Commit | Line | Data |
---|---|---|
b7cdc182 | 1 | /* SPDX-License-Identifier: (GPL-2.0-only OR LGPL-2.1-only) |
9f36eaed | 2 | * |
24591303 | 3 | * ringbuffer/backend.h |
f3bc08c5 MD |
4 | * |
5 | * Ring buffer backend (API). | |
6 | * | |
886d51a3 MD |
7 | * Copyright (C) 2010-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
8 | * | |
f3bc08c5 MD |
9 | * Credits to Steven Rostedt for proposing to use an extra-subbuffer owned by |
10 | * the reader in flight recorder mode. | |
11 | */ | |
12 | ||
9f36eaed MJ |
13 | #ifndef _LIB_RING_BUFFER_BACKEND_H |
14 | #define _LIB_RING_BUFFER_BACKEND_H | |
15 | ||
f3bc08c5 MD |
16 | #include <linux/types.h> |
17 | #include <linux/sched.h> | |
18 | #include <linux/timer.h> | |
19 | #include <linux/wait.h> | |
20 | #include <linux/poll.h> | |
21 | #include <linux/list.h> | |
22 | #include <linux/fs.h> | |
23 | #include <linux/mm.h> | |
80bb2600 | 24 | #include <wrapper/uaccess.h> |
f3bc08c5 MD |
25 | |
26 | /* Internal helpers */ | |
24591303 MD |
27 | #include <ringbuffer/backend_internal.h> |
28 | #include <ringbuffer/frontend_internal.h> | |
f3bc08c5 MD |
29 | |
30 | /* Ring buffer backend API */ | |
31 | ||
32 | /* Ring buffer backend access (read/write) */ | |
33 | ||
34 | extern size_t lib_ring_buffer_read(struct lib_ring_buffer_backend *bufb, | |
35 | size_t offset, void *dest, size_t len); | |
36 | ||
37 | extern int __lib_ring_buffer_copy_to_user(struct lib_ring_buffer_backend *bufb, | |
38 | size_t offset, void __user *dest, | |
39 | size_t len); | |
40 | ||
41 | extern int lib_ring_buffer_read_cstr(struct lib_ring_buffer_backend *bufb, | |
42 | size_t offset, void *dest, size_t len); | |
43 | ||
0112cb7b MD |
44 | extern unsigned long * |
45 | lib_ring_buffer_read_get_pfn(struct lib_ring_buffer_backend *bufb, size_t offset, | |
f3bc08c5 MD |
46 | void ***virt); |
47 | ||
48 | /* | |
49 | * Return the address where a given offset is located. | |
50 | * Should be used to get the current subbuffer header pointer. Given we know | |
51 | * it's never on a page boundary, it's safe to write directly to this address, | |
52 | * as long as the write is never bigger than a page size. | |
53 | */ | |
54 | extern void * | |
55 | lib_ring_buffer_offset_address(struct lib_ring_buffer_backend *bufb, | |
56 | size_t offset); | |
57 | extern void * | |
58 | lib_ring_buffer_read_offset_address(struct lib_ring_buffer_backend *bufb, | |
59 | size_t offset); | |
60 | ||
61 | /** | |
62 | * lib_ring_buffer_write - write data to a buffer backend | |
63 | * @config : ring buffer instance configuration | |
64 | * @ctx: ring buffer context. (input arguments only) | |
65 | * @src : source pointer to copy from | |
66 | * @len : length of data to copy | |
67 | * | |
68 | * This function copies "len" bytes of data from a source pointer to a buffer | |
69 | * backend, at the current context offset. This is more or less a buffer | |
70 | * backend-specific memcpy() operation. Calls the slow path (_ring_buffer_write) | |
71 | * if copy is crossing a page boundary. | |
72 | */ | |
8617eb9a | 73 | static inline __attribute__((always_inline)) |
f3bc08c5 MD |
74 | void lib_ring_buffer_write(const struct lib_ring_buffer_config *config, |
75 | struct lib_ring_buffer_ctx *ctx, | |
76 | const void *src, size_t len) | |
