kernel/trace/bpf

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
bpf_get_raw_tracepoint_module
bpf_get_raw_tracepoint_module
trace_call_bpf
BPF_CALL_2
BPF_CALL_3
BPF_CALL_3
bpf_get_probe_write_proto
BPF_CALL_5
bpf_get_trace_printk_proto
get_map_perf_counter
BPF_CALL_2
BPF_CALL_4
__bpf_perf_event_output
BPF_CALL_5
bpf_event_output
BPF_CALL_0
BPF_CALL_2
BPF_CALL_3
do_bpf_send_signal
BPF_CALL_1
tracing_func_proto
kprobe_prog_func_proto
kprobe_prog_is_valid_access
BPF_CALL_5
BPF_CALL_3
BPF_CALL_4
tp_prog_func_proto
tp_prog_is_valid_access
BPF_CALL_3
pe_prog_func_proto
get_bpf_raw_tp_regs
put_bpf_raw_tp_regs
BPF_CALL_5
BPF_CALL_3
BPF_CALL_4
raw_tp_prog_func_proto
raw_tp_prog_is_valid_access
raw_tp_writable_prog_is_valid_access
pe_prog_is_valid_access
pe_prog_convert_ctx_access
perf_event_attach_bpf_prog
perf_event_detach_bpf_prog
perf_event_query_prog_array
bpf_get_raw_tracepoint
bpf_put_raw_tracepoint
__bpf_trace_run
__bpf_probe_register
bpf_probe_register
bpf_probe_unregister
bpf_get_perf_event_info
send_signal_irq_work_init
bpf_event_notify
bpf_event_init
   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   3  * Copyright (c) 2016 Facebook
   4  */
   5 #include <linux/kernel.h>
   6 #include <linux/types.h>
   7 #include <linux/slab.h>
   8 #include <linux/bpf.h>
   9 #include <linux/bpf_perf_event.h>
  10 #include <linux/filter.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/ctype.h>
  13 #include <linux/kprobes.h>
  14 #include <linux/syscalls.h>
  15 #include <linux/error-injection.h>
  16 
  17 #include <asm/tlb.h>
  18 
  19 #include "trace_probe.h"
  20 #include "trace.h"
  21 
  22 #define bpf_event_rcu_dereference(p)                                    \
  23         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  24 
  25 #ifdef CONFIG_MODULES
  26 struct bpf_trace_module {
  27         struct module *module;
  28         struct list_head list;
  29 };
  30 
  31 static LIST_HEAD(bpf_trace_modules);
  32 static DEFINE_MUTEX(bpf_module_mutex);
  33 
  34 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  35 {
  36         struct bpf_raw_event_map *btp, *ret = NULL;
  37         struct bpf_trace_module *btm;
  38         unsigned int i;
  39 
  40         mutex_lock(&bpf_module_mutex);
  41         list_for_each_entry(btm, &bpf_trace_modules, list) {
  42                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  43                         btp = &btm->module->bpf_raw_events[i];
  44                         if (!strcmp(btp->tp->name, name)) {
  45                                 if (try_module_get(btm->module))
  46                                         ret = btp;
  47                                 goto out;
  48                         }
  49                 }
  50         }
  51 out:
  52         mutex_unlock(&bpf_module_mutex);
  53         return ret;
  54 }
  55 #else
  56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  57 {
  58         return NULL;
  59 }
  60 #endif /* CONFIG_MODULES */
  61 
  62 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  63 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  64 
  65 /**
  66  * trace_call_bpf - invoke BPF program
  67  * @call: tracepoint event
  68  * @ctx: opaque context pointer
  69  *
  70  * kprobe handlers execute BPF programs via this helper.
  71  * Can be used from static tracepoints in the future.
  72  *
  73  * Return: BPF programs always return an integer which is interpreted by
  74  * kprobe handler as:
  75  * 0 - return from kprobe (event is filtered out)
  76  * 1 - store kprobe event into ring buffer
  77  * Other values are reserved and currently alias to 1
  78  */
  79 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
  80 {
  81         unsigned int ret;
  82 
  83         if (in_nmi()) /* not supported yet */
  84                 return 1;
  85 
  86         preempt_disable();
  87 
  88         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
  89                 /*
  90                  * since some bpf program is already running on this cpu,
  91                  * don't call into another bpf program (same or different)
  92                  * and don't send kprobe event into ring-buffer,
  93                  * so return zero here
  94                  */
  95                 ret = 0;
  96                 goto out;
  97         }
  98 
  99         /*
 100          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 101          * to all call sites, we did a bpf_prog_array_valid() there to check
 102          * whether call->prog_array is empty or not, which is
 103          * a heurisitc to speed up execution.
 104          *
 105          * If bpf_prog_array_valid() fetched prog_array was
 106          * non-NULL, we go into trace_call_bpf() and do the actual
 107          * proper rcu_dereference() under RCU lock.
 108          * If it turns out that prog_array is NULL then, we bail out.
 109          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 110          * was NULL, you'll skip the prog_array with the risk of missing
 111          * out of events when it was updated in between this and the
 112          * rcu_dereference() which is accepted risk.
 113          */
 114         ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
 115 
 116  out:
 117         __this_cpu_dec(bpf_prog_active);
 118         preempt_enable();
 119 
 120         return ret;
 121 }
 122 EXPORT_SYMBOL_GPL(trace_call_bpf);
 123 
 124 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 125 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 126 {
 127         regs_set_return_value(regs, rc);
 128         override_function_with_return(regs);
 129         return 0;
 130 }
 131 
 132 static const struct bpf_func_proto bpf_override_return_proto = {
 133         .func           = bpf_override_return,
 134         .gpl_only       = true,
 135         .ret_type       = RET_INTEGER,
 136         .arg1_type      = ARG_PTR_TO_CTX,
 137         .arg2_type      = ARG_ANYTHING,
 138 };
 139 #endif
 140 
 141 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
 142 {
 143         int ret;
 144 
 145         ret = security_locked_down(LOCKDOWN_BPF_READ);
 146         if (ret < 0)
 147                 goto out;
 148 
 149         ret = probe_kernel_read(dst, unsafe_ptr, size);
 150         if (unlikely(ret < 0))
 151 out:
 152                 memset(dst, 0, size);
 153 
 154         return ret;
 155 }
 156 
 157 static const struct bpf_func_proto bpf_probe_read_proto = {
 158         .func           = bpf_probe_read,
 159         .gpl_only       = true,
 160         .ret_type       = RET_INTEGER,
 161         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 162         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 163         .arg3_type      = ARG_ANYTHING,
 164 };
 165 
 166 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 167            u32, size)
 168 {
 169         /*
 170          * Ensure we're in user context which is safe for the helper to
 171          * run. This helper has no business in a kthread.
