2 relayfs - a high-speed data relay filesystem
3 ============================================
5 relayfs is a filesystem designed to provide an efficient mechanism for
6 tools and facilities to relay large amounts of data from kernel space
9 The main idea behind relayfs is that every data flow is put into a
10 separate "channel" and each channel is a file. In practice, each
11 channel is a separate memory buffer allocated from within kernel space
12 upon channel instantiation. Software needing to relay data to user
13 space would open a channel or a number of channels, depending on its
14 needs, and would log data to that channel. All the buffering and
15 locking mechanics are taken care of by relayfs. The actual format and
16 protocol used for each channel is up to relayfs' clients.
18 relayfs makes no provisions for copying the same data to more than a
19 single channel. This is for the clients of the relay to take care of,
20 and so is any form of data filtering. The purpose is to keep relayfs
21 as simple as possible.
27 In addition to the relayfs kernel API described below, relayfs
28 implements basic file operations. Here are the file operations that
29 are available and some comments regarding their behavior:
31 open() enables user to open an _existing_ channel. A channel can be
32 opened in blocking or non-blocking mode, and can be opened
33 for reading as well as for writing. Readers will by default
36 mmap() results in channel's memory buffer being mmapped into the
37 caller's memory space.
39 read() since we are dealing with circular buffers, the user is only
40 allowed to read forward. Some apps may want to loop around
41 read() waiting for incoming data - if there is no data
42 available, read will put the reader on a wait queue until
43 data is available (blocking mode). Non-blocking reads return
44 -EAGAIN if data is not available.
47 write() writing from user space operates exactly as relay_write() does
50 poll() POLLIN/POLLRDNORM/POLLOUT/POLLWRNORM/POLLERR supported.
52 close() decrements the channel's refcount. When the refcount reaches
53 0 i.e. when no process or kernel client has the file open
54 (see relay_close() below), the channel buffer is freed.
57 In order for a user application to make use of relayfs files, the
58 relayfs filesystem must be mounted. For example,
60 mount -t relayfs relayfs /mountpoint
63 The relayfs kernel API
64 ======================
66 relayfs channels are implemented as circular buffers subdivided into
67 'sub-buffers'. kernel clients write data into the channel using
68 relay_write(), and are notified via a set of callbacks when
69 significant events occur within the channel. 'Significant events'
72 - a sub-buffer has been filled i.e. the current write won't fit into the
73 current sub-buffer, and a 'buffer-switch' is triggered, after which
74 the data is written into the next buffer (if the next buffer is
75 empty). The client is notified of this condition via two callbacks,
76 one providing an opportunity to perform start-of-buffer tasks, the
77 other end-of-buffer tasks.
79 - data is ready for the client to process. The client can choose to
80 be notified either on a per-sub-buffer basis (bulk delivery) or
81 per-write basis (packet delivery).
83 - data has been written to the channel from user space. The client can
84 use this notification to accept and process 'commands' sent to the
87 - the channel has been opened/closed/mapped/unmapped from user space.
88 The client can use this notification to trigger actions within the
89 kernel application, such as enabling/disabling logging to the
90 channel. It can also return result codes from the callback,
91 indicating that the operation should fail e.g. in order to restrict
92 more than one user space open or mmap.
94 - the channel needs resizing, or needs to update its
95 state based on the results of the resize. Resizing the channel is
96 up to the kernel client to actually perform. If the channel is
97 configured for resizing, the client is notified when the unread data
98 in the channel passes a preset threshold, giving it the opportunity
99 to allocate a new channel buffer and replace the old one.
104 Channel readers use an opaque rchan_reader object to read from
105 channels. For VFS readers (those using read(2) to read from a
106 channel), these objects are automatically created and used internally;
107 only kernel clients that need to directly read from channels, or whose
108 userspace applications use mmap to access channel data, need to know
109 anything about rchan_readers - others may skip this section.
111 A relay channel can have any number of readers, each represented by an
112 rchan_reader instance, which is used to encapsulate reader settings
113 and state. rchan_reader objects should be treated as opaque by kernel
114 clients. To create a reader object for directly accessing a channel
115 from kernel space, call the add_rchan_reader() kernel API function:
117 rchan_reader *add_rchan_reader(rchan_id, auto_consume)
119 This function returns an rchan_reader instance if successful, which
120 should then be passed to relay_read() when the kernel client is
121 interested in reading from the channel.
