relay.txt 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484
  1. relay interface (formerly relayfs)
  2. ==================================
  3. The relay interface provides a means for kernel applications to
  4. efficiently log and transfer large quantities of data from the kernel
  5. to userspace via user-defined 'relay channels'.
  6. A 'relay channel' is a kernel->user data relay mechanism implemented
  7. as a set of per-cpu kernel buffers ('channel buffers'), each
  8. represented as a regular file ('relay file') in user space. Kernel
  9. clients write into the channel buffers using efficient write
  10. functions; these automatically log into the current cpu's channel
  11. buffer. User space applications mmap() or read() from the relay files
  12. and retrieve the data as it becomes available. The relay files
  13. themselves are files created in a host filesystem, e.g. debugfs, and
  14. are associated with the channel buffers using the API described below.
  15. The format of the data logged into the channel buffers is completely
  16. up to the kernel client; the relay interface does however provide
  17. hooks which allow kernel clients to impose some structure on the
  18. buffer data. The relay interface doesn't implement any form of data
  19. filtering - this also is left to the kernel client. The purpose is to
  20. keep things as simple as possible.
  21. This document provides an overview of the relay interface API. The
  22. details of the function parameters are documented along with the
  23. functions in the relay interface code - please see that for details.
  24. Semantics
  25. =========
  26. Each relay channel has one buffer per CPU, each buffer has one or more
  27. sub-buffers. Messages are written to the first sub-buffer until it is
  28. too full to contain a new message, in which case it it is written to
  29. the next (if available). Messages are never split across sub-buffers.
  30. At this point, userspace can be notified so it empties the first
  31. sub-buffer, while the kernel continues writing to the next.
  32. When notified that a sub-buffer is full, the kernel knows how many
  33. bytes of it are padding i.e. unused space occurring because a complete
  34. message couldn't fit into a sub-buffer. Userspace can use this
  35. knowledge to copy only valid data.
  36. After copying it, userspace can notify the kernel that a sub-buffer
  37. has been consumed.
  38. A relay channel can operate in a mode where it will overwrite data not
  39. yet collected by userspace, and not wait for it to be consumed.
  40. The relay channel itself does not provide for communication of such
  41. data between userspace and kernel, allowing the kernel side to remain
  42. simple and not impose a single interface on userspace. It does
  43. provide a set of examples and a separate helper though, described
  44. below.
  45. The read() interface both removes padding and internally consumes the
  46. read sub-buffers; thus in cases where read(2) is being used to drain
  47. the channel buffers, special-purpose communication between kernel and
  48. user isn't necessary for basic operation.
  49. One of the major goals of the relay interface is to provide a low
  50. overhead mechanism for conveying kernel data to userspace. While the
  51. read() interface is easy to use, it's not as efficient as the mmap()
  52. approach; the example code attempts to make the tradeoff between the
  53. two approaches as small as possible.
  54. klog and relay-apps example code
  55. ================================
  56. The relay interface itself is ready to use, but to make things easier,
  57. a couple simple utility functions and a set of examples are provided.
  58. The relay-apps example tarball, available on the relay sourceforge
  59. site, contains a set of self-contained examples, each consisting of a
  60. pair of .c files containing boilerplate code for each of the user and
  61. kernel sides of a relay application. When combined these two sets of
  62. boilerplate code provide glue to easily stream data to disk, without
  63. having to bother with mundane housekeeping chores.
  64. The 'klog debugging functions' patch (klog.patch in the relay-apps
  65. tarball) provides a couple of high-level logging functions to the
  66. kernel which allow writing formatted text or raw data to a channel,
  67. regardless of whether a channel to write into exists or not, or even
  68. whether the relay interface is compiled into the kernel or not. These
  69. functions allow you to put unconditional 'trace' statements anywhere
  70. in the kernel or kernel modules; only when there is a 'klog handler'
  71. registered will data actually be logged (see the klog and kleak
  72. examples for details).