 172          *
 173          * access_ok() should prevent writing to non-user memory, but in
 174          * some situations (nommu, temporary switch, etc) access_ok() does
 175          * not provide enough validation, hence the check on KERNEL_DS.
 176          *
 177          * nmi_uaccess_okay() ensures the probe is not run in an interim
 178          * state, when the task or mm are switched. This is specifically
 179          * required to prevent the use of temporary mm.
 180          */
 181 
 182         if (unlikely(in_interrupt() ||
 183                      current->flags & (PF_KTHREAD | PF_EXITING)))
 184                 return -EPERM;
 185         if (unlikely(uaccess_kernel()))
 186                 return -EPERM;
 187         if (unlikely(!nmi_uaccess_okay()))
 188                 return -EPERM;
 189 
 190         return probe_user_write(unsafe_ptr, src, size);
 191 }
 192 
 193 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 194         .func           = bpf_probe_write_user,
 195         .gpl_only       = true,
 196         .ret_type       = RET_INTEGER,
 197         .arg1_type      = ARG_ANYTHING,
 198         .arg2_type      = ARG_PTR_TO_MEM,
 199         .arg3_type      = ARG_CONST_SIZE,
 200 };
 201 
 202 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 203 {
 204         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 205                             current->comm, task_pid_nr(current));
 206 
 207         return &bpf_probe_write_user_proto;
 208 }
 209 
 210 /*
 211  * Only limited trace_printk() conversion specifiers allowed:
 212  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
 213  */
 214 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 215            u64, arg2, u64, arg3)
 216 {
 217         bool str_seen = false;
 218         int mod[3] = {};
 219         int fmt_cnt = 0;
 220         u64 unsafe_addr;
 221         char buf[64];
 222         int i;
 223 
 224         /*
 225          * bpf_check()->check_func_arg()->check_stack_boundary()
 226          * guarantees that fmt points to bpf program stack,
 227          * fmt_size bytes of it were initialized and fmt_size > 0
 228          */
 229         if (fmt[--fmt_size] != 0)
 230                 return -EINVAL;
 231 
 232         /* check format string for allowed specifiers */
 233         for (i = 0; i < fmt_size; i++) {
 234                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 235                         return -EINVAL;
 236 
 237                 if (fmt[i] != '%')
 238                         continue;
 239 
 240                 if (fmt_cnt >= 3)
 241                         return -EINVAL;
 242 
 243                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 244                 i++;
 245                 if (fmt[i] == 'l') {
 246                         mod[fmt_cnt]++;
 247                         i++;
 248                 } else if (fmt[i] == 'p' || fmt[i] == 's') {
 249                         mod[fmt_cnt]++;
 250                         /* disallow any further format extensions */
 251                         if (fmt[i + 1] != 0 &&
 252                             !isspace(fmt[i + 1]) &&
 253                             !ispunct(fmt[i + 1]))
 254                                 return -EINVAL;
 255                         fmt_cnt++;
 256                         if (fmt[i] == 's') {
 257                                 if (str_seen)
 258                                         /* allow only one '%s' per fmt string */
 259                                         return -EINVAL;
 260                                 str_seen = true;
 261 
 262                                 switch (fmt_cnt) {
 263                                 case 1:
 264                                         unsafe_addr = arg1;
 265                                         arg1 = (long) buf;
 266                                         break;
 267                                 case 2:
 268                                         unsafe_addr = arg2;
 269                                         arg2 = (long) buf;
 270                                         break;
 271                                 case 3:
 272                                         unsafe_addr = arg3;
 273                                         arg3 = (long) buf;
 274                                         break;
 275                                 }
 276                                 buf[0] = 0;
 277                                 strncpy_from_unsafe(buf,
 278                                                     (void *) (long) unsafe_addr,
 279                                                     sizeof(buf));
 280                         }
 281                         continue;
 282                 }
 283 
 284                 if (fmt[i] == 'l') {
 285                         mod[fmt_cnt]++;
 286                         i++;
 287                 }
 288 
 289                 if (fmt[i] != 'i' && fmt[i] != 'd' &&
 290                     fmt[i] != 'u' && fmt[i] != 'x')
 291                         return -EINVAL;
 292                 fmt_cnt++;
 293         }
 294 
 295 /* Horrid workaround for getting va_list handling working with different
 296  * argument type combinations generically for 32 and 64 bit archs.