123 The auto_consume parameter indicates whether a read done by this
124 reader will automatically 'consume' that portion of the unread channel
125 buffer when relay_read() is called (see below for more details).
127 To close the reader, call
129 remove_rchan_reader(reader)
131 which will remove the reader from the list of current readers.
134 To create a reader object representing a userspace mmap reader in the
135 kernel application, call the add_map_reader() kernel API function:
137 rchan_reader *add_map_reader(rchan_id)
139 This function returns an rchan_reader instance if successful, whose
140 main purpose is as an argument to be passed into
141 relay_buffers_consumed() when the kernel client becomes aware that
142 data has been read by a user application using mmap to read from the
143 channel buffer. There is no auto_consume option in this case, since
144 only the kernel client/user application knows when data has been read.
146 To close the map reader, call
148 remove_map_reader(reader)
150 which will remove the reader from the list of current readers.
155 A relayfs channel is a circular buffer, which means that if there is
156 no reader reading from it or a reader reading too slowly, at some
157 point the channel writer will 'lap' the reader and data will be lost.
158 In normal use, readers will always be able to keep up with writers and
159 the buffer is thus never in danger of becoming full. In many
160 applications, it's sufficient to ensure that this is practically
161 speaking always the case, by making the buffers large enough. These
162 types of applications can basically open the channel as
163 RELAY_MODE_CONTINOUS (the default anyway) and not worry about the
164 meaning of 'consume' and skip the rest of this section.
166 If it's important for the application that a kernel client never allow
167 writers to overwrite unread data, the channel should be opened using
168 RELAY_MODE_NO_OVERWRITE and must be kept apprised of the count of
169 bytes actually read by the (typically) user-space channel readers.
170 This count is referred to as the 'consumed count'. read(2) channel
171 readers automatically update the channel's 'consumed count' as they
172 read. If the usage mode is to have only read(2) readers, which is
173 typically the case, the kernel client doesn't need to worry about any
174 of the relayfs functions having to do with 'bytes consumed' and can
175 skip the rest of this section. (Note that it is possible to have
176 multiple read(2) or auto-consuming readers, but like having multiple
177 readers on a pipe, these readers will race with each other i.e. it's
178 supported, but doesn't make much sense).
180 If the kernel client cannot rely on an auto-consuming reader to keep
181 the 'consumed count' up-to-date, then it must do so manually, by
182 making the appropriate calls to relay_buffers_consumed() or
183 relay_bytes_consumed(). In most cases, this should only be necessary
184 for bulk mmap clients - almost all packet clients should be covered by
185 having auto-consuming read(2) readers. For mmapped bulk clients, for
186 instance, there are no auto-consuming VFS readers, so the kernel
187 client needs to make the call to relay_buffers_consumed() after
188 sub-buffers are read.
193 Here's a summary of the API relayfs provides to in-kernel clients:
195 int relay_open(channel_path, bufsize, nbufs, channel_flags,
196 channel_callbacks, start_reserve, end_reserve,
197 rchan_start_reserve, resize_min, resize_max, mode,
198 init_buf, init_buf_size)
199 int relay_write(channel_id, *data_ptr, count, time_delta_offset, **wrote)
200 rchan_reader *add_rchan_reader(channel_id, auto_consume)
201 int remove_rchan_reader(rchan_reader *reader)
202 rchan_reader *add_map_reader(channel_id)
203 int remove_map_reader(rchan_reader *reader)
204 int relay_read(reader, buf, count, wait, *actual_read_offset)
205 void relay_buffers_consumed(reader, buffers_consumed)
206 void relay_bytes_consumed(reader, bytes_consumed, read_offset)
207 int relay_bytes_avail(reader)
208 int rchan_full(reader)
209 int rchan_empty(reader)
210 int relay_info(channel_id, *channel_info)
211 int relay_close(channel_id)
212 int relay_realloc_buffer(channel_id, nbufs, async)
213 int relay_replace_buffer(channel_id)
214 int relay_reset(int rchan_id)
217 int relay_open(channel_path, bufsize, nbufs,
218 channel_flags, channel_callbacks, start_reserve,
219 end_reserve, rchan_start_reserve, resize_min, resize_max, mode)
221 relay_open() is used to create a new entry in relayfs. This new entry
222 is created according to channel_path. channel_path contains the
223 absolute path to the channel file on relayfs. If, for example, the
224 caller sets channel_path to "/xlog/9", a "xlog/9" entry will appear
225 within relayfs automatically and the "xlog" directory will be created
226 in the filesystem's root. relayfs does not implement any policy on
227 its content, except to disallow the opening of two channels using the
228 same file. There are, nevertheless a set of guidelines for using
229 relayfs. Basically, each facility using relayfs should use a top-level
230 directory identifying it. The entry created above, for example,
231 presumably belongs to the "xlog" software.
