2 * I/O functions for libusb
3 * Copyright (C) 2007-2008 Daniel Drake <dsd@gentoo.org>
4 * Copyright (c) 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
36 * \page io Synchronous and asynchronous device I/O
38 * \section intro Introduction
40 * If you're using libusb in your application, you're probably wanting to
41 * perform I/O with devices - you want to perform USB data transfers.
43 * libusb offers two separate interfaces for device I/O. This page aims to
44 * introduce the two in order to help you decide which one is more suitable
45 * for your application. You can also choose to use both interfaces in your
46 * application by considering each transfer on a case-by-case basis.
48 * Once you have read through the following discussion, you should consult the
49 * detailed API documentation pages for the details:
53 * \section theory Transfers at a logical level
55 * At a logical level, USB transfers typically happen in two parts. For
56 * example, when reading data from a endpoint:
57 * -# A request for data is sent to the device
58 * -# Some time later, the incoming data is received by the host
60 * or when writing data to an endpoint:
62 * -# The data is sent to the device
63 * -# Some time later, the host receives acknowledgement from the device that
64 * the data has been transferred.
66 * There may be an indefinite delay between the two steps. Consider a
67 * fictional USB input device with a button that the user can press. In order
68 * to determine when the button is pressed, you would likely submit a request
69 * to read data on a bulk or interrupt endpoint and wait for data to arrive.
70 * Data will arrive when the button is pressed by the user, which is
71 * potentially hours later.
73 * libusb offers both a synchronous and an asynchronous interface to performing
74 * USB transfers. The main difference is that the synchronous interface
75 * combines both steps indicated above into a single function call, whereas
76 * the asynchronous interface separates them.
78 * \section sync The synchronous interface
80 * The synchronous I/O interface allows you to perform a USB transfer with
81 * a single function call. When the function call returns, the transfer has
82 * completed and you can parse the results.
84 * If you have used the libusb-0.1 before, this I/O style will seem familar to
85 * you. libusb-0.1 only offered a synchronous interface.
87 * In our input device example, to read button presses you might write code
88 * in the following style:
90 unsigned char data[4];
92 int r = libusb_bulk_transfer(handle, EP_IN, data, sizeof(data), &actual_length, 0);
93 if (r == 0 && actual_length == sizeof(data)) {
94 // results of the transaction can now be found in the data buffer
95 // parse them here and report button press
101 * The main advantage of this model is simplicity: you did everything with
102 * a single simple function call.
104 * However, this interface has its limitations. Your application will sleep
105 * inside libusb_bulk_transfer() until the transaction has completed. If it
106 * takes the user 3 hours to press the button, your application will be
107 * sleeping for that long. Execution will be tied up inside the library -
108 * the entire thread will be useless for that duration.
110 * Another issue is that by tieing up the thread with that single transaction
111 * there is no possibility of performing I/O with multiple endpoints and/or
112 * multiple devices simultaneously, unless you resort to creating one thread
115 * Additionally, there is no opportunity to cancel the transfer after the
116 * request has been submitted.
118 * For details on how to use the synchronous API, see the
119 * \ref syncio "synchronous I/O API documentation" pages.
121 * \section async The asynchronous interface
123 * Asynchronous I/O is the most significant new feature in libusb-1.0.
124 * Although it is a more complex interface, it solves all the issues detailed
127 * Instead of providing which functions that block until the I/O has complete,
128 * libusb's asynchronous interface presents non-blocking functions which
129 * begin a transfer and then return immediately. Your application passes a
130 * callback function pointer to this non-blocking function, which libusb will
131 * call with the results of the transaction when it has completed.
133 * Transfers which have been submitted through the non-blocking functions
134 * can be cancelled with a separate function call.
136 * The non-blocking nature of this interface allows you to be simultaneously
137 * performing I/O to multiple endpoints on multiple devices, without having
140 * This added flexibility does come with some complications though:
141 * - In the interest of being a lightweight library, libusb does not create
142 * threads and can only operate when your application is calling into it. Your
143 * application must call into libusb from it's main loop when events are ready
144 * to be handled, or you must use some other scheme to allow libusb to
145 * undertake whatever work needs to be done.
146 * - libusb also needs to be called into at certain fixed points in time in
147 * order to accurately handle transfer timeouts.
148 * - Memory handling becomes more complex. You cannot use stack memory unless
149 * the function with that stack is guaranteed not to return until the transfer
150 * callback has finished executing.
151 * - You generally lose some linearity from your code flow because submitting
152 * the transfer request is done in a separate function from where the transfer
153 * results are handled. This becomes particularly obvious when you want to
154 * submit a second transfer based on the results of an earlier transfer.
156 * Internally, libusb's synchronous interface is expressed in terms of function
157 * calls to the asynchronous interface.
159 * For details on how to use the asynchronous API, see the
160 * \ref asyncio "asynchronous I/O API" documentation pages.
165 * \page packetoverflow Packets and overflows
167 * \section packets Packet abstraction
169 * The USB specifications describe how data is transmitted in packets, with
170 * constraints on packet size defined by endpoint descriptors. The host must
171 * not send data payloads larger than the endpoint's maximum packet size.
173 * libusb and the underlying OS abstract out the packet concept, allowing you
174 * to request transfers of any size. Internally, the request will be divided
175 * up into correctly-sized packets. You do not have to be concerned with
176 * packet sizes, but there is one exception when considering overflows.
178 * \section overflow Bulk/interrupt transfer overflows
180 * When requesting data on a bulk endpoint, libusb requires you to supply a
181 * buffer and the maximum number of bytes of data that libusb can put in that
182 * buffer. However, the size of the buffer is not communicated to the device -
183 * the device is just asked to send any amount of data.
185 * There is no problem if the device sends an amount of data that is less than
186 * or equal to the buffer size. libusb reports this condition to you through
187 * the \ref libusb_transfer::actual_length "libusb_transfer.actual_length"
190 * Problems may occur if the device attempts to send more data than can fit in
191 * the buffer. libusb reports LIBUSB_TRANSFER_OVERFLOW for this condition but
192 * other behaviour is largely undefined: actual_length may or may not be
193 * accurate, the chunk of data that can fit in the buffer (before overflow)
194 * may or may not have been transferred.
196 * Overflows are nasty, but can be avoided. Even though you were told to
197 * ignore packets above, think about the lower level details: each transfer is
198 * split into packets (typically small, with a maximum size of 512 bytes).
199 * Overflows can only happen if the final packet in an incoming data transfer
200 * is smaller than the actual packet that the device wants to transfer.
201 * Therefore, you will never see an overflow if your transfer buffer size is a
202 * multiple of the endpoint's packet size: the final packet will either
203 * fill up completely or will be only partially filled.
207 * @defgroup asyncio Asynchronous device I/O
209 * This page details libusb's asynchronous (non-blocking) API for USB device
210 * I/O. This interface is very powerful but is also quite complex - you will
211 * need to read this page carefully to understand the necessary considerations
212 * and issues surrounding use of this interface. Simplistic applications
213 * may wish to consider the \ref syncio "synchronous I/O API" instead.
215 * The asynchronous interface is built around the idea of separating transfer
216 * submission and handling of transfer completion (the synchronous model
217 * combines both of these into one). There may be a long delay between
218 * submission and completion, however the asynchronous submission function
219 * is non-blocking so will return control to your application during that
220 * potentially long delay.
222 * \section asyncabstraction Transfer abstraction
224 * For the asynchronous I/O, libusb implements the concept of a generic
225 * transfer entity for all types of I/O (control, bulk, interrupt,
226 * isochronous). The generic transfer object must be treated slightly
227 * differently depending on which type of I/O you are performing with it.
229 * This is represented by the public libusb_transfer structure type.
231 * \section asynctrf Asynchronous transfers
233 * We can view asynchronous I/O as a 5 step process:
234 * -# <b>Allocation</b>: allocate a libusb_transfer
235 * -# <b>Filling</b>: populate the libusb_transfer instance with information
236 * about the transfer you wish to perform
237 * -# <b>Submission</b>: ask libusb to submit the transfer
238 * -# <b>Completion handling</b>: examine transfer results in the
239 * libusb_transfer structure
240 * -# <b>Deallocation</b>: clean up resources
243 * \subsection asyncalloc Allocation
245 * This step involves allocating memory for a USB transfer. This is the
246 * generic transfer object mentioned above. At this stage, the transfer
247 * is "blank" with no details about what type of I/O it will be used for.
249 * Allocation is done with the libusb_alloc_transfer() function. You must use
250 * this function rather than allocating your own transfers.
252 * \subsection asyncfill Filling
254 * This step is where you take a previously allocated transfer and fill it
255 * with information to determine the message type and direction, data buffer,
256 * callback function, etc.
258 * You can either fill the required fields yourself or you can use the
259 * helper functions: libusb_fill_control_transfer(), libusb_fill_bulk_transfer()
260 * and libusb_fill_interrupt_transfer().
262 * \subsection asyncsubmit Submission
264 * When you have allocated a transfer and filled it, you can submit it using
265 * libusb_submit_transfer(). This function returns immediately but can be
266 * regarded as firing off the I/O request in the background.
268 * \subsection asynccomplete Completion handling
270 * After a transfer has been submitted, one of four things can happen to it:
272 * - The transfer completes (i.e. some data was transferred)
273 * - The transfer has a timeout and the timeout expires before all data is
275 * - The transfer fails due to an error
276 * - The transfer is cancelled
278 * Each of these will cause the user-specified transfer callback function to
279 * be invoked. It is up to the callback function to determine which of the
280 * above actually happened and to act accordingly.
