lib/ringbuffer/backend.h

   1 /* SPDX-License-Identifier: (GPL-2.0-only OR LGPL-2.1-only)
   2  *
   3  * lib/ringbuffer/backend.h
   4  *
   5  * Ring buffer backend (API).
   6  *
   7  * Copyright (C) 2010-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   8  *
   9  * Credits to Steven Rostedt for proposing to use an extra-subbuffer owned by
  10  * the reader in flight recorder mode.
  11  */
  12
  13 #ifndef _LIB_RING_BUFFER_BACKEND_H
  14 #define _LIB_RING_BUFFER_BACKEND_H
  15
  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>
  24 #include <wrapper/uaccess.h>
  25
  26 /* Internal helpers */
  27 #include <wrapper/ringbuffer/backend_internal.h>
  28 #include <wrapper/ringbuffer/frontend_internal.h>
  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
  44 extern unsigned long *
  45 lib_ring_buffer_read_get_pfn(struct lib_ring_buffer_backend *bufb, size_t offset,
  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  */
  73 static inline __attribute__((always_inline))
  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 {
  78         struct lib_ring_buffer_backend *bufb = &ctx->buf->backend;
  79         struct channel_backend *chanb = &ctx->chan->backend;
  80         size_t index, pagecpy;
  81         size_t offset = ctx->buf_offset;
  82         struct lib_ring_buffer_backend_pages *backend_pages;
  83
  84         if (unlikely(!len))
  85                 return;
  86         backend_pages =
  87                 lib_ring_buffer_get_backend_pages_from_ctx(config, ctx);
  88         offset &= chanb->buf_size - 1;
  89         index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT;
  90         pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK);
  91         if (likely(pagecpy == len))
  92                 lib_ring_buffer_do_copy(config,
  93                                         backend_pages->p[index].virt
  94                                             + (offset & ~PAGE_MASK),
  95                                         src, len);
  96         else
  97                 _lib_ring_buffer_write(bufb, offset, src, len, 0);
  98         ctx->buf_offset += len;
  99 }
 100
 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
 119         struct lib_ring_buffer_backend *bufb = &ctx->buf->backend;
 120         struct channel_backend *chanb = &ctx->chan->backend;
 121         size_t index, pagecpy;
 122         size_t offset = ctx->buf_offset;
 123         struct lib_ring_buffer_backend_pages *backend_pages;
 124
 125         if (unlikely(!len))
 126                 return;
 127         backend_pages =
 128                 lib_ring_buffer_get_backend_pages_from_ctx(config, ctx);
 129         offset &= chanb->buf_size - 1;
 130         index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT;
 131         pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK);
 132         if (likely(pagecpy == len))
 133                 lib_ring_buffer_do_memset(backend_pages->p[index].virt
 134                                           + (offset & ~PAGE_MASK),
 135                                           c, len);
 136         else
 137                 _lib_ring_buffer_memset(bufb, offset, c, len, 0);
 138         ctx->buf_offset += len;
 139 }
 140
 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  */
 146 static inline __attribute__((always_inline))
 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                  */
 159                 c = READ_ONCE(src[count]);
 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  */
 177 static inline __attribute__((always_inline))
 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
 187                 ret = __copy_from_user_inatomic(&c, src + count, 1);
 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 {
 216         struct lib_ring_buffer_backend *bufb = &ctx->buf->backend;
 217         struct channel_backend *chanb = &ctx->chan->backend;
 218         size_t index, pagecpy;
 219         size_t offset = ctx->buf_offset;
 220         struct lib_ring_buffer_backend_pages *backend_pages;
 221
 222         if (unlikely(!len))
 223                 return;
 224         backend_pages =
 225                 lib_ring_buffer_get_backend_pages_from_ctx(config, ctx);
 226         offset &= chanb->buf_size - 1;
 227         index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT;
 228         pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK);
 229         if (likely(pagecpy == len)) {
 230                 size_t count;
 231
 232                 count = lib_ring_buffer_do_strcpy(config,
 233                                         backend_pages->p[index].virt
 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
 241                         lib_ring_buffer_do_memset(backend_pages->p[index].virt
 242                                                 + (offset & ~PAGE_MASK),
 243                                         pad, pad_len);
 244                         offset += pad_len;
 245                 }
 246                 /* Ending '\0' */
 247                 lib_ring_buffer_do_memset(backend_pages->p[index].virt
 248                                         + (offset & ~PAGE_MASK),
 249                                 '\0', 1);
 250         } else {
 251                 _lib_ring_buffer_strcpy(bufb, offset, src, len, 0, pad);
 252         }
 253         ctx->buf_offset += len;
 254 }
 255
 256 /**
 257  * lib_ring_buffer_copy_from_user_inatomic - write userspace data to a buffer backend
 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
 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.