233 The remaining parameters for relay_open() are as follows:
235 - channel_flags - an ORed combination of attribute values controlling
236 common channel characteristics:
238 - logging scheme - relayfs use 2 mutually exclusive schemes
239 for logging data to a channel. The 'lockless scheme'
240 reserves and writes data to a channel without the need of
241 any type of locking on the channel. This is the preferred
242 scheme, but may not be available on a given architecture (it
243 relies on the presence of a cmpxchg instruction). It's
244 specified by the RELAY_SCHEME_LOCKLESS flag. The 'locking
245 scheme' either obtains a lock on the channel for writing or
246 disables interrupts, depending on whether the channel was
247 opened for SMP or global usage (see below). It's specified
248 by the RELAY_SCHEME_LOCKING flag. While a client may want
249 to explicitly specify a particular scheme to use, it's more
250 convenient to specify RELAY_SCHEME_ANY for this flag, which
251 will allow relayfs to choose the best available scheme i.e.
252 lockless if supported.
254 - overwrite mode (default is RELAY_MODE_CONTINUOUS) -
255 If RELAY_MODE_CONTINUOUS is specified, writes to the channel
256 will succeed regardless of whether there are up-to-date
257 consumers or not. If RELAY_MODE_NO_OVERWRITE is specified,
258 the channel becomes 'full' when the total amount of buffer
259 space unconsumed by readers equals or exceeds the total
260 buffer size. With the buffer in this state, writes to the
261 buffer will fail - clients need to check the return code from
262 relay_write() to determine if this is the case and act
263 accordingly - 0 or a negative value indicate the write failed.
265 - SMP usage - this applies only when the locking scheme is in
266 use. If RELAY_USAGE_SMP is specified, it's assumed that the
267 channel will be used in a per-CPU fashion and consequently,
268 the only locking that will be done for writes is to disable
269 local irqs. If RELAY_USAGE_GLOBAL is specified, it's assumed
270 that writes to the buffer can occur within any CPU context,
271 and spinlock_irq_save will be used to lock the buffer.
273 - delivery mode - if RELAY_DELIVERY_BULK is specified, the
274 client will be notified via its deliver() callback whenever a
275 sub-buffer has been filled. Alternatively,
276 RELAY_DELIVERY_PACKET will cause delivery to occur after the
277 completion of each write. See the description of the channel
278 callbacks below for more details.
280 - timestamping - if RELAY_TIMESTAMP_TSC is specified and the
281 architecture supports it, efficient TSC 'timestamps' can be
282 associated with each write, otherwise more expensive
283 gettimeofday() timestamping is used. At the beginning of
284 each sub-buffer, a gettimeofday() timestamp and the current
285 TSC, if supported, are read, and are passed on to the client
286 via the buffer_start() callback. This allows correlation of
287 the current time with the current TSC for subsequent writes.
288 Each subsequent write is associated with a 'time delta',
289 which is either the current TSC, if the channel is using
290 TSCs, or the difference between the buffer_start gettimeofday
291 timestamp and the gettimeofday time read for the current
292 write. Note that relayfs never writes either a timestamp or
293 time delta into the buffer unless explicitly asked to (see
294 the description of relay_write() for details).
296 - bufsize - the size of the 'sub-buffers' making up the circular channel
297 buffer. For the lockless scheme, this must be a power of 2.