282 * The user-specified callback is passed a pointer to the libusb_transfer
283 * structure which was used to setup and submit the transfer. At completion
284 * time, libusb has populated this structure with results of the transfer:
285 * success or failure reason, number of bytes of data transferred, etc. See
286 * the libusb_transfer structure documentation for more information.
288 * \subsection Deallocation
290 * When a transfer has completed (i.e. the callback function has been invoked),
291 * you are advised to free the transfer (unless you wish to resubmit it, see
292 * below). Transfers are deallocated with libusb_free_transfer().
294 * It is undefined behaviour to free a transfer which has not completed.
296 * \section asyncresubmit Resubmission
298 * You may be wondering why allocation, filling, and submission are all
299 * separated above where they could reasonably be combined into a single
302 * The reason for separation is to allow you to resubmit transfers without
303 * having to allocate new ones every time. This is especially useful for
304 * common situations dealing with interrupt endpoints - you allocate one
305 * transfer, fill and submit it, and when it returns with results you just
306 * resubmit it for the next interrupt.
308 * \section asynccancel Cancellation
310 * Another advantage of using the asynchronous interface is that you have
311 * the ability to cancel transfers which have not yet completed. This is
312 * done by calling the libusb_cancel_transfer() function.
314 * libusb_cancel_transfer() is asynchronous/non-blocking in itself. When the
315 * cancellation actually completes, the transfer's callback function will
316 * be invoked, and the callback function should check the transfer status to
317 * determine that it was cancelled.
319 * Freeing the transfer after it has been cancelled but before cancellation
320 * has completed will result in undefined behaviour.
322 * \section bulk_overflows Overflows on device-to-host bulk/interrupt endpoints
324 * If your device does not have predictable transfer sizes (or it misbehaves),
325 * your application may submit a request for data on an IN endpoint which is
326 * smaller than the data that the device wishes to send. In some circumstances
327 * this will cause an overflow, which is a nasty condition to deal with. See
328 * the \ref packetoverflow page for discussion.
330 * \section asyncctrl Considerations for control transfers
332 * The <tt>libusb_transfer</tt> structure is generic and hence does not
333 * include specific fields for the control-specific setup packet structure.
335 * In order to perform a control transfer, you must place the 8-byte setup
336 * packet at the start of the data buffer. To simplify this, you could
337 * cast the buffer pointer to type struct libusb_control_setup, or you can
338 * use the helper function libusb_fill_control_setup().
340 * The wLength field placed in the setup packet must be the length you would
341 * expect to be sent in the setup packet: the length of the payload that
342 * follows (or the expected maximum number of bytes to receive). However,
343 * the length field of the libusb_transfer object must be the length of
344 * the data buffer - i.e. it should be wLength <em>plus</em> the size of
345 * the setup packet (LIBUSB_CONTROL_SETUP_SIZE).
347 * If you use the helper functions, this is simplified for you:
348 * -# Allocate a buffer of size LIBUSB_CONTROL_SETUP_SIZE plus the size of the
349 * data you are sending/requesting.
350 * -# Call libusb_fill_control_setup() on the data buffer, using the transfer
351 * request size as the wLength value (i.e. do not include the extra space you
352 * allocated for the control setup).
353 * -# If this is a host-to-device transfer, place the data to be transferred
354 * in the data buffer, starting at offset LIBUSB_CONTROL_SETUP_SIZE.
355 * -# Call libusb_fill_control_transfer() to associate the data buffer with
356 * the transfer (and to set the remaining details such as callback and timeout).
357 * - Note that there is no parameter to set the length field of the transfer.
358 * The length is automatically inferred from the wLength field of the setup
360 * -# Submit the transfer.
362 * The multi-byte control setup fields (wValue, wIndex and wLength) must
363 * be given in little-endian byte order (the endianness of the USB bus).
364 * Endianness conversion is transparently handled by
365 * libusb_fill_control_setup() which is documented to accept host-endian
368 * Further considerations are needed when handling transfer completion in
369 * your callback function:
370 * - As you might expect, the setup packet will still be sitting at the start
371 * of the data buffer.
372 * - If this was a device-to-host transfer, the received data will be sitting
373 * at offset LIBUSB_CONTROL_SETUP_SIZE into the buffer.
374 * - The actual_length field of the transfer structure is relative to the
375 * wLength of the setup packet, rather than the size of the data buffer. So,
376 * if your wLength was 4, your transfer's <tt>length</tt> was 12, then you
377 * should expect an <tt>actual_length</tt> of 4 to indicate that the data was
378 * transferred in entirity.
380 * To simplify parsing of setup packets and obtaining the data from the
381 * correct offset, you may wish to use the libusb_control_transfer_get_data()
382 * and libusb_control_transfer_get_setup() functions within your transfer
385 * Even though control endpoints do not halt, a completed control transfer
386 * may have a LIBUSB_TRANSFER_STALL status code. This indicates the control
387 * request was not supported.
389 * \section asyncintr Considerations for interrupt transfers
391 * All interrupt transfers are performed using the polling interval presented
392 * by the bInterval value of the endpoint descriptor.
394 * \section asynciso Considerations for isochronous transfers
396 * Isochronous transfers are more complicated than transfers to
397 * non-isochronous endpoints.
399 * To perform I/O to an isochronous endpoint, allocate the transfer by calling
400 * libusb_alloc_transfer() with an appropriate number of isochronous packets.
402 * During filling, set \ref libusb_transfer::type "type" to
403 * \ref libusb_transfer_type::LIBUSB_TRANSFER_TYPE_ISOCHRONOUS
404 * "LIBUSB_TRANSFER_TYPE_ISOCHRONOUS", and set
405 * \ref libusb_transfer::num_iso_packets "num_iso_packets" to a value less than
406 * or equal to the number of packets you requested during allocation.
407 * libusb_alloc_transfer() does not set either of these fields for you, given
408 * that you might not even use the transfer on an isochronous endpoint.
410 * Next, populate the length field for the first num_iso_packets entries in
411 * the \ref libusb_transfer::iso_packet_desc "iso_packet_desc" array. Section
412 * 5.6.3 of the USB2 specifications describe how the maximum isochronous
413 * packet length is determined by the wMaxPacketSize field in the endpoint
415 * Two functions can help you here:
417 * - libusb_get_max_iso_packet_size() is an easy way to determine the max
418 * packet size for an isochronous endpoint. Note that the maximum packet
419 * size is actually the maximum number of bytes that can be transmitted in
420 * a single microframe, therefore this function multiplies the maximum number
421 * of bytes per transaction by the number of transaction opportunities per
423 * - libusb_set_iso_packet_lengths() assigns the same length to all packets
424 * within a transfer, which is usually what you want.
426 * For outgoing transfers, you'll obviously fill the buffer and populate the
427 * packet descriptors in hope that all the data gets transferred. For incoming
428 * transfers, you must ensure the buffer has sufficient capacity for
429 * the situation where all packets transfer the full amount of requested data.
431 * Completion handling requires some extra consideration. The
432 * \ref libusb_transfer::actual_length "actual_length" field of the transfer
433 * is meaningless and should not be examined; instead you must refer to the
434 * \ref libusb_iso_packet_descriptor::actual_length "actual_length" field of
435 * each individual packet.
437 * The \ref libusb_transfer::status "status" field of the transfer is also a
439 * - If the packets were submitted and the isochronous data microframes
440 * completed normally, status will have value
441 * \ref libusb_transfer_status::LIBUSB_TRANSFER_COMPLETED
442 * "LIBUSB_TRANSFER_COMPLETED". Note that bus errors and software-incurred
443 * delays are not counted as transfer errors; the transfer.status field may
444 * indicate COMPLETED even if some or all of the packets failed. Refer to
445 * the \ref libusb_iso_packet_descriptor::status "status" field of each
446 * individual packet to determine packet failures.
447 * - The status field will have value
448 * \ref libusb_transfer_status::LIBUSB_TRANSFER_ERROR
449 * "LIBUSB_TRANSFER_ERROR" only when serious errors were encountered.
450 * - Other transfer status codes occur with normal behaviour.
452 * The data for each packet will be found at an offset into the buffer that
453 * can be calculated as if each prior packet completed in full. The
454 * libusb_get_iso_packet_buffer() and libusb_get_iso_packet_buffer_simple()
455 * functions may help you here.
457 * \section asyncmem Memory caveats
459 * In most circumstances, it is not safe to use stack memory for transfer
460 * buffers. This is because the function that fired off the asynchronous
461 * transfer may return before libusb has finished using the buffer, and when
462 * the function returns it's stack gets destroyed. This is true for both
463 * host-to-device and device-to-host transfers.
465 * The only case in which it is safe to use stack memory is where you can
466 * guarantee that the function owning the stack space for the buffer does not
467 * return until after the transfer's callback function has completed. In every
468 * other case, you need to use heap memory instead.
