| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Kprobes-based tracing events |
| 4 | * |
| 5 | * Created by Masami Hiramatsu <mhiramat@redhat.com> |
| 6 | * |
| 7 | */ |
| 8 | #define pr_fmt(fmt) "trace_kprobe: " fmt |
| 9 | |
| 10 | #include <linux/bpf-cgroup.h> |
| 11 | #include <linux/security.h> |
| 12 | #include <linux/module.h> |
| 13 | #include <linux/uaccess.h> |
| 14 | #include <linux/rculist.h> |
| 15 | #include <linux/error-injection.h> |
| 16 | |
| 17 | #include <asm/setup.h> /* for COMMAND_LINE_SIZE */ |
| 18 | |
| 19 | #include "trace_dynevent.h" |
| 20 | #include "trace_kprobe_selftest.h" |
| 21 | #include "trace_probe.h" |
| 22 | #include "trace_probe_tmpl.h" |
| 23 | #include "trace_probe_kernel.h" |
| 24 | |
| 25 | #define KPROBE_EVENT_SYSTEM "kprobes" |
| 26 | #define KRETPROBE_MAXACTIVE_MAX 4096 |
| 27 | |
| 28 | /* Kprobe early definition from command line */ |
| 29 | static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata; |
| 30 | |
| 31 | static int __init set_kprobe_boot_events(char *str) |
| 32 | { |
| 33 | strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE); |
| 34 | disable_tracing_selftest("running kprobe events"); |
| 35 | |
| 36 | return 1; |
| 37 | } |
| 38 | __setup("kprobe_event=", set_kprobe_boot_events); |
| 39 | |
| 40 | static int trace_kprobe_create(const char *raw_command); |
| 41 | static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev); |
| 42 | static int trace_kprobe_release(struct dyn_event *ev); |
| 43 | static bool trace_kprobe_is_busy(struct dyn_event *ev); |
| 44 | static bool trace_kprobe_match(const char *system, const char *event, |
| 45 | int argc, const char **argv, struct dyn_event *ev); |
| 46 | |
| 47 | static struct dyn_event_operations trace_kprobe_ops = { |
| 48 | .create = trace_kprobe_create, |
| 49 | .show = trace_kprobe_show, |
| 50 | .is_busy = trace_kprobe_is_busy, |
| 51 | .free = trace_kprobe_release, |
| 52 | .match = trace_kprobe_match, |
| 53 | }; |
| 54 | |
| 55 | /* |
| 56 | * Kprobe event core functions |
| 57 | */ |
| 58 | struct trace_kprobe { |
| 59 | struct dyn_event devent; |
| 60 | struct kretprobe rp; /* Use rp.kp for kprobe use */ |
| 61 | unsigned long __percpu *nhit; |
| 62 | const char *symbol; /* symbol name */ |
| 63 | struct trace_probe tp; |
| 64 | }; |
| 65 | |
| 66 | static bool is_trace_kprobe(struct dyn_event *ev) |
| 67 | { |
| 68 | return ev->ops == &trace_kprobe_ops; |
| 69 | } |
| 70 | |
| 71 | static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev) |
| 72 | { |
| 73 | return container_of(ev, struct trace_kprobe, devent); |
| 74 | } |
| 75 | |
| 76 | /** |
| 77 | * for_each_trace_kprobe - iterate over the trace_kprobe list |
| 78 | * @pos: the struct trace_kprobe * for each entry |
| 79 | * @dpos: the struct dyn_event * to use as a loop cursor |
| 80 | */ |
| 81 | #define for_each_trace_kprobe(pos, dpos) \ |
| 82 | for_each_dyn_event(dpos) \ |
| 83 | if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos))) |
| 84 | |
| 85 | static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk) |
| 86 | { |
| 87 | return tk->rp.handler != NULL; |
| 88 | } |
| 89 | |
| 90 | static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk) |
| 91 | { |
| 92 | return tk->symbol ? tk->symbol : "unknown"; |
| 93 | } |
| 94 | |
| 95 | static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk) |
| 96 | { |
| 97 | return tk->rp.kp.offset; |
| 98 | } |
| 99 | |
| 100 | static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk) |
| 101 | { |
| 102 | return kprobe_gone(&tk->rp.kp); |
| 103 | } |
| 104 | |
| 105 | static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk, |
| 106 | struct module *mod) |
| 107 | { |
| 108 | int len = strlen(module_name(mod)); |
| 109 | const char *name = trace_kprobe_symbol(tk); |
| 110 | |
| 111 | return strncmp(module_name(mod), name, len) == 0 && name[len] == ':'; |
| 112 | } |
| 113 | |
| 114 | #ifdef CONFIG_MODULES |
| 115 | static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk) |
| 116 | { |
| 117 | char *p; |
| 118 | bool ret; |
| 119 | |
| 120 | if (!tk->symbol) |
| 121 | return false; |
| 122 | p = strchr(tk->symbol, ':'); |
| 123 | if (!p) |
| 124 | return true; |
| 125 | *p = '\0'; |
| 126 | rcu_read_lock_sched(); |
| 127 | ret = !!find_module(tk->symbol); |
| 128 | rcu_read_unlock_sched(); |
| 129 | *p = ':'; |
| 130 | |
| 131 | return ret; |
| 132 | } |
| 133 | #else |
| 134 | static inline bool trace_kprobe_module_exist(struct trace_kprobe *tk) |
| 135 | { |
| 136 | return false; |
| 137 | } |
| 138 | #endif |
| 139 | |
| 140 | static bool trace_kprobe_is_busy(struct dyn_event *ev) |
| 141 | { |
| 142 | struct trace_kprobe *tk = to_trace_kprobe(ev); |
| 143 | |
| 144 | return trace_probe_is_enabled(&tk->tp); |
| 145 | } |
| 146 | |
| 147 | static bool trace_kprobe_match_command_head(struct trace_kprobe *tk, |
| 148 | int argc, const char **argv) |
| 149 | { |
| 150 | char buf[MAX_ARGSTR_LEN + 1]; |
| 151 | |
| 152 | if (!argc) |
| 153 | return true; |
| 154 | |
| 155 | if (!tk->symbol) |
| 156 | snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr); |
| 157 | else if (tk->rp.kp.offset) |
| 158 | snprintf(buf, sizeof(buf), "%s+%u", |
| 159 | trace_kprobe_symbol(tk), tk->rp.kp.offset); |
| 160 | else |
| 161 | snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk)); |
| 162 | if (strcmp(buf, argv[0])) |
| 163 | return false; |
| 164 | argc--; argv++; |
| 165 | |
| 166 | return trace_probe_match_command_args(&tk->tp, argc, argv); |
| 167 | } |
| 168 | |
| 169 | static bool trace_kprobe_match(const char *system, const char *event, |
| 170 | int argc, const char **argv, struct dyn_event *ev) |
| 171 | { |
| 172 | struct trace_kprobe *tk = to_trace_kprobe(ev); |
| 173 | |
| 174 | return (event[0] == '\0' || |
| 175 | strcmp(trace_probe_name(&tk->tp), event) == 0) && |
| 176 | (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) && |
| 177 | trace_kprobe_match_command_head(tk, argc, argv); |
| 178 | } |
| 179 | |
| 180 | static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk) |
| 181 | { |
| 182 | unsigned long nhit = 0; |
| 183 | int cpu; |
| 184 | |
| 185 | for_each_possible_cpu(cpu) |
| 186 | nhit += *per_cpu_ptr(tk->nhit, cpu); |
| 187 | |
| 188 | return nhit; |
| 189 | } |
| 190 | |
| 191 | static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk) |
| 192 | { |
| 193 | return !(list_empty(&tk->rp.kp.list) && |
| 194 | hlist_unhashed(&tk->rp.kp.hlist)); |
| 195 | } |
| 196 | |
| 197 | /* Return 0 if it fails to find the symbol address */ |
| 198 | static nokprobe_inline |
| 199 | unsigned long trace_kprobe_address(struct trace_kprobe *tk) |
| 200 | { |
| 201 | unsigned long addr; |
| 202 | |
| 203 | if (tk->symbol) { |
| 204 | addr = (unsigned long) |
| 205 | kallsyms_lookup_name(trace_kprobe_symbol(tk)); |
| 206 | if (addr) |
| 207 | addr += tk->rp.kp.offset; |
| 208 | } else { |
| 209 | addr = (unsigned long)tk->rp.kp.addr; |
| 210 | } |
| 211 | return addr; |
| 212 | } |
| 213 | |
| 214 | static nokprobe_inline struct trace_kprobe * |
| 215 | trace_kprobe_primary_from_call(struct trace_event_call *call) |
| 216 | { |
| 217 | struct trace_probe *tp; |
| 218 | |
| 219 | tp = trace_probe_primary_from_call(call); |
| 220 | if (WARN_ON_ONCE(!tp)) |
| 221 | return NULL; |
| 222 | |
| 223 | return container_of(tp, struct trace_kprobe, tp); |
| 224 | } |
| 225 | |
| 226 | bool trace_kprobe_on_func_entry(struct trace_event_call *call) |
| 227 | { |
| 228 | struct trace_kprobe *tk = trace_kprobe_primary_from_call(call); |
| 229 | |
| 230 | return tk ? (kprobe_on_func_entry(tk->rp.kp.addr, |
| 231 | tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name, |
| 232 | tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false; |
| 233 | } |
| 234 | |
| 235 | bool trace_kprobe_error_injectable(struct trace_event_call *call) |
| 236 | { |
| 237 | struct trace_kprobe *tk = trace_kprobe_primary_from_call(call); |
| 238 | |
| 239 | return tk ? within_error_injection_list(trace_kprobe_address(tk)) : |
| 240 | false; |
| 241 | } |
| 242 | |
| 243 | static int register_kprobe_event(struct trace_kprobe *tk); |
| 244 | static int unregister_kprobe_event(struct trace_kprobe *tk); |
| 245 | |
| 246 | static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs); |
| 247 | static int kretprobe_dispatcher(struct kretprobe_instance *ri, |
| 248 | struct pt_regs *regs); |
| 249 | |
| 250 | static void free_trace_kprobe(struct trace_kprobe *tk) |
| 251 | { |
| 252 | if (tk) { |
| 253 | trace_probe_cleanup(&tk->tp); |
| 254 | kfree(tk->symbol); |
| 255 | free_percpu(tk->nhit); |
| 256 | kfree(tk); |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | /* |
| 261 | * Allocate new trace_probe and initialize it (including kprobes). |
| 262 | */ |
| 263 | static struct trace_kprobe *alloc_trace_kprobe(const char *group, |
| 264 | const char *event, |
| 265 | void *addr, |
| 266 | const char *symbol, |
| 267 | unsigned long offs, |
| 268 | int maxactive, |