  73. It is of course possible to use the relay interface from scratch,
  74. i.e. without using any of the relay-apps example code or klog, but
  75. you'll have to implement communication between userspace and kernel,
  76. allowing both to convey the state of buffers (full, empty, amount of
  77. padding). The read() interface both removes padding and internally
  78. consumes the read sub-buffers; thus in cases where read(2) is being
  79. used to drain the channel buffers, special-purpose communication
  80. between kernel and user isn't necessary for basic operation. Things
  81. such as buffer-full conditions would still need to be communicated via
  82. some channel though.
  83. klog and the relay-apps examples can be found in the relay-apps
  84. tarball on http://relayfs.sourceforge.net
  85. The relay interface user space API
  86. ==================================
  87. The relay interface implements basic file operations for user space
  88. access to relay channel buffer data. Here are the file operations
  89. that are available and some comments regarding their behavior:
  90. open() enables user to open an _existing_ channel buffer.
  91. mmap() results in channel buffer being mapped into the caller's
  92. memory space. Note that you can't do a partial mmap - you
  93. must map the entire file, which is NRBUF * SUBBUFSIZE.
  94. read() read the contents of a channel buffer. The bytes read are
  95. 'consumed' by the reader, i.e. they won't be available
  96. again to subsequent reads. If the channel is being used
  97. in no-overwrite mode (the default), it can be read at any
  98. time even if there's an active kernel writer. If the
  99. channel is being used in overwrite mode and there are
  100. active channel writers, results may be unpredictable -
  101. users should make sure that all logging to the channel has
  102. ended before using read() with overwrite mode. Sub-buffer
  103. padding is automatically removed and will not be seen by
  104. the reader.
  105. sendfile() transfer data from a channel buffer to an output file
  106. descriptor. Sub-buffer padding is automatically removed
  107. and will not be seen by the reader.
  108. poll() POLLIN/POLLRDNORM/POLLERR supported. User applications are
  109. notified when sub-buffer boundaries are crossed.
  110. close() decrements the channel buffer's refcount. When the refcount
  111. reaches 0, i.e. when no process or kernel client has the
  112. buffer open, the channel buffer is freed.
  113. In order for a user application to make use of relay files, the
  114. host filesystem must be mounted. For example,
  115. mount -t debugfs debugfs /debug
  116. NOTE: the host filesystem doesn't need to be mounted for kernel
  117. clients to create or use channels - it only needs to be
  118. mounted when user space applications need access to the buffer
  119. data.
  120. The relay interface kernel API
  121. ==============================
  122. Here's a summary of the API the relay interface provides to in-kernel clients:
  123. TBD(curr. line MT:/API/)
  124. channel management functions:
  125. relay_open(base_filename, parent, subbuf_size, n_subbufs,
  126. callbacks, private_data)
  127. relay_close(chan)
  128. relay_flush(chan)
  129. relay_reset(chan)
  130. channel management typically called on instigation of userspace:
  131. relay_subbufs_consumed(chan, cpu, subbufs_consumed)
  132. write functions:
  133. relay_write(chan, data, length)
  134. __relay_write(chan, data, length)
  135. relay_reserve(chan, length)
  136. callbacks:
  137. subbuf_start(buf, subbuf, prev_subbuf, prev_padding)
  138. buf_mapped(buf, filp)
  139. buf_unmapped(buf, filp)
  140. create_buf_file(filename, parent, mode, buf, is_global)
  141. remove_buf_file(dentry)
  142. helper functions:
  143. relay_buf_full(buf)
  144. subbuf_start_reserve(buf, length)
  145. Creating a channel
  146. ------------------
  147. relay_open() is used to create a channel, along with its per-cpu
  148. channel buffers. Each channel buffer will have an associated file
  149. created for it in the host filesystem, which can be and mmapped or
  150. read from in user space. The files are named basename0...basenameN-1
  151. where N is the number of online cpus, and by default will be created
  152. in the root of the filesystem (if the parent param is NULL). If you
  153. want a directory structure to contain your relay files, you should
  154. create it using the host filesystem's directory creation function,
  155. e.g. debugfs_create_dir(), and pass the parent directory to
  156. relay_open(). Users are responsible for cleaning up any directory
  157. structure they create, when the channel is closed - again the host
  158. filesystem's directory removal functions should be used for that,
  159. e.g. debugfs_remove().