77 | { | |
b1199bd3 MD |
78 | struct lib_ring_buffer_backend *bufb = &ctx->priv.buf->backend; |
79 | struct channel_backend *chanb = &ctx->priv.chan->backend; | |
85a07c33 | 80 | size_t index, pagecpy; |
b1199bd3 | 81 | size_t offset = ctx->priv.buf_offset; |
85a07c33 | 82 | struct lib_ring_buffer_backend_pages *backend_pages; |
f3bc08c5 | 83 | |
61eb4c39 MD |
84 | if (unlikely(!len)) |
85 | return; | |
85a07c33 MD |
86 | backend_pages = |
87 | lib_ring_buffer_get_backend_pages_from_ctx(config, ctx); | |
f3bc08c5 | 88 | offset &= chanb->buf_size - 1; |
f3bc08c5 MD |
89 | index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT; |
90 | pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK); | |
f3bc08c5 MD |
91 | if (likely(pagecpy == len)) |
92 | lib_ring_buffer_do_copy(config, | |
85a07c33 | 93 | backend_pages->p[index].virt |
f3bc08c5 MD |
94 | + (offset & ~PAGE_MASK), |
95 | src, len); | |
96 | else | |
97 | _lib_ring_buffer_write(bufb, offset, src, len, 0); | |
b1199bd3 | 98 | ctx->priv.buf_offset += len; |
f3bc08c5 MD |
99 | } |
100 | ||
4ea00e4f JD |
101 | /** |
102 | * lib_ring_buffer_memset - write len bytes of c to a buffer backend | |
103 | * @config : ring buffer instance configuration | |
104 | * @bufb : ring buffer backend | |
105 | * @offset : offset within the buffer | |
106 | * @c : the byte to copy | |
107 | * @len : number of bytes to copy | |
108 | * | |
109 | * This function writes "len" bytes of "c" to a buffer backend, at a specific | |
110 | * offset. This is more or less a buffer backend-specific memset() operation. | |
111 | * Calls the slow path (_ring_buffer_memset) if write is crossing a page | |
112 | * boundary. | |
113 | */ | |
114 | static inline | |
115 | void lib_ring_buffer_memset(const struct lib_ring_buffer_config *config, | |
116 | struct lib_ring_buffer_ctx *ctx, int c, size_t len) | |
117 | { | |
118 | ||
b1199bd3 MD |
119 | struct lib_ring_buffer_backend *bufb = &ctx->priv.buf->backend; |
120 | struct channel_backend *chanb = &ctx->priv.chan->backend; | |
85a07c33 | 121 | size_t index, pagecpy; |
b1199bd3 | 122 | size_t offset = ctx->priv.buf_offset; |
85a07c33 | 123 | struct lib_ring_buffer_backend_pages *backend_pages; |
4ea00e4f | 124 | |
61eb4c39 MD |
125 | if (unlikely(!len)) |
126 | return; | |
85a07c33 MD |
127 | backend_pages = |
128 | lib_ring_buffer_get_backend_pages_from_ctx(config, ctx); | |
4ea00e4f | 129 | offset &= chanb->buf_size - 1; |
4ea00e4f JD |
130 | index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT; |
131 | pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK); | |
4ea00e4f | 132 | if (likely(pagecpy == len)) |
85a07c33 | 133 | lib_ring_buffer_do_memset(backend_pages->p[index].virt |
4ea00e4f JD |
134 | + (offset & ~PAGE_MASK), |
135 | c, len); | |
136 | else | |
137 | _lib_ring_buffer_memset(bufb, offset, c, len, 0); | |
b1199bd3 | 138 | ctx->priv.buf_offset += len; |
4ea00e4f JD |
139 | } |
140 | ||
16f78f3a MD |
141 | /* |
142 | * Copy up to @len string bytes from @src to @dest. Stop whenever a NULL | |
143 | * terminating character is found in @src. Returns the number of bytes | |
144 | * copied. Does *not* terminate @dest with NULL terminating character. | |
145 | */ | |
d723e905 | 146 | static inline __attribute__((always_inline)) |