| 269 | int nargs, bool is_return) |
| 270 | { |
| 271 | struct trace_kprobe *tk; |
| 272 | int ret = -ENOMEM; |
| 273 | |
| 274 | tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL); |
| 275 | if (!tk) |
| 276 | return ERR_PTR(ret); |
| 277 | |
| 278 | tk->nhit = alloc_percpu(unsigned long); |
| 279 | if (!tk->nhit) |
| 280 | goto error; |
| 281 | |
| 282 | if (symbol) { |
| 283 | tk->symbol = kstrdup(symbol, GFP_KERNEL); |
| 284 | if (!tk->symbol) |
| 285 | goto error; |
| 286 | tk->rp.kp.symbol_name = tk->symbol; |
| 287 | tk->rp.kp.offset = offs; |
| 288 | } else |
| 289 | tk->rp.kp.addr = addr; |
| 290 | |
| 291 | if (is_return) |
| 292 | tk->rp.handler = kretprobe_dispatcher; |
| 293 | else |
| 294 | tk->rp.kp.pre_handler = kprobe_dispatcher; |
| 295 | |
| 296 | tk->rp.maxactive = maxactive; |
| 297 | INIT_HLIST_NODE(&tk->rp.kp.hlist); |
| 298 | INIT_LIST_HEAD(&tk->rp.kp.list); |
| 299 | |
| 300 | ret = trace_probe_init(&tk->tp, event, group, false, nargs); |
| 301 | if (ret < 0) |
| 302 | goto error; |
| 303 | |
| 304 | dyn_event_init(&tk->devent, &trace_kprobe_ops); |
| 305 | return tk; |
| 306 | error: |
| 307 | free_trace_kprobe(tk); |
| 308 | return ERR_PTR(ret); |
| 309 | } |
| 310 | |
| 311 | static struct trace_kprobe *find_trace_kprobe(const char *event, |
| 312 | const char *group) |
| 313 | { |
| 314 | struct dyn_event *pos; |
| 315 | struct trace_kprobe *tk; |
| 316 | |
| 317 | for_each_trace_kprobe(tk, pos) |
| 318 | if (strcmp(trace_probe_name(&tk->tp), event) == 0 && |
| 319 | strcmp(trace_probe_group_name(&tk->tp), group) == 0) |
| 320 | return tk; |
| 321 | return NULL; |
| 322 | } |
| 323 | |
| 324 | static inline int __enable_trace_kprobe(struct trace_kprobe *tk) |
| 325 | { |
| 326 | int ret = 0; |
| 327 | |
| 328 | if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) { |
| 329 | if (trace_kprobe_is_return(tk)) |
| 330 | ret = enable_kretprobe(&tk->rp); |
| 331 | else |
| 332 | ret = enable_kprobe(&tk->rp.kp); |
| 333 | } |
| 334 | |
| 335 | return ret; |
| 336 | } |
| 337 | |
| 338 | static void __disable_trace_kprobe(struct trace_probe *tp) |
| 339 | { |
| 340 | struct trace_kprobe *tk; |
| 341 | |
| 342 | list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) { |
| 343 | if (!trace_kprobe_is_registered(tk)) |
| 344 | continue; |
| 345 | if (trace_kprobe_is_return(tk)) |
| 346 | disable_kretprobe(&tk->rp); |
| 347 | else |
| 348 | disable_kprobe(&tk->rp.kp); |
| 349 | } |
| 350 | } |
| 351 | |
| 352 | /* |
| 353 | * Enable trace_probe |
| 354 | * if the file is NULL, enable "perf" handler, or enable "trace" handler. |
| 355 | */ |
| 356 | static int enable_trace_kprobe(struct trace_event_call *call, |
| 357 | struct trace_event_file *file) |
| 358 | { |
| 359 | struct trace_probe *tp; |
| 360 | struct trace_kprobe *tk; |
| 361 | bool enabled; |
| 362 | int ret = 0; |
| 363 | |
| 364 | tp = trace_probe_primary_from_call(call); |
| 365 | if (WARN_ON_ONCE(!tp)) |
| 366 | return -ENODEV; |
| 367 | enabled = trace_probe_is_enabled(tp); |
| 368 | |
| 369 | /* This also changes "enabled" state */ |
| 370 | if (file) { |
| 371 | ret = trace_probe_add_file(tp, file); |
| 372 | if (ret) |
| 373 | return ret; |
| 374 | } else |
| 375 | trace_probe_set_flag(tp, TP_FLAG_PROFILE); |
| 376 | |
| 377 | if (enabled) |
| 378 | return 0; |
| 379 | |
| 380 | list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) { |
| 381 | if (trace_kprobe_has_gone(tk)) |
| 382 | continue; |
| 383 | ret = __enable_trace_kprobe(tk); |
| 384 | if (ret) |
| 385 | break; |
| 386 | enabled = true; |
| 387 | } |
| 388 | |
| 389 | if (ret) { |
| 390 | /* Failed to enable one of them. Roll back all */ |
| 391 | if (enabled) |
| 392 | __disable_trace_kprobe(tp); |
| 393 | if (file) |
| 394 | trace_probe_remove_file(tp, file); |
| 395 | else |
| 396 | trace_probe_clear_flag(tp, TP_FLAG_PROFILE); |
| 397 | } |
| 398 | |
| 399 | return ret; |
| 400 | } |
| 401 | |
| 402 | /* |
| 403 | * Disable trace_probe |
| 404 | * if the file is NULL, disable "perf" handler, or disable "trace" handler. |
| 405 | */ |
| 406 | static int disable_trace_kprobe(struct trace_event_call *call, |
| 407 | struct trace_event_file *file) |
| 408 | { |
| 409 | struct trace_probe *tp; |
| 410 | |
| 411 | tp = trace_probe_primary_from_call(call); |
| 412 | if (WARN_ON_ONCE(!tp)) |
| 413 | return -ENODEV; |
| 414 | |
| 415 | if (file) { |
| 416 | if (!trace_probe_get_file_link(tp, file)) |
| 417 | return -ENOENT; |
| 418 | if (!trace_probe_has_single_file(tp)) |
| 419 | goto out; |
| 420 | trace_probe_clear_flag(tp, TP_FLAG_TRACE); |
| 421 | } else |
| 422 | trace_probe_clear_flag(tp, TP_FLAG_PROFILE); |
| 423 | |
| 424 | if (!trace_probe_is_enabled(tp)) |
| 425 | __disable_trace_kprobe(tp); |
| 426 | |
| 427 | out: |
| 428 | if (file) |
| 429 | /* |
| 430 | * Synchronization is done in below function. For perf event, |
| 431 | * file == NULL and perf_trace_event_unreg() calls |
| 432 | * tracepoint_synchronize_unregister() to ensure synchronize |
| 433 | * event. We don't need to care about it. |
| 434 | */ |
| 435 | trace_probe_remove_file(tp, file); |
| 436 | |
| 437 | return 0; |
| 438 | } |
| 439 | |
| 440 | #if defined(CONFIG_DYNAMIC_FTRACE) && \ |
| 441 | !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE) |
| 442 | static bool __within_notrace_func(unsigned long addr) |
| 443 | { |
| 444 | unsigned long offset, size; |
| 445 | |
| 446 | if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset)) |
| 447 | return false; |
| 448 | |
| 449 | /* Get the entry address of the target function */ |
| 450 | addr -= offset; |
| 451 | |
| 452 | /* |
| 453 | * Since ftrace_location_range() does inclusive range check, we need |
| 454 | * to subtract 1 byte from the end address. |
| 455 | */ |
| 456 | return !ftrace_location_range(addr, addr + size - 1); |
| 457 | } |
| 458 | |
| 459 | static bool within_notrace_func(struct trace_kprobe *tk) |
| 460 | { |
| 461 | unsigned long addr = trace_kprobe_address(tk); |
| 462 | char symname[KSYM_NAME_LEN], *p; |
| 463 | |
| 464 | if (!__within_notrace_func(addr)) |
| 465 | return false; |
| 466 | |
| 467 | /* Check if the address is on a suffixed-symbol */ |
| 468 | if (!lookup_symbol_name(addr, symname)) { |
| 469 | p = strchr(symname, '.'); |
| 470 | if (!p) |
| 471 | return true; |
| 472 | *p = '\0'; |
| 473 | addr = (unsigned long)kprobe_lookup_name(symname, 0); |
| 474 | if (addr) |
| 475 | return __within_notrace_func(addr); |
| 476 | } |
| 477 | |
| 478 | return true; |
| 479 | } |
| 480 | #else |
| 481 | #define within_notrace_func(tk) (false) |
| 482 | #endif |
| 483 | |
| 484 | /* Internal register function - just handle k*probes and flags */ |
| 485 | static int __register_trace_kprobe(struct trace_kprobe *tk) |
| 486 | { |
| 487 | int i, ret; |
| 488 | |
| 489 | ret = security_locked_down(LOCKDOWN_KPROBES); |
| 490 | if (ret) |
| 491 | return ret; |
| 492 | |
| 493 | if (trace_kprobe_is_registered(tk)) |
| 494 | return -EINVAL; |
| 495 | |
| 496 | if (within_notrace_func(tk)) { |
| 497 | pr_warn("Could not probe notrace function %ps\n", |
| 498 | (void *)trace_kprobe_address(tk)); |
| 499 | return -EINVAL; |
| 500 | } |
| 501 | |
| 502 | for (i = 0; i < tk->tp.nr_args; i++) { |
| 503 | ret = traceprobe_update_arg(&tk->tp.args[i]); |
| 504 | if (ret) |
| 505 | return ret; |
| 506 | } |
| 507 | |
| 508 | /* Set/clear disabled flag according to tp->flag */ |
| 509 | if (trace_probe_is_enabled(&tk->tp)) |
| 510 | tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED; |
| 511 | else |
| 512 | tk->rp.kp.flags |= KPROBE_FLAG_DISABLED; |
| 513 | |
| 514 | if (trace_kprobe_is_return(tk)) |
| 515 | ret = register_kretprobe(&tk->rp); |
| 516 | else |
| 517 | ret = register_kprobe(&tk->rp.kp); |
| 518 | |
| 519 | return ret; |
| 520 | } |
| 521 | |
| 522 | /* Internal unregister function - just handle k*probes and flags */ |
| 523 | static void __unregister_trace_kprobe(struct trace_kprobe *tk) |
| 524 | { |
| 525 | if (trace_kprobe_is_registered(tk)) { |
| 526 | if (trace_kprobe_is_return(tk)) |
| 527 | unregister_kretprobe(&tk->rp); |
| 528 | else |
| 529 | unregister_kprobe(&tk->rp.kp); |
| 530 | /* Cleanup kprobe for reuse and mark it unregistered */ |
| 531 | INIT_HLIST_NODE(&tk->rp.kp.hlist); |
| 532 | INIT_LIST_HEAD(&tk->rp.kp.list); |
| 533 | if (tk->rp.kp.symbol_name) |
| 534 | tk->rp.kp.addr = NULL; |
| 535 | } |
| 536 | } |
| 537 | |
| 538 | /* Unregister a trace_probe and probe_event */ |
| 539 | static int unregister_trace_kprobe(struct trace_kprobe *tk) |
| 540 | { |
| 541 | /* If other probes are on the event, just unregister kprobe */ |
| 542 | if (trace_probe_has_sibling(&tk->tp)) |
| 543 | goto unreg; |
| 544 | |
| 545 | /* Enabled event can not be unregistered */ |
| 546 | if (trace_probe_is_enabled(&tk->tp)) |
| 547 | return -EBUSY; |
| 548 | |
| 549 | /* If there's a reference to the dynamic event */ |
| 550 | if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp))) |
| 551 | return -EBUSY; |