 297  */
 298 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
 299 #define __BPF_TP(...)                                                   \
 300         __trace_printk(0 /* Fake ip */,                                 \
 301                        fmt, ##__VA_ARGS__)
 302 
 303 #define __BPF_ARG1_TP(...)                                              \
 304         ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))        \
 305           ? __BPF_TP(arg1, ##__VA_ARGS__)                               \
 306           : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))    \
 307               ? __BPF_TP((long)arg1, ##__VA_ARGS__)                     \
 308               : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
 309 
 310 #define __BPF_ARG2_TP(...)                                              \
 311         ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))        \
 312           ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)                          \
 313           : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))    \
 314               ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)                \
 315               : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
 316 
 317 #define __BPF_ARG3_TP(...)                                              \
 318         ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))        \
 319           ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)                          \
 320           : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))    \
 321               ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)                \
 322               : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
 323 
 324         return __BPF_TP_EMIT();
 325 }
 326 
 327 static const struct bpf_func_proto bpf_trace_printk_proto = {
 328         .func           = bpf_trace_printk,
 329         .gpl_only       = true,
 330         .ret_type       = RET_INTEGER,
 331         .arg1_type      = ARG_PTR_TO_MEM,
 332         .arg2_type      = ARG_CONST_SIZE,
 333 };
 334 
 335 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 336 {
 337         /*
 338          * this program might be calling bpf_trace_printk,
 339          * so allocate per-cpu printk buffers
 340          */
 341         trace_printk_init_buffers();
 342 
 343         return &bpf_trace_printk_proto;
 344 }
 345 
 346 static __always_inline int
 347 get_map_perf_counter(struct bpf_map *map, u64 flags,
 348                      u64 *value, u64 *enabled, u64 *running)
 349 {
 350         struct bpf_array *array = container_of(map, struct bpf_array, map);
 351         unsigned int cpu = smp_processor_id();
 352         u64 index = flags & BPF_F_INDEX_MASK;
 353         struct bpf_event_entry *ee;
 354 
 355         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 356                 return -EINVAL;
 357         if (index == BPF_F_CURRENT_CPU)
 358                 index = cpu;
 359         if (unlikely(index >= array->map.max_entries))
 360                 return -E2BIG;
 361 
 362         ee = READ_ONCE(array->ptrs[index]);
 363         if (!ee)
 364                 return -ENOENT;
 365 
 366         return perf_event_read_local(ee->event, value, enabled, running);
 367 }
 368 
 369 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 370 {
 371         u64 value = 0;
 372         int err;
 373 
 374         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 375         /*
 376          * this api is ugly since we miss [-22..-2] range of valid
 377          * counter values, but that's uapi
 378          */
 379         if (err)
 380                 return err;
 381         return value;
 382 }
 383 
 384 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 385         .func           = bpf_perf_event_read,
 386         .gpl_only       = true,
 387         .ret_type       = RET_INTEGER,
 388         .arg1_type      = ARG_CONST_MAP_PTR,
 389         .arg2_type      = ARG_ANYTHING,
 390 };
 391 
 392 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 393            struct bpf_perf_event_value *, buf, u32, size)
 394 {
 395         int err = -EINVAL;
 396 
 397         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 398                 goto clear;
 399         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 400                                    &buf->running);
 401         if (unlikely(err))
 402                 goto clear;
 403         return 0;
 404 clear:
 405         memset(buf, 0, size);
 406         return err;
 407 }
 408 
 409 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 410         .func           = bpf_perf_event_read_value,
 411         .gpl_only       = true,
 412         .ret_type       = RET_INTEGER,
 413         .arg1_type      = ARG_CONST_MAP_PTR,
 414         .arg2_type      = ARG_ANYTHING,
 415         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 416         .arg4_type      = ARG_CONST_SIZE,
 417 };
 418 
 419 static __always_inline u64
 420 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 421                         u64 flags, struct perf_sample_data *sd)
 422 {
 423         struct bpf_array *array = container_of(map, struct bpf_array, map);
 424         unsigned int cpu = smp_processor_id();
 425         u64 index = flags & BPF_F_INDEX_MASK;
 426         struct bpf_event_entry *ee;
 427         struct perf_event *event;
 428 
 429         if (index == BPF_F_CURRENT_CPU)
 430                 index = cpu;
 431         if (unlikely(index >= array->map.max_entries))
 432                 return -E2BIG;
 433 
 434         ee = READ_ONCE(array->ptrs[index]);
 435         if (!ee)
 436                 return -ENOENT;
 437 
 438         event = ee->event;
 439         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 440                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 441                 return -EINVAL;
 442 
 443         if (unlikely(event->oncpu != cpu))
 444                 return -EOPNOTSUPP;
 445 
 446         return perf_event_output(event, sd, regs);
 447 }
 448 
 449 /*
 450  * Support executing tracepoints in normal, irq, and nmi context that each call
 451  * bpf_perf_event_output
 452  */
 453 struct bpf_trace_sample_data {
 454         struct perf_sample_data sds[3];
 455 };
 456 
 457 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 458 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 459 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 460            u64, flags, void *, data, u64, size)
 461 {
 462         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 463         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 464         struct perf_raw_record raw = {
 465                 .frag = {
 466                         .size = size,
 467                         .data = data,
 468                 },
 469         };
 470         struct perf_sample_data *sd;
 471         int err;
 472 
 473         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 474                 err = -EBUSY;
 475                 goto out;
 476         }
 477 
 478         sd = &sds->sds[nest_level - 1];
 479 
 480         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 481                 err = -EINVAL;
 482                 goto out;
 483         }
 484 
 485         perf_sample_data_init(sd, 0, 0);
 486         sd->raw = &raw;
 487 
 488         err = __bpf_perf_event_output(regs, map, flags, sd);
 489 
 490 out:
 491         this_cpu_dec(bpf_trace_nest_level);
 492         return err;
 493 }
 494 
 495 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 496         .func           = bpf_perf_event_output,
 497         .gpl_only       = true,
 498         .ret_type       = RET_INTEGER,
 499         .arg1_type      = ARG_PTR_TO_CTX,
 500         .arg2_type      = ARG_CONST_MAP_PTR,
 501         .arg3_type      = ARG_ANYTHING,
 502         .arg4_type      = ARG_PTR_TO_MEM,
 503         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 504 };
 505 
 506 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 507 struct bpf_nested_pt_regs {
 508         struct pt_regs regs[3];
 509 };
 510 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 511 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 512 
 513 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 514                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 515 {
 516         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 517         struct perf_raw_frag frag = {
 518                 .copy           = ctx_copy,
 519                 .size           = ctx_size,
 520                 .data           = ctx,
 521         };
 522         struct perf_raw_record raw = {
 523                 .frag = {
 524                         {
 525                                 .next   = ctx_size ? &frag : NULL,
 526                         },
 527                         .size   = meta_size,
 528                         .data   = meta,
 529                 },
 530         };
 531         struct perf_sample_data *sd;
 532         struct pt_regs *regs;
 533         u64 ret;
 534 
 535         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 536                 ret = -EBUSY;
 537                 goto out;
 538         }
 539         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 540         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 541 
 542         perf_fetch_caller_regs(regs);
 543         perf_sample_data_init(sd, 0, 0);
 544         sd->raw = &raw;
 545 
 546         ret = __bpf_perf_event_output(regs, map, flags, sd);
 547 out:
 548         this_cpu_dec(bpf_event_output_nest_level);
 549         return ret;
 550 }
 551 
 552 BPF_CALL_0(bpf_get_current_task)
 553 {
 554         return (long) current;
 555 }
 556 
 557 static const struct bpf_func_proto bpf_get_current_task_proto = {
 558         .func           = bpf_get_current_task,
 559         .gpl_only       = true,
 560         .ret_type       = RET_INTEGER,
 561 };
 562 
 563 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 564 {
 565         struct bpf_array *array = container_of(map, struct bpf_array, map);
 566         struct cgroup *cgrp;
 567 
 568         if (unlikely(idx >= array->map.max_entries))
 569                 return -E2BIG;
 570 
 571         cgrp = READ_ONCE(array->ptrs[idx]);
 572         if (unlikely(!cgrp))
 573                 return -EAGAIN;
 574 
 575         return task_under_cgroup_hierarchy(current, cgrp);
 576 }
 577 
 578 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 579         .func           = bpf_current_task_under_cgroup,
 580         .gpl_only       = false,
 581         .ret_type       = RET_INTEGER,
 582         .arg1_type      = ARG_CONST_MAP_PTR,
 583         .arg2_type      = ARG_ANYTHING,
 584 };
 585 
 586 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
 587            const void *, unsafe_ptr)
 588 {
 589         int ret;
 590 
 591         ret = security_locked_down(LOCKDOWN_BPF_READ);
 592         if (ret < 0)
 593                 goto out;
 594 
 595         /*
 596          * The strncpy_from_unsafe() call will likely not fill the entire
 597          * buffer, but that's okay in this circumstance as we're probing
 598          * arbitrary memory anyway similar to bpf_probe_read() and might
 599          * as well probe the stack. Thus, memory is explicitly cleared
 600          * only in error case, so that improper users ignoring return
 601          * code altogether don't copy garbage; otherwise length of string
 602          * is returned that can be used for bpf_perf_event_output() et al.