470 * \section asyncflags Fine control
472 * Through using this asynchronous interface, you may find yourself repeating
473 * a few simple operations many times. You can apply a bitwise OR of certain
474 * flags to a transfer to simplify certain things:
475 * - \ref libusb_transfer_flags::LIBUSB_TRANSFER_SHORT_NOT_OK
476 * "LIBUSB_TRANSFER_SHORT_NOT_OK" results in transfers which transferred
477 * less than the requested amount of data being marked with status
478 * \ref libusb_transfer_status::LIBUSB_TRANSFER_ERROR "LIBUSB_TRANSFER_ERROR"
479 * (they would normally be regarded as COMPLETED)
480 * - \ref libusb_transfer_flags::LIBUSB_TRANSFER_FREE_BUFFER
481 * "LIBUSB_TRANSFER_FREE_BUFFER" allows you to ask libusb to free the transfer
482 * buffer when freeing the transfer.
483 * - \ref libusb_transfer_flags::LIBUSB_TRANSFER_FREE_TRANSFER
484 * "LIBUSB_TRANSFER_FREE_TRANSFER" causes libusb to automatically free the
485 * transfer after the transfer callback returns.
487 * \section asyncevent Event handling
489 * In accordance of the aim of being a lightweight library, libusb does not
490 * create threads internally. This means that libusb code does not execute
491 * at any time other than when your application is calling a libusb function.
492 * However, an asynchronous model requires that libusb perform work at various
493 * points in time - namely processing the results of previously-submitted
494 * transfers and invoking the user-supplied callback function.
496 * This gives rise to the libusb_handle_events() function which your
497 * application must call into when libusb has work do to. This gives libusb
498 * the opportunity to reap pending transfers, invoke callbacks, etc.
500 * The first issue to discuss here is how your application can figure out
501 * when libusb has work to do. In fact, there are two naive options which
502 * do not actually require your application to know this:
503 * -# Periodically call libusb_handle_events() in non-blocking mode at fixed
504 * short intervals from your main loop
505 * -# Repeatedly call libusb_handle_events() in blocking mode from a dedicated
508 * The first option is plainly not very nice, and will cause unnecessary
509 * CPU wakeups leading to increased power usage and decreased battery life.
510 * The second option is not very nice either, but may be the nicest option
511 * available to you if the "proper" approach can not be applied to your
512 * application (read on...).
514 * The recommended option is to integrate libusb with your application main
515 * event loop. libusb exposes a set of file descriptors which allow you to do
516 * this. Your main loop is probably already calling poll() or select() or a
517 * variant on a set of file descriptors for other event sources (e.g. keyboard
518 * button presses, mouse movements, network sockets, etc). You then add
519 * libusb's file descriptors to your poll()/select() calls, and when activity
520 * is detected on such descriptors you know it is time to call
521 * libusb_handle_events().
523 * There is one final event handling complication. libusb supports
524 * asynchronous transfers which time out after a specified time period, and
525 * this requires that libusb is called into at or after the timeout so that
526 * the timeout can be handled. So, in addition to considering libusb's file
527 * descriptors in your main event loop, you must also consider that libusb
528 * sometimes needs to be called into at fixed points in time even when there
529 * is no file descriptor activity.
531 * For the details on retrieving the set of file descriptors and determining
532 * the next timeout, see the \ref poll "polling and timing" API documentation.
536 * @defgroup poll Polling and timing
538 * This page documents libusb's functions for polling events and timing.
539 * These functions are only necessary for users of the
540 * \ref asyncio "asynchronous API". If you are only using the simpler
541 * \ref syncio "synchronous API" then you do not need to ever call these
544 * The justification for the functionality described here has already been
545 * discussed in the \ref asyncevent "event handling" section of the
546 * asynchronous API documentation. In summary, libusb does not create internal
547 * threads for event processing and hence relies on your application calling
548 * into libusb at certain points in time so that pending events can be handled.
549 * In order to know precisely when libusb needs to be called into, libusb
550 * offers you a set of pollable file descriptors and information about when
551 * the next timeout expires.
553 * If you are using the asynchronous I/O API, you must take one of the two
554 * following options, otherwise your I/O will not complete.
556 * \section pollsimple The simple option
558 * If your application revolves solely around libusb and does not need to
559 * handle other event sources, you can have a program structure as follows:
562 // find and open device
563 // maybe fire off some initial async I/O
565 while (user_has_not_requested_exit)
566 libusb_handle_events(ctx);
571 * With such a simple main loop, you do not have to worry about managing
572 * sets of file descriptors or handling timeouts. libusb_handle_events() will
573 * handle those details internally.
575 * \section pollmain The more advanced option
577 * In more advanced applications, you will already have a main loop which
578 * is monitoring other event sources: network sockets, X11 events, mouse
579 * movements, etc. Through exposing a set of file descriptors, libusb is
580 * designed to cleanly integrate into such main loops.
582 * In addition to polling file descriptors for the other event sources, you
583 * take a set of file descriptors from libusb and monitor those too. When you
584 * detect activity on libusb's file descriptors, you call
585 * libusb_handle_events_timeout() in non-blocking mode.
587 * You must also consider the fact that libusb sometimes has to handle events
588 * at certain known times which do not generate activity on file descriptors.
589 * Your main loop must also consider these times, modify it's poll()/select()
590 * timeout accordingly, and track time so that libusb_handle_events_timeout()
591 * is called in non-blocking mode when timeouts expire.
593 * In pseudo-code, you want something that looks like:
597 libusb_get_pollfds(ctx)
598 while (user has not requested application exit) {
599 libusb_get_next_timeout(ctx);
600 select(on libusb file descriptors plus any other event sources of interest,
601 using a timeout no larger than the value libusb just suggested)
602 if (select() indicated activity on libusb file descriptors)
603 libusb_handle_events_timeout(ctx, 0);
604 if (time has elapsed to or beyond the libusb timeout)
605 libusb_handle_events_timeout(ctx, 0);
611 * The set of file descriptors that libusb uses as event sources may change
612 * during the life of your application. Rather than having to repeatedly
613 * call libusb_get_pollfds(), you can set up notification functions for when
614 * the file descriptor set changes using libusb_set_pollfd_notifiers().
616 * \section mtissues Multi-threaded considerations
618 * Unfortunately, the situation is complicated further when multiple threads
619 * come into play. If two threads are monitoring the same file descriptors,
620 * the fact that only one thread will be woken up when an event occurs causes
623 * The events lock, event waiters lock, and libusb_handle_events_locked()
624 * entities are added to solve these problems. You do not need to be concerned
625 * with these entities otherwise.
627 * See the extra documentation: \ref mtasync
630 /** \page mtasync Multi-threaded applications and asynchronous I/O
632 * libusb is a thread-safe library, but extra considerations must be applied
633 * to applications which interact with libusb from multiple threads.
635 * The underlying issue that must be addressed is that all libusb I/O
636 * revolves around monitoring file descriptors through the poll()/select()
637 * system calls. This is directly exposed at the
638 * \ref asyncio "asynchronous interface" but it is important to note that the
639 * \ref syncio "synchronous interface" is implemented on top of the
640 * asynchonrous interface, therefore the same considerations apply.
642 * The issue is that if two or more threads are concurrently calling poll()
643 * or select() on libusb's file descriptors then only one of those threads
644 * will be woken up when an event arrives. The others will be completely
645 * oblivious that anything has happened.
647 * Consider the following pseudo-code, which submits an asynchronous transfer
648 * then waits for its completion. This style is one way you could implement a
649 * synchronous interface on top of the asynchronous interface (and libusb
650 * does something similar, albeit more advanced due to the complications
651 * explained on this page).
654 void cb(struct libusb_transfer *transfer)
656 int *completed = transfer->user_data;
661 struct libusb_transfer *transfer;
662 unsigned char buffer[LIBUSB_CONTROL_SETUP_SIZE];
665 transfer = libusb_alloc_transfer(0);
666 libusb_fill_control_setup(buffer,
667 LIBUSB_REQUEST_TYPE_VENDOR | LIBUSB_ENDPOINT_OUT, 0x04, 0x01, 0, 0);
668 libusb_fill_control_transfer(transfer, dev, buffer, cb, &completed, 1000);
669 libusb_submit_transfer(transfer);
672 poll(libusb file descriptors, 120*1000);
673 if (poll indicates activity)
674 libusb_handle_events_timeout(ctx, 0);
676 printf("completed!");
681 * Here we are <em>serializing</em> completion of an asynchronous event
682 * against a condition - the condition being completion of a specific transfer.
683 * The poll() loop has a long timeout to minimize CPU usage during situations
684 * when nothing is happening (it could reasonably be unlimited).
686 * If this is the only thread that is polling libusb's file descriptors, there
687 * is no problem: there is no danger that another thread will swallow up the
688 * event that we are interested in. On the other hand, if there is another
689 * thread polling the same descriptors, there is a chance that it will receive
690 * the event that we were interested in. In this situation, <tt>myfunc()</tt>
691 * will only realise that the transfer has completed on the next iteration of
692 * the loop, <em>up to 120 seconds later.</em> Clearly a two-minute delay is
693 * undesirable, and don't even think about using short timeouts to circumvent
696 * The solution here is to ensure that no two threads are ever polling the
697 * file descriptors at the same time. A naive implementation of this would
698 * impact the capabilities of the library, so libusb offers the scheme
699 * documented below to ensure no loss of functionality.
701 * Before we go any further, it is worth mentioning that all libusb-wrapped
702 * event handling procedures fully adhere to the scheme documented below.
703 * This includes libusb_handle_events() and all the synchronous I/O functions -
704 * libusb hides this headache from you. You do not need to worry about any
705 * of these issues if you stick to that level.