| 552 | |
| 553 | /* Will fail if probe is being used by ftrace or perf */ |
| 554 | if (unregister_kprobe_event(tk)) |
| 555 | return -EBUSY; |
| 556 | |
| 557 | unreg: |
| 558 | __unregister_trace_kprobe(tk); |
| 559 | dyn_event_remove(&tk->devent); |
| 560 | trace_probe_unlink(&tk->tp); |
| 561 | |
| 562 | return 0; |
| 563 | } |
| 564 | |
| 565 | static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig, |
| 566 | struct trace_kprobe *comp) |
| 567 | { |
| 568 | struct trace_probe_event *tpe = orig->tp.event; |
| 569 | int i; |
| 570 | |
| 571 | list_for_each_entry(orig, &tpe->probes, tp.list) { |
| 572 | if (strcmp(trace_kprobe_symbol(orig), |
| 573 | trace_kprobe_symbol(comp)) || |
| 574 | trace_kprobe_offset(orig) != trace_kprobe_offset(comp)) |
| 575 | continue; |
| 576 | |
| 577 | /* |
| 578 | * trace_probe_compare_arg_type() ensured that nr_args and |
| 579 | * each argument name and type are same. Let's compare comm. |
| 580 | */ |
| 581 | for (i = 0; i < orig->tp.nr_args; i++) { |
| 582 | if (strcmp(orig->tp.args[i].comm, |
| 583 | comp->tp.args[i].comm)) |
| 584 | break; |
| 585 | } |
| 586 | |
| 587 | if (i == orig->tp.nr_args) |
| 588 | return true; |
| 589 | } |
| 590 | |
| 591 | return false; |
| 592 | } |
| 593 | |
| 594 | static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to) |
| 595 | { |
| 596 | int ret; |
| 597 | |
| 598 | ret = trace_probe_compare_arg_type(&tk->tp, &to->tp); |
| 599 | if (ret) { |
| 600 | /* Note that argument starts index = 2 */ |
| 601 | trace_probe_log_set_index(ret + 1); |
| 602 | trace_probe_log_err(0, DIFF_ARG_TYPE); |
| 603 | return -EEXIST; |
| 604 | } |
| 605 | if (trace_kprobe_has_same_kprobe(to, tk)) { |
| 606 | trace_probe_log_set_index(0); |
| 607 | trace_probe_log_err(0, SAME_PROBE); |
| 608 | return -EEXIST; |
| 609 | } |
| 610 | |
| 611 | /* Append to existing event */ |
| 612 | ret = trace_probe_append(&tk->tp, &to->tp); |
| 613 | if (ret) |
| 614 | return ret; |
| 615 | |
| 616 | /* Register k*probe */ |
| 617 | ret = __register_trace_kprobe(tk); |
| 618 | if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) { |
| 619 | pr_warn("This probe might be able to register after target module is loaded. Continue.\n"); |
| 620 | ret = 0; |
| 621 | } |
| 622 | |
| 623 | if (ret) |
| 624 | trace_probe_unlink(&tk->tp); |
| 625 | else |
| 626 | dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp)); |
| 627 | |
| 628 | return ret; |
| 629 | } |
| 630 | |
| 631 | /* Register a trace_probe and probe_event */ |
| 632 | static int register_trace_kprobe(struct trace_kprobe *tk) |
| 633 | { |
| 634 | struct trace_kprobe *old_tk; |
| 635 | int ret; |
| 636 | |
| 637 | mutex_lock(&event_mutex); |
| 638 | |
| 639 | old_tk = find_trace_kprobe(trace_probe_name(&tk->tp), |
| 640 | trace_probe_group_name(&tk->tp)); |
| 641 | if (old_tk) { |
| 642 | if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) { |
| 643 | trace_probe_log_set_index(0); |
| 644 | trace_probe_log_err(0, DIFF_PROBE_TYPE); |
| 645 | ret = -EEXIST; |
| 646 | } else { |
| 647 | ret = append_trace_kprobe(tk, old_tk); |
| 648 | } |
| 649 | goto end; |
| 650 | } |
| 651 | |
| 652 | /* Register new event */ |
| 653 | ret = register_kprobe_event(tk); |
| 654 | if (ret) { |
| 655 | if (ret == -EEXIST) { |
| 656 | trace_probe_log_set_index(0); |
| 657 | trace_probe_log_err(0, EVENT_EXIST); |
| 658 | } else |
| 659 | pr_warn("Failed to register probe event(%d)\n", ret); |
| 660 | goto end; |
| 661 | } |
| 662 | |
| 663 | /* Register k*probe */ |
| 664 | ret = __register_trace_kprobe(tk); |
| 665 | if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) { |
| 666 | pr_warn("This probe might be able to register after target module is loaded. Continue.\n"); |
| 667 | ret = 0; |
| 668 | } |
| 669 | |
| 670 | if (ret < 0) |
| 671 | unregister_kprobe_event(tk); |
| 672 | else |
| 673 | dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp)); |
| 674 | |
| 675 | end: |
| 676 | mutex_unlock(&event_mutex); |
| 677 | return ret; |
| 678 | } |
| 679 | |
| 680 | #ifdef CONFIG_MODULES |
| 681 | /* Module notifier call back, checking event on the module */ |
| 682 | static int trace_kprobe_module_callback(struct notifier_block *nb, |
| 683 | unsigned long val, void *data) |
| 684 | { |
| 685 | struct module *mod = data; |
| 686 | struct dyn_event *pos; |
| 687 | struct trace_kprobe *tk; |
| 688 | int ret; |
| 689 | |
| 690 | if (val != MODULE_STATE_COMING) |
| 691 | return NOTIFY_DONE; |
| 692 | |
| 693 | /* Update probes on coming module */ |
| 694 | mutex_lock(&event_mutex); |
| 695 | for_each_trace_kprobe(tk, pos) { |
| 696 | if (trace_kprobe_within_module(tk, mod)) { |
| 697 | /* Don't need to check busy - this should have gone. */ |
| 698 | __unregister_trace_kprobe(tk); |
| 699 | ret = __register_trace_kprobe(tk); |
| 700 | if (ret) |
| 701 | pr_warn("Failed to re-register probe %s on %s: %d\n", |
| 702 | trace_probe_name(&tk->tp), |
| 703 | module_name(mod), ret); |
| 704 | } |
| 705 | } |
| 706 | mutex_unlock(&event_mutex); |
| 707 | |
| 708 | return NOTIFY_DONE; |
| 709 | } |
| 710 | |
| 711 | static struct notifier_block trace_kprobe_module_nb = { |
| 712 | .notifier_call = trace_kprobe_module_callback, |
| 713 | .priority = 1 /* Invoked after kprobe module callback */ |
| 714 | }; |
| 715 | static int trace_kprobe_register_module_notifier(void) |
| 716 | { |
| 717 | return register_module_notifier(&trace_kprobe_module_nb); |
| 718 | } |
| 719 | #else |
| 720 | static int trace_kprobe_register_module_notifier(void) |
| 721 | { |
| 722 | return 0; |
| 723 | } |
| 724 | #endif /* CONFIG_MODULES */ |
| 725 | |
| 726 | static int count_symbols(void *data, unsigned long unused) |
| 727 | { |
| 728 | unsigned int *count = data; |
| 729 | |
| 730 | (*count)++; |
| 731 | |
| 732 | return 0; |
| 733 | } |
| 734 | |
| 735 | struct sym_count_ctx { |
| 736 | unsigned int count; |
| 737 | const char *name; |
| 738 | }; |
| 739 | |
| 740 | static int count_mod_symbols(void *data, const char *name, unsigned long unused) |
| 741 | { |
| 742 | struct sym_count_ctx *ctx = data; |
| 743 | |
| 744 | if (strcmp(name, ctx->name) == 0) |
| 745 | ctx->count++; |
| 746 | |
| 747 | return 0; |
| 748 | } |
| 749 | |
| 750 | static unsigned int number_of_same_symbols(char *func_name) |
| 751 | { |
| 752 | struct sym_count_ctx ctx = { .count = 0, .name = func_name }; |
| 753 | |
| 754 | kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count); |
| 755 | |
| 756 | module_kallsyms_on_each_symbol(NULL, count_mod_symbols, &ctx); |
| 757 | |
| 758 | return ctx.count; |
| 759 | } |
| 760 | |
| 761 | static int trace_kprobe_entry_handler(struct kretprobe_instance *ri, |
| 762 | struct pt_regs *regs); |
| 763 | |
| 764 | static int __trace_kprobe_create(int argc, const char *argv[]) |
| 765 | { |
| 766 | /* |
| 767 | * Argument syntax: |
| 768 | * - Add kprobe: |
| 769 | * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS] |
| 770 | * - Add kretprobe: |
| 771 | * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS] |
| 772 | * Or |
| 773 | * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS] |
| 774 | * |
| 775 | * Fetch args: |
| 776 | * $retval : fetch return value |
| 777 | * $stack : fetch stack address |
| 778 | * $stackN : fetch Nth of stack (N:0-) |
| 779 | * $comm : fetch current task comm |
| 780 | * @ADDR : fetch memory at ADDR (ADDR should be in kernel) |
| 781 | * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol) |
| 782 | * %REG : fetch register REG |
| 783 | * Dereferencing memory fetch: |
| 784 | * +|-offs(ARG) : fetch memory at ARG +|- offs address. |
| 785 | * Alias name of args: |
| 786 | * NAME=FETCHARG : set NAME as alias of FETCHARG. |
| 787 | * Type of args: |
| 788 | * FETCHARG:TYPE : use TYPE instead of unsigned long. |
| 789 | */ |
| 790 | struct trace_kprobe *tk = NULL; |
| 791 | int i, len, new_argc = 0, ret = 0; |
| 792 | bool is_return = false; |
| 793 | char *symbol = NULL, *tmp = NULL; |
| 794 | const char **new_argv = NULL; |
| 795 | const char *event = NULL, *group = KPROBE_EVENT_SYSTEM; |
| 796 | enum probe_print_type ptype; |
| 797 | int maxactive = 0; |
| 798 | long offset = 0; |
| 799 | void *addr = NULL; |
| 800 | char buf[MAX_EVENT_NAME_LEN]; |
| 801 | char gbuf[MAX_EVENT_NAME_LEN]; |
| 802 | char abuf[MAX_BTF_ARGS_LEN]; |
| 803 | char *dbuf = NULL; |
| 804 | struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL }; |
| 805 | |
| 806 | switch (argv[0][0]) { |
| 807 | case 'r': |
| 808 | is_return = true; |
| 809 | break; |
| 810 | case 'p': |
| 811 | break; |
| 812 | default: |
| 813 | return -ECANCELED; |
| 814 | } |
| 815 | if (argc < 2) |
| 816 | return -ECANCELED; |
| 817 | |
| 818 | trace_probe_log_init("trace_kprobe", argc, argv); |
| 819 | |
| 820 | event = strchr(&argv[0][1], ':'); |
| 821 | if (event) |
| 822 | event++; |
| 823 | |
| 824 | if (isdigit(argv[0][1])) { |
| 825 | if (!is_return) { |
| 826 | trace_probe_log_err(1, BAD_MAXACT_TYPE); |
| 827 | goto parse_error; |
| 828 | } |