16f78f3a MD |
147 | size_t lib_ring_buffer_do_strcpy(const struct lib_ring_buffer_config *config, |
148 | char *dest, const char *src, size_t len) | |
149 | { | |
150 | size_t count; | |
151 | ||
152 | for (count = 0; count < len; count++) { | |
153 | char c; | |
154 | ||
155 | /* | |
156 | * Only read source character once, in case it is | |
157 | * modified concurrently. | |
158 | */ | |
585e5dcc | 159 | c = LTTNG_READ_ONCE(src[count]); |
16f78f3a MD |
160 | if (!c) |
161 | break; | |
162 | lib_ring_buffer_do_copy(config, &dest[count], &c, 1); | |
163 | } | |
164 | return count; | |
165 | } | |
166 | ||
167 | /* | |
168 | * Copy up to @len string bytes from @src to @dest. Stop whenever a NULL | |
169 | * terminating character is found in @src, or when a fault occurs. | |
170 | * Returns the number of bytes copied. Does *not* terminate @dest with | |
171 | * NULL terminating character. | |
172 | * | |
173 | * This function deals with userspace pointers, it should never be called | |
174 | * directly without having the src pointer checked with access_ok() | |
175 | * previously. | |
176 | */ | |
d723e905 | 177 | static inline __attribute__((always_inline)) |
16f78f3a MD |
178 | size_t lib_ring_buffer_do_strcpy_from_user_inatomic(const struct lib_ring_buffer_config *config, |
179 | char *dest, const char __user *src, size_t len) | |
180 | { | |
181 | size_t count; | |
182 | ||
183 | for (count = 0; count < len; count++) { | |
184 | int ret; | |
185 | char c; | |
186 | ||
f127e61e | 187 | ret = __copy_from_user_inatomic(&c, src + count, 1); |
16f78f3a MD |
188 | if (ret || !c) |
189 | break; | |
190 | lib_ring_buffer_do_copy(config, &dest[count], &c, 1); | |
191 | } | |
192 | return count; | |
193 | } | |
194 | ||
195 | /** | |
196 | * lib_ring_buffer_strcpy - write string data to a buffer backend | |
197 | * @config : ring buffer instance configuration | |
198 | * @ctx: ring buffer context. (input arguments only) | |
199 | * @src : source pointer to copy from | |
200 | * @len : length of data to copy | |
201 | * @pad : character to use for padding | |
202 | * | |
203 | * This function copies @len - 1 bytes of string data from a source | |
204 | * pointer to a buffer backend, followed by a terminating '\0' | |
205 | * character, at the current context offset. This is more or less a | |
206 | * buffer backend-specific strncpy() operation. If a terminating '\0' | |
207 | * character is found in @src before @len - 1 characters are copied, pad | |
208 | * the buffer with @pad characters (e.g. '#'). Calls the slow path | |
209 | * (_ring_buffer_strcpy) if copy is crossing a page boundary. | |
210 | */ | |
211 | static inline | |
212 | void lib_ring_buffer_strcpy(const struct lib_ring_buffer_config *config, | |
213 | struct lib_ring_buffer_ctx *ctx, | |
214 | const char *src, size_t len, int pad) | |
215 | { | |
b1199bd3 MD |
216 | struct lib_ring_buffer_backend *bufb = &ctx->priv.buf->backend; |
217 | struct channel_backend *chanb = &ctx->priv.chan->backend; | |
85a07c33 | 218 | size_t index, pagecpy; |
b1199bd3 | 219 | size_t offset = ctx->priv.buf_offset; |
85a07c33 | 220 | struct lib_ring_buffer_backend_pages *backend_pages; |
16f78f3a MD |
221 | |
222 | if (unlikely(!len)) | |
223 | return; | |