  160. In order for a channel to be created and the host filesystem's files
  161. associated with its channel buffers, the user must provide definitions
  162. for two callback functions, create_buf_file() and remove_buf_file().
  163. create_buf_file() is called once for each per-cpu buffer from
  164. relay_open() and allows the user to create the file which will be used
  165. to represent the corresponding channel buffer. The callback should
  166. return the dentry of the file created to represent the channel buffer.
  167. remove_buf_file() must also be defined; it's responsible for deleting
  168. the file(s) created in create_buf_file() and is called during
  169. relay_close().
  170. Here are some typical definitions for these callbacks, in this case
  171. using debugfs:
  172. /*
  173. * create_buf_file() callback. Creates relay file in debugfs.
  174. */
  175. static struct dentry *create_buf_file_handler(const char *filename,
  176. struct dentry *parent,
  177. int mode,
  178. struct rchan_buf *buf,
  179. int *is_global)
  180. {
  181. return debugfs_create_file(filename, mode, parent, buf,
  182. &relay_file_operations);
  183. }
  184. /*
  185. * remove_buf_file() callback. Removes relay file from debugfs.
  186. */
  187. static int remove_buf_file_handler(struct dentry *dentry)
  188. {
  189. debugfs_remove(dentry);
  190. return 0;
  191. }
  192. /*
  193. * relay interface callbacks
  194. */
  195. static struct rchan_callbacks relay_callbacks =
  196. {
  197. .create_buf_file = create_buf_file_handler,
  198. .remove_buf_file = remove_buf_file_handler,
  199. };
  200. And an example relay_open() invocation using them:
  201. chan = relay_open("cpu", NULL, SUBBUF_SIZE, N_SUBBUFS, &relay_callbacks, NULL);
  202. If the create_buf_file() callback fails, or isn't defined, channel
  203. creation and thus relay_open() will fail.
  204. The total size of each per-cpu buffer is calculated by multiplying the
  205. number of sub-buffers by the sub-buffer size passed into relay_open().
  206. The idea behind sub-buffers is that they're basically an extension of
  207. double-buffering to N buffers, and they also allow applications to
  208. easily implement random-access-on-buffer-boundary schemes, which can
  209. be important for some high-volume applications. The number and size
  210. of sub-buffers is completely dependent on the application and even for
  211. the same application, different conditions will warrant different
  212. values for these parameters at different times. Typically, the right
  213. values to use are best decided after some experimentation; in general,
  214. though, it's safe to assume that having only 1 sub-buffer is a bad
  215. idea - you're guaranteed to either overwrite data or lose events
  216. depending on the channel mode being used.
  217. The create_buf_file() implementation can also be defined in such a way
  218. as to allow the creation of a single 'global' buffer instead of the
  219. default per-cpu set. This can be useful for applications interested
  220. mainly in seeing the relative ordering of system-wide events without
  221. the need to bother with saving explicit timestamps for the purpose of
  222. merging/sorting per-cpu files in a postprocessing step.
  223. To have relay_open() create a global buffer, the create_buf_file()
  224. implementation should set the value of the is_global outparam to a
  225. non-zero value in addition to creating the file that will be used to
  226. represent the single buffer. In the case of a global buffer,
  227. create_buf_file() and remove_buf_file() will be called only once. The
  228. normal channel-writing functions, e.g. relay_write(), can still be
  229. used - writes from any cpu will transparently end up in the global
  230. buffer - but since it is a global buffer, callers should make sure
  231. they use the proper locking for such a buffer, either by wrapping
  232. writes in a spinlock, or by copying a write function from relay.h and
  233. creating a local version that internally does the proper locking.
  234. The private_data passed into relay_open() allows clients to associate
  235. user-defined data with a channel, and is immediately available
  236. (including in create_buf_file()) via chan->private_data or
  237. buf->chan->private_data.