| 829 | if (event) |
| 830 | len = event - &argv[0][1] - 1; |
| 831 | else |
| 832 | len = strlen(&argv[0][1]); |
| 833 | if (len > MAX_EVENT_NAME_LEN - 1) { |
| 834 | trace_probe_log_err(1, BAD_MAXACT); |
| 835 | goto parse_error; |
| 836 | } |
| 837 | memcpy(buf, &argv[0][1], len); |
| 838 | buf[len] = '\0'; |
| 839 | ret = kstrtouint(buf, 0, &maxactive); |
| 840 | if (ret || !maxactive) { |
| 841 | trace_probe_log_err(1, BAD_MAXACT); |
| 842 | goto parse_error; |
| 843 | } |
| 844 | /* kretprobes instances are iterated over via a list. The |
| 845 | * maximum should stay reasonable. |
| 846 | */ |
| 847 | if (maxactive > KRETPROBE_MAXACTIVE_MAX) { |
| 848 | trace_probe_log_err(1, MAXACT_TOO_BIG); |
| 849 | goto parse_error; |
| 850 | } |
| 851 | } |
| 852 | |
| 853 | /* try to parse an address. if that fails, try to read the |
| 854 | * input as a symbol. */ |
| 855 | if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) { |
| 856 | trace_probe_log_set_index(1); |
| 857 | /* Check whether uprobe event specified */ |
| 858 | if (strchr(argv[1], '/') && strchr(argv[1], ':')) { |
| 859 | ret = -ECANCELED; |
| 860 | goto error; |
| 861 | } |
| 862 | /* a symbol specified */ |
| 863 | symbol = kstrdup(argv[1], GFP_KERNEL); |
| 864 | if (!symbol) |
| 865 | return -ENOMEM; |
| 866 | |
| 867 | tmp = strchr(symbol, '%'); |
| 868 | if (tmp) { |
| 869 | if (!strcmp(tmp, "%return")) { |
| 870 | *tmp = '\0'; |
| 871 | is_return = true; |
| 872 | } else { |
| 873 | trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX); |
| 874 | goto parse_error; |
| 875 | } |
| 876 | } |
| 877 | |
| 878 | /* TODO: support .init module functions */ |
| 879 | ret = traceprobe_split_symbol_offset(symbol, &offset); |
| 880 | if (ret || offset < 0 || offset > UINT_MAX) { |
| 881 | trace_probe_log_err(0, BAD_PROBE_ADDR); |
| 882 | goto parse_error; |
| 883 | } |
| 884 | if (is_return) |
| 885 | ctx.flags |= TPARG_FL_RETURN; |
| 886 | ret = kprobe_on_func_entry(NULL, symbol, offset); |
| 887 | if (ret == 0 && !is_return) |
| 888 | ctx.flags |= TPARG_FL_FENTRY; |
| 889 | /* Defer the ENOENT case until register kprobe */ |
| 890 | if (ret == -EINVAL && is_return) { |
| 891 | trace_probe_log_err(0, BAD_RETPROBE); |
| 892 | goto parse_error; |
| 893 | } |
| 894 | } |
| 895 | |
| 896 | if (symbol && !strchr(symbol, ':')) { |
| 897 | unsigned int count; |
| 898 | |
| 899 | count = number_of_same_symbols(symbol); |
| 900 | if (count > 1) { |
| 901 | /* |
| 902 | * Users should use ADDR to remove the ambiguity of |
| 903 | * using KSYM only. |
| 904 | */ |
| 905 | trace_probe_log_err(0, NON_UNIQ_SYMBOL); |
| 906 | ret = -EADDRNOTAVAIL; |
| 907 | |
| 908 | goto error; |
| 909 | } else if (count == 0) { |
| 910 | /* |
| 911 | * We can return ENOENT earlier than when register the |
| 912 | * kprobe. |
| 913 | */ |
| 914 | trace_probe_log_err(0, BAD_PROBE_ADDR); |
| 915 | ret = -ENOENT; |
| 916 | |
| 917 | goto error; |
| 918 | } |
| 919 | } |
| 920 | |
| 921 | trace_probe_log_set_index(0); |
| 922 | if (event) { |
| 923 | ret = traceprobe_parse_event_name(&event, &group, gbuf, |
| 924 | event - argv[0]); |
| 925 | if (ret) |
| 926 | goto parse_error; |
| 927 | } |
| 928 | |
| 929 | if (!event) { |
| 930 | /* Make a new event name */ |
| 931 | if (symbol) |
| 932 | snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld", |
| 933 | is_return ? 'r' : 'p', symbol, offset); |
| 934 | else |
| 935 | snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p", |
| 936 | is_return ? 'r' : 'p', addr); |
| 937 | sanitize_event_name(buf); |
| 938 | event = buf; |
| 939 | } |
| 940 | |
| 941 | argc -= 2; argv += 2; |
| 942 | ctx.funcname = symbol; |
| 943 | new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc, |
| 944 | abuf, MAX_BTF_ARGS_LEN, &ctx); |
| 945 | if (IS_ERR(new_argv)) { |
| 946 | ret = PTR_ERR(new_argv); |
| 947 | new_argv = NULL; |
| 948 | goto out; |
| 949 | } |
| 950 | if (new_argv) { |
| 951 | argc = new_argc; |
| 952 | argv = new_argv; |
| 953 | } |
| 954 | |
| 955 | ret = traceprobe_expand_dentry_args(argc, argv, &dbuf); |
| 956 | if (ret) |
| 957 | goto out; |
| 958 | |
| 959 | /* setup a probe */ |
| 960 | tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive, |
| 961 | argc, is_return); |
| 962 | if (IS_ERR(tk)) { |
| 963 | ret = PTR_ERR(tk); |
| 964 | /* This must return -ENOMEM, else there is a bug */ |
| 965 | WARN_ON_ONCE(ret != -ENOMEM); |
| 966 | goto out; /* We know tk is not allocated */ |
| 967 | } |
| 968 | |
| 969 | /* parse arguments */ |
| 970 | for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) { |
| 971 | trace_probe_log_set_index(i + 2); |
| 972 | ctx.offset = 0; |
| 973 | ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx); |
| 974 | if (ret) |
| 975 | goto error; /* This can be -ENOMEM */ |
| 976 | } |
| 977 | /* entry handler for kretprobe */ |
| 978 | if (is_return && tk->tp.entry_arg) { |
| 979 | tk->rp.entry_handler = trace_kprobe_entry_handler; |
| 980 | tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp); |
| 981 | } |
| 982 | |
| 983 | ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL; |
| 984 | ret = traceprobe_set_print_fmt(&tk->tp, ptype); |
| 985 | if (ret < 0) |
| 986 | goto error; |
| 987 | |
| 988 | ret = register_trace_kprobe(tk); |
| 989 | if (ret) { |
| 990 | trace_probe_log_set_index(1); |
| 991 | if (ret == -EILSEQ) |
| 992 | trace_probe_log_err(0, BAD_INSN_BNDRY); |
| 993 | else if (ret == -ENOENT) |
| 994 | trace_probe_log_err(0, BAD_PROBE_ADDR); |
| 995 | else if (ret != -ENOMEM && ret != -EEXIST) |
| 996 | trace_probe_log_err(0, FAIL_REG_PROBE); |
| 997 | goto error; |
| 998 | } |
| 999 | |
| 1000 | out: |
| 1001 | traceprobe_finish_parse(&ctx); |
| 1002 | trace_probe_log_clear(); |
| 1003 | kfree(new_argv); |
| 1004 | kfree(symbol); |
| 1005 | kfree(dbuf); |
| 1006 | return ret; |
| 1007 | |
| 1008 | parse_error: |
| 1009 | ret = -EINVAL; |
| 1010 | error: |
| 1011 | free_trace_kprobe(tk); |
| 1012 | goto out; |
| 1013 | } |
| 1014 | |
| 1015 | static int trace_kprobe_create(const char *raw_command) |
| 1016 | { |
| 1017 | return trace_probe_create(raw_command, __trace_kprobe_create); |
| 1018 | } |
| 1019 | |
| 1020 | static int create_or_delete_trace_kprobe(const char *raw_command) |
| 1021 | { |
| 1022 | int ret; |
| 1023 | |
| 1024 | if (raw_command[0] == '-') |
| 1025 | return dyn_event_release(raw_command, &trace_kprobe_ops); |
| 1026 | |
| 1027 | ret = trace_kprobe_create(raw_command); |
| 1028 | return ret == -ECANCELED ? -EINVAL : ret; |
| 1029 | } |
| 1030 | |
| 1031 | static int trace_kprobe_run_command(struct dynevent_cmd *cmd) |
| 1032 | { |
| 1033 | return create_or_delete_trace_kprobe(cmd->seq.buffer); |
| 1034 | } |
| 1035 | |
| 1036 | /** |
| 1037 | * kprobe_event_cmd_init - Initialize a kprobe event command object |
| 1038 | * @cmd: A pointer to the dynevent_cmd struct representing the new event |
| 1039 | * @buf: A pointer to the buffer used to build the command |
| 1040 | * @maxlen: The length of the buffer passed in @buf |
| 1041 | * |
| 1042 | * Initialize a synthetic event command object. Use this before |
| 1043 | * calling any of the other kprobe_event functions. |
| 1044 | */ |
| 1045 | void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen) |
| 1046 | { |
| 1047 | dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE, |
| 1048 | trace_kprobe_run_command); |
| 1049 | } |
| 1050 | EXPORT_SYMBOL_GPL(kprobe_event_cmd_init); |
| 1051 | |
| 1052 | /** |
| 1053 | * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list |
| 1054 | * @cmd: A pointer to the dynevent_cmd struct representing the new event |
| 1055 | * @kretprobe: Is this a return probe? |
| 1056 | * @name: The name of the kprobe event |
| 1057 | * @loc: The location of the kprobe event |
| 1058 | * @...: Variable number of arg (pairs), one pair for each field |
| 1059 | * |
| 1060 | * NOTE: Users normally won't want to call this function directly, but |
| 1061 | * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically |
| 1062 | * adds a NULL to the end of the arg list. If this function is used |
| 1063 | * directly, make sure the last arg in the variable arg list is NULL. |
| 1064 | * |
| 1065 | * Generate a kprobe event command to be executed by |
| 1066 | * kprobe_event_gen_cmd_end(). This function can be used to generate the |
| 1067 | * complete command or only the first part of it; in the latter case, |
| 1068 | * kprobe_event_add_fields() can be used to add more fields following this. |
| 1069 | * |
| 1070 | * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This |
| 1071 | * returns -EINVAL if @loc == NULL. |
| 1072 | * |
| 1073 | * Return: 0 if successful, error otherwise. |
| 1074 | */ |
| 1075 | int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe, |
| 1076 | const char *name, const char *loc, ...) |
| 1077 | { |
| 1078 | char buf[MAX_EVENT_NAME_LEN]; |
| 1079 | struct dynevent_arg arg; |
| 1080 | va_list args; |
| 1081 | int ret; |
| 1082 | |
