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 and potentially sustained streams
7 of data from kernel space to user space.
9 The main abstraction of relayfs is the 'channel'. A channel consists
10 of a set of per-cpu kernel buffers each represented by a file in the
11 relayfs filesystem. Kernel clients write into a channel using
12 efficient write functions which automatically log to the current cpu's
13 channel buffer. User space applications mmap() the per-cpu files and
14 retrieve the data as it becomes available.
16 The format of the data logged into the channel buffers is completely
17 up to the relayfs client; relayfs does however provide hooks which
18 allow clients to impose some structure on the buffer data. Nor does
19 relayfs implement any form of data filtering - this also is left to
20 the client. The purpose is to keep relayfs as simple as possible.
22 This document provides an overview of the relayfs API. The details of
23 the function parameters are documented along with the functions in the
24 filesystem code - please see that for details.
29 Each relayfs channel has one buffer per CPU, each buffer has one or
30 more sub-buffers. Messages are written to the first sub-buffer until
31 it is too full to contain a new message, in which case it it is
32 written to the next (if available). Messages are never split across
33 sub-buffers. At this point, userspace can be notified so it empties
34 the first sub-buffer, while the kernel continues writing to the next.
36 When notified that a sub-buffer is full, the kernel knows how many
37 bytes of it are padding i.e. unused. Userspace can use this knowledge
38 to copy only valid data.
40 After copying it, userspace can notify the kernel that a sub-buffer
43 relayfs can operate in a mode where it will overwrite data not yet
44 collected by userspace, and not wait for it to consume it.
46 relayfs itself does not provide for communication of such data between
47 userspace and kernel, allowing the kernel side to remain simple and not
48 impose a single interface on userspace. It does provide a separate
49 helper though, described below.
51 klog, relay-app & librelay
52 ==========================
54 relayfs itself is ready to use, but to make things easier, two
55 additional systems are provided. klog is a simple wrapper to make
56 writing formatted text or raw data to a channel simpler, regardless of
57 whether a channel to write into exists or not, or whether relayfs is
58 compiled into the kernel or is configured as a module. relay-app is
59 the kernel counterpart of userspace librelay.c, combined these two
60 files provide glue to easily stream data to disk, without having to
61 bother with housekeeping. klog and relay-app can be used together,
62 with klog providing high-level logging functions to the kernel and
63 relay-app taking care of kernel-user control and disk-logging chores.
65 It is possible to use relayfs without relay-app & librelay, but you'll
66 have to implement communication between userspace and kernel, allowing
67 both to convey the state of buffers (full, empty, amount of padding).
69 klog, relay-app and librelay can be found in the relay-apps tarball on
70 http://relayfs.sourceforge.net
72 The relayfs user space API
73 ==========================
75 relayfs implements basic file operations for user space access to
76 relayfs channel buffer data. Here are the file operations that are
77 available and some comments regarding their behavior:
79 open() enables user to open an _existing_ buffer.
81 mmap() results in channel buffer being mapped into the caller's
82 memory space. Note that you can't do a partial mmap - you must
83 map the entire file, which is NRBUF * SUBBUFSIZE.
85 read() read the contents of a channel buffer. The bytes read are
86 'consumed' by the reader i.e. they won't be available again
87 to subsequent reads. If the channel is being used in
88 no-overwrite mode (the default), it can be read at any time
89 even if there's an active kernel writer. If the channel is
90 being used in overwrite mode and there are active channel
91 writers, results may be unpredictable - users should make
92 sure that all logging to the channel has ended before using
93 read() with overwrite mode.
95 poll() POLLIN/POLLRDNORM/POLLERR supported. User applications are
96 notified when sub-buffer boundaries are crossed.
98 close() decrements the channel buffer's refcount. When the refcount
99 reaches 0 i.e. when no process or kernel client has the buffer
100 open, the channel buffer is freed.
103 In order for a user application to make use of relayfs files, the
104 relayfs filesystem must be mounted. For example,
106 mount -t relayfs relayfs /mnt/relay
108 NOTE: relayfs doesn't need to be mounted for kernel clients to create
109 or use channels - it only needs to be mounted when user space
110 applications need access to the buffer data.