 268  */
 269 static inline __attribute__((always_inline))
 270 void lib_ring_buffer_copy_from_user_inatomic(const struct lib_ring_buffer_config *config,
 271                                     struct lib_ring_buffer_ctx *ctx,
 272                                     const void __user *src, size_t len)
 273 {
 274         struct lib_ring_buffer_backend *bufb = &ctx->buf->backend;
 275         struct channel_backend *chanb = &ctx->chan->backend;
 276         size_t index, pagecpy;
 277         size_t offset = ctx->buf_offset;
 278         struct lib_ring_buffer_backend_pages *backend_pages;
 279         unsigned long ret;
 280         mm_segment_t old_fs = get_fs();
 281
 282         if (unlikely(!len))
 283                 return;
 284         backend_pages =
 285                 lib_ring_buffer_get_backend_pages_from_ctx(config, ctx);
 286         offset &= chanb->buf_size - 1;
 287         index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT;
 288         pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK);
 289
 290         set_fs(KERNEL_DS);
 291         pagefault_disable();
 292         if (unlikely(!lttng_access_ok(VERIFY_READ, src, len)))
 293                 goto fill_buffer;
 294
 295         if (likely(pagecpy == len)) {
 296                 ret = lib_ring_buffer_do_copy_from_user_inatomic(
 297                         backend_pages->p[index].virt + (offset & ~PAGE_MASK),
 298                         src, len);
 299                 if (unlikely(ret > 0)) {
 300                         /* Copy failed. */
 301                         goto fill_buffer;
 302                 }
 303         } else {
 304                 _lib_ring_buffer_copy_from_user_inatomic(bufb, offset, src, len, 0);
 305         }
 306         pagefault_enable();
 307         set_fs(old_fs);
 308         ctx->buf_offset += len;
 309
 310         return;
 311
 312 fill_buffer:
 313         pagefault_enable();
 314         set_fs(old_fs);
 315         /*
 316          * In the error path we call the slow path version to avoid
 317          * the pollution of static inline code.
 318          */
 319         _lib_ring_buffer_memset(bufb, offset, 0, len, 0);
 320 }
 321
 322 /**
 323  * lib_ring_buffer_strcpy_from_user_inatomic - write userspace string data to a buffer backend
 324  * @config : ring buffer instance configuration
 325  * @ctx: ring buffer context (input arguments only)
 326  * @src : userspace source pointer to copy from
 327  * @len : length of data to copy
 328  * @pad : character to use for padding
 329  *
 330  * This function copies @len - 1 bytes of string data from a userspace
 331  * source pointer to a buffer backend, followed by a terminating '\0'
 332  * character, at the current context offset. This is more or less a
 333  * buffer backend-specific strncpy() operation. If a terminating '\0'
 334  * character is found in @src before @len - 1 characters are copied, pad
 335  * the buffer with @pad characters (e.g. '#'). Calls the slow path
 336  * (_ring_buffer_strcpy_from_user_inatomic) if copy is crossing a page
 337  * boundary. Disable the page fault handler to ensure we never try to
 338  * take the mmap_sem.