| 1083 | if (cmd->type != DYNEVENT_TYPE_KPROBE) |
| 1084 | return -EINVAL; |
| 1085 | |
| 1086 | if (!loc) |
| 1087 | return -EINVAL; |
| 1088 | |
| 1089 | if (kretprobe) |
| 1090 | snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name); |
| 1091 | else |
| 1092 | snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name); |
| 1093 | |
| 1094 | ret = dynevent_str_add(cmd, buf); |
| 1095 | if (ret) |
| 1096 | return ret; |
| 1097 | |
| 1098 | dynevent_arg_init(&arg, 0); |
| 1099 | arg.str = loc; |
| 1100 | ret = dynevent_arg_add(cmd, &arg, NULL); |
| 1101 | if (ret) |
| 1102 | return ret; |
| 1103 | |
| 1104 | va_start(args, loc); |
| 1105 | for (;;) { |
| 1106 | const char *field; |
| 1107 | |
| 1108 | field = va_arg(args, const char *); |
| 1109 | if (!field) |
| 1110 | break; |
| 1111 | |
| 1112 | if (++cmd->n_fields > MAX_TRACE_ARGS) { |
| 1113 | ret = -EINVAL; |
| 1114 | break; |
| 1115 | } |
| 1116 | |
| 1117 | arg.str = field; |
| 1118 | ret = dynevent_arg_add(cmd, &arg, NULL); |
| 1119 | if (ret) |
| 1120 | break; |
| 1121 | } |
| 1122 | va_end(args); |
| 1123 | |
| 1124 | return ret; |
| 1125 | } |
| 1126 | EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start); |
| 1127 | |
| 1128 | /** |
| 1129 | * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list |
| 1130 | * @cmd: A pointer to the dynevent_cmd struct representing the new event |
| 1131 | * @...: Variable number of arg (pairs), one pair for each field |
| 1132 | * |
| 1133 | * NOTE: Users normally won't want to call this function directly, but |
| 1134 | * rather use the kprobe_event_add_fields() wrapper, which |
| 1135 | * automatically adds a NULL to the end of the arg list. If this |
| 1136 | * function is used directly, make sure the last arg in the variable |
| 1137 | * arg list is NULL. |
| 1138 | * |
| 1139 | * Add probe fields to an existing kprobe command using a variable |
| 1140 | * list of args. Fields are added in the same order they're listed. |
| 1141 | * |
| 1142 | * Return: 0 if successful, error otherwise. |
| 1143 | */ |
| 1144 | int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...) |
| 1145 | { |
| 1146 | struct dynevent_arg arg; |
| 1147 | va_list args; |
| 1148 | int ret = 0; |
| 1149 | |
| 1150 | if (cmd->type != DYNEVENT_TYPE_KPROBE) |
| 1151 | return -EINVAL; |
| 1152 | |
| 1153 | dynevent_arg_init(&arg, 0); |
| 1154 | |
| 1155 | va_start(args, cmd); |
| 1156 | for (;;) { |
| 1157 | const char *field; |
| 1158 | |
| 1159 | field = va_arg(args, const char *); |
| 1160 | if (!field) |
| 1161 | break; |
| 1162 | |
| 1163 | if (++cmd->n_fields > MAX_TRACE_ARGS) { |
| 1164 | ret = -EINVAL; |
| 1165 | break; |
| 1166 | } |
| 1167 | |
| 1168 | arg.str = field; |
| 1169 | ret = dynevent_arg_add(cmd, &arg, NULL); |
| 1170 | if (ret) |
| 1171 | break; |
| 1172 | } |
| 1173 | va_end(args); |
| 1174 | |
| 1175 | return ret; |
| 1176 | } |
| 1177 | EXPORT_SYMBOL_GPL(__kprobe_event_add_fields); |
| 1178 | |
| 1179 | /** |
| 1180 | * kprobe_event_delete - Delete a kprobe event |
| 1181 | * @name: The name of the kprobe event to delete |
| 1182 | * |
| 1183 | * Delete a kprobe event with the give @name from kernel code rather |
| 1184 | * than directly from the command line. |
| 1185 | * |
| 1186 | * Return: 0 if successful, error otherwise. |
| 1187 | */ |
| 1188 | int kprobe_event_delete(const char *name) |
| 1189 | { |
| 1190 | char buf[MAX_EVENT_NAME_LEN]; |
| 1191 | |
| 1192 | snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name); |
| 1193 | |
| 1194 | return create_or_delete_trace_kprobe(buf); |
| 1195 | } |
| 1196 | EXPORT_SYMBOL_GPL(kprobe_event_delete); |
| 1197 | |
| 1198 | static int trace_kprobe_release(struct dyn_event *ev) |
| 1199 | { |
| 1200 | struct trace_kprobe *tk = to_trace_kprobe(ev); |
| 1201 | int ret = unregister_trace_kprobe(tk); |
| 1202 | |
| 1203 | if (!ret) |
| 1204 | free_trace_kprobe(tk); |
| 1205 | return ret; |
| 1206 | } |
| 1207 | |
| 1208 | static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev) |
| 1209 | { |
| 1210 | struct trace_kprobe *tk = to_trace_kprobe(ev); |
| 1211 | int i; |
| 1212 | |
| 1213 | seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p'); |
| 1214 | if (trace_kprobe_is_return(tk) && tk->rp.maxactive) |
| 1215 | seq_printf(m, "%d", tk->rp.maxactive); |
| 1216 | seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp), |
| 1217 | trace_probe_name(&tk->tp)); |
| 1218 | |
| 1219 | if (!tk->symbol) |
| 1220 | seq_printf(m, " 0x%p", tk->rp.kp.addr); |
| 1221 | else if (tk->rp.kp.offset) |
| 1222 | seq_printf(m, " %s+%u", trace_kprobe_symbol(tk), |
| 1223 | tk->rp.kp.offset); |
| 1224 | else |
| 1225 | seq_printf(m, " %s", trace_kprobe_symbol(tk)); |
| 1226 | |
| 1227 | for (i = 0; i < tk->tp.nr_args; i++) |
| 1228 | seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm); |
| 1229 | seq_putc(m, '\n'); |
| 1230 | |
| 1231 | return 0; |
| 1232 | } |
| 1233 | |
| 1234 | static int probes_seq_show(struct seq_file *m, void *v) |
| 1235 | { |
| 1236 | struct dyn_event *ev = v; |
| 1237 | |
| 1238 | if (!is_trace_kprobe(ev)) |
| 1239 | return 0; |
| 1240 | |
| 1241 | return trace_kprobe_show(m, ev); |
| 1242 | } |
| 1243 | |
| 1244 | static const struct seq_operations probes_seq_op = { |
| 1245 | .start = dyn_event_seq_start, |
| 1246 | .next = dyn_event_seq_next, |
| 1247 | .stop = dyn_event_seq_stop, |
| 1248 | .show = probes_seq_show |
| 1249 | }; |
| 1250 | |
| 1251 | static int probes_open(struct inode *inode, struct file *file) |
| 1252 | { |
| 1253 | int ret; |
| 1254 | |
| 1255 | ret = security_locked_down(LOCKDOWN_TRACEFS); |
| 1256 | if (ret) |
| 1257 | return ret; |
| 1258 | |
| 1259 | if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) { |
| 1260 | ret = dyn_events_release_all(&trace_kprobe_ops); |
| 1261 | if (ret < 0) |
| 1262 | return ret; |
| 1263 | } |
| 1264 | |
| 1265 | return seq_open(file, &probes_seq_op); |
| 1266 | } |
| 1267 | |
| 1268 | static ssize_t probes_write(struct file *file, const char __user *buffer, |
| 1269 | size_t count, loff_t *ppos) |
| 1270 | { |
| 1271 | return trace_parse_run_command(file, buffer, count, ppos, |
| 1272 | create_or_delete_trace_kprobe); |
| 1273 | } |
| 1274 | |
| 1275 | static const struct file_operations kprobe_events_ops = { |
| 1276 | .owner = THIS_MODULE, |
| 1277 | .open = probes_open, |
| 1278 | .read = seq_read, |
| 1279 | .llseek = seq_lseek, |
| 1280 | .release = seq_release, |
| 1281 | .write = probes_write, |
| 1282 | }; |
| 1283 | |
| 1284 | static unsigned long trace_kprobe_missed(struct trace_kprobe *tk) |
| 1285 | { |
| 1286 | return trace_kprobe_is_return(tk) ? |
| 1287 | tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed; |
| 1288 | } |
| 1289 | |
| 1290 | /* Probes profiling interfaces */ |
| 1291 | static int probes_profile_seq_show(struct seq_file *m, void *v) |
| 1292 | { |
| 1293 | struct dyn_event *ev = v; |
| 1294 | struct trace_kprobe *tk; |
| 1295 | unsigned long nmissed; |
| 1296 | |
| 1297 | if (!is_trace_kprobe(ev)) |
| 1298 | return 0; |
| 1299 | |
| 1300 | tk = to_trace_kprobe(ev); |
| 1301 | nmissed = trace_kprobe_missed(tk); |
| 1302 | seq_printf(m, " %-44s %15lu %15lu\n", |
| 1303 | trace_probe_name(&tk->tp), |
| 1304 | trace_kprobe_nhit(tk), |
| 1305 | nmissed); |
| 1306 | |
| 1307 | return 0; |
| 1308 | } |
| 1309 | |
| 1310 | static const struct seq_operations profile_seq_op = { |
| 1311 | .start = dyn_event_seq_start, |
| 1312 | .next = dyn_event_seq_next, |
| 1313 | .stop = dyn_event_seq_stop, |
| 1314 | .show = probes_profile_seq_show |
| 1315 | }; |
| 1316 | |
| 1317 | static int profile_open(struct inode *inode, struct file *file) |
| 1318 | { |
| 1319 | int ret; |
| 1320 | |
| 1321 | ret = security_locked_down(LOCKDOWN_TRACEFS); |
| 1322 | if (ret) |
| 1323 | return ret; |
| 1324 | |
| 1325 | return seq_open(file, &profile_seq_op); |
| 1326 | } |
| 1327 | |
| 1328 | static const struct file_operations kprobe_profile_ops = { |
| 1329 | .owner = THIS_MODULE, |
| 1330 | .open = profile_open, |
| 1331 | .read = seq_read, |
| 1332 | .llseek = seq_lseek, |
| 1333 | .release = seq_release, |
| 1334 | }; |
| 1335 | |
| 1336 | /* Note that we don't verify it, since the code does not come from user space */ |
| 1337 | static int |
| 1338 | process_fetch_insn(struct fetch_insn *code, void *rec, void *edata, |
| 1339 | void *dest, void *base) |
| 1340 | { |
| 1341 | struct pt_regs *regs = rec; |
| 1342 | unsigned long val; |
| 1343 | int ret; |
| 1344 | |
| 1345 | retry: |
| 1346 | /* 1st stage: get value from context */ |
| 1347 | switch (code->op) { |
| 1348 | case FETCH_OP_REG: |
| 1349 | val = regs_get_register(regs, code->param); |
| 1350 | break; |
| 1351 | case FETCH_OP_STACK: |
| 1352 | val = regs_get_kernel_stack_nth(regs, code->param); |
| 1353 | break; |
| 1354 | case FETCH_OP_STACKP: |
| 1355 | val = kernel_stack_pointer(regs); |
| 1356 | break; |
| 1357 | case FETCH_OP_RETVAL: |
| 1358 | val = regs_return_value(regs); |
| 1359 | break; |