299 - nbufs - the number of 'sub-buffers' making up the circular
300 channel buffer. This must be a power of 2.
302 The total size of the channel buffer is bufsize * nbufs rounded up
303 to the next kernel page size. If the lockless scheme is used, both
304 bufsize and nbufs must be a power of 2. If the locking scheme is
305 used, the bufsize can be anything and nbufs must be a power of 2. If
306 RELAY_SCHEME_ANY is used, the bufsize and nbufs should be a power of 2.
308 NOTE: if nbufs is 1, relayfs will bypass the normal size
309 checks and will allocate an rvmalloced buffer of size bufsize.
310 This buffer will be freed when relay_close() is called, if the channel
311 isn't still being referenced.
313 - callbacks - a table of callback functions called when events occur
314 within the data relay that clients need to know about:
316 - int buffer_start(channel_id, current_write_pos, buffer_id,
317 start_time, start_tsc, using_tsc) -
319 called at the beginning of a new sub-buffer, the
320 buffer_start() callback gives the client an opportunity to
321 write data into space reserved at the beginning of a
322 sub-buffer. The client should only write into the buffer
323 if it specified a value for start_reserve and/or
324 channel_start_reserve (see below) when the channel was
325 opened. In the latter case, the client can determine
326 whether to write its one-time rchan_start_reserve data by
327 examining the value of buffer_id, which will be 0 for the
328 first sub-buffer. The address that the client can write
329 to is contained in current_write_pos (the client by
330 definition knows how much it can write i.e. the value it
331 passed to relay_open() for start_reserve/
332 channel_start_reserve). start_time contains the
333 gettimeofday() value for the start of the buffer and start
334 TSC contains the TSC read at the same time. The using_tsc
335 param indicates whether or not start_tsc is valid (it
336 wouldn't be if TSC timestamping isn't being used).
338 The client should return the number of bytes it wrote to
339 the channel, 0 if none.
341 - int buffer_end(channel_id, current_write_pos, end_of_buffer,
342 end_time, end_tsc, using_tsc)
344 called at the end of a sub-buffer, the buffer_end()
345 callback gives the client an opportunity to perform
346 end-of-buffer processing. Note that the current_write_pos
347 is the position where the next write would occur, but
348 since the current write wouldn't fit (which is the trigger
349 for the buffer_end event), the buffer is considered full
350 even though there may be unused space at the end. The
351 end_of_buffer param pointer value can be used to determine
352 exactly the size of the unused space. The client should
353 only write into the buffer if it specified a value for
354 end_reserve when the channel was opened. If the client
355 doesn't write anything i.e. returns 0, the unused space at
356 the end of the sub-buffer is available via relay_info() -
357 this data may be needed by the client later if it needs to
358 process raw sub-buffers (an alternative would be to save
359 the unused bytes count value in end_reserve space at the
360 end of each sub-buffer during buffer_end processing and
361 read it when needed at a later time. The other
362 alternative would be to use read(2), which makes the
363 unused count invisible to the caller). end_time contains
364 the gettimeofday() value for the end of the buffer and end
365 TSC contains the TSC read at the same time. The using_tsc
366 param indicates whether or not end_tsc is valid (it
367 wouldn't be if TSC timestamping isn't being used).
369 The client should return the number of bytes it wrote to
370 the channel, 0 if none.
372 - void deliver(channel_id, from, len)
374 called when data is ready for the client. This callback
375 is used to notify a client when a sub-buffer is complete
376 (in the case of bulk delivery) or a single write is
377 complete (packet delivery). A bulk delivery client might
378 wish to then signal a daemon that a sub-buffer is ready.
379 A packet delivery client might wish to process the packet
380 or send it elsewhere. The from param is a pointer to the
381 delivered data and len specifies how many bytes are ready.
383 - void user_deliver(channel_id, from, len)
385 called when data has been written to the channel from user
386 space. This callback is used to notify a client when a
387 successful write from userspace has occurred, independent
388 of whether bulk or packet delivery is in use. This can be
389 used to allow userspace programs to communicate with the
390 kernel client through the channel via out-of-band write(2)
391 'commands' instead of via ioctls, for instance. The from
392 param is a pointer to the delivered data and len specifies
393 how many bytes are ready. Note that this callback occurs
394 after the bytes have been successfully written into the
395 channel, which means that channel readers must be able to
396 deal with the 'command' data which will appear in the
397 channel data stream just as any other userspace or
398 non-userspace write would.