 339  */
 340 static inline
 341 void lib_ring_buffer_strcpy_from_user_inatomic(const struct lib_ring_buffer_config *config,
 342                 struct lib_ring_buffer_ctx *ctx,
 343                 const void __user *src, size_t len, int pad)
 344 {
 345         struct lib_ring_buffer_backend *bufb = &ctx->buf->backend;
 346         struct channel_backend *chanb = &ctx->chan->backend;
 347         size_t index, pagecpy;
 348         size_t offset = ctx->buf_offset;
 349         struct lib_ring_buffer_backend_pages *backend_pages;
 350         mm_segment_t old_fs = get_fs();
 351
 352         if (unlikely(!len))
 353                 return;
 354         backend_pages =
 355                 lib_ring_buffer_get_backend_pages_from_ctx(config, ctx);
 356         offset &= chanb->buf_size - 1;
 357         index = (offset & (chanb->subbuf_size - 1)) >> PAGE_SHIFT;
 358         pagecpy = min_t(size_t, len, (-offset) & ~PAGE_MASK);
 359
 360         set_fs(KERNEL_DS);
 361         pagefault_disable();
 362         if (unlikely(!lttng_access_ok(VERIFY_READ, src, len)))
 363                 goto fill_buffer;
 364
 365         if (likely(pagecpy == len)) {
 366                 size_t count;
 367
 368                 count = lib_ring_buffer_do_strcpy_from_user_inatomic(config,
 369                                         backend_pages->p[index].virt
 370                                             + (offset & ~PAGE_MASK),
 371                                         src, len - 1);
 372                 offset += count;
 373                 /* Padding */
 374                 if (unlikely(count < len - 1)) {
 375                         size_t pad_len = len - 1 - count;
 376
 377                         lib_ring_buffer_do_memset(backend_pages->p[index].virt
 378                                                 + (offset & ~PAGE_MASK),
 379                                         pad, pad_len);
 380                         offset += pad_len;
 381                 }
 382                 /* Ending '\0' */
 383                 lib_ring_buffer_do_memset(backend_pages->p[index].virt
 384                                         + (offset & ~PAGE_MASK),
 385                                 '\0', 1);
 386         } else {
 387                 _lib_ring_buffer_strcpy_from_user_inatomic(bufb, offset, src,
 388                                         len, 0, pad);
 389         }
 390         pagefault_enable();
 391         set_fs(old_fs);
 392         ctx->buf_offset += len;
 393
 394         return;
 395
 396 fill_buffer:
 397         pagefault_enable();
 398         set_fs(old_fs);
 399         /*
 400          * In the error path we call the slow path version to avoid
 401          * the pollution of static inline code.
 402          */
 403         _lib_ring_buffer_memset(bufb, offset, pad, len - 1, 0);
 404         offset += len - 1;
 405         _lib_ring_buffer_memset(bufb, offset, '\0', 1, 0);
 406 }
 407
 408 /*
 409  * This accessor counts the number of unread records in a buffer.
 410  * It only provides a consistent value if no reads not writes are performed
 411  * concurrently.
 412  */
 413 static inline
 414 unsigned long lib_ring_buffer_get_records_unread(
 415                                 const struct lib_ring_buffer_config *config,
 416                                 struct lib_ring_buffer *buf)
 417 {
 418         struct lib_ring_buffer_backend *bufb = &buf->backend;
 419         struct lib_ring_buffer_backend_pages *pages;
 420         unsigned long records_unread = 0, sb_bindex, id;
 421         unsigned int i;
 422
 423         for (i = 0; i < bufb->chan->backend.num_subbuf; i++) {
 424                 id = bufb->buf_wsb[i].id;
 425                 sb_bindex = subbuffer_id_get_index(config, id);
 426                 pages = bufb->array[sb_bindex];
 427                 records_unread += v_read(config, &pages->records_unread);
 428         }
 429         if (config->mode == RING_BUFFER_OVERWRITE) {
 430                 id = bufb->buf_rsb.id;
 431                 sb_bindex = subbuffer_id_get_index(config, id);
 432                 pages = bufb->array[sb_bindex];
 433                 records_unread += v_read(config, &pages->records_unread);
 434         }
 435         return records_unread;
 436 }
 437
 438 /*
 439  * We use __copy_from_user_inatomic to copy userspace data after
 440  * checking with access_ok() and disabling page faults.
 441  *
 442  * Return 0 if OK, nonzero on error.
 443  */
 444 static inline
 445 unsigned long lib_ring_buffer_copy_from_user_check_nofault(void *dest,
 446                                                 const void __user *src,
 447                                                 unsigned long len)
 448 {
 449         unsigned long ret;
 450         mm_segment_t old_fs;
 451
 452         if (!lttng_access_ok(VERIFY_READ, src, len))
 453                 return 1;
 454         old_fs = get_fs();
 455         set_fs(KERNEL_DS);
 456         pagefault_disable();
 457         ret = __copy_from_user_inatomic(dest, src, len);
 458         pagefault_enable();
 459         set_fs(old_fs);
 460         return ret;
 461 }
 462
 463 #endif /* _LIB_RING_BUFFER_BACKEND_H */