 603          */
 604         ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
 605         if (unlikely(ret < 0))
 606 out:
 607                 memset(dst, 0, size);
 608 
 609         return ret;
 610 }
 611 
 612 static const struct bpf_func_proto bpf_probe_read_str_proto = {
 613         .func           = bpf_probe_read_str,
 614         .gpl_only       = true,
 615         .ret_type       = RET_INTEGER,
 616         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 617         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 618         .arg3_type      = ARG_ANYTHING,
 619 };
 620 
 621 struct send_signal_irq_work {
 622         struct irq_work irq_work;
 623         struct task_struct *task;
 624         u32 sig;
 625 };
 626 
 627 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 628 
 629 static void do_bpf_send_signal(struct irq_work *entry)
 630 {
 631         struct send_signal_irq_work *work;
 632 
 633         work = container_of(entry, struct send_signal_irq_work, irq_work);
 634         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
 635 }
 636 
 637 BPF_CALL_1(bpf_send_signal, u32, sig)
 638 {
 639         struct send_signal_irq_work *work = NULL;
 640 
 641         /* Similar to bpf_probe_write_user, task needs to be
 642          * in a sound condition and kernel memory access be
 643          * permitted in order to send signal to the current
 644          * task.
 645          */
 646         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 647                 return -EPERM;
 648         if (unlikely(uaccess_kernel()))
 649                 return -EPERM;
 650         if (unlikely(!nmi_uaccess_okay()))
 651                 return -EPERM;
 652 
 653         if (irqs_disabled()) {
 654                 /* Do an early check on signal validity. Otherwise,
 655                  * the error is lost in deferred irq_work.
 656                  */
 657                 if (unlikely(!valid_signal(sig)))
 658                         return -EINVAL;
 659 
 660                 work = this_cpu_ptr(&send_signal_work);
 661                 if (work->irq_work.flags & IRQ_WORK_BUSY)
 662                         return -EBUSY;
 663 
 664                 /* Add the current task, which is the target of sending signal,
 665                  * to the irq_work. The current task may change when queued
 666                  * irq works get executed.
 667                  */
 668                 work->task = current;
 669                 work->sig = sig;
 670                 irq_work_queue(&work->irq_work);
 671                 return 0;
 672         }
 673 
 674         return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
 675 }
 676 
 677 static const struct bpf_func_proto bpf_send_signal_proto = {
 678         .func           = bpf_send_signal,
 679         .gpl_only       = false,
 680         .ret_type       = RET_INTEGER,
 681         .arg1_type      = ARG_ANYTHING,
 682 };
 683 
 684 static const struct bpf_func_proto *
 685 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 686 {
 687         switch (func_id) {
 688         case BPF_FUNC_map_lookup_elem:
 689                 return &bpf_map_lookup_elem_proto;
 690         case BPF_FUNC_map_update_elem:
 691                 return &bpf_map_update_elem_proto;
 692         case BPF_FUNC_map_delete_elem:
 693                 return &bpf_map_delete_elem_proto;
 694         case BPF_FUNC_map_push_elem:
 695                 return &bpf_map_push_elem_proto;
 696         case BPF_FUNC_map_pop_elem:
 697                 return &bpf_map_pop_elem_proto;
 698         case BPF_FUNC_map_peek_elem:
 699                 return &bpf_map_peek_elem_proto;
 700         case BPF_FUNC_probe_read:
 701                 return &bpf_probe_read_proto;
 702         case BPF_FUNC_ktime_get_ns:
 703                 return &bpf_ktime_get_ns_proto;
 704         case BPF_FUNC_tail_call:
 705                 return &bpf_tail_call_proto;
 706         case BPF_FUNC_get_current_pid_tgid:
 707                 return &bpf_get_current_pid_tgid_proto;
 708         case BPF_FUNC_get_current_task:
 709                 return &bpf_get_current_task_proto;
 710         case BPF_FUNC_get_current_uid_gid:
 711                 return &bpf_get_current_uid_gid_proto;
 712         case BPF_FUNC_get_current_comm:
 713                 return &bpf_get_current_comm_proto;
 714         case BPF_FUNC_trace_printk:
 715                 return bpf_get_trace_printk_proto();
 716         case BPF_FUNC_get_smp_processor_id:
 717                 return &bpf_get_smp_processor_id_proto;
 718         case BPF_FUNC_get_numa_node_id:
 719                 return &bpf_get_numa_node_id_proto;
 720         case BPF_FUNC_perf_event_read:
 721                 return &bpf_perf_event_read_proto;
 722         case BPF_FUNC_probe_write_user:
 723                 return bpf_get_probe_write_proto();
 724         case BPF_FUNC_current_task_under_cgroup:
 725                 return &bpf_current_task_under_cgroup_proto;
 726         case BPF_FUNC_get_prandom_u32:
 727                 return &bpf_get_prandom_u32_proto;
 728         case BPF_FUNC_probe_read_str:
 729                 return &bpf_probe_read_str_proto;
 730 #ifdef CONFIG_CGROUPS
 731         case BPF_FUNC_get_current_cgroup_id:
 732                 return &bpf_get_current_cgroup_id_proto;
 733 #endif
 734         case BPF_FUNC_send_signal:
 735                 return &bpf_send_signal_proto;
 736         default:
 737                 return NULL;
 738         }
 739 }
 740 
 741 static const struct bpf_func_proto *
 742 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 743 {
 744         switch (func_id) {
 745         case BPF_FUNC_perf_event_output:
 746                 return &bpf_perf_event_output_proto;
 747         case BPF_FUNC_get_stackid:
 748                 return &bpf_get_stackid_proto;
 749         case BPF_FUNC_get_stack:
 750                 return &bpf_get_stack_proto;
 751         case BPF_FUNC_perf_event_read_value:
 752                 return &bpf_perf_event_read_value_proto;
 753 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 754         case BPF_FUNC_override_return:
 755                 return &bpf_override_return_proto;
 756 #endif
 757         default:
 758                 return tracing_func_proto(func_id, prog);
 759         }
 760 }
 761 
 762 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
 763 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 764                                         const struct bpf_prog *prog,
 765                                         struct bpf_insn_access_aux *info)
 766 {
 767         if (off < 0 || off >= sizeof(struct pt_regs))
