xref: /openbmc/linux/tools/lib/bpf/usdt.c (revision e2815b71)
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2022 Meta Platforms, Inc. and affiliates. */
3 #include <ctype.h>
4 #include <stdio.h>
5 #include <stdlib.h>
6 #include <string.h>
7 #include <libelf.h>
8 #include <gelf.h>
9 #include <unistd.h>
10 #include <linux/ptrace.h>
11 #include <linux/kernel.h>
12 
13 /* s8 will be marked as poison while it's a reg of riscv */
14 #if defined(__riscv)
15 #define rv_s8 s8
16 #endif
17 
18 #include "bpf.h"
19 #include "libbpf.h"
20 #include "libbpf_common.h"
21 #include "libbpf_internal.h"
22 #include "hashmap.h"
23 
24 /* libbpf's USDT support consists of BPF-side state/code and user-space
25  * state/code working together in concert. BPF-side parts are defined in
26  * usdt.bpf.h header library. User-space state is encapsulated by struct
27  * usdt_manager and all the supporting code centered around usdt_manager.
28  *
29  * usdt.bpf.h defines two BPF maps that usdt_manager expects: USDT spec map
30  * and IP-to-spec-ID map, which is auxiliary map necessary for kernels that
31  * don't support BPF cookie (see below). These two maps are implicitly
32  * embedded into user's end BPF object file when user's code included
33  * usdt.bpf.h. This means that libbpf doesn't do anything special to create
34  * these USDT support maps. They are created by normal libbpf logic of
35  * instantiating BPF maps when opening and loading BPF object.
36  *
37  * As such, libbpf is basically unaware of the need to do anything
38  * USDT-related until the very first call to bpf_program__attach_usdt(), which
39  * can be called by user explicitly or happen automatically during skeleton
40  * attach (or, equivalently, through generic bpf_program__attach() call). At
41  * this point, libbpf will instantiate and initialize struct usdt_manager and
42  * store it in bpf_object. USDT manager is per-BPF object construct, as each
43  * independent BPF object might or might not have USDT programs, and thus all
44  * the expected USDT-related state. There is no coordination between two
45  * bpf_object in parts of USDT attachment, they are oblivious of each other's
46  * existence and libbpf is just oblivious, dealing with bpf_object-specific
47  * USDT state.
48  *
49  * Quick crash course on USDTs.
50  *
51  * From user-space application's point of view, USDT is essentially just
52  * a slightly special function call that normally has zero overhead, unless it
53  * is being traced by some external entity (e.g, BPF-based tool). Here's how
54  * a typical application can trigger USDT probe:
55  *
56  * #include <sys/sdt.h>  // provided by systemtap-sdt-devel package
57  * // folly also provide similar functionality in folly/tracing/StaticTracepoint.h
58  *
59  * STAP_PROBE3(my_usdt_provider, my_usdt_probe_name, 123, x, &y);
60  *
61  * USDT is identified by it's <provider-name>:<probe-name> pair of names. Each
62  * individual USDT has a fixed number of arguments (3 in the above example)
63  * and specifies values of each argument as if it was a function call.
64  *
65  * USDT call is actually not a function call, but is instead replaced by
66  * a single NOP instruction (thus zero overhead, effectively). But in addition
67  * to that, those USDT macros generate special SHT_NOTE ELF records in
68  * .note.stapsdt ELF section. Here's an example USDT definition as emitted by
69  * `readelf -n <binary>`:
70  *
71  *   stapsdt              0x00000089       NT_STAPSDT (SystemTap probe descriptors)
72  *   Provider: test
73  *   Name: usdt12
74  *   Location: 0x0000000000549df3, Base: 0x00000000008effa4, Semaphore: 0x0000000000a4606e
75  *   Arguments: -4@-1204(%rbp) -4@%edi -8@-1216(%rbp) -8@%r8 -4@$5 -8@%r9 8@%rdx 8@%r10 -4@$-9 -2@%cx -2@%ax -1@%sil
76  *
77  * In this case we have USDT test:usdt12 with 12 arguments.
78  *
79  * Location and base are offsets used to calculate absolute IP address of that
80  * NOP instruction that kernel can replace with an interrupt instruction to
81  * trigger instrumentation code (BPF program for all that we care about).
82  *
83  * Semaphore above is and optional feature. It records an address of a 2-byte
84  * refcount variable (normally in '.probes' ELF section) used for signaling if
85  * there is anything that is attached to USDT. This is useful for user
86  * applications if, for example, they need to prepare some arguments that are
87  * passed only to USDTs and preparation is expensive. By checking if USDT is
88  * "activated", an application can avoid paying those costs unnecessarily.
89  * Recent enough kernel has built-in support for automatically managing this
90  * refcount, which libbpf expects and relies on. If USDT is defined without
91  * associated semaphore, this value will be zero. See selftests for semaphore
92  * examples.