400 - int needs_resize(channel_id, resize_type,
401 suggested_buf_size, suggested_n_bufs)
403 called when a channel's buffers are in danger of becoming
404 full i.e. the number of unread bytes in the channel passes
405 a preset threshold, or when the current capacity of a
406 channel's buffer is no longer needed. Also called to
407 notify the client when a channel's buffer has been
408 replaced. If resize_type is RELAY_RESIZE_EXPAND or
409 RELAY_RESIZE_SHRINK, the kernel client should arrange to
410 call relay_realloc_buffer() with the suggested buffer size
411 and buffer count, which will allocate (but will not
412 replace the old one) a new buffer of the recommended size
413 for the channel. When the allocation has completed,
414 needs_resize() is again called, this time with a
415 resize_type of RELAY_RESIZE_REPLACE. The kernel client
416 should then arrange to call relay_replace_buffer() to
417 actually replace the old channel buffer with the newly
418 allocated buffer. Finally, once the buffer replacement
419 has completed, needs_resize() is again called, this time
420 with a resize_type of RELAY_RESIZE_REPLACED, to inform the
421 client that the replacement is complete and additionally
422 confirming the current sub-buffer size and number of
423 sub-buffers. Note that a resize can be canceled if
424 relay_realloc_buffer() is called with the async param
425 non-zero and the resize conditions no longer hold. In
426 this case, the RELAY_RESIZE_REPLACED suggested number of
427 sub-buffers will be the same as the number of sub-buffers
428 that existed before the RELAY_RESIZE_SHRINK or EXPAND i.e.
429 values indicating that the resize didn't actually occur.
431 - int fileop_notify(channel_id, struct file *filp, enum relay_fileop)
433 called when a userspace file operation has occurred or
434 will occur on a relayfs channel file. These notifications
435 can be used by the kernel client to trigger actions within
436 the kernel client when the corresponding event occurs,
437 such as enabling logging only when a userspace application
438 opens or mmaps a relayfs file and disabling it again when
439 the file is closed or unmapped. The kernel client can
440 also return its own return value, which can affect the
441 outcome of file operation - returning 0 indicates that the
442 operation should succeed, and returning a negative value
443 indicates that the operation should be failed, and that
444 the returned value should be returned to the ultimate
445 caller e.g. returning -EPERM from the open fileop will
446 cause the open to fail with -EPERM. Among other things,
447 the return value can be used to restrict a relayfs file
448 from being opened or mmap'ed more than once. The currently
449 implemented fileops are:
451 RELAY_FILE_OPEN - a relayfs file is being opened. Return
452 0 to allow it to succeed, negative to
453 have it fail. A negative return value will
454 be passed on unmodified to the open fileop.
455 RELAY_FILE_CLOSE- a relayfs file is being closed. The return
457 RELAY_FILE_MAP - a relayfs file is being mmap'ed. Return 0
458 to allow it to succeed, negative to have
459 it fail. A negative return value will be
460 passed on unmodified to the mmap fileop.
461 RELAY_FILE_UNMAP- a relayfs file is being unmapped. The return
464 - void ioctl(rchan_id, cmd, arg)
466 called when an ioctl call is made using a relayfs file
467 descriptor. The cmd and arg are passed along to this
468 callback unmodified for it to do as it wishes with. The
469 return value from this callback is used as the return value
472 If the callbacks param passed to relay_open() is NULL, a set of
473 default do-nothing callbacks will be defined for the channel.
474 Likewise, any NULL rchan_callback function contained in a non-NULL
475 callbacks struct will be filled in with a default callback function
478 - start_reserve - the number of bytes to be reserved at the start of
479 each sub-buffer. The client can do what it wants with this number
480 of bytes when the buffer_start() callback is invoked. Typically
481 clients would use this to write per-sub-buffer header data.
483 - end_reserve - the number of bytes to be reserved at the end of each
484 sub-buffer. The client can do what it wants with this number of
485 bytes when the buffer_end() callback is invoked. Typically clients
486 would use this to write per-sub-buffer footer data.