  238. Channel 'modes'
  239. ---------------
  240. relay channels can be used in either of two modes - 'overwrite' or
  241. 'no-overwrite'. The mode is entirely determined by the implementation
  242. of the subbuf_start() callback, as described below. The default if no
  243. subbuf_start() callback is defined is 'no-overwrite' mode. If the
  244. default mode suits your needs, and you plan to use the read()
  245. interface to retrieve channel data, you can ignore the details of this
  246. section, as it pertains mainly to mmap() implementations.
  247. In 'overwrite' mode, also known as 'flight recorder' mode, writes
  248. continuously cycle around the buffer and will never fail, but will
  249. unconditionally overwrite old data regardless of whether it's actually
  250. been consumed. In no-overwrite mode, writes will fail, i.e. data will
  251. be lost, if the number of unconsumed sub-buffers equals the total
  252. number of sub-buffers in the channel. It should be clear that if
  253. there is no consumer or if the consumer can't consume sub-buffers fast
  254. enough, data will be lost in either case; the only difference is
  255. whether data is lost from the beginning or the end of a buffer.
  256. As explained above, a relay channel is made of up one or more
  257. per-cpu channel buffers, each implemented as a circular buffer
  258. subdivided into one or more sub-buffers. Messages are written into
  259. the current sub-buffer of the channel's current per-cpu buffer via the
  260. write functions described below. Whenever a message can't fit into
  261. the current sub-buffer, because there's no room left for it, the
  262. client is notified via the subbuf_start() callback that a switch to a
  263. new sub-buffer is about to occur. The client uses this callback to 1)
  264. initialize the next sub-buffer if appropriate 2) finalize the previous
  265. sub-buffer if appropriate and 3) return a boolean value indicating
  266. whether or not to actually move on to the next sub-buffer.
  267. To implement 'no-overwrite' mode, the userspace client would provide
  268. an implementation of the subbuf_start() callback something like the
  269. following:
  270. static int subbuf_start(struct rchan_buf *buf,
  271. void *subbuf,
  272. void *prev_subbuf,
  273. unsigned int prev_padding)
  274. {
  275. if (prev_subbuf)
  276. *((unsigned *)prev_subbuf) = prev_padding;
  277. if (relay_buf_full(buf))
  278. return 0;
  279. subbuf_start_reserve(buf, sizeof(unsigned int));
  280. return 1;
  281. }
  282. If the current buffer is full, i.e. all sub-buffers remain unconsumed,
  283. the callback returns 0 to indicate that the buffer switch should not
  284. occur yet, i.e. until the consumer has had a chance to read the
  285. current set of ready sub-buffers. For the relay_buf_full() function
  286. to make sense, the consumer is reponsible for notifying the relay
  287. interface when sub-buffers have been consumed via
  288. relay_subbufs_consumed(). Any subsequent attempts to write into the
  289. buffer will again invoke the subbuf_start() callback with the same
  290. parameters; only when the consumer has consumed one or more of the
  291. ready sub-buffers will relay_buf_full() return 0, in which case the
  292. buffer switch can continue.
  293. The implementation of the subbuf_start() callback for 'overwrite' mode
  294. would be very similar:
  295. static int subbuf_start(struct rchan_buf *buf,
  296. void *subbuf,
  297. void *prev_subbuf,
  298. unsigned int prev_padding)
  299. {
  300. if (prev_subbuf)
  301. *((unsigned *)prev_subbuf) = prev_padding;
  302. subbuf_start_reserve(buf, sizeof(unsigned int));
  303. return 1;
  304. }
  305. In this case, the relay_buf_full() check is meaningless and the
  306. callback always returns 1, causing the buffer switch to occur
  307. unconditionally. It's also meaningless for the client to use the
  308. relay_subbufs_consumed() function in this mode, as it's never
  309. consulted.
  310. The default subbuf_start() implementation, used if the client doesn't
  311. define any callbacks, or doesn't define the subbuf_start() callback,
  312. implements the simplest possible 'no-overwrite' mode, i.e. it does
  313. nothing but return 0.
