfer->transferred;
1505 usbi_dbg("transfer %p has callback %p", transfer, transfer->callback);
1506 if (transfer->callback)
1507 transfer->callback(transfer);
1508 /* transfer might have been freed by the above call, do not use from
1510 if (flags & LIBUSB_TRANSFER_FREE_TRANSFER)
1511 libusb_free_transfer(transfer);
1512 usbi_mutex_lock(&ctx->event_waiters_lock);
1513 usbi_cond_broadcast(&ctx->event_waiters_cond);
1514 usbi_mutex_unlock(&ctx->event_waiters_lock);
1518 /* Similar to usbi_handle_transfer_completion() but exclusively for transfers
1519 * that were asynchronously cancelled. The same concerns w.r.t. freeing of
1520 * transfers exist here.
1521 * Do not call this function with the usbi_transfer lock held. User-specified
1522 * callback functions may attempt to directly resubmit the transfer, which
1523 * will attempt to take the lock. */
1524 int usbi_handle_transfer_cancellation(struct usbi_transfer *transfer)
1526 /* if the URB was cancelled due to timeout, report timeout to the user */
1527 if (transfer->flags & USBI_TRANSFER_TIMED_OUT) {
1528 usbi_dbg("detected timeout cancellation");
1529 return usbi_handle_transfer_completion(transfer, LIBUSB_TRANSFER_TIMED_OUT);
1532 /* otherwise its a normal async cancel */
1533 return usbi_handle_transfer_completion(transfer, LIBUSB_TRANSFER_CANCELLED);
1537 * Attempt to acquire the event handling lock. This lock is used to ensure that
1538 * only one thread is monitoring libusbx event sources at any one time.
1540 * You only need to use this lock if you are developing an application
1541 * which calls poll() or select() on libusbx's file descriptors directly.
1542 * If you stick to libusbx's event handling loop functions (e.g.
1543 * libusb_handle_events()) then you do not need to be concerned with this
1546 * While holding this lock, you are trusted to actually be handling events.
1547 * If you are no longer handling events, you must call libusb_unlock_events()
1548 * as soon as possible.
1550 * \param ctx the context to operate on, or NULL for the default context
1551 * \returns 0 if the lock was obtained successfully
1552 * \returns 1 if the lock was not obtained (i.e. another thread holds the lock)
1555 int API_EXPORTED libusb_try_lock_events(libusb_context *ctx)
1559 USBI_GET_CONTEXT(ctx);
1561 /* is someone else waiting to modify poll fds? if so, don't let this thread
1562 * start event handling */
1563 usbi_mutex_lock(&ctx->pollfd_modify_lock);
1564 ru = ctx->pollfd_modify;
1565 usbi_mutex_unlock(&ctx->pollfd_modify_lock);
1567 usbi_dbg("someone else is modifying poll fds");
1571 r = usbi_mutex_trylock(&ctx->events_lock);
1575 ctx->event_handler_active = 1;
1580 * Acquire the event handling lock, blocking until successful acquisition if
1581 * it is contended. This lock is used to ensure that only one thread is
1582 * monitoring libusbx event sources at any one time.
1584 * You only need to use this lock if you are developing an application
1585 * which calls poll() or select() on libusbx's file descriptors directly.
1586 * If you stick to libusbx's event handling loop functions (e.g.
1587 * libusb_handle_events()) then you do not need to be concerned with this
1590 * While holding this lock, you are trusted to actually be handling events.
1591 * If you are no longer handling events, you must call libusb_unlock_events()
1592 * as soon as possible.
1594 * \param ctx the context to operate on, or NULL for the default context
1597 void API_EXPORTED libusb_lock_events(libusb_context *ctx)
1599 USBI_GET_CONTEXT(ctx);
1600 usbi_mutex_lock(&ctx->events_lock);
1601 ctx->event_handler_active = 1;
1605 * Release the lock previously acquired with libusb_try_lock_events() or
1606 * libusb_lock_events(). Releasing this lock will wake up any threads blocked
1607 * on libusb_wait_for_event().