85a07c33 MD |
224 | backend_pages = |
225 | lib_ring_buffer_get_backend_pages_from_ctx(config, ctx); | |
16f78f3a | 226 | offset &= chanb->buf_size - 1; |
16f78f3a MD |
227 | index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT; |
228 | pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK); | |
16f78f3a MD |
229 | if (likely(pagecpy == len)) { |
230 | size_t count; | |
231 | ||
232 | count = lib_ring_buffer_do_strcpy(config, | |
85a07c33 | 233 | backend_pages->p[index].virt |
16f78f3a MD |
234 | + (offset & ~PAGE_MASK), |
235 | src, len - 1); | |
236 | offset += count; | |
237 | /* Padding */ | |
238 | if (unlikely(count < len - 1)) { | |
239 | size_t pad_len = len - 1 - count; | |
240 | ||
85a07c33 | 241 | lib_ring_buffer_do_memset(backend_pages->p[index].virt |
16f78f3a MD |
242 | + (offset & ~PAGE_MASK), |
243 | pad, pad_len); | |
244 | offset += pad_len; | |
245 | } | |
246 | /* Ending '\0' */ | |
85a07c33 | 247 | lib_ring_buffer_do_memset(backend_pages->p[index].virt |
16f78f3a MD |
248 | + (offset & ~PAGE_MASK), |
249 | '\0', 1); | |
250 | } else { | |
251 | _lib_ring_buffer_strcpy(bufb, offset, src, len, 0, pad); | |
252 | } | |
b1199bd3 | 253 | ctx->priv.buf_offset += len; |
16f78f3a MD |
254 | } |
255 | ||
4ea00e4f | 256 | /** |
7b8ea3a5 | 257 | * lib_ring_buffer_copy_from_user_inatomic - write userspace data to a buffer backend |
4ea00e4f JD |
258 | * @config : ring buffer instance configuration |
259 | * @ctx: ring buffer context. (input arguments only) | |
260 | * @src : userspace source pointer to copy from | |
261 | * @len : length of data to copy | |
262 | * | |
263 | * This function copies "len" bytes of data from a userspace pointer to a | |
264 | * buffer backend, at the current context offset. This is more or less a buffer | |
265 | * backend-specific memcpy() operation. Calls the slow path | |
7b8ea3a5 MD |
266 | * (_ring_buffer_write_from_user_inatomic) if copy is crossing a page boundary. |
267 | * Disable the page fault handler to ensure we never try to take the mmap_sem. | |
4ea00e4f | 268 | */ |
d723e905 | 269 | static inline __attribute__((always_inline)) |
7b8ea3a5 | 270 | void lib_ring_buffer_copy_from_user_inatomic(const struct lib_ring_buffer_config *config, |
4ea00e4f JD |
271 | struct lib_ring_buffer_ctx *ctx, |
272 | const void __user *src, size_t len) | |
273 | { | |
b1199bd3 MD |
274 | struct lib_ring_buffer_backend *bufb = &ctx->priv.buf->backend; |
275 | struct channel_backend *chanb = &ctx->priv.chan->backend; | |
85a07c33 | 276 | size_t index, pagecpy; |
b1199bd3 | 277 | size_t offset = ctx->priv.buf_offset; |
85a07c33 | 278 | struct lib_ring_buffer_backend_pages *backend_pages; |
4ea00e4f JD |
279 | unsigned long ret; |
280 | ||
61eb4c39 MD |
281 | if (unlikely(!len)) |
282 | return; | |
85a07c33 MD |
283 | backend_pages = |
284 | lib_ring_buffer_get_backend_pages_from_ctx(config, ctx); | |
4ea00e4f | 285 | offset &= chanb->buf_size - 1; |
4ea00e4f JD |
286 | index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT; |
287 | pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK); | |
4ea00e4f | 288 | |
7b8ea3a5 | 289 | pagefault_disable(); |
80bb2600 | 290 | if (unlikely(!lttng_access_ok(VERIFY_READ, src, len))) |
4ea00e4f JD |
291 | goto fill_buffer; |
292 | ||