 768                 return false;
 769         if (type != BPF_READ)
 770                 return false;
 771         if (off % size != 0)
 772                 return false;
 773         /*
 774          * Assertion for 32 bit to make sure last 8 byte access
 775          * (BPF_DW) to the last 4 byte member is disallowed.
 776          */
 777         if (off + size > sizeof(struct pt_regs))
 778                 return false;
 779 
 780         return true;
 781 }
 782 
 783 const struct bpf_verifier_ops kprobe_verifier_ops = {
 784         .get_func_proto  = kprobe_prog_func_proto,
 785         .is_valid_access = kprobe_prog_is_valid_access,
 786 };
 787 
 788 const struct bpf_prog_ops kprobe_prog_ops = {
 789 };
 790 
 791 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
 792            u64, flags, void *, data, u64, size)
 793 {
 794         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
 795 
 796         /*
 797          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
 798          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
 799          * from there and call the same bpf_perf_event_output() helper inline.
 800          */
 801         return ____bpf_perf_event_output(regs, map, flags, data, size);
 802 }
 803 
 804 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
 805         .func           = bpf_perf_event_output_tp,
 806         .gpl_only       = true,
 807         .ret_type       = RET_INTEGER,
 808         .arg1_type      = ARG_PTR_TO_CTX,
 809         .arg2_type      = ARG_CONST_MAP_PTR,
 810         .arg3_type      = ARG_ANYTHING,
 811         .arg4_type      = ARG_PTR_TO_MEM,
 812         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 813 };
 814 
 815 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
 816            u64, flags)
 817 {
 818         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
 819 
 820         /*
 821          * Same comment as in bpf_perf_event_output_tp(), only that this time
 822          * the other helper's function body cannot be inlined due to being
 823          * external, thus we need to call raw helper function.
 824          */
 825         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
 826                                flags, 0, 0);
 827 }
 828 
 829 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
 830         .func           = bpf_get_stackid_tp,
 831         .gpl_only       = true,
 832         .ret_type       = RET_INTEGER,
 833         .arg1_type      = ARG_PTR_TO_CTX,
 834         .arg2_type      = ARG_CONST_MAP_PTR,
 835         .arg3_type      = ARG_ANYTHING,
 836 };
 837 
 838 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
 839            u64, flags)
 840 {
 841         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
 842 
 843         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
 844                              (unsigned long) size, flags, 0);
 845 }
 846 
 847 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
 848         .func           = bpf_get_stack_tp,
 849         .gpl_only       = true,
 850         .ret_type       = RET_INTEGER,
 851         .arg1_type      = ARG_PTR_TO_CTX,
 852         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 853         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 854         .arg4_type      = ARG_ANYTHING,
 855 };
 856 
 857 static const struct bpf_func_proto *
 858 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 859 {
 860         switch (func_id) {
 861         case BPF_FUNC_perf_event_output:
 862                 return &bpf_perf_event_output_proto_tp;
 863         case BPF_FUNC_get_stackid:
 864                 return &bpf_get_stackid_proto_tp;
 865         case BPF_FUNC_get_stack:
 866                 return &bpf_get_stack_proto_tp;
 867         default:
 868                 return tracing_func_proto(func_id, prog);
 869         }
 870 }
 871 
 872 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 873                                     const struct bpf_prog *prog,
 874                                     struct bpf_insn_access_aux *info)
 875 {
 876         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
 877                 return false;
 878         if (type != BPF_READ)
 879                 return false;
 880         if (off % size != 0)
 881                 return false;
 882 
 883         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
 884         return true;
 885 }
 886 
 887 const struct bpf_verifier_ops tracepoint_verifier_ops = {
 888         .get_func_proto  = tp_prog_func_proto,
 889         .is_valid_access = tp_prog_is_valid_access,
 890 };
 891 
 892 const struct bpf_prog_ops tracepoint_prog_ops = {
 893 };
 894 
 895 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
 896            struct bpf_perf_event_value *, buf, u32, size)
 897 {
 898         int err = -EINVAL;
 899 
 900         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 901                 goto clear;
 902         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
 903                                     &buf->running);
 904         if (unlikely(err))
 905                 goto clear;
 906         return 0;
 907 clear:
 908         memset(buf, 0, size);
 909         return err;
 910 }
 911 
 912 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
 913          .func           = bpf_perf_prog_read_value,
 914          .gpl_only       = true,
 915          .ret_type       = RET_INTEGER,
 916          .arg1_type      = ARG_PTR_TO_CTX,
 917          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
 918          .arg3_type      = ARG_CONST_SIZE,
 919 };
 920 
 921 static const struct bpf_func_proto *
 922 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 923 {
 924         switch (func_id) {
 925         case BPF_FUNC_perf_event_output:
 926                 return &bpf_perf_event_output_proto_tp;
 927         case BPF_FUNC_get_stackid:
 928                 return &bpf_get_stackid_proto_tp;
 929         case BPF_FUNC_get_stack:
 930                 return &bpf_get_stack_proto_tp;
 931         case BPF_FUNC_perf_prog_read_value:
 932                 return &bpf_perf_prog_read_value_proto;
 933         default:
 934                 return tracing_func_proto(func_id, prog);
 935         }
 936 }
 937 
 938 /*
 939  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
 940  * to avoid potential recursive reuse issue when/if tracepoints are added
 941  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
 942  *
 943  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
 944  * in normal, irq, and nmi context.