1609 * \param ctx the context to operate on, or NULL for the default context
1612 void API_EXPORTED libusb_unlock_events(libusb_context *ctx)
1614 USBI_GET_CONTEXT(ctx);
1615 ctx->event_handler_active = 0;
1616 usbi_mutex_unlock(&ctx->events_lock);
1618 /* FIXME: perhaps we should be a bit more efficient by not broadcasting
1619 * the availability of the events lock when we are modifying pollfds
1620 * (check ctx->pollfd_modify)? */
1621 usbi_mutex_lock(&ctx->event_waiters_lock);
1622 usbi_cond_broadcast(&ctx->event_waiters_cond);
1623 usbi_mutex_unlock(&ctx->event_waiters_lock);
1627 * Determine if it is still OK for this thread to be doing event handling.
1629 * Sometimes, libusbx needs to temporarily pause all event handlers, and this
1630 * is the function you should use before polling file descriptors to see if
1633 * If this function instructs your thread to give up the events lock, you
1634 * should just continue the usual logic that is documented in \ref mtasync.
1635 * On the next iteration, your thread will fail to obtain the events lock,
1636 * and will hence become an event waiter.
1638 * This function should be called while the events lock is held: you don't
1639 * need to worry about the results of this function if your thread is not
1640 * the current event handler.
1642 * \param ctx the context to operate on, or NULL for the default context
1643 * \returns 1 if event handling can start or continue
1644 * \returns 0 if this thread must give up the events lock
1645 * \see \ref fullstory "Multi-threaded I/O: the full story"
1647 int API_EXPORTED libusb_event_handling_ok(libusb_context *ctx)
1650 USBI_GET_CONTEXT(ctx);
1652 /* is someone else waiting to modify poll fds? if so, don't let this thread
1653 * continue event handling */
1654 usbi_mutex_lock(&ctx->pollfd_modify_lock);
1655 r = ctx->pollfd_modify;
1656 usbi_mutex_unlock(&ctx->pollfd_modify_lock);
1658 usbi_dbg("someone else is modifying poll fds");
1667 * Determine if an active thread is handling events (i.e. if anyone is holding
1668 * the event handling lock).
1670 * \param ctx the context to operate on, or NULL for the default context
1671 * \returns 1 if a thread is handling events
1672 * \returns 0 if there are no threads currently handling events
1675 int API_EXPORTED libusb_event_handler_active(libusb_context *ctx)
1678 USBI_GET_CONTEXT(ctx);
1680 /* is someone else waiting to modify poll fds? if so, don't let this thread
1681 * start event handling -- indicate that event handling is happening */
1682 usbi_mutex_lock(&ctx->pollfd_modify_lock);
1683 r = ctx->pollfd_modify;
1684 usbi_mutex_unlock(&ctx->pollfd_modify_lock);
1686 usbi_dbg("someone else is modifying poll fds");
1690 return ctx->event_handler_active;
1694 * Acquire the event waiters lock. This lock is designed to be obtained under
1695 * the situation where you want to be aware when events are completed, but
1696 * some other thread is event handling so calling libusb_handle_events() is not
1699 * You then obtain this lock, re-check that another thread is still handling
1700 * events, then call libusb_wait_for_event().
1702 * You only need to use this lock if you are developing an application
1703 * which calls poll() or select() on libusbx's file descriptors directly,
1704 * <b>and</b> may potentially be handling events from 2 threads simultaenously.
1705 * If you stick to libusbx's event handling loop functions (e.g.
1706 * libusb_handle_events()) then you do not need to be concerned with this
1709 * \param ctx the context to operate on, or NULL for the default context
1712 void API_EXPORTED libusb_lock_event_waiters(libusb_context *ctx)
1714 USBI_GET_CONTEXT(ctx);
1715 usbi_mutex_lock(&ctx->event_waiters_lock);
1719 * Release the event waiters lock.
1720 * \param ctx the context to operate on, or NULL for the default context
1723 void API_EXPORTED libusb_unlock_event_waiters(libusb_context *ctx)
1725 USBI_GET_CONTEXT(ctx);
1726 usbi_mutex_unlock(&ctx->event_waiters_lock);
1730 * Wait for another thread to signal completion of an event. Must be called
1731 * with the event waiters lock held, see libusb_lock_event_waiters().