488 - channel_start_reserve - the number of bytes to be reserved, in
489 addition to start_reserve, at the beginning of the first sub-buffer
490 in the channel. The client can do what it wants with this number of
491 bytes when the buffer_start() callback is invoked. Typically
492 clients would use this to write per-channel header data.
494 - resize_min - if set, this signifies that the channel is
495 auto-resizeable. The value specifies the size that the channel will
496 try to maintain as a normal working size, and that it won't go
497 below. The client makes use of the resizing callbacks and
498 relay_realloc_buffer() and relay_replace_buffer() to actually effect
501 - resize_max - if set, this signifies that the channel is
502 auto-resizeable. The value specifies the maximum size the channel
503 can have as a result of resizing.
505 - mode - if non-zero, specifies the file permissions that will be given
506 to the channel file. If 0, the default rw user perms will be used.
508 - init_buf - if non-NULL, rather than allocating the channel buffer,
509 this buffer will be used as the initial channel buffer. The kernel
510 API function relay_discard_init_buf() can later be used to have
511 relayfs allocate a normal mmappable channel buffer and switch over
512 to using it after copying the init_buf contents into it. Currently,
513 the size of init_buf must be exactly buf_size * n_bufs. The caller
514 is responsible for managing the init_buf memory. This feature is
515 typically used for init-time channel use and should normally be
518 - init_buf_size - the total size of init_buf, if init_buf is specified
519 as non-NULL. Currently, the size of init_buf must be exactly
522 Upon successful completion, relay_open() returns a channel id
523 to be used for all other operations with the relay. All buffers
524 managed by the relay are allocated using rvmalloc/rvfree to allow
525 for easy mmapping to user-space.
528 int relay_write(channel_id, *data_ptr, count, time_delta_offset, **wrote_pos)
530 relay_write() reserves space in the channel and writes count bytes of
531 data pointed to by data_ptr to it. Automatically performs any
532 necessary locking, depending on the scheme and SMP usage in effect (no
533 locking is done for the lockless scheme regardless of usage). It
534 returns the number of bytes written, or 0/negative on failure. If
535 time_delta_offset is >= 0, the internal time delta, the internal time
536 delta calculated when the slot was reserved will be written at that
537 offset. This is the TSC or gettimeofday() delta between the current
538 write and the beginning of the buffer, whichever method is being used
539 by the channel. Trying to write a count larger than the bufsize
540 specified to relay_open() (taking into account the reserved
541 start-of-buffer and end-of-buffer space as well) will fail. If
542 wrote_pos is non-NULL, it will receive the location the data was
543 written to, which may be needed for some applications but is not
544 normally interesting. Most applications should pass in NULL for this
548 struct rchan_reader *add_rchan_reader(int rchan_id, int auto_consume)
550 add_rchan_reader creates and initializes a reader object for a
551 channel. An opaque rchan_reader object is returned on success, and is
552 passed to relay_read() when reading the channel. If the boolean
553 auto_consume parameter is 1, the reader is defined to be
554 auto-consuming. auto-consuming reader objects are automatically
555 created and used for VFS read(2) readers.
558 void remove_rchan_reader(struct rchan_reader *reader)
560 remove_rchan_reader finds and removes the given reader from the
561 channel. This function is used only by non-VFS read(2) readers. VFS
562 read(2) readers are automatically removed when the corresponding file
566 reader add_map_reader(int rchan_id)
568 Creates and initializes an rchan_reader object for channel map
569 readers, and is needed for updating relay_bytes/buffers_consumed()
570 when kernel clients become aware of the need to do so by their mmap
574 int remove_map_reader(reader)
576 Finds and removes the given map reader from the channel. This function
577 is useful only for map readers.