113 The relayfs kernel API
114 ======================
116 Here's a summary of the API relayfs provides to in-kernel clients:
119 channel management functions:
121 relay_open(base_filename, parent, subbuf_size, n_subbufs,
126 relayfs_create_dir(name, parent)
127 relayfs_remove_dir(dentry)
128 relayfs_create_file(name, parent, mode, fops, data)
129 relayfs_remove_file(dentry)
131 channel management typically called on instigation of userspace:
133 relay_subbufs_consumed(chan, cpu, subbufs_consumed)
137 relay_write(chan, data, length)
138 __relay_write(chan, data, length)
139 relay_reserve(chan, length)
143 subbuf_start(buf, subbuf, prev_subbuf, prev_padding)
144 buf_mapped(buf, filp)
145 buf_unmapped(buf, filp)
150 subbuf_start_reserve(buf, length)
156 relay_open() is used to create a channel, along with its per-cpu
157 channel buffers. Each channel buffer will have an associated file
158 created for it in the relayfs filesystem, which can be opened and
159 mmapped from user space if desired. The files are named
160 basename0...basenameN-1 where N is the number of online cpus, and by
161 default will be created in the root of the filesystem. If you want a
162 directory structure to contain your relayfs files, you can create it
163 with relayfs_create_dir() and pass the parent directory to
164 relay_open(). Clients are responsible for cleaning up any directory
165 structure they create when the channel is closed - use
166 relayfs_remove_dir() for that.
168 The total size of each per-cpu buffer is calculated by multiplying the
169 number of sub-buffers by the sub-buffer size passed into relay_open().
170 The idea behind sub-buffers is that they're basically an extension of
171 double-buffering to N buffers, and they also allow applications to
172 easily implement random-access-on-buffer-boundary schemes, which can
173 be important for some high-volume applications. The number and size
174 of sub-buffers is completely dependent on the application and even for
175 the same application, different conditions will warrant different
176 values for these parameters at different times. Typically, the right
177 values to use are best decided after some experimentation; in general,
178 though, it's safe to assume that having only 1 sub-buffer is a bad
179 idea - you're guaranteed to either overwrite data or lose events
180 depending on the channel mode being used.
185 relayfs channels can be used in either of two modes - 'overwrite' or
186 'no-overwrite'. The mode is entirely determined by the implementation
187 of the subbuf_start() callback, as described below. In 'overwrite'
188 mode, also known as 'flight recorder' mode, writes continuously cycle
189 around the buffer and will never fail, but will unconditionally
190 overwrite old data regardless of whether it's actually been consumed.
191 In no-overwrite mode, writes will fail i.e. data will be lost, if the
192 number of unconsumed sub-buffers equals the total number of
193 sub-buffers in the channel. It should be clear that if there is no
194 consumer or if the consumer can't consume sub-buffers fast enought,
195 data will be lost in either case; the only difference is whether data
196 is lost from the beginning or the end of a buffer.
198 As explained above, a relayfs channel is made of up one or more
199 per-cpu channel buffers, each implemented as a circular buffer
200 subdivided into one or more sub-buffers. Messages are written into
201 the current sub-buffer of the channel's current per-cpu buffer via the
202 write functions described below. Whenever a message can't fit into
203 the current sub-buffer, because there's no room left for it, the
204 client is notified via the subbuf_start() callback that a switch to a
205 new sub-buffer is about to occur. The client uses this callback to 1)
206 initialize the next sub-buffer if appropriate 2) finalize the previous
207 sub-buffer if appropriate and 3) return a boolean value indicating
208 whether or not to actually go ahead with the sub-buffer switch.
210 To implement 'no-overwrite' mode, the userspace client would provide
211 an implementation of the subbuf_start() callback something like the
214 static int subbuf_start(struct rchan_buf *buf,
217 unsigned int prev_padding)
220 *((unsigned *)prev_subbuf) = prev_padding;
222 if (relay_buf_full(buf))
225 subbuf_start_reserve(buf, sizeof(unsigned int));
230 If the current buffer is full i.e. all sub-buffers remain unconsumed,
231 the callback returns 0 to indicate that the buffer switch should not
232 occur yet i.e. until the consumer has had a chance to read the current
233 set of ready sub-buffers. For the relay_buf_full() function to make
234 sense, the consumer is reponsible for notifying relayfs when
235 sub-buffers have been consumed via relay_subbufs_consumed(). Any
236 subsequent attempts to write into the buffer will again invoke the
237 subbuf_start() callback with the same parameters; only when the
238 consumer has consumed one or more of the ready sub-buffers will
239 relay_buf_full() return 0, in which case the buffer switch can
242 The implementation of the subbuf_start() callback for 'overwrite' mode
243 would be very similar:
245 static int subbuf_start(struct rchan_buf *buf,
248 unsigned int prev_padding)
251 *((unsigned *)prev_subbuf) = prev_padding;
253 subbuf_start_reserve(buf, sizeof(unsigned int));
258 In this case, the relay_buf_full() check is meaningless and the
259 callback always returns 1, causing the buffer switch to occur
260 unconditionally. It's also meaningless for the client to use the
261 relay_subbufs_consumed() function in this mode, as it's never
264 The default subbuf_start() implementation, used if the client doesn't
265 define any callbacks, or doesn't define the subbuf_start() callback,
266 implements the simplest possible 'no-overwrite' mode i.e. it does
267 nothing but return 0.