| 1360 | #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API |
| 1361 | case FETCH_OP_ARG: |
| 1362 | val = regs_get_kernel_argument(regs, code->param); |
| 1363 | break; |
| 1364 | case FETCH_OP_EDATA: |
| 1365 | val = *(unsigned long *)((unsigned long)edata + code->offset); |
| 1366 | break; |
| 1367 | #endif |
| 1368 | case FETCH_NOP_SYMBOL: /* Ignore a place holder */ |
| 1369 | code++; |
| 1370 | goto retry; |
| 1371 | default: |
| 1372 | ret = process_common_fetch_insn(code, &val); |
| 1373 | if (ret < 0) |
| 1374 | return ret; |
| 1375 | } |
| 1376 | code++; |
| 1377 | |
| 1378 | return process_fetch_insn_bottom(code, val, dest, base); |
| 1379 | } |
| 1380 | NOKPROBE_SYMBOL(process_fetch_insn) |
| 1381 | |
| 1382 | /* Kprobe handler */ |
| 1383 | static nokprobe_inline void |
| 1384 | __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs, |
| 1385 | struct trace_event_file *trace_file) |
| 1386 | { |
| 1387 | struct kprobe_trace_entry_head *entry; |
| 1388 | struct trace_event_call *call = trace_probe_event_call(&tk->tp); |
| 1389 | struct trace_event_buffer fbuffer; |
| 1390 | int dsize; |
| 1391 | |
| 1392 | WARN_ON(call != trace_file->event_call); |
| 1393 | |
| 1394 | if (trace_trigger_soft_disabled(trace_file)) |
| 1395 | return; |
| 1396 | |
| 1397 | dsize = __get_data_size(&tk->tp, regs, NULL); |
| 1398 | |
| 1399 | entry = trace_event_buffer_reserve(&fbuffer, trace_file, |
| 1400 | sizeof(*entry) + tk->tp.size + dsize); |
| 1401 | if (!entry) |
| 1402 | return; |
| 1403 | |
| 1404 | fbuffer.regs = regs; |
| 1405 | entry->ip = (unsigned long)tk->rp.kp.addr; |
| 1406 | store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize); |
| 1407 | |
| 1408 | trace_event_buffer_commit(&fbuffer); |
| 1409 | } |
| 1410 | |
| 1411 | static void |
| 1412 | kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs) |
| 1413 | { |
| 1414 | struct event_file_link *link; |
| 1415 | |
| 1416 | trace_probe_for_each_link_rcu(link, &tk->tp) |
| 1417 | __kprobe_trace_func(tk, regs, link->file); |
| 1418 | } |
| 1419 | NOKPROBE_SYMBOL(kprobe_trace_func); |
| 1420 | |
| 1421 | /* Kretprobe handler */ |
| 1422 | |
| 1423 | static int trace_kprobe_entry_handler(struct kretprobe_instance *ri, |
| 1424 | struct pt_regs *regs) |
| 1425 | { |
| 1426 | struct kretprobe *rp = get_kretprobe(ri); |
| 1427 | struct trace_kprobe *tk; |
| 1428 | |
| 1429 | /* |
| 1430 | * There is a small chance that get_kretprobe(ri) returns NULL when |
| 1431 | * the kretprobe is unregister on another CPU between kretprobe's |
| 1432 | * trampoline_handler and this function. |
| 1433 | */ |
| 1434 | if (unlikely(!rp)) |
| 1435 | return -ENOENT; |
| 1436 | |
| 1437 | tk = container_of(rp, struct trace_kprobe, rp); |
| 1438 | |
| 1439 | /* store argument values into ri->data as entry data */ |
| 1440 | if (tk->tp.entry_arg) |
| 1441 | store_trace_entry_data(ri->data, &tk->tp, regs); |
| 1442 | |
| 1443 | return 0; |
| 1444 | } |
| 1445 | |
| 1446 | |
| 1447 | static nokprobe_inline void |
| 1448 | __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri, |
| 1449 | struct pt_regs *regs, |
| 1450 | struct trace_event_file *trace_file) |
| 1451 | { |
| 1452 | struct kretprobe_trace_entry_head *entry; |
| 1453 | struct trace_event_buffer fbuffer; |
| 1454 | struct trace_event_call *call = trace_probe_event_call(&tk->tp); |
| 1455 | int dsize; |
| 1456 | |
| 1457 | WARN_ON(call != trace_file->event_call); |
| 1458 | |
| 1459 | if (trace_trigger_soft_disabled(trace_file)) |
| 1460 | return; |
| 1461 | |
| 1462 | dsize = __get_data_size(&tk->tp, regs, ri->data); |
| 1463 | |
| 1464 | entry = trace_event_buffer_reserve(&fbuffer, trace_file, |
| 1465 | sizeof(*entry) + tk->tp.size + dsize); |
| 1466 | if (!entry) |
| 1467 | return; |
| 1468 | |
| 1469 | fbuffer.regs = regs; |
| 1470 | entry->func = (unsigned long)tk->rp.kp.addr; |
| 1471 | entry->ret_ip = get_kretprobe_retaddr(ri); |
| 1472 | store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize); |
| 1473 | |
| 1474 | trace_event_buffer_commit(&fbuffer); |
| 1475 | } |
| 1476 | |
| 1477 | static void |
| 1478 | kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri, |
| 1479 | struct pt_regs *regs) |
| 1480 | { |
| 1481 | struct event_file_link *link; |
| 1482 | |
| 1483 | trace_probe_for_each_link_rcu(link, &tk->tp) |
| 1484 | __kretprobe_trace_func(tk, ri, regs, link->file); |
| 1485 | } |
| 1486 | NOKPROBE_SYMBOL(kretprobe_trace_func); |
| 1487 | |
| 1488 | /* Event entry printers */ |
| 1489 | static enum print_line_t |
| 1490 | print_kprobe_event(struct trace_iterator *iter, int flags, |
| 1491 | struct trace_event *event) |
| 1492 | { |
| 1493 | struct kprobe_trace_entry_head *field; |
| 1494 | struct trace_seq *s = &iter->seq; |
| 1495 | struct trace_probe *tp; |
| 1496 | |
| 1497 | field = (struct kprobe_trace_entry_head *)iter->ent; |
| 1498 | tp = trace_probe_primary_from_call( |
| 1499 | container_of(event, struct trace_event_call, event)); |
| 1500 | if (WARN_ON_ONCE(!tp)) |
| 1501 | goto out; |
| 1502 | |
| 1503 | trace_seq_printf(s, "%s: (", trace_probe_name(tp)); |
| 1504 | |
| 1505 | if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET)) |
| 1506 | goto out; |
| 1507 | |
| 1508 | trace_seq_putc(s, ')'); |
| 1509 | |
| 1510 | if (trace_probe_print_args(s, tp->args, tp->nr_args, |
| 1511 | (u8 *)&field[1], field) < 0) |
| 1512 | goto out; |
| 1513 | |
| 1514 | trace_seq_putc(s, '\n'); |
| 1515 | out: |
| 1516 | return trace_handle_return(s); |
| 1517 | } |
| 1518 | |
| 1519 | static enum print_line_t |
| 1520 | print_kretprobe_event(struct trace_iterator *iter, int flags, |
| 1521 | struct trace_event *event) |
| 1522 | { |
| 1523 | struct kretprobe_trace_entry_head *field; |
| 1524 | struct trace_seq *s = &iter->seq; |
| 1525 | struct trace_probe *tp; |
| 1526 | |
| 1527 | field = (struct kretprobe_trace_entry_head *)iter->ent; |
| 1528 | tp = trace_probe_primary_from_call( |
| 1529 | container_of(event, struct trace_event_call, event)); |
| 1530 | if (WARN_ON_ONCE(!tp)) |
| 1531 | goto out; |
| 1532 | |
| 1533 | trace_seq_printf(s, "%s: (", trace_probe_name(tp)); |
| 1534 | |
| 1535 | if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET)) |
| 1536 | goto out; |
| 1537 | |
| 1538 | trace_seq_puts(s, " <- "); |
| 1539 | |
| 1540 | if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET)) |
| 1541 | goto out; |
| 1542 | |
| 1543 | trace_seq_putc(s, ')'); |
| 1544 | |
| 1545 | if (trace_probe_print_args(s, tp->args, tp->nr_args, |
| 1546 | (u8 *)&field[1], field) < 0) |
| 1547 | goto out; |
| 1548 | |
| 1549 | trace_seq_putc(s, '\n'); |
| 1550 | |
| 1551 | out: |
| 1552 | return trace_handle_return(s); |
| 1553 | } |
| 1554 | |
| 1555 | |
| 1556 | static int kprobe_event_define_fields(struct trace_event_call *event_call) |
| 1557 | { |
| 1558 | int ret; |
| 1559 | struct kprobe_trace_entry_head field; |
| 1560 | struct trace_probe *tp; |
| 1561 | |
| 1562 | tp = trace_probe_primary_from_call(event_call); |
| 1563 | if (WARN_ON_ONCE(!tp)) |
| 1564 | return -ENOENT; |
| 1565 | |
| 1566 | DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0); |
| 1567 | |
| 1568 | return traceprobe_define_arg_fields(event_call, sizeof(field), tp); |
| 1569 | } |
| 1570 | |
| 1571 | static int kretprobe_event_define_fields(struct trace_event_call *event_call) |
| 1572 | { |
| 1573 | int ret; |
| 1574 | struct kretprobe_trace_entry_head field; |
| 1575 | struct trace_probe *tp; |
| 1576 | |
| 1577 | tp = trace_probe_primary_from_call(event_call); |
| 1578 | if (WARN_ON_ONCE(!tp)) |
| 1579 | return -ENOENT; |
| 1580 | |
| 1581 | DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0); |
| 1582 | DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0); |
| 1583 | |
| 1584 | return traceprobe_define_arg_fields(event_call, sizeof(field), tp); |
| 1585 | } |
| 1586 | |
| 1587 | #ifdef CONFIG_PERF_EVENTS |
| 1588 | |
| 1589 | /* Kprobe profile handler */ |
| 1590 | static int |
| 1591 | kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs) |
| 1592 | { |
| 1593 | struct trace_event_call *call = trace_probe_event_call(&tk->tp); |
| 1594 | struct kprobe_trace_entry_head *entry; |
| 1595 | struct hlist_head *head; |
| 1596 | int size, __size, dsize; |
| 1597 | int rctx; |
| 1598 | |
| 1599 | if (bpf_prog_array_valid(call)) { |
| 1600 | unsigned long orig_ip = instruction_pointer(regs); |
| 1601 | int ret; |
| 1602 | |
| 1603 | ret = trace_call_bpf(call, regs); |
| 1604 | |
| 1605 | /* |
| 1606 | * We need to check and see if we modified the pc of the |
| 1607 | * pt_regs, and if so return 1 so that we don't do the |
| 1608 | * single stepping. |
| 1609 | */ |
| 1610 | if (orig_ip != instruction_pointer(regs)) |
| 1611 | return 1; |
| 1612 | if (!ret) |
| 1613 | return 0; |
| 1614 | } |
| 1615 | |
| 1616 | head = this_cpu_ptr(call->perf_events); |
| 1617 | if (hlist_empty(head)) |
| 1618 | return 0; |
| 1619 | |
| 1620 | dsize = __get_data_size(&tk->tp, regs, NULL); |
| 1621 | __size = sizeof(*entry) + tk->tp.size + dsize; |