580 int relay_read(reader, buf, count, wait, *actual_read_offset)
582 Reads count bytes from the channel, or as much as is available within
583 the sub-buffer currently being read. The read offset that will be
584 read from is the position contained within the reader object. If the
585 wait flag is set, buf is non-NULL, and there is nothing available, it
586 will wait until there is. If the wait flag is 0 and there is nothing
587 available, -EAGAIN is returned. If buf is NULL, the value returned is
588 the number of bytes that would have been read. actual_read_offset is
589 the value that should be passed as the read offset to
590 relay_bytes_consumed, needed only if the reader is not auto-consuming
591 and the channel is MODE_NO_OVERWRITE, but in any case, it must not be
596 int relay_bytes_avail(reader)
598 Returns the number of bytes available relative to the reader's current
599 read position within the corresponding sub-buffer, 0 if there is
600 nothing available. Note that this doesn't return the total bytes
601 available in the channel buffer - this is enough though to know if
602 anything is available, however, or how many bytes might be returned
606 void relay_buffers_consumed(reader, buffers_consumed)
608 Adds to the channel's consumed buffer count. buffers_consumed should
609 be the number of buffers newly consumed, not the total number
610 consumed. NOTE: kernel clients don't need to call this function if
611 the reader is auto-consuming or the channel is MODE_CONTINUOUS.
613 In order for the relay to detect the 'buffers full' condition for a
614 channel, it must be kept up-to-date with respect to the number of
615 buffers consumed by the client. If the addition of the value of the
616 bufs_consumed param to the current bufs_consumed count for the channel
617 would exceed the bufs_produced count for the channel, the channel's
618 bufs_consumed count will be set to the bufs_produced count for the
619 channel. This allows clients to 'catch up' if necessary.
622 void relay_bytes_consumed(reader, bytes_consumed, read_offset)
624 Adds to the channel's consumed count. bytes_consumed should be the
625 number of bytes actually read e.g. return value of relay_read() and
626 the read_offset should be the actual offset the bytes were read from
627 e.g. the actual_read_offset set by relay_read(). NOTE: kernel clients
628 don't need to call this function if the reader is auto-consuming or
629 the channel is MODE_CONTINUOUS.
631 In order for the relay to detect the 'buffers full' condition for a
632 channel, it must be kept up-to-date with respect to the number of
633 bytes consumed by the client. For packet clients, it makes more sense
634 to update after each read rather than after each complete sub-buffer
635 read. The bytes_consumed count updates bufs_consumed when a buffer
636 has been consumed so this count remains consistent.
639 int relay_info(channel_id, *channel_info)
641 relay_info() fills in an rchan_info struct with channel status and
642 attribute information such as usage modes, sub-buffer size and count,
643 the allocated size of the entire buffer, buffers produced and
644 consumed, current buffer id, count of writes lost due to buffers full
647 The virtual address of the channel buffer is also available here, for
648 those clients that need it.
650 Clients may need to know how many 'unused' bytes there are at the end
651 of a given sub-buffer. This would only be the case if the client 1)
652 didn't either write this count to the end of the sub-buffer or
653 otherwise note it (it's available as the difference between the buffer
654 end and current write pos params in the buffer_end callback) (if the
655 client returned 0 from the buffer_end callback, it's assumed that this
656 is indeed the case) 2) isn't using the read() system call to read the
657 buffer. In other words, if the client isn't annotating the stream and
658 is reading the buffer by mmaping it, this information would be needed
659 in order for the client to 'skip over' the unused bytes at the ends of
662 Additionally, for the lockless scheme, clients may need to know
663 whether a particular sub-buffer is actually complete. An array of
664 boolean values, one per sub-buffer, contains non-zero if the buffer is
665 complete, non-zero otherwise.
668 int relay_close(channel_id)
670 relay_close() is used to close the channel. It finalizes the last
671 sub-buffer (the one currently being written to) and marks the channel
672 as finalized. The channel buffer and channel data structure are then
673 freed automatically when the last reference to the channel is given
677 int relay_realloc_buffer(channel_id, nbufs, async)
679 Allocates a new channel buffer using the specified sub-buffer count
680 (note that resizing can't change sub-buffer sizes). If async is
681 non-zero, the allocation is done in the background using a work queue.
682 When the allocation has completed, the needs_resize() callback is
683 called with a resize_type of RELAY_RESIZE_REPLACE. This function
684 doesn't replace the old buffer with the new - see
685 relay_replace_buffer().