 945  */
 946 struct bpf_raw_tp_regs {
 947         struct pt_regs regs[3];
 948 };
 949 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
 950 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
 951 static struct pt_regs *get_bpf_raw_tp_regs(void)
 952 {
 953         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
 954         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
 955 
 956         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
 957                 this_cpu_dec(bpf_raw_tp_nest_level);
 958                 return ERR_PTR(-EBUSY);
 959         }
 960 
 961         return &tp_regs->regs[nest_level - 1];
 962 }
 963 
 964 static void put_bpf_raw_tp_regs(void)
 965 {
 966         this_cpu_dec(bpf_raw_tp_nest_level);
 967 }
 968 
 969 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
 970            struct bpf_map *, map, u64, flags, void *, data, u64, size)
 971 {
 972         struct pt_regs *regs = get_bpf_raw_tp_regs();
 973         int ret;
 974 
 975         if (IS_ERR(regs))
 976                 return PTR_ERR(regs);
 977 
 978         perf_fetch_caller_regs(regs);
 979         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
 980 
 981         put_bpf_raw_tp_regs();
 982         return ret;
 983 }
 984 
 985 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
 986         .func           = bpf_perf_event_output_raw_tp,
 987         .gpl_only       = true,
 988         .ret_type       = RET_INTEGER,
 989         .arg1_type      = ARG_PTR_TO_CTX,
 990         .arg2_type      = ARG_CONST_MAP_PTR,
 991         .arg3_type      = ARG_ANYTHING,
 992         .arg4_type      = ARG_PTR_TO_MEM,
 993         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 994 };
 995 
 996 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
 997            struct bpf_map *, map, u64, flags)
 998 {
 999         struct pt_regs *regs = get_bpf_raw_tp_regs();
1000         int ret;
1001 
1002         if (IS_ERR(regs))
1003                 return PTR_ERR(regs);
1004 
1005         perf_fetch_caller_regs(regs);
1006         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1007         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1008                               flags, 0, 0);
1009         put_bpf_raw_tp_regs();
1010         return ret;
1011 }
1012 
1013 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1014         .func           = bpf_get_stackid_raw_tp,
1015         .gpl_only       = true,
1016         .ret_type       = RET_INTEGER,
1017         .arg1_type      = ARG_PTR_TO_CTX,
1018         .arg2_type      = ARG_CONST_MAP_PTR,
1019         .arg3_type      = ARG_ANYTHING,
1020 };
1021 
1022 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1023            void *, buf, u32, size, u64, flags)
1024 {
1025         struct pt_regs *regs = get_bpf_raw_tp_regs();
1026         int ret;
1027 
1028         if (IS_ERR(regs))
1029                 return PTR_ERR(regs);
1030 
1031         perf_fetch_caller_regs(regs);
1032         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1033                             (unsigned long) size, flags, 0);
1034         put_bpf_raw_tp_regs();
1035         return ret;
1036 }
1037 
1038 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1039         .func           = bpf_get_stack_raw_tp,
1040         .gpl_only       = true,
1041         .ret_type       = RET_INTEGER,
1042         .arg1_type      = ARG_PTR_TO_CTX,
1043         .arg2_type      = ARG_PTR_TO_MEM,
1044         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1045         .arg4_type      = ARG_ANYTHING,
1046 };
1047 
1048 static const struct bpf_func_proto *
1049 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1050 {
1051         switch (func_id) {
1052         case BPF_FUNC_perf_event_output:
1053                 return &bpf_perf_event_output_proto_raw_tp;
1054         case BPF_FUNC_get_stackid:
1055                 return &bpf_get_stackid_proto_raw_tp;
1056         case BPF_FUNC_get_stack:
1057                 return &bpf_get_stack_proto_raw_tp;
1058         default:
1059                 return tracing_func_proto(func_id, prog);
1060         }
1061 }
1062 
1063 static bool raw_tp_prog_is_valid_access(int off, int size,
1064                                         enum bpf_access_type type,
1065                                         const struct bpf_prog *prog,
1066                                         struct bpf_insn_access_aux *info)
1067 {
1068         /* largest tracepoint in the kernel has 12 args */
1069         if (off < 0 || off >= sizeof(__u64) * 12)
1070                 return false;
1071         if (type != BPF_READ)
1072                 return false;
1073         if (off % size != 0)
1074                 return false;
1075         return true;
1076 }
1077 
1078 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1079         .get_func_proto  = raw_tp_prog_func_proto,
1080         .is_valid_access = raw_tp_prog_is_valid_access,
1081 };
1082 
1083 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1084 };
1085 
1086 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1087                                                  enum bpf_access_type type,
1088                                                  const struct bpf_prog *prog,
1089                                                  struct bpf_insn_access_aux *info)
1090 {
1091         if (off == 0) {
1092                 if (size != sizeof(u64) || type != BPF_READ)
1093                         return false;
1094                 info->reg_type = PTR_TO_TP_BUFFER;
1095         }
1096         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1097 }
1098 
1099 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1100         .get_func_proto  = raw_tp_prog_func_proto,
1101         .is_valid_access = raw_tp_writable_prog_is_valid_access,
1102 };
1103 
1104 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1105 };
1106 
1107 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1108                                     const struct bpf_prog *prog,
1109                                     struct bpf_insn_access_aux *info)
1110 {
1111         const int size_u64 = sizeof(u64);
1112 
1113         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1114                 return false;
1115         if (type != BPF_READ)
1116                 return false;
1117         if (off % size != 0) {
1118                 if (sizeof(unsigned long) != 4)
1119                         return false;
1120                 if (size != 8)
1121                         return false;
1122                 if (off % size != 4)
1123                         return false;
1124         }
1125 
1126         switch (off) {