707 * The problem is when we consider the fact that libusb exposes file
708 * descriptors to allow for you to integrate asynchronous USB I/O into
709 * existing main loops, effectively allowing you to do some work behind
710 * libusb's back. If you do take libusb's file descriptors and pass them to
711 * poll()/select() yourself, you need to be aware of the associated issues.
713 * \section eventlock The events lock
715 * The first concept to be introduced is the events lock. The events lock
716 * is used to serialize threads that want to handle events, such that only
717 * one thread is handling events at any one time.
719 * You must take the events lock before polling libusb file descriptors,
720 * using libusb_lock_events(). You must release the lock as soon as you have
721 * aborted your poll()/select() loop, using libusb_unlock_events().
723 * \section threadwait Letting other threads do the work for you
725 * Although the events lock is a critical part of the solution, it is not
726 * enough on it's own. You might wonder if the following is sufficient...
728 libusb_lock_events(ctx);
730 poll(libusb file descriptors, 120*1000);
731 if (poll indicates activity)
732 libusb_handle_events_timeout(ctx, 0);
734 libusb_unlock_events(ctx);
736 * ...and the answer is that it is not. This is because the transfer in the
737 * code shown above may take a long time (say 30 seconds) to complete, and
738 * the lock is not released until the transfer is completed.
740 * Another thread with similar code that wants to do event handling may be
741 * working with a transfer that completes after a few milliseconds. Despite
742 * having such a quick completion time, the other thread cannot check that
743 * status of its transfer until the code above has finished (30 seconds later)
744 * due to contention on the lock.
746 * To solve this, libusb offers you a mechanism to determine when another
747 * thread is handling events. It also offers a mechanism to block your thread
748 * until the event handling thread has completed an event (and this mechanism
749 * does not involve polling of file descriptors).
751 * After determining that another thread is currently handling events, you
752 * obtain the <em>event waiters</em> lock using libusb_lock_event_waiters().
753 * You then re-check that some other thread is still handling events, and if
754 * so, you call libusb_wait_for_event().
756 * libusb_wait_for_event() puts your application to sleep until an event
757 * occurs, or until a thread releases the events lock. When either of these
758 * things happen, your thread is woken up, and should re-check the condition
759 * it was waiting on. It should also re-check that another thread is handling
760 * events, and if not, it should start handling events itself.
762 * This looks like the following, as pseudo-code:
765 if (libusb_try_lock_events(ctx) == 0) {
766 // we obtained the event lock: do our own event handling
768 if (!libusb_event_handling_ok(ctx)) {
769 libusb_unlock_events(ctx);
772 poll(libusb file descriptors, 120*1000);
773 if (poll indicates activity)
774 libusb_handle_events_locked(ctx, 0);
776 libusb_unlock_events(ctx);
778 // another thread is doing event handling. wait for it to signal us that
779 // an event has completed
780 libusb_lock_event_waiters(ctx);
783 // now that we have the event waiters lock, double check that another
784 // thread is still handling events for us. (it may have ceased handling
785 // events in the time it took us to reach this point)
786 if (!libusb_event_handler_active(ctx)) {
787 // whoever was handling events is no longer doing so, try again
788 libusb_unlock_event_waiters(ctx);
792 libusb_wait_for_event(ctx);
794 libusb_unlock_event_waiters(ctx);
796 printf("completed!\n");
799 * A naive look at the above code may suggest that this can only support
800 * one event waiter (hence a total of 2 competing threads, the other doing
801 * event handling), because the event waiter seems to have taken the event
802 * waiters lock while waiting for an event. However, the system does support
803 * multiple event waiters, because libusb_wait_for_event() actually drops
804 * the lock while waiting, and reaquires it before continuing.
806 * We have now implemented code which can dynamically handle situations where
807 * nobody is handling events (so we should do it ourselves), and it can also
808 * handle situations where another thread is doing event handling (so we can
809 * piggyback onto them). It is also equipped to handle a combination of
810 * the two, for example, another thread is doing event handling, but for
811 * whatever reason it stops doing so before our condition is met, so we take
812 * over the event handling.
814 * Four functions were introduced in the above pseudo-code. Their importance
815 * should be apparent from the code shown above.
816 * -# libusb_try_lock_events() is a non-blocking function which attempts
817 * to acquire the events lock but returns a failure code if it is contended.
818 * -# libusb_event_handling_ok() checks that libusb is still happy for your
819 * thread to be performing event handling. Sometimes, libusb needs to
820 * interrupt the event handler, and this is how you can check if you have
821 * been interrupted. If this function returns 0, the correct behaviour is
822 * for you to give up the event handling lock, and then to repeat the cycle.
823 * The following libusb_try_lock_events() will fail, so you will become an
824 * events waiter. For more information on this, read \ref fullstory below.
825 * -# libusb_handle_events_locked() is a variant of
826 * libusb_handle_events_timeout() that you can call while holding the
827 * events lock. libusb_handle_events_timeout() itself implements similar
828 * logic to the above, so be sure not to call it when you are
829 * "working behind libusb's back", as is the case here.
830 * -# libusb_event_handler_active() determines if someone is currently
831 * holding the events lock
833 * You might be wondering why there is no function to wake up all threads
834 * blocked on libusb_wait_for_event(). This is because libusb can do this
835 * internally: it will wake up all such threads when someone calls
836 * libusb_unlock_events() or when a transfer completes (at the point after its
837 * callback has returned).
839 * \subsection fullstory The full story
841 * The above explanation should be enough to get you going, but if you're
842 * really thinking through the issues then you may be left with some more
843 * questions regarding libusb's internals. If you're curious, read on, and if
844 * not, skip to the next section to avoid confusing yourself!
846 * The immediate question that may spring to mind is: what if one thread
847 * modifies the set of file descriptors that need to be polled while another
848 * thread is doing event handling?
850 * There are 2 situations in which this may happen.
851 * -# libusb_open() will add another file descriptor to the poll set,
852 * therefore it is desirable to interrupt the event handler so that it
853 * restarts, picking up the new descriptor.
854 * -# libusb_close() will remove a file descriptor from the poll set. There
855 * are all kinds of race conditions that could arise here, so it is
856 * important that nobody is doing event handling at this time.
858 * libusb handles these issues internally, so application developers do not
859 * have to stop their event handlers while opening/closing devices. Here's how
860 * it works, focusing on the libusb_close() situation first:
862 * -# During initialization, libusb opens an internal pipe, and it adds the read
863 * end of this pipe to the set of file descriptors to be polled.
864 * -# During libusb_close(), libusb writes some dummy data on this control pipe.
865 * This immediately interrupts the event handler. libusb also records
866 * internally that it is trying to interrupt event handlers for this
867 * high-priority event.
868 * -# At this point, some of the functions described above start behaving
870 * - libusb_event_handling_ok() starts returning 1, indicating that it is NOT
871 * OK for event handling to continue.
872 * - libusb_try_lock_events() starts returning 1, indicating that another
873 * thread holds the event handling lock, even if the lock is uncontended.
874 * - libusb_event_handler_active() starts returning 1, indicating that
875 * another thread is doing event handling, even if that is not true.
876 * -# The above changes in behaviour result in the event handler stopping and
877 * giving up the events lock very quickly, giving the high-priority
878 * libusb_close() operation a "free ride" to acquire the events lock. All
879 * threads that are competing to do event handling become event waiters.
880 * -# With the events lock held inside libusb_close(), libusb can safely remove
881 * a file descriptor from the poll set, in the safety of knowledge that
882 * nobody is polling those descriptors or trying to access the poll set.
883 * -# After obtaining the events lock, the close operation completes very
884 * quickly (usually a matter of milliseconds) and then immediately releases
886 * -# At the same time, the behaviour of libusb_event_handling_ok() and friends
887 * reverts to the original, documented behaviour.
888 * -# The release of the events lock causes the threads that are waiting for
889 * events to be woken up and to start competing to become event handlers
890 * again. One of them will succeed; it will then re-obtain the list of poll
891 * descriptors, and USB I/O will then continue as normal.
893 * libusb_open() is similar, and is actually a more simplistic case. Upon a
894 * call to libusb_open():
896 * -# The device is opened and a file descriptor is added to the poll set.
897 * -# libusb sends some dummy data on the control pipe, and records that it
898 * is trying to modify the poll descriptor set.
899 * -# The event handler is interrupted, and the same behaviour change as for
900 * libusb_close() takes effect, causing all event handling threads to become
902 * -# The libusb_open() implementation takes its free ride to the events lock.
903 * -# Happy that it has successfully paused the events handler, libusb_open()
904 * releases the events lock.
905 * -# The event waiter threads are all woken up and compete to become event
906 * handlers again. The one that succeeds will obtain the list of poll
907 * descriptors again, which will include the addition of the new device.
909 * \subsection concl Closing remarks
911 * The above may seem a little complicated, but hopefully I have made it clear
912 * why such complications are necessary. Also, do not forget that this only
913 * applies to applications that take libusb's file descriptors and integrate
914 * them into their own polling loops.
916 * You may decide that it is OK for your multi-threaded application to ignore
917 * some of the rules and locks detailed above, because you don't think that
918 * two threads can ever be polling the descriptors at the same time. If that
919 * is the case, then that's good news for you because you don't have to worry.