293 | if (likely(pagecpy == len)) { | |
7b8ea3a5 | 294 | ret = lib_ring_buffer_do_copy_from_user_inatomic( |
85a07c33 | 295 | backend_pages->p[index].virt + (offset & ~PAGE_MASK), |
4ea00e4f JD |
296 | src, len); |
297 | if (unlikely(ret > 0)) { | |
d87a9f03 | 298 | /* Copy failed. */ |
4ea00e4f JD |
299 | goto fill_buffer; |
300 | } | |
301 | } else { | |
7b8ea3a5 | 302 | _lib_ring_buffer_copy_from_user_inatomic(bufb, offset, src, len, 0); |
4ea00e4f | 303 | } |
7b8ea3a5 | 304 | pagefault_enable(); |
b1199bd3 | 305 | ctx->priv.buf_offset += len; |
4ea00e4f JD |
306 | |
307 | return; | |
308 | ||
309 | fill_buffer: | |
7b8ea3a5 | 310 | pagefault_enable(); |
4ea00e4f JD |
311 | /* |
312 | * In the error path we call the slow path version to avoid | |
313 | * the pollution of static inline code. | |
314 | */ | |
315 | _lib_ring_buffer_memset(bufb, offset, 0, len, 0); | |
316 | } | |
317 | ||
16f78f3a MD |
318 | /** |
319 | * lib_ring_buffer_strcpy_from_user_inatomic - write userspace string data to a buffer backend | |
320 | * @config : ring buffer instance configuration | |
321 | * @ctx: ring buffer context (input arguments only) | |
322 | * @src : userspace source pointer to copy from | |
323 | * @len : length of data to copy | |
324 | * @pad : character to use for padding | |
325 | * | |
326 | * This function copies @len - 1 bytes of string data from a userspace | |
327 | * source pointer to a buffer backend, followed by a terminating '\0' | |
328 | * character, at the current context offset. This is more or less a | |
329 | * buffer backend-specific strncpy() operation. If a terminating '\0' | |
330 | * character is found in @src before @len - 1 characters are copied, pad | |
331 | * the buffer with @pad characters (e.g. '#'). Calls the slow path | |
332 | * (_ring_buffer_strcpy_from_user_inatomic) if copy is crossing a page | |
333 | * boundary. Disable the page fault handler to ensure we never try to | |
334 | * take the mmap_sem. | |
335 | */ | |
336 | static inline | |
337 | void lib_ring_buffer_strcpy_from_user_inatomic(const struct lib_ring_buffer_config *config, | |
338 | struct lib_ring_buffer_ctx *ctx, | |
339 | const void __user *src, size_t len, int pad) | |
340 | { | |
b1199bd3 MD |
341 | struct lib_ring_buffer_backend *bufb = &ctx->priv.buf->backend; |
342 | struct channel_backend *chanb = &ctx->priv.chan->backend; | |
85a07c33 | 343 | size_t index, pagecpy; |
b1199bd3 | 344 | size_t offset = ctx->priv.buf_offset; |
85a07c33 | 345 | struct lib_ring_buffer_backend_pages *backend_pages; |
16f78f3a MD |
346 | |
347 | if (unlikely(!len)) | |
348 | return; | |
85a07c33 MD |
349 | backend_pages = |
350 | lib_ring_buffer_get_backend_pages_from_ctx(config, ctx); | |
16f78f3a | 351 | offset &= chanb->buf_size - 1; |
16f78f3a MD |
352 | index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT; |
353 | pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK); | |
16f78f3a | 354 | |
16f78f3a | 355 | pagefault_disable(); |
80bb2600 | 356 | if (unlikely(!lttng_access_ok(VERIFY_READ, src, len))) |
16f78f3a MD |
357 | goto fill_buffer; |
358 | ||
359 | if (likely(pagecpy == len)) { | |
360 | size_t count; | |
361 | ||
362 | count = lib_ring_buffer_do_strcpy_from_user_inatomic(config, | |