1733 * This function will block until any of the following conditions are met:
1734 * -# The timeout expires
1735 * -# A transfer completes
1736 * -# A thread releases the event handling lock through libusb_unlock_events()
1738 * Condition 1 is obvious. Condition 2 unblocks your thread <em>after</em>
1739 * the callback for the transfer has completed. Condition 3 is important
1740 * because it means that the thread that was previously handling events is no
1741 * longer doing so, so if any events are to complete, another thread needs to
1742 * step up and start event handling.
1744 * This function releases the event waiters lock before putting your thread
1745 * to sleep, and reacquires the lock as it is being woken up.
1747 * \param ctx the context to operate on, or NULL for the default context
1748 * \param tv maximum timeout for this blocking function. A NULL value
1749 * indicates unlimited timeout.
1750 * \returns 0 after a transfer completes or another thread stops event handling
1751 * \returns 1 if the timeout expired
1754 int API_EXPORTED libusb_wait_for_event(libusb_context *ctx, struct timeval *tv)
1756 struct timespec timeout;
1759 USBI_GET_CONTEXT(ctx);
1761 usbi_cond_wait(&ctx->event_waiters_cond, &ctx->event_waiters_lock);
1765 r = usbi_backend->clock_gettime(USBI_CLOCK_REALTIME, &timeout);
1767 usbi_err(ctx, "failed to read realtime clock, error %d", errno);
1768 return LIBUSB_ERROR_OTHER;
1771 timeout.tv_sec += tv->tv_sec;
1772 timeout.tv_nsec += tv->tv_usec * 1000;
1773 while (timeout.tv_nsec >= 1000000000) {
1774 timeout.tv_nsec -= 1000000000;
1778 r = usbi_cond_timedwait(&ctx->event_waiters_cond,
1779 &ctx->event_waiters_lock, &timeout);
1780 return (r == ETIMEDOUT);
1783 static void handle_timeout(struct usbi_transfer *itransfer)
1785 struct libusb_transfer *transfer =
1786 USBI_TRANSFER_TO_LIBUSB_TRANSFER(itransfer);
1789 itransfer->flags |= USBI_TRANSFER_TIMED_OUT;
1790 r = libusb_cancel_transfer(transfer);
1792 usbi_warn(TRANSFER_CTX(transfer),
1793 "async cancel failed %d errno=%d", r, errno);
1796 static int handle_timeouts_locked(struct libusb_context *ctx)
1799 struct timespec systime_ts;
1800 struct timeval systime;
1801 struct usbi_transfer *transfer;
1803 if (list_empty(&ctx->flying_transfers))
1806 /* get current time */
1807 r = usbi_backend->clock_gettime(USBI_CLOCK_MONOTONIC, &systime_ts);
1811 TIMESPEC_TO_TIMEVAL(&systime, &systime_ts);
1813 /* iterate through flying transfers list, finding all transfers that
1814 * have expired timeouts */
1815 list_for_each_entry(transfer, &ctx->flying_transfers, list, struct usbi_transfer) {
1816 struct timeval *cur_tv = &transfer->timeout;
1818 /* if we've reached transfers of infinite timeout, we're all done */
1819 if (!timerisset(cur_tv))
1822 /* ignore timeouts we've already handled */
1823 if (transfer->flags & (USBI_TRANSFER_TIMED_OUT | USBI_TRANSFER_OS_HANDLES_TIMEOUT))
1826 /* if transfer has non-expired timeout, nothing more to do */
1827 if ((cur_tv->tv_sec > systime.tv_sec) ||
1828 (cur_tv->tv_sec == systime.tv_sec &&
1829 cur_tv->tv_usec > systime.tv_usec))
1832 /* otherwise, we've got an expired timeout to handle */
1833 handle_timeout(transfer);
1838 static int handle_timeouts(struct libusb_context *ctx)
1841 USBI_GET_CONTEXT(ctx);
1842 usbi_mutex_lock(&ctx->flying_transfers_lock);
1843 r = handle_timeouts_locked(ctx);
1844 usbi_mutex_unlock(&ctx->flying_transfers_lock);
1848 #ifdef USBI_TIMERFD_AVAILABLE
1849 static int handle_timerfd_trigger(struct libusb_context *ctx)
1853 usbi_mutex_lock(&ctx->flying_transfers_lock);
1855 /* process the timeout that just happened */