920 * But be careful here; remember that the synchronous I/O functions do event
921 * handling internally. If you have one thread doing event handling in a loop
922 * (without implementing the rules and locking semantics documented above)
923 * and another trying to send a synchronous USB transfer, you will end up with
924 * two threads monitoring the same descriptors, and the above-described
925 * undesirable behaviour occuring. The solution is for your polling thread to
926 * play by the rules; the synchronous I/O functions do so, and this will result
927 * in them getting along in perfect harmony.
929 * If you do have a dedicated thread doing event handling, it is perfectly
930 * legal for it to take the event handling lock for long periods of time. Any
931 * synchronous I/O functions you call from other threads will transparently
932 * fall back to the "event waiters" mechanism detailed above. The only
933 * consideration that your event handling thread must apply is the one related
934 * to libusb_event_handling_ok(): you must call this before every poll(), and
935 * give up the events lock if instructed.
938 int usbi_io_init(struct libusb_context *ctx)
942 pthread_mutex_init(&ctx->flying_transfers_lock, NULL);
943 pthread_mutex_init(&ctx->pollfds_lock, NULL);
944 pthread_mutex_init(&ctx->pollfd_modify_lock, NULL);
945 pthread_mutex_init(&ctx->events_lock, NULL);
946 pthread_mutex_init(&ctx->event_waiters_lock, NULL);
947 pthread_cond_init(&ctx->event_waiters_cond, NULL);
948 list_init(&ctx->flying_transfers);
949 list_init(&ctx->pollfds);
951 /* FIXME should use an eventfd on kernels that support it */
952 r = pipe(ctx->ctrl_pipe);
954 return LIBUSB_ERROR_OTHER;
956 r = usbi_add_pollfd(ctx, ctx->ctrl_pipe[0], POLLIN);
963 void usbi_io_exit(struct libusb_context *ctx)
965 usbi_remove_pollfd(ctx, ctx->ctrl_pipe[0]);
966 close(ctx->ctrl_pipe[0]);
967 close(ctx->ctrl_pipe[1]);
970 static int calculate_timeout(struct usbi_transfer *transfer)
973 struct timespec current_time;
974 unsigned int timeout =
975 __USBI_TRANSFER_TO_LIBUSB_TRANSFER(transfer)->timeout;
980 r = usbi_backend->clock_gettime(USBI_CLOCK_MONOTONIC, ¤t_time);
982 usbi_err(ITRANSFER_CTX(transfer),
983 "failed to read monotonic clock, errno=%d", errno);
987 current_time.tv_sec += timeout / 1000;
988 current_time.tv_nsec += (timeout % 1000) * 1000000;
990 if (current_time.tv_nsec > 1000000000) {
991 current_time.tv_nsec -= 1000000000;
992 current_time.tv_sec++;
995 TIMESPEC_TO_TIMEVAL(&transfer->timeout, ¤t_time);
999 static void add_to_flying_list(struct usbi_transfer *transfer)
1001 struct usbi_transfer *cur;
1002 struct timeval *timeout = &transfer->timeout;
1003 struct libusb_context *ctx = ITRANSFER_CTX(transfer);
1005 pthread_mutex_lock(&ctx->flying_transfers_lock);
1007 /* if we have no other flying transfers, start the list with this one */
1008 if (list_empty(&ctx->flying_transfers)) {
1009 list_add(&transfer->list, &ctx->flying_transfers);
1013 /* if we have infinite timeout, append to end of list */
1014 if (!timerisset(timeout)) {
1015 list_add_tail(&transfer->list, &ctx->flying_transfers);
1019 /* otherwise, find appropriate place in list */
1020 list_for_each_entry(cur, &ctx->flying_transfers, list) {
1021 /* find first timeout that occurs after the transfer in question */
1022 struct timeval *cur_tv = &cur->timeout;
1024 if (!timerisset(cur_tv) || (cur_tv->tv_sec > timeout->tv_sec) ||
1025 (cur_tv->tv_sec == timeout->tv_sec &&
1026 cur_tv->tv_usec > timeout->tv_usec)) {
1027 list_add_tail(&transfer->list, &cur->list);
1032 /* otherwise we need to be inserted at the end */
1033 list_add_tail(&transfer->list, &ctx->flying_transfers);
1035 pthread_mutex_unlock(&ctx->flying_transfers_lock);
1038 /** \ingroup asyncio
1039 * Allocate a libusb transfer with a specified number of isochronous packet
1040 * descriptors. The returned transfer is pre-initialized for you. When the new
1041 * transfer is no longer needed, it should be freed with
1042 * libusb_free_transfer().
1044 * Transfers intended for non-isochronous endpoints (e.g. control, bulk,
1045 * interrupt) should specify an iso_packets count of zero.
1047 * For transfers intended for isochronous endpoints, specify an appropriate
1048 * number of packet descriptors to be allocated as part of the transfer.
1049 * The returned transfer is not specially initialized for isochronous I/O;
1050 * you are still required to set the
1051 * \ref libusb_transfer::num_iso_packets "num_iso_packets" and
1052 * \ref libusb_transfer::type "type" fields accordingly.
1054 * It is safe to allocate a transfer with some isochronous packets and then
1055 * use it on a non-isochronous endpoint. If you do this, ensure that at time
1056 * of submission, num_iso_packets is 0 and that type is set appropriately.
1058 * \param iso_packets number of isochronous packet descriptors to allocate
1059 * \returns a newly allocated transfer, or NULL on error
1061 API_EXPORTED struct libusb_transfer *libusb_alloc_transfer(int iso_packets)
1063 size_t os_alloc_size = usbi_backend->transfer_priv_size
1064 + (usbi_backend->add_iso_packet_size * iso_packets);
1065 int alloc_size = sizeof(struct usbi_transfer)
1066 + sizeof(struct libusb_transfer)
1067 + (sizeof(struct libusb_iso_packet_descriptor) * iso_packets)
1069 struct usbi_transfer *itransfer = malloc(alloc_size);
1073 memset(itransfer, 0, alloc_size);
1074 itransfer->num_iso_packets = iso_packets;
1075 return __USBI_TRANSFER_TO_LIBUSB_TRANSFER(itransfer);
1078 /** \ingroup asyncio
1079 * Free a transfer structure. This should be called for all transfers
1080 * allocated with libusb_alloc_transfer().
1082 * If the \ref libusb_transfer_flags::LIBUSB_TRANSFER_FREE_BUFFER
1083 * "LIBUSB_TRANSFER_FREE_BUFFER" flag is set and the transfer buffer is
1084 * non-NULL, this function will also free the transfer buffer using the
1085 * standard system memory allocator (e.g. free()).
1087 * It is legal to call this function with a NULL transfer. In this case,
1088 * the function will simply return safely.
1090 * \param transfer the transfer to free
1092 API_EXPORTED void libusb_free_transfer(struct libusb_transfer *transfer)
1094 struct usbi_transfer *itransfer;
1098 if (transfer->flags & LIBUSB_TRANSFER_FREE_BUFFER && transfer->buffer)
1099 free(transfer->buffer);
1101 itransfer = __LIBUSB_TRANSFER_TO_USBI_TRANSFER(transfer);
1105 /** \ingroup asyncio
1106 * Submit a transfer. This function will fire off the USB transfer and then
1107 * return immediately.
1109 * \param transfer the transfer to submit
1110 * \returns 0 on success
1111 * \returns LIBUSB_ERROR_NO_DEVICE if the device has been disconnected
1112 * \returns LIBUSB_ERROR_BUSY if the transfer has already been submitted.
1113 * \returns another LIBUSB_ERROR code on other failure
1115 API_EXPORTED int libusb_submit_transfer(struct libusb_transfer *transfer)
1117 struct usbi_transfer *itransfer =
1118 __LIBUSB_TRANSFER_TO_USBI_TRANSFER(transfer);
1121 itransfer->transferred = 0;
1122 itransfer->flags = 0;
1123 r = calculate_timeout(itransfer);
1125 return LIBUSB_ERROR_OTHER;
1127 add_to_flying_list(itransfer);
1128 r = usbi_backend->submit_transfer(itransfer);
1130 pthread_mutex_lock(&TRANSFER_CTX(transfer)->flying_transfers_lock);
1131 list_del(&itransfer->list);
1132 pthread_mutex_unlock(&TRANSFER_CTX(transfer)->flying_transfers_lock);
1138 /** \ingroup asyncio
1139 * Asynchronously cancel a previously submitted transfer.
1140 * This function returns immediately, but this does not indicate cancellation
1141 * is complete. Your callback function will be invoked at some later time
1142 * with a transfer status of
1143 * \ref libusb_transfer_status::LIBUSB_TRANSFER_CANCELLED
1144 * "LIBUSB_TRANSFER_CANCELLED."
1146 * \param transfer the transfer to cancel
1147 * \returns 0 on success
1148 * \returns LIBUSB_ERROR_NOT_FOUND if the transfer is already complete or
1150 * \returns a LIBUSB_ERROR code on failure
1152 API_EXPORTED int libusb_cancel_transfer(struct libusb_transfer *transfer)
1154 struct usbi_transfer *itransfer =
1155 __LIBUSB_TRANSFER_TO_USBI_TRANSFER(transfer);
1159 r = usbi_backend->cancel_transfer(itransfer);
1161 usbi_err(TRANSFER_CTX(transfer),
1162 "cancel transfer failed error %d", r);
1166 /* Handle completion of a transfer (completion might be an error condition).