85a07c33 | 363 | backend_pages->p[index].virt |
16f78f3a MD |
364 | + (offset & ~PAGE_MASK), |
365 | src, len - 1); | |
366 | offset += count; | |
367 | /* Padding */ | |
368 | if (unlikely(count < len - 1)) { | |
369 | size_t pad_len = len - 1 - count; | |
370 | ||
85a07c33 | 371 | lib_ring_buffer_do_memset(backend_pages->p[index].virt |
16f78f3a MD |
372 | + (offset & ~PAGE_MASK), |
373 | pad, pad_len); | |
374 | offset += pad_len; | |
375 | } | |
376 | /* Ending '\0' */ | |
85a07c33 | 377 | lib_ring_buffer_do_memset(backend_pages->p[index].virt |
16f78f3a MD |
378 | + (offset & ~PAGE_MASK), |
379 | '\0', 1); | |
380 | } else { | |
381 | _lib_ring_buffer_strcpy_from_user_inatomic(bufb, offset, src, | |
382 | len, 0, pad); | |
383 | } | |
384 | pagefault_enable(); | |
b1199bd3 | 385 | ctx->priv.buf_offset += len; |
16f78f3a MD |
386 | |
387 | return; | |
388 | ||
389 | fill_buffer: | |
390 | pagefault_enable(); | |
16f78f3a MD |
391 | /* |
392 | * In the error path we call the slow path version to avoid | |
393 | * the pollution of static inline code. | |
394 | */ | |
395 | _lib_ring_buffer_memset(bufb, offset, pad, len - 1, 0); | |
396 | offset += len - 1; | |
397 | _lib_ring_buffer_memset(bufb, offset, '\0', 1, 0); | |
398 | } | |
399 | ||
f3bc08c5 MD |
400 | /* |
401 | * This accessor counts the number of unread records in a buffer. | |
402 | * It only provides a consistent value if no reads not writes are performed | |
403 | * concurrently. | |
404 | */ | |
405 | static inline | |
406 | unsigned long lib_ring_buffer_get_records_unread( | |
407 | const struct lib_ring_buffer_config *config, | |
408 | struct lib_ring_buffer *buf) | |
409 | { | |
410 | struct lib_ring_buffer_backend *bufb = &buf->backend; | |
411 | struct lib_ring_buffer_backend_pages *pages; | |
412 | unsigned long records_unread = 0, sb_bindex, id; | |
413 | unsigned int i; | |
414 | ||
415 | for (i = 0; i < bufb->chan->backend.num_subbuf; i++) { | |
416 | id = bufb->buf_wsb[i].id; | |
417 | sb_bindex = subbuffer_id_get_index(config, id); | |
418 | pages = bufb->array[sb_bindex]; | |
419 | records_unread += v_read(config, &pages->records_unread); | |
420 | } | |
421 | if (config->mode == RING_BUFFER_OVERWRITE) { | |
422 | id = bufb->buf_rsb.id; | |
423 | sb_bindex = subbuffer_id_get_index(config, id); | |
424 | pages = bufb->array[sb_bindex]; | |
425 | records_unread += v_read(config, &pages->records_unread); | |
426 | } | |
427 | return records_unread; | |
428 | } | |
429 | ||
3c8ebbc8 MD |
430 | /* |
431 | * We use __copy_from_user_inatomic to copy userspace data after | |
432 | * checking with access_ok() and disabling page faults. | |
433 | * | |
434 | * Return 0 if OK, nonzero on error. | |
435 | */ | |
436 | static inline | |
437 | unsigned long lib_ring_buffer_copy_from_user_check_nofault(void *dest, | |
438 | const void __user *src, | |
439 | unsigned long len) | |
440 | { | |
441 | unsigned long ret; | |
3c8ebbc8 | 442 | |
80bb2600 | 443 | if (!lttng_access_ok(VERIFY_READ, src, len)) |
3c8ebbc8 | 444 | return 1; |
3c8ebbc8 MD |
445 | pagefault_disable(); |
446 | ret = __copy_from_user_inatomic(dest, src, len); | |
447 | pagefault_enable(); | |
3c8ebbc8 MD |
448 | return ret; |
449 | } | |
450 | ||
886d51a3 | 451 | #endif /* _LIB_RING_BUFFER_BACKEND_H */ |