1856 r = handle_timeouts_locked(ctx);
1860 /* arm for next timeout*/
1861 r = arm_timerfd_for_next_timeout(ctx);
1864 usbi_mutex_unlock(&ctx->flying_transfers_lock);
1869 /* do the actual event handling. assumes that no other thread is concurrently
1870 * doing the same thing. */
1871 static int handle_events(struct libusb_context *ctx, struct timeval *tv)
1874 struct usbi_pollfd *ipollfd;
1875 POLL_NFDS_TYPE nfds = 0;
1876 struct pollfd *fds = NULL;
1880 usbi_mutex_lock(&ctx->pollfds_lock);
1881 list_for_each_entry(ipollfd, &ctx->pollfds, list, struct usbi_pollfd)
1884 /* TODO: malloc when number of fd's changes, not on every poll */
1886 fds = malloc(sizeof(*fds) * nfds);
1888 usbi_mutex_unlock(&ctx->pollfds_lock);
1889 return LIBUSB_ERROR_NO_MEM;
1892 list_for_each_entry(ipollfd, &ctx->pollfds, list, struct usbi_pollfd) {
1893 struct libusb_pollfd *pollfd = &ipollfd->pollfd;
1894 int fd = pollfd->fd;
1897 fds[i].events = pollfd->events;
1900 usbi_mutex_unlock(&ctx->pollfds_lock);
1902 timeout_ms = (int)(tv->tv_sec * 1000) + (tv->tv_usec / 1000);
1904 /* round up to next millisecond */
1905 if (tv->tv_usec % 1000)
1908 usbi_dbg("poll() %d fds with timeout in %dms", nfds, timeout_ms);
1909 r = usbi_poll(fds, nfds, timeout_ms);
1910 usbi_dbg("poll() returned %d", r);
1913 return handle_timeouts(ctx);
1914 } else if (r == -1 && errno == EINTR) {
1916 return LIBUSB_ERROR_INTERRUPTED;
1919 usbi_err(ctx, "poll failed %d err=%d\n", r, errno);
1920 return LIBUSB_ERROR_IO;
1923 /* fd[0] is always the ctrl pipe */
1924 if (fds[0].revents) {
1925 /* another thread wanted to interrupt event handling, and it succeeded!
1926 * handle any other events that cropped up at the same time, and
1928 usbi_dbg("caught a fish on the control pipe");
1934 /* prevent OS backend from trying to handle events on ctrl pipe */
1940 /* fd[1] is always the hotplug pipe */
1941 if (libusb_has_capability(LIBUSB_CAP_HAS_HOTPLUG) && fds[1].revents) {
1942 libusb_hotplug_message message;
1945 /* read the message from the hotplug thread */
1946 ret = usbi_read(ctx->hotplug_pipe[0], &message, sizeof (message));
1947 if (ret < sizeof(message)) {
1948 ret = LIBUSB_ERROR_OTHER;
1952 usbi_hotplug_match(message.device, message.event);
1954 /* the device left. dereference the device */
1955 if (LIBUSB_HOTPLUG_EVENT_DEVICE_LEFT == message.event)
1956 libusb_unref_device(message.device);
1961 } /* else there shouldn't be anything on this pipe */
1963 #ifdef USBI_TIMERFD_AVAILABLE
1964 /* on timerfd configurations, fds[2] is the timerfd */
1965 if (usbi_using_timerfd(ctx) && fds[2].revents) {
1966 /* timerfd indicates that a timeout has expired */
1968 usbi_dbg("timerfd triggered");
1970 ret = handle_timerfd_trigger(ctx);
1972 /* return error code */
1975 } else if (r == 1) {
1976 /* no more active file descriptors, nothing more to do */
1980 /* more events pending...
1981 * prevent OS backend from trying to handle events on timerfd */
1988 r = usbi_backend->handle_events(ctx, fds, nfds, r);
1990 usbi_err(ctx, "backend handle_events failed with error %d", r);
1997 /* returns the smallest of:
1998 * 1. timeout of next URB
1999 * 2. user-supplied timeout
2000 * returns 1 if there is an already-expired timeout, otherwise returns 0
2003 static int get_next_timeout(libusb_context *ctx, struct timeval *tv,
2004 struct timeval *out)
2006 struct timeval timeout;
2007 int r = libusb_get_next_timeout(ctx, &timeout);
2009 /* timeout already expired? */
2010 if (!timerisset(&timeout))
2013 /* choose the smallest of next URB timeout or user specified timeout */
2014 if (timercmp(&timeout, tv, <))
2025 * Handle any pending events.