1167 * This will invoke the user-supplied callback function, which may end up
1168 * freeing the transfer. Therefore you cannot use the transfer structure
1169 * after calling this function, and you should free all backend-specific
1170 * data before calling it. */
1171 void usbi_handle_transfer_completion(struct usbi_transfer *itransfer,
1172 enum libusb_transfer_status status)
1174 struct libusb_transfer *transfer =
1175 __USBI_TRANSFER_TO_LIBUSB_TRANSFER(itransfer);
1176 struct libusb_context *ctx = TRANSFER_CTX(transfer);
1179 pthread_mutex_lock(&ctx->flying_transfers_lock);
1180 list_del(&itransfer->list);
1181 pthread_mutex_unlock(&ctx->flying_transfers_lock);
1183 if (status == LIBUSB_TRANSFER_COMPLETED
1184 && transfer->flags & LIBUSB_TRANSFER_SHORT_NOT_OK) {
1185 int rqlen = transfer->length;
1186 if (transfer->type == LIBUSB_TRANSFER_TYPE_CONTROL)
1187 rqlen -= LIBUSB_CONTROL_SETUP_SIZE;
1188 if (rqlen != itransfer->transferred) {
1189 usbi_dbg("interpreting short transfer as error");
1190 status = LIBUSB_TRANSFER_ERROR;
1194 flags = transfer->flags;
1195 transfer->status = status;
1196 transfer->actual_length = itransfer->transferred;
1197 if (transfer->callback)
1198 transfer->callback(transfer);
1199 /* transfer might have been freed by the above call, do not use from
1201 if (flags & LIBUSB_TRANSFER_FREE_TRANSFER)
1202 libusb_free_transfer(transfer);
1203 pthread_mutex_lock(&ctx->event_waiters_lock);
1204 pthread_cond_broadcast(&ctx->event_waiters_cond);
1205 pthread_mutex_unlock(&ctx->event_waiters_lock);
1208 /* Similar to usbi_handle_transfer_completion() but exclusively for transfers
1209 * that were asynchronously cancelled. The same concerns w.r.t. freeing of
1210 * transfers exist here.
1212 void usbi_handle_transfer_cancellation(struct usbi_transfer *transfer)
1214 /* if the URB was cancelled due to timeout, report timeout to the user */
1215 if (transfer->flags & USBI_TRANSFER_TIMED_OUT) {
1216 usbi_dbg("detected timeout cancellation");
1217 usbi_handle_transfer_completion(transfer, LIBUSB_TRANSFER_TIMED_OUT);
1221 /* otherwise its a normal async cancel */
1222 usbi_handle_transfer_completion(transfer, LIBUSB_TRANSFER_CANCELLED);
1226 * Attempt to acquire the event handling lock. This lock is used to ensure that
1227 * only one thread is monitoring libusb event sources at any one time.
1229 * You only need to use this lock if you are developing an application
1230 * which calls poll() or select() on libusb's file descriptors directly.
1231 * If you stick to libusb's event handling loop functions (e.g.
1232 * libusb_handle_events()) then you do not need to be concerned with this
1235 * While holding this lock, you are trusted to actually be handling events.
1236 * If you are no longer handling events, you must call libusb_unlock_events()
1237 * as soon as possible.
1239 * \param ctx the context to operate on, or NULL for the default context
1240 * \returns 0 if the lock was obtained successfully
1241 * \returns 1 if the lock was not obtained (i.e. another thread holds the lock)
1244 API_EXPORTED int libusb_try_lock_events(libusb_context *ctx)
1247 USBI_GET_CONTEXT(ctx);
1249 /* is someone else waiting to modify poll fds? if so, don't let this thread
1250 * start event handling */
1251 pthread_mutex_lock(&ctx->pollfd_modify_lock);
1252 r = ctx->pollfd_modify;
1253 pthread_mutex_unlock(&ctx->pollfd_modify_lock);
1255 usbi_dbg("someone else is modifying poll fds");
1259 r = pthread_mutex_trylock(&ctx->events_lock);
1263 ctx->event_handler_active = 1;
1268 * Acquire the event handling lock, blocking until successful acquisition if
1269 * it is contended. This lock is used to ensure that only one thread is
1270 * monitoring libusb event sources at any one time.
1272 * You only need to use this lock if you are developing an application
1273 * which calls poll() or select() on libusb's file descriptors directly.
1274 * If you stick to libusb's event handling loop functions (e.g.
1275 * libusb_handle_events()) then you do not need to be concerned with this
1278 * While holding this lock, you are trusted to actually be handling events.
1279 * If you are no longer handling events, you must call libusb_unlock_events()
1280 * as soon as possible.
1282 * \param ctx the context to operate on, or NULL for the default context
1285 API_EXPORTED void libusb_lock_events(libusb_context *ctx)
1287 USBI_GET_CONTEXT(ctx);
1288 pthread_mutex_lock(&ctx->events_lock);
1289 ctx->event_handler_active = 1;
1293 * Release the lock previously acquired with libusb_try_lock_events() or
1294 * libusb_lock_events(). Releasing this lock will wake up any threads blocked
1295 * on libusb_wait_for_event().
1297 * \param ctx the context to operate on, or NULL for the default context
1300 API_EXPORTED void libusb_unlock_events(libusb_context *ctx)
1302 USBI_GET_CONTEXT(ctx);
1303 ctx->event_handler_active = 0;
1304 pthread_mutex_unlock(&ctx->events_lock);
1306 /* FIXME: perhaps we should be a bit more efficient by not broadcasting
1307 * the availability of the events lock when we are modifying pollfds
1308 * (check ctx->pollfd_modify)? */
1309 pthread_mutex_lock(&ctx->event_waiters_lock);
1310 pthread_cond_broadcast(&ctx->event_waiters_cond);
1311 pthread_mutex_unlock(&ctx->event_waiters_lock);
1315 * Determine if it is still OK for this thread to be doing event handling.
1317 * Sometimes, libusb needs to temporarily pause all event handlers, and this
1318 * is the function you should use before polling file descriptors to see if
1321 * If this function instructs your thread to give up the events lock, you
1322 * should just continue the usual logic that is documented in \ref mtasync.
1323 * On the next iteration, your thread will fail to obtain the events lock,
1324 * and will hence become an event waiter.
1326 * This function should be called while the events lock is held: you don't
1327 * need to worry about the results of this function if your thread is not
1328 * the current event handler.
1330 * \param ctx the context to operate on, or NULL for the default context
1331 * \returns 1 if event handling can start or continue
1332 * \returns 0 if this thread must give up the events lock
1333 * \see \ref fullstory "Multi-threaded I/O: the full story"
1335 API_EXPORTED int libusb_event_handling_ok(libusb_context *ctx)
1338 USBI_GET_CONTEXT(ctx);
1340 /* is someone else waiting to modify poll fds? if so, don't let this thread
1341 * continue event handling */
1342 pthread_mutex_lock(&ctx->pollfd_modify_lock);
1343 r = ctx->pollfd_modify;
1344 pthread_mutex_unlock(&ctx->pollfd_modify_lock);
1346 usbi_dbg("someone else is modifying poll fds");
1355 * Determine if an active thread is handling events (i.e. if anyone is holding
1356 * the event handling lock).
1358 * \param ctx the context to operate on, or NULL for the default context
1359 * \returns 1 if a thread is handling events
1360 * \returns 0 if there are no threads currently handling events
1363 API_EXPORTED int libusb_event_handler_active(libusb_context *ctx)
1366 USBI_GET_CONTEXT(ctx);
1368 /* is someone else waiting to modify poll fds? if so, don't let this thread
1369 * start event handling -- indicate that event handling is happening */
1370 pthread_mutex_lock(&ctx->pollfd_modify_lock);
1371 r = ctx->pollfd_modify;
1372 pthread_mutex_unlock(&ctx->pollfd_modify_lock);
1374 usbi_dbg("someone else is modifying poll fds");
1378 return ctx->event_handler_active;
1382 * Acquire the event waiters lock. This lock is designed to be obtained under
1383 * the situation where you want to be aware when events are completed, but
1384 * some other thread is event handling so calling libusb_handle_events() is not
1387 * You then obtain this lock, re-check that another thread is still handling
1388 * events, then call libusb_wait_for_event().
1390 * You only need to use this lock if you are developing an application
1391 * which calls poll() or select() on libusb's file descriptors directly,
1392 * <b>and</b> may potentially be handling events from 2 threads simultaenously.
1393 * If you stick to libusb's event handling loop functions (e.g.
1394 * libusb_handle_events()) then you do not need to be concerned with this
1397 * \param ctx the context to operate on, or NULL for the default context
1400 API_EXPORTED void libusb_lock_event_waiters(libusb_context *ctx)
1402 USBI_GET_CONTEXT(ctx);
1403 pthread_mutex_lock(&ctx->event_waiters_lock);
1407 * Release the event waiters lock.
1408 * \param ctx the context to operate on, or NULL for the default context
1411 API_EXPORTED void libusb_unlock_event_waiters(libusb_context *ctx)
1413 USBI_GET_CONTEXT(ctx);
1414 pthread_mutex_unlock(&ctx->event_waiters_lock);
1418 * Wait for another thread to signal completion of an event. Must be called
1419 * with the event waiters lock held, see libusb_lock_event_waiters().