1127         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1128                 bpf_ctx_record_field_size(info, size_u64);
1129                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1130                         return false;
1131                 break;
1132         case bpf_ctx_range(struct bpf_perf_event_data, addr):
1133                 bpf_ctx_record_field_size(info, size_u64);
1134                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1135                         return false;
1136                 break;
1137         default:
1138                 if (size != sizeof(long))
1139                         return false;
1140         }
1141 
1142         return true;
1143 }
1144 
1145 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1146                                       const struct bpf_insn *si,
1147                                       struct bpf_insn *insn_buf,
1148                                       struct bpf_prog *prog, u32 *target_size)
1149 {
1150         struct bpf_insn *insn = insn_buf;
1151 
1152         switch (si->off) {
1153         case offsetof(struct bpf_perf_event_data, sample_period):
1154                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1155                                                        data), si->dst_reg, si->src_reg,
1156                                       offsetof(struct bpf_perf_event_data_kern, data));
1157                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1158                                       bpf_target_off(struct perf_sample_data, period, 8,
1159                                                      target_size));
1160                 break;
1161         case offsetof(struct bpf_perf_event_data, addr):
1162                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1163                                                        data), si->dst_reg, si->src_reg,
1164                                       offsetof(struct bpf_perf_event_data_kern, data));
1165                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1166                                       bpf_target_off(struct perf_sample_data, addr, 8,
1167                                                      target_size));
1168                 break;
1169         default:
1170                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1171                                                        regs), si->dst_reg, si->src_reg,
1172                                       offsetof(struct bpf_perf_event_data_kern, regs));
1173                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1174                                       si->off);
1175                 break;
1176         }
1177 
1178         return insn - insn_buf;
1179 }
1180 
1181 const struct bpf_verifier_ops perf_event_verifier_ops = {
1182         .get_func_proto         = pe_prog_func_proto,
1183         .is_valid_access        = pe_prog_is_valid_access,
1184         .convert_ctx_access     = pe_prog_convert_ctx_access,
1185 };
1186 
1187 const struct bpf_prog_ops perf_event_prog_ops = {
1188 };
1189 
1190 static DEFINE_MUTEX(bpf_event_mutex);
1191 
1192 #define BPF_TRACE_MAX_PROGS 64
1193 
1194 int perf_event_attach_bpf_prog(struct perf_event *event,
1195                                struct bpf_prog *prog)
1196 {
1197         struct bpf_prog_array *old_array;
1198         struct bpf_prog_array *new_array;
1199         int ret = -EEXIST;
1200 
1201         /*
1202          * Kprobe override only works if they are on the function entry,
1203          * and only if they are on the opt-in list.
1204          */
1205         if (prog->kprobe_override &&
1206             (!trace_kprobe_on_func_entry(event->tp_event) ||
1207              !trace_kprobe_error_injectable(event->tp_event)))
1208                 return -EINVAL;
1209 
1210         mutex_lock(&bpf_event_mutex);
1211 
1212         if (event->prog)
1213                 goto unlock;
1214 
1215         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1216         if (old_array &&
1217             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1218                 ret = -E2BIG;
1219                 goto unlock;
1220         }
1221 
1222         ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1223         if (ret < 0)
1224                 goto unlock;
1225 
1226         /* set the new array to event->tp_event and set event->prog */
1227         event->prog = prog;
1228         rcu_assign_pointer(event->tp_event->prog_array, new_array);
1229         bpf_prog_array_free(old_array);
1230 
1231 unlock:
1232         mutex_unlock(&bpf_event_mutex);
1233         return ret;
1234 }
1235 
1236 void perf_event_detach_bpf_prog(struct perf_event *event)
1237 {
1238         struct bpf_prog_array *old_array;
1239         struct bpf_prog_array *new_array;
1240         int ret;
1241 
1242         mutex_lock(&bpf_event_mutex);
1243 
1244         if (!event->prog)
1245                 goto unlock;
1246 
1247         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1248         ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1249         if (ret == -ENOENT)
1250                 goto unlock;
1251         if (ret < 0) {
1252                 bpf_prog_array_delete_safe(old_array, event->prog);
1253         } else {
1254                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1255                 bpf_prog_array_free(old_array);
1256         }
1257 
1258         bpf_prog_put(event->prog);
1259         event->prog = NULL;
1260 
1261 unlock:
1262         mutex_unlock(&bpf_event_mutex);
1263 }
1264 
1265 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1266 {
1267         struct perf_event_query_bpf __user *uquery = info;
1268         struct perf_event_query_bpf query = {};
1269         struct bpf_prog_array *progs;
1270         u32 *ids, prog_cnt, ids_len;
1271         int ret;
1272 
1273         if (!capable(CAP_SYS_ADMIN))
1274                 return -EPERM;
1275         if (event->attr.type != PERF_TYPE_TRACEPOINT)
1276                 return -EINVAL;
1277         if (copy_from_user(&query, uquery, sizeof(query)))
1278                 return -EFAULT;
1279 
1280         ids_len = query.ids_len;
1281         if (ids_len > BPF_TRACE_MAX_PROGS)
1282                 return -E2BIG;
1283         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1284         if (!ids)
1285                 return -ENOMEM;
1286         /*
1287          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1288          * is required when user only wants to check for uquery->prog_cnt.
1289          * There is no need to check for it since the case is handled
1290          * gracefully in bpf_prog_array_copy_info.