2027 * libusbx determines "pending events" by checking if any timeouts have expired
2028 * and by checking the set of file descriptors for activity.
2030 * If a zero timeval is passed, this function will handle any already-pending
2031 * events and then immediately return in non-blocking style.
2033 * If a non-zero timeval is passed and no events are currently pending, this
2034 * function will block waiting for events to handle up until the specified
2035 * timeout. If an event arrives or a signal is raised, this function will
2038 * If the parameter completed is not NULL then <em>after obtaining the event
2039 * handling lock</em> this function will return immediately if the integer
2040 * pointed to is not 0. This allows for race free waiting for the completion
2041 * of a specific transfer.
2043 * \param ctx the context to operate on, or NULL for the default context
2044 * \param tv the maximum time to block waiting for events, or an all zero
2045 * timeval struct for non-blocking mode
2046 * \param completed pointer to completion integer to check, or NULL
2047 * \returns 0 on success, or a LIBUSB_ERROR code on failure
2050 int API_EXPORTED libusb_handle_events_timeout_completed(libusb_context *ctx,
2051 struct timeval *tv, int *completed)
2054 struct timeval poll_timeout;
2056 USBI_GET_CONTEXT(ctx);
2057 r = get_next_timeout(ctx, tv, &poll_timeout);
2059 /* timeout already expired */
2060 return handle_timeouts(ctx);
2064 if (libusb_try_lock_events(ctx) == 0) {
2065 if (completed == NULL || !*completed) {
2066 /* we obtained the event lock: do our own event handling */
2067 usbi_dbg("doing our own event handling");
2068 r = handle_events(ctx, &poll_timeout);
2070 libusb_unlock_events(ctx);
2074 /* another thread is doing event handling. wait for thread events that
2075 * notify event completion. */
2076 libusb_lock_event_waiters(ctx);
2078 if (completed && *completed)
2081 if (!libusb_event_handler_active(ctx)) {
2082 /* we hit a race: whoever was event handling earlier finished in the
2083 * time it took us to reach this point. try the cycle again. */
2084 libusb_unlock_event_waiters(ctx);
2085 usbi_dbg("event handler was active but went away, retrying");
2089 usbi_dbg("another thread is doing event handling");
2090 r = libusb_wait_for_event(ctx, &poll_timeout);
2093 libusb_unlock_event_waiters(ctx);
2098 return handle_timeouts(ctx);
2104 * Handle any pending events
2106 * Like libusb_handle_events_timeout_completed(), but without the completed
2107 * parameter, calling this function is equivalent to calling
2108 * libusb_handle_events_timeout_completed() with a NULL completed parameter.
2110 * This function is kept primarily for backwards compatibility.
2111 * All new code should call libusb_handle_events_completed() or
2112 * libusb_handle_events_timeout_completed() to avoid race conditions.
2114 * \param ctx the context to operate on, or NULL for the default context
2115 * \param tv the maximum time to block waiting for events, or an all zero
2116 * timeval struct for non-blocking mode
2117 * \returns 0 on success, or a LIBUSB_ERROR code on failure
2119 int API_EXPORTED libusb_handle_events_timeout(libusb_context *ctx,
2122 return libusb_handle_events_timeout_completed(ctx, tv, NULL);
2126 * Handle any pending events in blocking mode. There is currently a timeout
2127 * hardcoded at 60 seconds but we plan to make it unlimited in future. For
2128 * finer control over whether this function is blocking or non-blocking, or
2129 * for control over the timeout, use libusb_handle_events_timeout_completed()
2132 * This function is kept primarily for backwards compatibility.
2133 * All new code should call libusb_handle_events_completed() or
2134 * libusb_handle_events_timeout_completed() to avoid race conditions.
2136 * \param ctx the context to operate on, or NULL for the default context
2137 * \returns 0 on success, or a LIBUSB_ERROR code on failure
2139 int API_EXPORTED libusb_handle_events(libusb_context *ctx)
2144 return libusb_handle_events_timeout_completed(ctx, &tv, NULL);
2148 * Handle any pending events in blocking mode.