1421 * This function will block until any of the following conditions are met:
1422 * -# The timeout expires
1423 * -# A transfer completes
1424 * -# A thread releases the event handling lock through libusb_unlock_events()
1426 * Condition 1 is obvious. Condition 2 unblocks your thread <em>after</em>
1427 * the callback for the transfer has completed. Condition 3 is important
1428 * because it means that the thread that was previously handling events is no
1429 * longer doing so, so if any events are to complete, another thread needs to
1430 * step up and start event handling.
1432 * This function releases the event waiters lock before putting your thread
1433 * to sleep, and reacquires the lock as it is being woken up.
1435 * \param ctx the context to operate on, or NULL for the default context
1436 * \param tv maximum timeout for this blocking function. A NULL value
1437 * indicates unlimited timeout.
1438 * \returns 0 after a transfer completes or another thread stops event handling
1439 * \returns 1 if the timeout expired
1442 API_EXPORTED int libusb_wait_for_event(libusb_context *ctx, struct timeval *tv)
1444 struct timespec timeout;
1447 USBI_GET_CONTEXT(ctx);
1449 pthread_cond_wait(&ctx->event_waiters_cond, &ctx->event_waiters_lock);
1453 r = usbi_backend->clock_gettime(USBI_CLOCK_REALTIME, &timeout);
1455 usbi_err(ctx, "failed to read realtime clock, error %d", errno);
1456 return LIBUSB_ERROR_OTHER;
1459 timeout.tv_sec += tv->tv_sec;
1460 timeout.tv_nsec += tv->tv_usec * 1000;
1461 if (timeout.tv_nsec > 1000000000) {
1462 timeout.tv_nsec -= 1000000000;
1466 r = pthread_cond_timedwait(&ctx->event_waiters_cond,
1467 &ctx->event_waiters_lock, &timeout);
1468 return (r == ETIMEDOUT);
1471 static void handle_timeout(struct usbi_transfer *itransfer)
1473 struct libusb_transfer *transfer =
1474 __USBI_TRANSFER_TO_LIBUSB_TRANSFER(itransfer);
1477 itransfer->flags |= USBI_TRANSFER_TIMED_OUT;
1478 r = libusb_cancel_transfer(transfer);
1480 usbi_warn(TRANSFER_CTX(transfer),
1481 "async cancel failed %d errno=%d", r, errno);
1484 static int handle_timeouts(struct libusb_context *ctx)
1487 #ifndef USBI_OS_HANDLES_TIMEOUT
1488 struct timespec systime_ts;
1489 struct timeval systime;
1490 struct usbi_transfer *transfer;
1492 USBI_GET_CONTEXT(ctx);
1493 pthread_mutex_lock(&ctx->flying_transfers_lock);
1494 if (list_empty(&ctx->flying_transfers))
1497 /* get current time */
1498 r = usbi_backend->clock_gettime(USBI_CLOCK_MONOTONIC, &systime_ts);
1502 TIMESPEC_TO_TIMEVAL(&systime, &systime_ts);
1504 /* iterate through flying transfers list, finding all transfers that
1505 * have expired timeouts */
1506 list_for_each_entry(transfer, &ctx->flying_transfers, list) {
1507 struct timeval *cur_tv = &transfer->timeout;
1509 /* if we've reached transfers of infinite timeout, we're all done */
1510 if (!timerisset(cur_tv))
1513 /* ignore timeouts we've already handled */
1514 if (transfer->flags & USBI_TRANSFER_TIMED_OUT)
1517 /* if transfer has non-expired timeout, nothing more to do */
1518 if ((cur_tv->tv_sec > systime.tv_sec) ||
1519 (cur_tv->tv_sec == systime.tv_sec &&
1520 cur_tv->tv_usec > systime.tv_usec))
1523 /* otherwise, we've got an expired timeout to handle */
1524 handle_timeout(transfer);
1528 pthread_mutex_unlock(&ctx->flying_transfers_lock);
1534 /* do the actual event handling. assumes that no other thread is concurrently
1535 * doing the same thing. */
1536 static int handle_events(struct libusb_context *ctx, struct timeval *tv)
1539 struct usbi_pollfd *ipollfd;
1545 pthread_mutex_lock(&ctx->pollfds_lock);
1546 list_for_each_entry(ipollfd, &ctx->pollfds, list)
1549 /* TODO: malloc when number of fd's changes, not on every poll */
1550 fds = malloc(sizeof(*fds) * nfds);
1552 return LIBUSB_ERROR_NO_MEM;
1554 list_for_each_entry(ipollfd, &ctx->pollfds, list) {
1555 struct libusb_pollfd *pollfd = &ipollfd->pollfd;
1556 int fd = pollfd->fd;
1559 fds[i].events = pollfd->events;
1562 pthread_mutex_unlock(&ctx->pollfds_lock);
1564 timeout_ms = (tv->tv_sec * 1000) + (tv->tv_usec / 1000);
1566 /* round up to next millisecond */
1567 if (tv->tv_usec % 1000)
1570 usbi_dbg("poll() %d fds with timeout in %dms", nfds, timeout_ms);
1571 r = poll(fds, nfds, timeout_ms);
1572 usbi_dbg("poll() returned %d", r);
1575 return handle_timeouts(ctx);
1576 } else if (r == -1 && errno == EINTR) {
1578 return LIBUSB_ERROR_INTERRUPTED;
1581 usbi_err(ctx, "poll failed %d err=%d\n", r, errno);
1582 return LIBUSB_ERROR_IO;
1585 /* fd[0] is always the ctrl pipe */
1586 if (fds[0].revents) {
1587 /* another thread wanted to interrupt event handling, and it succeeded!
1588 * handle any other events that cropped up at the same time, and
1590 usbi_dbg("caught a fish on the control pipe");
1596 /* prevent OS backend from trying to handle events on ctrl pipe */
1602 r = usbi_backend->handle_events(ctx, fds, nfds, r);
1604 usbi_err(ctx, "backend handle_events failed with error %d", r);
1611 /* returns the smallest of:
1612 * 1. timeout of next URB
1613 * 2. user-supplied timeout
1614 * returns 1 if there is an already-expired timeout, otherwise returns 0
1617 static int get_next_timeout(libusb_context *ctx, struct timeval *tv,
1618 struct timeval *out)
1620 struct timeval timeout;
1621 int r = libusb_get_next_timeout(ctx, &timeout);
1623 /* timeout already expired? */
1624 if (!timerisset(&timeout))
1627 /* choose the smallest of next URB timeout or user specified timeout */
1628 if (timercmp(&timeout, tv, <))
1639 * Handle any pending events.
1641 * libusb determines "pending events" by checking if any timeouts have expired
1642 * and by checking the set of file descriptors for activity.
1644 * If a zero timeval is passed, this function will handle any already-pending
1645 * events and then immediately return in non-blocking style.
1647 * If a non-zero timeval is passed and no events are currently pending, this
1648 * function will block waiting for events to handle up until the specified
1649 * timeout. If an event arrives or a signal is raised, this function will
1652 * \param ctx the context to operate on, or NULL for the default context
1653 * \param tv the maximum time to block waiting for events, or zero for
1655 * \returns 0 on success, or a LIBUSB_ERROR code on failure
1657 API_EXPORTED int libusb_handle_events_timeout(libusb_context *ctx,
1661 struct timeval poll_timeout;
1663 USBI_GET_CONTEXT(ctx);
1664 r = get_next_timeout(ctx, tv, &poll_timeout);
1666 /* timeout already expired */
1667 return handle_timeouts(ctx);
1671 if (libusb_try_lock_events(ctx) == 0) {
1672 /* we obtained the event lock: do our own event handling */
1673 r = handle_events(ctx, &poll_timeout);
1674 libusb_unlock_events(ctx);
1678 /* another thread is doing event handling. wait for pthread events that
1679 * notify event completion. */
1680 libusb_lock_event_waiters(ctx);
1682 if (!libusb_event_handler_active(ctx)) {
1683 /* we hit a race: whoever was event handling earlier finished in the
1684 * time it took us to reach this point. try the cycle again. */
1685 libusb_unlock_event_waiters(ctx);
1686 usbi_dbg("event handler was active but went away, retrying");
1690 usbi_dbg("another thread is doing event handling");
1691 r = libusb_wait_for_event(ctx, &poll_timeout);
1692 libusb_unlock_event_waiters(ctx);
1697 return handle_timeouts(ctx);
1703 * Handle any pending events in blocking mode with a sensible timeout. This
1704 * timeout is currently hardcoded at 2 seconds but we may change this if we
1705 * decide other values are more sensible. For finer control over whether this
1706 * function is blocking or non-blocking, or the maximum timeout, use
1707 * libusb_handle_events_timeout() instead.
1709 * \param ctx the context to operate on, or NULL for the default context
1710 * \returns 0 on success, or a LIBUSB_ERROR code on failure
1712 API_EXPORTED int libusb_handle_events(libusb_context *ctx)
1717 return libusb_handle_events_timeout(ctx, &tv);
1721 * Handle any pending events by polling file descriptors, without checking if
1722 * any other threads are already doing so. Must be called with the event lock
1723 * held, see libusb_lock_events().
1725 * This function is designed to be called under the situation where you have
1726 * taken the event lock and are calling poll()/select() directly on libusb's
1727 * file descriptors (as opposed to using libusb_handle_events() or similar).
1728 * You detect events on libusb's descriptors, so you then call this function
1729 * with a zero timeout value (while still holding the event lock).