| 1622 | size = ALIGN(__size + sizeof(u32), sizeof(u64)); |
| 1623 | size -= sizeof(u32); |
| 1624 | |
| 1625 | entry = perf_trace_buf_alloc(size, NULL, &rctx); |
| 1626 | if (!entry) |
| 1627 | return 0; |
| 1628 | |
| 1629 | entry->ip = (unsigned long)tk->rp.kp.addr; |
| 1630 | memset(&entry[1], 0, dsize); |
| 1631 | store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize); |
| 1632 | perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs, |
| 1633 | head, NULL); |
| 1634 | return 0; |
| 1635 | } |
| 1636 | NOKPROBE_SYMBOL(kprobe_perf_func); |
| 1637 | |
| 1638 | /* Kretprobe profile handler */ |
| 1639 | static void |
| 1640 | kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri, |
| 1641 | struct pt_regs *regs) |
| 1642 | { |
| 1643 | struct trace_event_call *call = trace_probe_event_call(&tk->tp); |
| 1644 | struct kretprobe_trace_entry_head *entry; |
| 1645 | struct hlist_head *head; |
| 1646 | int size, __size, dsize; |
| 1647 | int rctx; |
| 1648 | |
| 1649 | if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs)) |
| 1650 | return; |
| 1651 | |
| 1652 | head = this_cpu_ptr(call->perf_events); |
| 1653 | if (hlist_empty(head)) |
| 1654 | return; |
| 1655 | |
| 1656 | dsize = __get_data_size(&tk->tp, regs, ri->data); |
| 1657 | __size = sizeof(*entry) + tk->tp.size + dsize; |
| 1658 | size = ALIGN(__size + sizeof(u32), sizeof(u64)); |
| 1659 | size -= sizeof(u32); |
| 1660 | |
| 1661 | entry = perf_trace_buf_alloc(size, NULL, &rctx); |
| 1662 | if (!entry) |
| 1663 | return; |
| 1664 | |
| 1665 | entry->func = (unsigned long)tk->rp.kp.addr; |
| 1666 | entry->ret_ip = get_kretprobe_retaddr(ri); |
| 1667 | store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize); |
| 1668 | perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs, |
| 1669 | head, NULL); |
| 1670 | } |
| 1671 | NOKPROBE_SYMBOL(kretprobe_perf_func); |
| 1672 | |
| 1673 | int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type, |
| 1674 | const char **symbol, u64 *probe_offset, |
| 1675 | u64 *probe_addr, unsigned long *missed, |
| 1676 | bool perf_type_tracepoint) |
| 1677 | { |
| 1678 | const char *pevent = trace_event_name(event->tp_event); |
| 1679 | const char *group = event->tp_event->class->system; |
| 1680 | struct trace_kprobe *tk; |
| 1681 | |
| 1682 | if (perf_type_tracepoint) |
| 1683 | tk = find_trace_kprobe(pevent, group); |
| 1684 | else |
| 1685 | tk = trace_kprobe_primary_from_call(event->tp_event); |
| 1686 | if (!tk) |
| 1687 | return -EINVAL; |
| 1688 | |
| 1689 | *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE |
| 1690 | : BPF_FD_TYPE_KPROBE; |
| 1691 | *probe_offset = tk->rp.kp.offset; |
| 1692 | *probe_addr = kallsyms_show_value(current_cred()) ? |
| 1693 | (unsigned long)tk->rp.kp.addr : 0; |
| 1694 | *symbol = tk->symbol; |
| 1695 | if (missed) |
| 1696 | *missed = trace_kprobe_missed(tk); |
| 1697 | return 0; |
| 1698 | } |
| 1699 | #endif /* CONFIG_PERF_EVENTS */ |
| 1700 | |
| 1701 | /* |
| 1702 | * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex. |
| 1703 | * |
| 1704 | * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe |
| 1705 | * lockless, but we can't race with this __init function. |
| 1706 | */ |
| 1707 | static int kprobe_register(struct trace_event_call *event, |
| 1708 | enum trace_reg type, void *data) |
| 1709 | { |
| 1710 | struct trace_event_file *file = data; |
| 1711 | |
| 1712 | switch (type) { |
| 1713 | case TRACE_REG_REGISTER: |
| 1714 | return enable_trace_kprobe(event, file); |
| 1715 | case TRACE_REG_UNREGISTER: |
| 1716 | return disable_trace_kprobe(event, file); |
| 1717 | |
| 1718 | #ifdef CONFIG_PERF_EVENTS |
| 1719 | case TRACE_REG_PERF_REGISTER: |
| 1720 | return enable_trace_kprobe(event, NULL); |
| 1721 | case TRACE_REG_PERF_UNREGISTER: |
| 1722 | return disable_trace_kprobe(event, NULL); |
| 1723 | case TRACE_REG_PERF_OPEN: |
| 1724 | case TRACE_REG_PERF_CLOSE: |
| 1725 | case TRACE_REG_PERF_ADD: |
| 1726 | case TRACE_REG_PERF_DEL: |
| 1727 | return 0; |
| 1728 | #endif |
| 1729 | } |
| 1730 | return 0; |
| 1731 | } |
| 1732 | |
| 1733 | static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs) |
| 1734 | { |
| 1735 | struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp); |
| 1736 | int ret = 0; |
| 1737 | |
| 1738 | raw_cpu_inc(*tk->nhit); |
| 1739 | |
| 1740 | if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE)) |
| 1741 | kprobe_trace_func(tk, regs); |
| 1742 | #ifdef CONFIG_PERF_EVENTS |
| 1743 | if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE)) |
| 1744 | ret = kprobe_perf_func(tk, regs); |
| 1745 | #endif |
| 1746 | return ret; |
| 1747 | } |
| 1748 | NOKPROBE_SYMBOL(kprobe_dispatcher); |
| 1749 | |
| 1750 | static int |
| 1751 | kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs) |
| 1752 | { |
| 1753 | struct kretprobe *rp = get_kretprobe(ri); |
| 1754 | struct trace_kprobe *tk; |
| 1755 | |
| 1756 | /* |
| 1757 | * There is a small chance that get_kretprobe(ri) returns NULL when |
| 1758 | * the kretprobe is unregister on another CPU between kretprobe's |
| 1759 | * trampoline_handler and this function. |
| 1760 | */ |
| 1761 | if (unlikely(!rp)) |
| 1762 | return 0; |
| 1763 | |
| 1764 | tk = container_of(rp, struct trace_kprobe, rp); |
| 1765 | raw_cpu_inc(*tk->nhit); |
| 1766 | |
| 1767 | if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE)) |
| 1768 | kretprobe_trace_func(tk, ri, regs); |
| 1769 | #ifdef CONFIG_PERF_EVENTS |
| 1770 | if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE)) |
| 1771 | kretprobe_perf_func(tk, ri, regs); |
| 1772 | #endif |
| 1773 | return 0; /* We don't tweak kernel, so just return 0 */ |
| 1774 | } |
| 1775 | NOKPROBE_SYMBOL(kretprobe_dispatcher); |
| 1776 | |
| 1777 | static struct trace_event_functions kretprobe_funcs = { |
| 1778 | .trace = print_kretprobe_event |
| 1779 | }; |
| 1780 | |
| 1781 | static struct trace_event_functions kprobe_funcs = { |
| 1782 | .trace = print_kprobe_event |
| 1783 | }; |
| 1784 | |
| 1785 | static struct trace_event_fields kretprobe_fields_array[] = { |
| 1786 | { .type = TRACE_FUNCTION_TYPE, |
| 1787 | .define_fields = kretprobe_event_define_fields }, |
| 1788 | {} |
| 1789 | }; |
| 1790 | |
| 1791 | static struct trace_event_fields kprobe_fields_array[] = { |
| 1792 | { .type = TRACE_FUNCTION_TYPE, |
| 1793 | .define_fields = kprobe_event_define_fields }, |
| 1794 | {} |
| 1795 | }; |
| 1796 | |
| 1797 | static inline void init_trace_event_call(struct trace_kprobe *tk) |
| 1798 | { |
| 1799 | struct trace_event_call *call = trace_probe_event_call(&tk->tp); |
| 1800 | |
| 1801 | if (trace_kprobe_is_return(tk)) { |
| 1802 | call->event.funcs = &kretprobe_funcs; |
| 1803 | call->class->fields_array = kretprobe_fields_array; |
| 1804 | } else { |
| 1805 | call->event.funcs = &kprobe_funcs; |
| 1806 | call->class->fields_array = kprobe_fields_array; |
| 1807 | } |
| 1808 | |
| 1809 | call->flags = TRACE_EVENT_FL_KPROBE; |
| 1810 | call->class->reg = kprobe_register; |
| 1811 | } |
| 1812 | |
| 1813 | static int register_kprobe_event(struct trace_kprobe *tk) |
| 1814 | { |
| 1815 | init_trace_event_call(tk); |
| 1816 | |
| 1817 | return trace_probe_register_event_call(&tk->tp); |
| 1818 | } |
| 1819 | |
| 1820 | static int unregister_kprobe_event(struct trace_kprobe *tk) |
| 1821 | { |
| 1822 | return trace_probe_unregister_event_call(&tk->tp); |
| 1823 | } |
| 1824 | |
| 1825 | #ifdef CONFIG_PERF_EVENTS |
| 1826 | |
| 1827 | /* create a trace_kprobe, but don't add it to global lists */ |
| 1828 | struct trace_event_call * |
| 1829 | create_local_trace_kprobe(char *func, void *addr, unsigned long offs, |
| 1830 | bool is_return) |
| 1831 | { |
| 1832 | enum probe_print_type ptype; |
| 1833 | struct trace_kprobe *tk; |
| 1834 | int ret; |
| 1835 | char *event; |
| 1836 | |
| 1837 | if (func) { |
| 1838 | unsigned int count; |
| 1839 | |
| 1840 | count = number_of_same_symbols(func); |
| 1841 | if (count > 1) |
| 1842 | /* |
| 1843 | * Users should use addr to remove the ambiguity of |
| 1844 | * using func only. |
| 1845 | */ |
| 1846 | return ERR_PTR(-EADDRNOTAVAIL); |
| 1847 | else if (count == 0) |
| 1848 | /* |
| 1849 | * We can return ENOENT earlier than when register the |
| 1850 | * kprobe. |
| 1851 | */ |
| 1852 | return ERR_PTR(-ENOENT); |
| 1853 | } |
| 1854 | |
| 1855 | /* |
| 1856 | * local trace_kprobes are not added to dyn_event, so they are never |
| 1857 | * searched in find_trace_kprobe(). Therefore, there is no concern of |
| 1858 | * duplicated name here. |
| 1859 | */ |
| 1860 | event = func ? func : "DUMMY_EVENT"; |
| 1861 | |
| 1862 | tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func, |
| 1863 | offs, 0 /* maxactive */, 0 /* nargs */, |
| 1864 | is_return); |
| 1865 | |
| 1866 | if (IS_ERR(tk)) { |
| 1867 | pr_info("Failed to allocate trace_probe.(%d)\n", |
| 1868 | (int)PTR_ERR(tk)); |
| 1869 | return ERR_CAST(tk); |
| 1870 | } |
| 1871 | |
| 1872 | init_trace_event_call(tk); |
| 1873 | |