2150 * Like libusb_handle_events(), with the addition of a completed parameter
2151 * to allow for race free waiting for the completion of a specific transfer.
2153 * See libusb_handle_events_timeout_completed() for details on the completed
2156 * \param ctx the context to operate on, or NULL for the default context
2157 * \param completed pointer to completion integer to check, or NULL
2158 * \returns 0 on success, or a LIBUSB_ERROR code on failure
2161 int API_EXPORTED libusb_handle_events_completed(libusb_context *ctx,
2167 return libusb_handle_events_timeout_completed(ctx, &tv, completed);
2171 * Handle any pending events by polling file descriptors, without checking if
2172 * any other threads are already doing so. Must be called with the event lock
2173 * held, see libusb_lock_events().
2175 * This function is designed to be called under the situation where you have
2176 * taken the event lock and are calling poll()/select() directly on libusbx's
2177 * file descriptors (as opposed to using libusb_handle_events() or similar).
2178 * You detect events on libusbx's descriptors, so you then call this function
2179 * with a zero timeout value (while still holding the event lock).
2181 * \param ctx the context to operate on, or NULL for the default context
2182 * \param tv the maximum time to block waiting for events, or zero for
2184 * \returns 0 on success, or a LIBUSB_ERROR code on failure
2187 int API_EXPORTED libusb_handle_events_locked(libusb_context *ctx,
2191 struct timeval poll_timeout;
2193 USBI_GET_CONTEXT(ctx);
2194 r = get_next_timeout(ctx, tv, &poll_timeout);
2196 /* timeout already expired */
2197 return handle_timeouts(ctx);
2200 return handle_events(ctx, &poll_timeout);
2204 * Determines whether your application must apply special timing considerations
2205 * when monitoring libusbx's file descriptors.
2207 * This function is only useful for applications which retrieve and poll
2208 * libusbx's file descriptors in their own main loop (\ref pollmain).
2210 * Ordinarily, libusbx's event handler needs to be called into at specific
2211 * moments in time (in addition to times when there is activity on the file
2212 * descriptor set). The usual approach is to use libusb_get_next_timeout()
2213 * to learn about when the next timeout occurs, and to adjust your
2214 * poll()/select() timeout accordingly so that you can make a call into the
2215 * library at that time.
2217 * Some platforms supported by libusbx do not come with this baggage - any
2218 * events relevant to timing will be represented by activity on the file
2219 * descriptor set, and libusb_get_next_timeout() will always return 0.
2220 * This function allows you to detect whether you are running on such a
2225 * \param ctx the context to operate on, or NULL for the default context
2226 * \returns 0 if you must call into libusbx at times determined by
2227 * libusb_get_next_timeout(), or 1 if all timeout events are handled internally
2228 * or through regular activity on the file descriptors.
2229 * \see \ref pollmain "Polling libusbx file descriptors for event handling"
2231 int API_EXPORTED libusb_pollfds_handle_timeouts(libusb_context *ctx)
2233 #if defined(USBI_TIMERFD_AVAILABLE)
2234 USBI_GET_CONTEXT(ctx);
2235 return usbi_using_timerfd(ctx);
2243 * Determine the next internal timeout that libusbx needs to handle. You only
2244 * need to use this function if you are calling poll() or select() or similar
2245 * on libusbx's file descriptors yourself - you do not need to use it if you
2246 * are calling libusb_handle_events() or a variant directly.
2248 * You should call this function in your main loop in order to determine how
2249 * long to wait for select() or poll() to return results. libusbx needs to be
2250 * called into at this timeout, so you should use it as an upper bound on
2251 * your select() or poll() call.
2253 * When the timeout has expired, call into libusb_handle_events_timeout()
2254 * (perhaps in non-blocking mode) so that libusbx can handle the timeout.
2256 * This function may return 1 (success) and an all-zero timeval. If this is
2257 * the case, it indicates that libusbx has a timeout that has already expired
2258 * so you should call libusb_handle_events_timeout() or similar immediately.
2259 * A return code of 0 indicates that there are no pending timeouts.