1291          */
1292 
1293         mutex_lock(&bpf_event_mutex);
1294         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1295         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1296         mutex_unlock(&bpf_event_mutex);
1297 
1298         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1299             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1300                 ret = -EFAULT;
1301 
1302         kfree(ids);
1303         return ret;
1304 }
1305 
1306 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1307 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1308 
1309 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1310 {
1311         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1312 
1313         for (; btp < __stop__bpf_raw_tp; btp++) {
1314                 if (!strcmp(btp->tp->name, name))
1315                         return btp;
1316         }
1317 
1318         return bpf_get_raw_tracepoint_module(name);
1319 }
1320 
1321 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1322 {
1323         struct module *mod = __module_address((unsigned long)btp);
1324 
1325         if (mod)
1326                 module_put(mod);
1327 }
1328 
1329 static __always_inline
1330 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1331 {
1332         rcu_read_lock();
1333         preempt_disable();
1334         (void) BPF_PROG_RUN(prog, args);
1335         preempt_enable();
1336         rcu_read_unlock();
1337 }
1338 
1339 #define UNPACK(...)                     __VA_ARGS__
1340 #define REPEAT_1(FN, DL, X, ...)        FN(X)
1341 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1342 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1343 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1344 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1345 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1346 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1347 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1348 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1349 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1350 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1351 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1352 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
1353 
1354 #define SARG(X)         u64 arg##X
1355 #define COPY(X)         args[X] = arg##X
1356 
1357 #define __DL_COM        (,)
1358 #define __DL_SEM        (;)
1359 
1360 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1361 
1362 #define BPF_TRACE_DEFN_x(x)                                             \
1363         void bpf_trace_run##x(struct bpf_prog *prog,                    \
1364                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
1365         {                                                               \
1366                 u64 args[x];                                            \
1367                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
1368                 __bpf_trace_run(prog, args);                            \
1369         }                                                               \
1370         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1371 BPF_TRACE_DEFN_x(1);
1372 BPF_TRACE_DEFN_x(2);
1373 BPF_TRACE_DEFN_x(3);
1374 BPF_TRACE_DEFN_x(4);
1375 BPF_TRACE_DEFN_x(5);
1376 BPF_TRACE_DEFN_x(6);
1377 BPF_TRACE_DEFN_x(7);
1378 BPF_TRACE_DEFN_x(8);
1379 BPF_TRACE_DEFN_x(9);
1380 BPF_TRACE_DEFN_x(10);
1381 BPF_TRACE_DEFN_x(11);
1382 BPF_TRACE_DEFN_x(12);
1383 
1384 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1385 {
1386         struct tracepoint *tp = btp->tp;
1387 
1388         /*
1389          * check that program doesn't access arguments beyond what's
1390          * available in this tracepoint
1391          */
1392         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1393                 return -EINVAL;
1394 
1395         if (prog->aux->max_tp_access > btp->writable_size)
1396                 return -EINVAL;
1397 
1398         return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1399 }
1400 
1401 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1402 {
1403         return __bpf_probe_register(btp, prog);
1404 }
1405 
1406 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1407 {
1408         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1409 }
1410 
1411 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1412                             u32 *fd_type, const char **buf,
1413                             u64 *probe_offset, u64 *probe_addr)
1414 {
1415         bool is_tracepoint, is_syscall_tp;
1416         struct bpf_prog *prog;
1417         int flags, err = 0;
1418 
1419         prog = event->prog;
1420         if (!prog)
1421                 return -ENOENT;
1422 
1423         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1424         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1425                 return -EOPNOTSUPP;
1426 
1427         *prog_id = prog->aux->id;
1428         flags = event->tp_event->flags;
1429         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1430         is_syscall_tp = is_syscall_trace_event(event->tp_event);
1431 
1432         if (is_tracepoint || is_syscall_tp) {
1433                 *buf = is_tracepoint ? event->tp_event->tp->name
1434                                      : event->tp_event->name;
1435                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1436                 *probe_offset = 0x0;
1437                 *probe_addr = 0x0;
1438         } else {
1439                 /* kprobe/uprobe */
1440                 err = -EOPNOTSUPP;
1441 #ifdef CONFIG_KPROBE_EVENTS
1442                 if (flags & TRACE_EVENT_FL_KPROBE)
1443                         err = bpf_get_kprobe_info(event, fd_type, buf,
1444                                                   probe_offset, probe_addr,
1445                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
1446 #endif
1447 #ifdef CONFIG_UPROBE_EVENTS
1448                 if (flags & TRACE_EVENT_FL_UPROBE)
1449                         err = bpf_get_uprobe_info(event, fd_type, buf,
1450                                                   probe_offset,
1451                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
1452 #endif
1453         }
1454 
1455         return err;
1456 }
1457 
1458 static int __init send_signal_irq_work_init(void)
1459 {
1460         int cpu;
1461         struct send_signal_irq_work *work;
1462 
1463         for_each_possible_cpu(cpu) {
1464                 work = per_cpu_ptr(&send_signal_work, cpu);
1465                 init_irq_work(&work->irq_work, do_bpf_send_signal);
1466         }
1467         return 0;
1468 }
1469 
1470 subsys_initcall(send_signal_irq_work_init);
1471 
1472 #ifdef CONFIG_MODULES
1473 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1474                             void *module)
1475 {
1476         struct bpf_trace_module *btm, *tmp;
1477         struct module *mod = module;
1478 
1479         if (mod->num_bpf_raw_events == 0 ||
1480             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1481                 return 0;
1482 
1483         mutex_lock(&bpf_module_mutex);
1484 
1485         switch (op) {
1486         case MODULE_STATE_COMING:
1487                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1488                 if (btm) {
1489                         btm->module = module;
1490                         list_add(&btm->list, &bpf_trace_modules);
1491                 }
1492                 break;
1493         case MODULE_STATE_GOING:
1494                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1495                         if (btm->module == module) {
1496                                 list_del(&btm->list);
1497                                 kfree(btm);
1498                                 break;
1499                         }
1500                 }
1501                 break;
1502         }
1503 
1504         mutex_unlock(&bpf_module_mutex);
1505 
1506         return 0;
1507 }
1508 
1509 static struct notifier_block bpf_module_nb = {
1510         .notifier_call = bpf_event_notify,
1511 };
1512 
1513 static int __init bpf_event_init(void)
1514 {
1515         register_module_notifier(&bpf_module_nb);
1516         return 0;
1517 }
1518 
1519 fs_initcall(bpf_event_init);
1520 #endif /* CONFIG_MODULES */
/* [<][>][^][v][top][bottom][index][help] */
root/kernel/trace/bpf_trace.c

DEFINITIONS