687 This function is called by kernel clients in response to a
688 needs_resize() callback call with a resize type of RELAY_RESIZE_EXPAND
689 or RELAY_RESIZE_SHRINK. That callback also includes a suggested
690 new_bufsize and new_nbufs which should be used when calling this
693 Returns 0 on success, or errcode if the channel is busy or if
694 the allocation couldn't happen for some reason.
696 NOTE: if async is not set, this function should not be called with a
697 lock held, as it may sleep.
700 int relay_replace_buffer(channel_id)
702 Replaces the current channel buffer with the new buffer allocated by
703 relay_realloc_buffer and contained in the channel struct. When the
704 replacement is complete, the needs_resize() callback is called with
705 RELAY_RESIZE_REPLACED. This function is called by kernel clients in
706 response to a needs_resize() callback having a resize type of
707 RELAY_RESIZE_REPLACE.
709 Returns 0 on success, or errcode if the channel is busy or if the
710 replacement or previous allocation didn't happen for some reason.
712 NOTE: This function will not sleep, so can called in any context and
713 with locks held. The client should, however, ensure that the channel
714 isn't actively being read from or written to.
717 int relay_reset(rchan_id)
719 relay_reset() has the effect of erasing all data from the buffer and
720 restarting the channel in its initial state. The buffer itself is not
721 freed, so any mappings are still in effect. NOTE: Care should be
722 taken that the channnel isn't actually being used by anything when
726 int rchan_full(reader)
728 returns 1 if the channel is full with respect to the reader, 0 if not.
731 int rchan_empty(reader)
733 returns 1 if the channel is empty with respect to the reader, 0 if not.
736 int relay_discard_init_buf(rchan_id)
738 allocates an mmappable channel buffer, copies the contents of init_buf
739 into it, and sets the current channel buffer to the newly allocated
740 buffer. This function is used only in conjunction with the init_buf
741 and init_buf_size params to relay_open(), and is typically used when
742 the ability to write into the channel at init-time is needed. The
743 basic usage is to specify an init_buf and init_buf_size to relay_open,
744 then call this function when it's safe to switch over to a normally
745 allocated channel buffer. 'Safe' means that the caller is in a
746 context that can sleep and that nothing is actively writing to the
747 channel. Returns 0 if successful, negative otherwise.
750 Writing directly into the channel
751 =================================
753 Using the relay_write() API function as described above is the
754 preferred means of writing into a channel. In some cases, however,
755 in-kernel clients might want to write directly into a relay channel
756 rather than have relay_write() copy it into the buffer on the client's
757 behalf. Clients wishing to do this should follow the model used to
758 implement relay_write itself. The general sequence is:
760 - get a pointer to the channel via rchan_get(). This increments the
761 channel's reference count.
762 - call relay_lock_channel(). This will perform the proper locking for
763 the channel given the scheme in use and the SMP usage.
764 - reserve a slot in the channel via relay_reserve()
765 - write directly to the reserved address
766 - call relay_commit() to commit the write
767 - call relay_unlock_channel()
768 - call rchan_put() to release the channel reference
770 In particular, clients should make sure they call rchan_get() and
771 rchan_put() and not hold on to references to the channel pointer.
772 Also, forgetting to use relay_lock_channel()/relay_unlock_channel()
773 has no effect if the lockless scheme is being used, but could result
774 in corrupted buffer contents if the locking scheme is used.
780 Writes made via the write() system call are currently limited to 2
781 pages worth of data. There is no such limit on the in-kernel API
782 function relay_write().
784 User applications can currently only mmap the complete buffer (it
785 doesn't really make sense to mmap only part of it, given its purpose).
791 The latest version can be found at:
793 http://www.opersys.com/relayfs
795 Example relayfs clients, such as dynamic printk and the Linux Trace
796 Toolkit, can also be found there.
802 The ideas and specs for relayfs came about as a result of discussions
803 on tracing involving the following:
805 Michel Dagenais <michel.dagenais@polymtl.ca>
806 Richard Moore <richardj_moore@uk.ibm.com>
807 Bob Wisniewski <bob@watson.ibm.com>
808 Karim Yaghmour <karim@opersys.com>
809 Tom Zanussi <zanussi@us.ibm.com>
811 Also thanks to Hubertus Franke for a lot of useful suggestions and bug
812 reports, and for contributing the klog code.