2261 * On some platforms, this function will always returns 0 (no pending
2262 * timeouts). See \ref polltime.
2264 * \param ctx the context to operate on, or NULL for the default context
2265 * \param tv output location for a relative time against the current
2266 * clock in which libusbx must be called into in order to process timeout events
2267 * \returns 0 if there are no pending timeouts, 1 if a timeout was returned,
2268 * or LIBUSB_ERROR_OTHER on failure
2270 int API_EXPORTED libusb_get_next_timeout(libusb_context *ctx,
2273 struct usbi_transfer *transfer;
2274 struct timespec cur_ts;
2275 struct timeval cur_tv;
2276 struct timeval *next_timeout;
2280 USBI_GET_CONTEXT(ctx);
2281 if (usbi_using_timerfd(ctx))
2284 usbi_mutex_lock(&ctx->flying_transfers_lock);
2285 if (list_empty(&ctx->flying_transfers)) {
2286 usbi_mutex_unlock(&ctx->flying_transfers_lock);
2287 usbi_dbg("no URBs, no timeout!");
2291 /* find next transfer which hasn't already been processed as timed out */
2292 list_for_each_entry(transfer, &ctx->flying_transfers, list, struct usbi_transfer) {
2293 if (transfer->flags & (USBI_TRANSFER_TIMED_OUT | USBI_TRANSFER_OS_HANDLES_TIMEOUT))
2296 /* no timeout for this transfer? */
2297 if (!timerisset(&transfer->timeout))
2303 usbi_mutex_unlock(&ctx->flying_transfers_lock);
2306 usbi_dbg("no URB with timeout or all handled by OS; no timeout!");
2310 next_timeout = &transfer->timeout;
2312 r = usbi_backend->clock_gettime(USBI_CLOCK_MONOTONIC, &cur_ts);
2314 usbi_err(ctx, "failed to read monotonic clock, errno=%d", errno);
2317 TIMESPEC_TO_TIMEVAL(&cur_tv, &cur_ts);
2319 if (!timercmp(&cur_tv, next_timeout, <)) {
2320 usbi_dbg("first timeout already expired");
2323 timersub(next_timeout, &cur_tv, tv);
2324 usbi_dbg("next timeout in %d.%06ds", tv->tv_sec, tv->tv_usec);
2331 * Register notification functions for file descriptor additions/removals.
2332 * These functions will be invoked for every new or removed file descriptor
2333 * that libusbx uses as an event source.
2335 * To remove notifiers, pass NULL values for the function pointers.
2337 * Note that file descriptors may have been added even before you register
2338 * these notifiers (e.g. at libusb_init() time).
2340 * Additionally, note that the removal notifier may be called during
2341 * libusb_exit() (e.g. when it is closing file descriptors that were opened
2342 * and added to the poll set at libusb_init() time). If you don't want this,
2343 * remove the notifiers immediately before calling libusb_exit().
2345 * \param ctx the context to operate on, or NULL for the default context
2346 * \param added_cb pointer to function for addition notifications
2347 * \param removed_cb pointer to function for removal notifications
2348 * \param user_data User data to be passed back to callbacks (useful for
2349 * passing context information)
2351 void API_EXPORTED libusb_set_pollfd_notifiers(libusb_context *ctx,
2352 libusb_pollfd_added_cb added_cb, libusb_pollfd_removed_cb removed_cb,
2355 USBI_GET_CONTEXT(ctx);
2356 ctx->fd_added_cb = added_cb;
2357 ctx->fd_removed_cb = removed_cb;
2358 ctx->fd_cb_user_data = user_data;
2361 /* Add a file descriptor to the list of file descriptors to be monitored.
2362 * events should be specified as a bitmask of events passed to poll(), e.g.