| 1874 | ptype = trace_kprobe_is_return(tk) ? |
| 1875 | PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL; |
| 1876 | if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) { |
| 1877 | ret = -ENOMEM; |
| 1878 | goto error; |
| 1879 | } |
| 1880 | |
| 1881 | ret = __register_trace_kprobe(tk); |
| 1882 | if (ret < 0) |
| 1883 | goto error; |
| 1884 | |
| 1885 | return trace_probe_event_call(&tk->tp); |
| 1886 | error: |
| 1887 | free_trace_kprobe(tk); |
| 1888 | return ERR_PTR(ret); |
| 1889 | } |
| 1890 | |
| 1891 | void destroy_local_trace_kprobe(struct trace_event_call *event_call) |
| 1892 | { |
| 1893 | struct trace_kprobe *tk; |
| 1894 | |
| 1895 | tk = trace_kprobe_primary_from_call(event_call); |
| 1896 | if (unlikely(!tk)) |
| 1897 | return; |
| 1898 | |
| 1899 | if (trace_probe_is_enabled(&tk->tp)) { |
| 1900 | WARN_ON(1); |
| 1901 | return; |
| 1902 | } |
| 1903 | |
| 1904 | __unregister_trace_kprobe(tk); |
| 1905 | |
| 1906 | free_trace_kprobe(tk); |
| 1907 | } |
| 1908 | #endif /* CONFIG_PERF_EVENTS */ |
| 1909 | |
| 1910 | static __init void enable_boot_kprobe_events(void) |
| 1911 | { |
| 1912 | struct trace_array *tr = top_trace_array(); |
| 1913 | struct trace_event_file *file; |
| 1914 | struct trace_kprobe *tk; |
| 1915 | struct dyn_event *pos; |
| 1916 | |
| 1917 | mutex_lock(&event_mutex); |
| 1918 | for_each_trace_kprobe(tk, pos) { |
| 1919 | list_for_each_entry(file, &tr->events, list) |
| 1920 | if (file->event_call == trace_probe_event_call(&tk->tp)) |
| 1921 | trace_event_enable_disable(file, 1, 0); |
| 1922 | } |
| 1923 | mutex_unlock(&event_mutex); |
| 1924 | } |
| 1925 | |
| 1926 | static __init void setup_boot_kprobe_events(void) |
| 1927 | { |
| 1928 | char *p, *cmd = kprobe_boot_events_buf; |
| 1929 | int ret; |
| 1930 | |
| 1931 | strreplace(kprobe_boot_events_buf, ',', ' '); |
| 1932 | |
| 1933 | while (cmd && *cmd != '\0') { |
| 1934 | p = strchr(cmd, ';'); |
| 1935 | if (p) |
| 1936 | *p++ = '\0'; |
| 1937 | |
| 1938 | ret = create_or_delete_trace_kprobe(cmd); |
| 1939 | if (ret) |
| 1940 | pr_warn("Failed to add event(%d): %s\n", ret, cmd); |
| 1941 | |
| 1942 | cmd = p; |
| 1943 | } |
| 1944 | |
| 1945 | enable_boot_kprobe_events(); |
| 1946 | } |
| 1947 | |
| 1948 | /* |
| 1949 | * Register dynevent at core_initcall. This allows kernel to setup kprobe |
| 1950 | * events in postcore_initcall without tracefs. |
| 1951 | */ |
| 1952 | static __init int init_kprobe_trace_early(void) |
| 1953 | { |
| 1954 | int ret; |
| 1955 | |
| 1956 | ret = dyn_event_register(&trace_kprobe_ops); |
| 1957 | if (ret) |
| 1958 | return ret; |
| 1959 | |
| 1960 | if (trace_kprobe_register_module_notifier()) |
| 1961 | return -EINVAL; |
| 1962 | |
| 1963 | return 0; |
| 1964 | } |
| 1965 | core_initcall(init_kprobe_trace_early); |
| 1966 | |
| 1967 | /* Make a tracefs interface for controlling probe points */ |
| 1968 | static __init int init_kprobe_trace(void) |
| 1969 | { |
| 1970 | int ret; |
| 1971 | |
| 1972 | ret = tracing_init_dentry(); |
| 1973 | if (ret) |
| 1974 | return 0; |
| 1975 | |
| 1976 | /* Event list interface */ |
| 1977 | trace_create_file("kprobe_events", TRACE_MODE_WRITE, |
| 1978 | NULL, NULL, &kprobe_events_ops); |
| 1979 | |
| 1980 | /* Profile interface */ |
| 1981 | trace_create_file("kprobe_profile", TRACE_MODE_READ, |
| 1982 | NULL, NULL, &kprobe_profile_ops); |
| 1983 | |
| 1984 | setup_boot_kprobe_events(); |
| 1985 | |
| 1986 | return 0; |
| 1987 | } |
| 1988 | fs_initcall(init_kprobe_trace); |
| 1989 | |
| 1990 | |
| 1991 | #ifdef CONFIG_FTRACE_STARTUP_TEST |
| 1992 | static __init struct trace_event_file * |
| 1993 | find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr) |
| 1994 | { |
| 1995 | struct trace_event_file *file; |
| 1996 | |
| 1997 | list_for_each_entry(file, &tr->events, list) |
| 1998 | if (file->event_call == trace_probe_event_call(&tk->tp)) |
| 1999 | return file; |
| 2000 | |
| 2001 | return NULL; |
| 2002 | } |
| 2003 | |
| 2004 | /* |
| 2005 | * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this |
| 2006 | * stage, we can do this lockless. |
| 2007 | */ |
| 2008 | static __init int kprobe_trace_self_tests_init(void) |
| 2009 | { |
| 2010 | int ret, warn = 0; |
| 2011 | int (*target)(int, int, int, int, int, int); |
| 2012 | struct trace_kprobe *tk; |
| 2013 | struct trace_event_file *file; |
| 2014 | |
| 2015 | if (tracing_is_disabled()) |
| 2016 | return -ENODEV; |
| 2017 | |
| 2018 | if (tracing_selftest_disabled) |
| 2019 | return 0; |
| 2020 | |
| 2021 | target = kprobe_trace_selftest_target; |
| 2022 | |
| 2023 | pr_info("Testing kprobe tracing: "); |
| 2024 | |
| 2025 | ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)"); |
| 2026 | if (WARN_ON_ONCE(ret)) { |
| 2027 | pr_warn("error on probing function entry.\n"); |
| 2028 | warn++; |
| 2029 | } else { |
| 2030 | /* Enable trace point */ |
| 2031 | tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM); |
| 2032 | if (WARN_ON_ONCE(tk == NULL)) { |
| 2033 | pr_warn("error on getting new probe.\n"); |
| 2034 | warn++; |
| 2035 | } else { |
| 2036 | file = find_trace_probe_file(tk, top_trace_array()); |
| 2037 | if (WARN_ON_ONCE(file == NULL)) { |
| 2038 | pr_warn("error on getting probe file.\n"); |
| 2039 | warn++; |
| 2040 | } else |
| 2041 | enable_trace_kprobe( |
| 2042 | trace_probe_event_call(&tk->tp), file); |
| 2043 | } |
| 2044 | } |
| 2045 | |
| 2046 | ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval"); |
| 2047 | if (WARN_ON_ONCE(ret)) { |
| 2048 | pr_warn("error on probing function return.\n"); |
| 2049 | warn++; |
| 2050 | } else { |
| 2051 | /* Enable trace point */ |
| 2052 | tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM); |
| 2053 | if (WARN_ON_ONCE(tk == NULL)) { |
| 2054 | pr_warn("error on getting 2nd new probe.\n"); |
| 2055 | warn++; |
| 2056 | } else { |
| 2057 | file = find_trace_probe_file(tk, top_trace_array()); |
| 2058 | if (WARN_ON_ONCE(file == NULL)) { |
| 2059 | pr_warn("error on getting probe file.\n"); |
| 2060 | warn++; |
| 2061 | } else |
| 2062 | enable_trace_kprobe( |
| 2063 | trace_probe_event_call(&tk->tp), file); |
| 2064 | } |
| 2065 | } |
| 2066 | |
| 2067 | if (warn) |
| 2068 | goto end; |
| 2069 | |
| 2070 | ret = target(1, 2, 3, 4, 5, 6); |
| 2071 | |
| 2072 | /* |
| 2073 | * Not expecting an error here, the check is only to prevent the |
| 2074 | * optimizer from removing the call to target() as otherwise there |
| 2075 | * are no side-effects and the call is never performed. |
| 2076 | */ |
| 2077 | if (ret != 21) |
| 2078 | warn++; |
| 2079 | |
| 2080 | /* Disable trace points before removing it */ |
| 2081 | tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM); |
| 2082 | if (WARN_ON_ONCE(tk == NULL)) { |
| 2083 | pr_warn("error on getting test probe.\n"); |
| 2084 | warn++; |
| 2085 | } else { |
| 2086 | if (trace_kprobe_nhit(tk) != 1) { |
| 2087 | pr_warn("incorrect number of testprobe hits\n"); |
| 2088 | warn++; |
| 2089 | } |
| 2090 | |
| 2091 | file = find_trace_probe_file(tk, top_trace_array()); |
| 2092 | if (WARN_ON_ONCE(file == NULL)) { |
| 2093 | pr_warn("error on getting probe file.\n"); |
| 2094 | warn++; |
| 2095 | } else |
| 2096 | disable_trace_kprobe( |
| 2097 | trace_probe_event_call(&tk->tp), file); |
| 2098 | } |
| 2099 | |
| 2100 | tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM); |
| 2101 | if (WARN_ON_ONCE(tk == NULL)) { |
| 2102 | pr_warn("error on getting 2nd test probe.\n"); |
| 2103 | warn++; |
| 2104 | } else { |
| 2105 | if (trace_kprobe_nhit(tk) != 1) { |
| 2106 | pr_warn("incorrect number of testprobe2 hits\n"); |
| 2107 | warn++; |
| 2108 | } |
| 2109 | |
| 2110 | file = find_trace_probe_file(tk, top_trace_array()); |
| 2111 | if (WARN_ON_ONCE(file == NULL)) { |
| 2112 | pr_warn("error on getting probe file.\n"); |
| 2113 | warn++; |
| 2114 | } else |
| 2115 | disable_trace_kprobe( |
| 2116 | trace_probe_event_call(&tk->tp), file); |
| 2117 | } |
| 2118 | |
| 2119 | ret = create_or_delete_trace_kprobe("-:testprobe"); |
| 2120 | if (WARN_ON_ONCE(ret)) { |
| 2121 | pr_warn("error on deleting a probe.\n"); |
| 2122 | warn++; |
| 2123 | } |
| 2124 | |
| 2125 | ret = create_or_delete_trace_kprobe("-:testprobe2"); |
| 2126 | if (WARN_ON_ONCE(ret)) { |
| 2127 | pr_warn("error on deleting a probe.\n"); |
| 2128 | warn++; |
| 2129 | } |
| 2130 | |
| 2131 | end: |
| 2132 | ret = dyn_events_release_all(&trace_kprobe_ops); |
| 2133 | if (WARN_ON_ONCE(ret)) { |
| 2134 | pr_warn("error on cleaning up probes.\n"); |
| 2135 | warn++; |
| 2136 | } |
| 2137 | /* |
| 2138 | * Wait for the optimizer work to finish. Otherwise it might fiddle |
| 2139 | * with probes in already freed __init text. |
| 2140 | */ |
| 2141 | wait_for_kprobe_optimizer(); |
| 2142 | if (warn) |
| 2143 | pr_cont("NG: Some tests are failed. Please check them.\n"); |
| 2144 | else |
| 2145 | pr_cont("OK\n"); |
| 2146 | return 0; |
| 2147 | } |
| 2148 | |
| 2149 | late_initcall(kprobe_trace_self_tests_init); |
| 2150 | |
| 2151 | #endif |