1731 * \param ctx the context to operate on, or NULL for the default context
1732 * \param tv the maximum time to block waiting for events, or zero for
1734 * \returns 0 on success, or a LIBUSB_ERROR code on failure
1737 API_EXPORTED int libusb_handle_events_locked(libusb_context *ctx,
1741 struct timeval poll_timeout;
1743 USBI_GET_CONTEXT(ctx);
1744 r = get_next_timeout(ctx, tv, &poll_timeout);
1746 /* timeout already expired */
1747 return handle_timeouts(ctx);
1750 return handle_events(ctx, &poll_timeout);
1754 * Determine the next internal timeout that libusb needs to handle. You only
1755 * need to use this function if you are calling poll() or select() or similar
1756 * on libusb's file descriptors yourself - you do not need to use it if you
1757 * are calling libusb_handle_events() or a variant directly.
1759 * You should call this function in your main loop in order to determine how
1760 * long to wait for select() or poll() to return results. libusb needs to be
1761 * called into at this timeout, so you should use it as an upper bound on
1762 * your select() or poll() call.
1764 * When the timeout has expired, call into libusb_handle_events_timeout()
1765 * (perhaps in non-blocking mode) so that libusb can handle the timeout.
1767 * This function may return 1 (success) and an all-zero timeval. If this is
1768 * the case, it indicates that libusb has a timeout that has already expired
1769 * so you should call libusb_handle_events_timeout() or similar immediately.
1770 * A return code of 0 indicates that there are no pending timeouts.
1772 * \param ctx the context to operate on, or NULL for the default context
1773 * \param tv output location for a relative time against the current
1774 * clock in which libusb must be called into in order to process timeout events
1775 * \returns 0 if there are no pending timeouts, 1 if a timeout was returned,
1776 * or LIBUSB_ERROR_OTHER on failure
1778 API_EXPORTED int libusb_get_next_timeout(libusb_context *ctx,
1781 #ifndef USBI_OS_HANDLES_TIMEOUT
1782 struct usbi_transfer *transfer;
1783 struct timespec cur_ts;
1784 struct timeval cur_tv;
1785 struct timeval *next_timeout;
1789 USBI_GET_CONTEXT(ctx);
1790 pthread_mutex_lock(&ctx->flying_transfers_lock);
1791 if (list_empty(&ctx->flying_transfers)) {
1792 pthread_mutex_unlock(&ctx->flying_transfers_lock);
1793 usbi_dbg("no URBs, no timeout!");
1797 /* find next transfer which hasn't already been processed as timed out */
1798 list_for_each_entry(transfer, &ctx->flying_transfers, list) {
1799 if (!(transfer->flags & USBI_TRANSFER_TIMED_OUT)) {
1804 pthread_mutex_unlock(&ctx->flying_transfers_lock);
1807 usbi_dbg("all URBs have already been processed for timeouts");
1811 next_timeout = &transfer->timeout;
1813 /* no timeout for next transfer */
1814 if (!timerisset(next_timeout)) {
1815 usbi_dbg("no URBs with timeouts, no timeout!");
1819 r = usbi_backend->clock_gettime(USBI_CLOCK_MONOTONIC, &cur_ts);
1821 usbi_err(ctx, "failed to read monotonic clock, errno=%d", errno);
1822 return LIBUSB_ERROR_OTHER;
1824 TIMESPEC_TO_TIMEVAL(&cur_tv, &cur_ts);
1826 if (timercmp(&cur_tv, next_timeout, >=)) {
1827 usbi_dbg("first timeout already expired");
1830 timersub(next_timeout, &cur_tv, tv);
1831 usbi_dbg("next timeout in %d.%06ds", tv->tv_sec, tv->tv_usec);
1841 * Register notification functions for file descriptor additions/removals.
1842 * These functions will be invoked for every new or removed file descriptor
1843 * that libusb uses as an event source.
1845 * To remove notifiers, pass NULL values for the function pointers.
1847 * Note that file descriptors may have been added even before you register
1848 * these notifiers (e.g. at libusb_init() time).
1850 * Additionally, note that the removal notifier may be called during
1851 * libusb_exit() (e.g. when it is closing file descriptors that were opened
1852 * and added to the poll set at libusb_init() time). If you don't want this,
1853 * remove the notifiers immediately before calling libusb_exit().
1855 * \param ctx the context to operate on, or NULL for the default context
1856 * \param added_cb pointer to function for addition notifications
1857 * \param removed_cb pointer to function for removal notifications
1858 * \param user_data User data to be passed back to callbacks (useful for
1859 * passing context information)
1861 API_EXPORTED void libusb_set_pollfd_notifiers(libusb_context *ctx,
1862 libusb_pollfd_added_cb added_cb, libusb_pollfd_removed_cb removed_cb,
1865 USBI_GET_CONTEXT(ctx);
1866 ctx->fd_added_cb = added_cb;
1867 ctx->fd_removed_cb = removed_cb;
1868 ctx->fd_cb_user_data = user_data;
1871 /* Add a file descriptor to the list of file descriptors to be monitored.
1872 * events should be specified as a bitmask of events passed to poll(), e.g.
1873 * POLLIN and/or POLLOUT. */
1874 int usbi_add_pollfd(struct libusb_context *ctx, int fd, short events)
1876 struct usbi_pollfd *ipollfd = malloc(sizeof(*ipollfd));
1878 return LIBUSB_ERROR_NO_MEM;
1880 usbi_dbg("add fd %d events %d", fd, events);
1881 ipollfd->pollfd.fd = fd;
1882 ipollfd->pollfd.events = events;
1883 pthread_mutex_lock(&ctx->pollfds_lock);
1884 list_add_tail(&ipollfd->list, &ctx->pollfds);
1885 pthread_mutex_unlock(&ctx->pollfds_lock);
1887 if (ctx->fd_added_cb)
1888 ctx->fd_added_cb(fd, events, ctx->fd_cb_user_data);
1892 /* Remove a file descriptor from the list of file descriptors to be polled. */
1893 void usbi_remove_pollfd(struct libusb_context *ctx, int fd)
1895 struct usbi_pollfd *ipollfd;
1898 usbi_dbg("remove fd %d", fd);
1899 pthread_mutex_lock(&ctx->pollfds_lock);
1900 list_for_each_entry(ipollfd, &ctx->pollfds, list)
1901 if (ipollfd->pollfd.fd == fd) {
1907 usbi_dbg("couldn't find fd %d to remove", fd);
1908 pthread_mutex_unlock(&ctx->pollfds_lock);
1912 list_del(&ipollfd->list);
1913 pthread_mutex_unlock(&ctx->pollfds_lock);
1915 if (ctx->fd_removed_cb)
1916 ctx->fd_removed_cb(fd, ctx->fd_cb_user_data);
1920 * Retrieve a list of file descriptors that should be polled by your main loop
1921 * as libusb event sources.
1923 * The returned list is NULL-terminated and should be freed with free() when
1924 * done. The actual list contents must not be touched.
1926 * \param ctx the context to operate on, or NULL for the default context
1927 * \returns a NULL-terminated list of libusb_pollfd structures, or NULL on
1930 API_EXPORTED const struct libusb_pollfd **libusb_get_pollfds(
1931 libusb_context *ctx)
1933 struct libusb_pollfd **ret = NULL;
1934 struct usbi_pollfd *ipollfd;
1937 USBI_GET_CONTEXT(ctx);
1939 pthread_mutex_lock(&ctx->pollfds_lock);
1940 list_for_each_entry(ipollfd, &ctx->pollfds, list)
1943 ret = calloc(cnt + 1, sizeof(struct libusb_pollfd *));
1947 list_for_each_entry(ipollfd, &ctx->pollfds, list)
1948 ret[i++] = (struct libusb_pollfd *) ipollfd;
1952 pthread_mutex_unlock(&ctx->pollfds_lock);
1953 return (const struct libusb_pollfd **) ret;
1956 /* Backends call this from handle_events to report disconnection of a device.
1957 * The transfers get cancelled appropriately.
1959 void usbi_handle_disconnect(struct libusb_device_handle *handle)
1961 struct usbi_transfer *cur;
1962 struct usbi_transfer *to_cancel;
1964 usbi_dbg("device %d.%d",
1965 handle->dev->bus_number, handle->dev->device_address);
1967 /* terminate all pending transfers with the LIBUSB_TRANSFER_NO_DEVICE
1970 * this is a bit tricky because:
1971 * 1. we can't do transfer completion while holding flying_transfers_lock
1972 * 2. the transfers list can change underneath us - if we were to build a
1973 * list of transfers to complete (while holding look), the situation
1974 * might be different by the time we come to free them
1976 * so we resort to a loop-based approach as below
1977 * FIXME: is this still potentially racy?
1981 pthread_mutex_lock(&HANDLE_CTX(handle)->flying_transfers_lock);
1983 list_for_each_entry(cur, &HANDLE_CTX(handle)->flying_transfers, list)
1984 if (__USBI_TRANSFER_TO_LIBUSB_TRANSFER(cur)->dev_handle == handle) {
1988 pthread_mutex_unlock(&HANDLE_CTX(handle)->flying_transfers_lock);
1993 usbi_backend->clear_transfer_priv(to_cancel);
1994 usbi_handle_transfer_completion(to_cancel, LIBUSB_TRANSFER_NO_DEVICE);