  314. Header information can be reserved at the beginning of each sub-buffer
  315. by calling the subbuf_start_reserve() helper function from within the
  316. subbuf_start() callback. This reserved area can be used to store
  317. whatever information the client wants. In the example above, room is
  318. reserved in each sub-buffer to store the padding count for that
  319. sub-buffer. This is filled in for the previous sub-buffer in the
  320. subbuf_start() implementation; the padding value for the previous
  321. sub-buffer is passed into the subbuf_start() callback along with a
  322. pointer to the previous sub-buffer, since the padding value isn't
  323. known until a sub-buffer is filled. The subbuf_start() callback is
  324. also called for the first sub-buffer when the channel is opened, to
  325. give the client a chance to reserve space in it. In this case the
  326. previous sub-buffer pointer passed into the callback will be NULL, so
  327. the client should check the value of the prev_subbuf pointer before
  328. writing into the previous sub-buffer.
  329. Writing to a channel
  330. --------------------
  331. Kernel clients write data into the current cpu's channel buffer using
  332. relay_write() or __relay_write(). relay_write() is the main logging
  333. function - it uses local_irqsave() to protect the buffer and should be
  334. used if you might be logging from interrupt context. If you know
  335. you'll never be logging from interrupt context, you can use
  336. __relay_write(), which only disables preemption. These functions
  337. don't return a value, so you can't determine whether or not they
  338. failed - the assumption is that you wouldn't want to check a return
  339. value in the fast logging path anyway, and that they'll always succeed
  340. unless the buffer is full and no-overwrite mode is being used, in
  341. which case you can detect a failed write in the subbuf_start()
  342. callback by calling the relay_buf_full() helper function.
  343. relay_reserve() is used to reserve a slot in a channel buffer which
  344. can be written to later. This would typically be used in applications
  345. that need to write directly into a channel buffer without having to
  346. stage data in a temporary buffer beforehand. Because the actual write
  347. may not happen immediately after the slot is reserved, applications
  348. using relay_reserve() can keep a count of the number of bytes actually
  349. written, either in space reserved in the sub-buffers themselves or as
  350. a separate array. See the 'reserve' example in the relay-apps tarball
  351. at http://relayfs.sourceforge.net for an example of how this can be
  352. done. Because the write is under control of the client and is
  353. separated from the reserve, relay_reserve() doesn't protect the buffer
  354. at all - it's up to the client to provide the appropriate
  355. synchronization when using relay_reserve().
  356. Closing a channel
  357. -----------------
  358. The client calls relay_close() when it's finished using the channel.
  359. The channel and its associated buffers are destroyed when there are no
  360. longer any references to any of the channel buffers. relay_flush()
  361. forces a sub-buffer switch on all the channel buffers, and can be used
  362. to finalize and process the last sub-buffers before the channel is
  363. closed.
  364. Misc
  365. ----
  366. Some applications may want to keep a channel around and re-use it
  367. rather than open and close a new channel for each use. relay_reset()
  368. can be used for this purpose - it resets a channel to its initial
  369. state without reallocating channel buffer memory or destroying
  370. existing mappings. It should however only be called when it's safe to
  371. do so, i.e. when the channel isn't currently being written to.
  372. Finally, there are a couple of utility callbacks that can be used for
  373. different purposes. buf_mapped() is called whenever a channel buffer
  374. is mmapped from user space and buf_unmapped() is called when it's
  375. unmapped. The client can use this notification to trigger actions
  376. within the kernel application, such as enabling/disabling logging to
  377. the channel.
  378. Resources
  379. =========
  380. For news, example code, mailing list, etc. see the relay interface homepage:
  381. http://relayfs.sourceforge.net
  382. Credits
  383. =======
  384. The ideas and specs for the relay interface came about as a result of
  385. discussions on tracing involving the following:
  386. Michel Dagenais <michel.dagenais@polymtl.ca>
  387. Richard Moore <richardj_moore@uk.ibm.com>
  388. Bob Wisniewski <bob@watson.ibm.com>
  389. Karim Yaghmour <karim@opersys.com>
  390. Tom Zanussi <zanussi@us.ibm.com>
  391. Also thanks to Hubertus Franke for a lot of useful suggestions and bug
  392. reports.