269 Header information can be reserved at the beginning of each sub-buffer
270 by calling the subbuf_start_reserve() helper function from within the
271 subbuf_start() callback. This reserved area can be used to store
272 whatever information the client wants. In the example above, room is
273 reserved in each sub-buffer to store the padding count for that
274 sub-buffer. This is filled in for the previous sub-buffer in the
275 subbuf_start() implementation; the padding value for the previous
276 sub-buffer is passed into the subbuf_start() callback along with a
277 pointer to the previous sub-buffer, since the padding value isn't
278 known until a sub-buffer is filled. The subbuf_start() callback is
279 also called for the first sub-buffer when the channel is opened, to
280 give the client a chance to reserve space in it. In this case the
281 previous sub-buffer pointer passed into the callback will be NULL, so
282 the client should check the value of the prev_subbuf pointer before
283 writing into the previous sub-buffer.
288 kernel clients write data into the current cpu's channel buffer using
289 relay_write() or __relay_write(). relay_write() is the main logging
290 function - it uses local_irqsave() to protect the buffer and should be
291 used if you might be logging from interrupt context. If you know
292 you'll never be logging from interrupt context, you can use
293 __relay_write(), which only disables preemption. These functions
294 don't return a value, so you can't determine whether or not they
295 failed - the assumption is that you wouldn't want to check a return
296 value in the fast logging path anyway, and that they'll always succeed
297 unless the buffer is full and no-overwrite mode is being used, in
298 which case you can detect a failed write in the subbuf_start()
299 callback by calling the relay_buf_full() helper function.
301 relay_reserve() is used to reserve a slot in a channel buffer which
302 can be written to later. This would typically be used in applications
303 that need to write directly into a channel buffer without having to
304 stage data in a temporary buffer beforehand. Because the actual write
305 may not happen immediately after the slot is reserved, applications
306 using relay_reserve() can keep a count of the number of bytes actually
307 written, either in space reserved in the sub-buffers themselves or as
308 a separate array. See the 'reserve' example in the relay-apps tarball
309 at http://relayfs.sourceforge.net for an example of how this can be
310 done. Because the write is under control of the client and is
311 separated from the reserve, relay_reserve() doesn't protect the buffer
312 at all - it's up to the client to provide the appropriate
313 synchronization when using relay_reserve().
318 The client calls relay_close() when it's finished using the channel.
319 The channel and its associated buffers are destroyed when there are no
320 longer any references to any of the channel buffers. relay_flush()
321 forces a sub-buffer switch on all the channel buffers, and can be used
322 to finalize and process the last sub-buffers before the channel is
325 Creating non-relay files
326 ------------------------
328 relay_open() automatically creates files in the relayfs filesystem to
329 represent the per-cpu kernel buffers; it's often useful for
330 applications to be able to create their own files alongside the relay
331 files in the relayfs filesystem as well e.g. 'control' files much like
332 those created in /proc or debugfs for similar purposes, used to
333 communicate control information between the kernel and user sides of a
334 relayfs application. For this purpose the relayfs_create_file() and
335 relayfs_remove_file() API functions exist. For relayfs_create_file(),
336 the caller passes in a set of user-defined file operations to be used
337 for the file and an optional void * to a user-specified data item,
338 which will be accessible via inode->u.generic_ip (see the relay-apps
339 tarball for examples). The file_operations are a required parameter
340 to relayfs_create_file() and thus the semantics of these files are
341 completely defined by the caller.
343 See the relay-apps tarball at http://relayfs.sourceforge.net for
344 examples of how these non-relay files are meant to be used.
349 Some applications may want to keep a channel around and re-use it
350 rather than open and close a new channel for each use. relay_reset()
351 can be used for this purpose - it resets a channel to its initial
352 state without reallocating channel buffer memory or destroying
353 existing mappings. It should however only be called when it's safe to
354 do so i.e. when the channel isn't currently being written to.
356 Finally, there are a couple of utility callbacks that can be used for
357 different purposes. buf_mapped() is called whenever a channel buffer
358 is mmapped from user space and buf_unmapped() is called when it's
359 unmapped. The client can use this notification to trigger actions
360 within the kernel application, such as enabling/disabling logging to
367 For news, example code, mailing list, etc. see the relayfs homepage:
369 http://relayfs.sourceforge.net
375 The ideas and specs for relayfs came about as a result of discussions
376 on tracing involving the following:
378 Michel Dagenais <michel.dagenais@polymtl.ca>
379 Richard Moore <richardj_moore@uk.ibm.com>
380 Bob Wisniewski <bob@watson.ibm.com>
381 Karim Yaghmour <karim@opersys.com>
382 Tom Zanussi <zanussi@us.ibm.com>
384 Also thanks to Hubertus Franke for a lot of useful suggestions and bug