2363 * POLLIN and/or POLLOUT. */
2364 int usbi_add_pollfd(struct libusb_context *ctx, int fd, short events)
2366 struct usbi_pollfd *ipollfd = malloc(sizeof(*ipollfd));
2368 return LIBUSB_ERROR_NO_MEM;
2370 usbi_dbg("add fd %d events %d", fd, events);
2371 ipollfd->pollfd.fd = fd;
2372 ipollfd->pollfd.events = events;
2373 usbi_mutex_lock(&ctx->pollfds_lock);
2374 list_add_tail(&ipollfd->list, &ctx->pollfds);
2375 usbi_mutex_unlock(&ctx->pollfds_lock);
2377 if (ctx->fd_added_cb)
2378 ctx->fd_added_cb(fd, events, ctx->fd_cb_user_data);
2382 /* Remove a file descriptor from the list of file descriptors to be polled. */
2383 void usbi_remove_pollfd(struct libusb_context *ctx, int fd)
2385 struct usbi_pollfd *ipollfd;
2388 usbi_dbg("remove fd %d", fd);
2389 usbi_mutex_lock(&ctx->pollfds_lock);
2390 list_for_each_entry(ipollfd, &ctx->pollfds, list, struct usbi_pollfd)
2391 if (ipollfd->pollfd.fd == fd) {
2397 usbi_dbg("couldn't find fd %d to remove", fd);
2398 usbi_mutex_unlock(&ctx->pollfds_lock);
2402 list_del(&ipollfd->list);
2403 usbi_mutex_unlock(&ctx->pollfds_lock);
2405 if (ctx->fd_removed_cb)
2406 ctx->fd_removed_cb(fd, ctx->fd_cb_user_data);
2410 * Retrieve a list of file descriptors that should be polled by your main loop
2411 * as libusbx event sources.
2413 * The returned list is NULL-terminated and should be freed with free() when
2414 * done. The actual list contents must not be touched.
2416 * As file descriptors are a Unix-specific concept, this function is not
2417 * available on Windows and will always return NULL.
2419 * \param ctx the context to operate on, or NULL for the default context
2420 * \returns a NULL-terminated list of libusb_pollfd structures
2421 * \returns NULL on error
2422 * \returns NULL on platforms where the functionality is not available
2425 const struct libusb_pollfd ** LIBUSB_CALL libusb_get_pollfds(
2426 libusb_context *ctx)
2429 struct libusb_pollfd **ret = NULL;
2430 struct usbi_pollfd *ipollfd;
2433 USBI_GET_CONTEXT(ctx);
2435 usbi_mutex_lock(&ctx->pollfds_lock);
2436 list_for_each_entry(ipollfd, &ctx->pollfds, list, struct usbi_pollfd)
2439 ret = calloc(cnt + 1, sizeof(struct libusb_pollfd *));
2443 list_for_each_entry(ipollfd, &ctx->pollfds, list, struct usbi_pollfd)
2444 ret[i++] = (struct libusb_pollfd *) ipollfd;
2448 usbi_mutex_unlock(&ctx->pollfds_lock);
2449 return (const struct libusb_pollfd **) ret;
2451 usbi_err(ctx, "external polling of libusbx's internal descriptors "\
2452 "is not yet supported on Windows platforms");
2457 /* Backends call this from handle_events to report disconnection of a device.
2458 * The transfers get cancelled appropriately.
2460 void usbi_handle_disconnect(struct libusb_device_handle *handle)
2462 struct usbi_transfer *cur;
2463 struct usbi_transfer *to_cancel;
2465 usbi_dbg("device %d.%d",
2466 handle->dev->bus_number, handle->dev->device_address);
2468 /* terminate all pending transfers with the LIBUSB_TRANSFER_NO_DEVICE
2471 * this is a bit tricky because:
2472 * 1. we can't do transfer completion while holding flying_transfers_lock
2473 * 2. the transfers list can change underneath us - if we were to build a
2474 * list of transfers to complete (while holding look), the situation
2475 * might be different by the time we come to free them
2477 * so we resort to a loop-based approach as below
2478 * FIXME: is this still potentially racy?
2482 usbi_mutex_lock(&HANDLE_CTX(handle)->flying_transfers_lock);
2484 list_for_each_entry(cur, &HANDLE_CTX(handle)->flying_transfers, list, struct usbi_transfer)
2485 if (USBI_TRANSFER_TO_LIBUSB_TRANSFER(cur)->dev_handle == handle) {
2489 usbi_mutex_unlock(&HANDLE_CTX(handle)->flying_transfers_lock);
2494 usbi_backend->clear_transfer_priv(to_cancel);
2495 usbi_handle_transfer_completion(to_cancel, LIBUSB_TRANSFER_NO_DEVICE);