93  *
94  * Arguments is the most interesting part. This USDT specification string is
95  * providing information about all the USDT arguments and their locations. The
96  * part before @ sign defined byte size of the argument (1, 2, 4, or 8) and
97  * whether the argument is signed or unsigned (negative size means signed).
98  * The part after @ sign is assembly-like definition of argument location
99  * (see [0] for more details). Technically, assembler can provide some pretty
100  * advanced definitions, but libbpf is currently supporting three most common
101  * cases:
102  *   1) immediate constant, see 5th and 9th args above (-4@$5 and -4@-9);
103  *   2) register value, e.g., 8@%rdx, which means "unsigned 8-byte integer
104  *      whose value is in register %rdx";
105  *   3) memory dereference addressed by register, e.g., -4@-1204(%rbp), which
106  *      specifies signed 32-bit integer stored at offset -1204 bytes from
107  *      memory address stored in %rbp.
108  *
109  *   [0] https://sourceware.org/systemtap/wiki/UserSpaceProbeImplementation
110  *
111  * During attachment, libbpf parses all the relevant USDT specifications and
112  * prepares `struct usdt_spec` (USDT spec), which is then provided to BPF-side
113  * code through spec map. This allows BPF applications to quickly fetch the
114  * actual value at runtime using a simple BPF-side code.
115  *
116  * With basics out of the way, let's go over less immediately obvious aspects
117  * of supporting USDTs.
118  *
119  * First, there is no special USDT BPF program type. It is actually just
120  * a uprobe BPF program (which for kernel, at least currently, is just a kprobe
121  * program, so BPF_PROG_TYPE_KPROBE program type). With the only difference
122  * that uprobe is usually attached at the function entry, while USDT will
123  * normally will be somewhere inside the function. But it should always be
124  * pointing to NOP instruction, which makes such uprobes the fastest uprobe
125  * kind.
126  *
127  * Second, it's important to realize that such STAP_PROBEn(provider, name, ...)
128  * macro invocations can end up being inlined many-many times, depending on
129  * specifics of each individual user application. So single conceptual USDT
130  * (identified by provider:name pair of identifiers) is, generally speaking,
131  * multiple uprobe locations (USDT call sites) in different places in user
132  * application. Further, again due to inlining, each USDT call site might end
133  * up having the same argument #N be located in a different place. In one call
134  * site it could be a constant, in another will end up in a register, and in
135  * yet another could be some other register or even somewhere on the stack.
136  *
137  * As such, "attaching to USDT" means (in general case) attaching the same
138  * uprobe BPF program to multiple target locations in user application, each
139  * potentially having a completely different USDT spec associated with it.
140  * To wire all this up together libbpf allocates a unique integer spec ID for
141  * each unique USDT spec. Spec IDs are allocated as sequential small integers
142  * so that they can be used as keys in array BPF map (for performance reasons).
143  * Spec ID allocation and accounting is big part of what usdt_manager is
144  * about. This state has to be maintained per-BPF object and coordinate
145  * between different USDT attachments within the same BPF object.
146  *
147  * Spec ID is the key in spec BPF map, value is the actual USDT spec layed out
148  * as struct usdt_spec. Each invocation of BPF program at runtime needs to
149  * know its associated spec ID. It gets it either through BPF cookie, which
150  * libbpf sets to spec ID during attach time, or, if kernel is too old to
151  * support BPF cookie, through IP-to-spec-ID map that libbpf maintains in such
152  * case. The latter means that some modes of operation can't be supported
153  * without BPF cookie. Such mode is attaching to shared library "generically",
154  * without specifying target process. In such case, it's impossible to
155  * calculate absolute IP addresses for IP-to-spec-ID map, and thus such mode
156  * is not supported without BPF cookie support.
157  *
158  * Note that libbpf is using BPF cookie functionality for its own internal
159  * needs, so user itself can't rely on BPF cookie feature. To that end, libbpf
160  * provides conceptually equivalent USDT cookie support. It's still u64
161  * user-provided value that can be associated with USDT attachment. Note that
162  * this will be the same value for all USDT call sites within the same single
163  * *logical* USDT attachment. This makes sense because to user attaching to
164  * USDT is a single BPF program triggered for singular USDT probe. The fact
165  * that this is done at multiple actual locations is a mostly hidden
166  * implementation details. This USDT cookie value can be fetched with
167  * bpf_usdt_cookie(ctx) API provided by usdt.bpf.h
168  *
169  * Lastly, while single USDT can have tons of USDT call sites, it doesn't
170  * necessarily have that many different USDT specs. It very well might be
171  * that 1000 USDT call sites only need 5 different USDT specs, because all the
172  * arguments are typically contained in a small set of registers or stack
173  * locations. As such, it's wasteful to allocate as many USDT spec IDs as
174  * there are USDT call sites. So libbpf tries to be frugal and performs
175  * on-the-fly deduplication during a single USDT attachment to only allocate
176  * the minimal required amount of unique USDT specs (and thus spec IDs). This
177  * is trivially achieved by using USDT spec string (Arguments string from USDT
178  * note) as a lookup key in a hashmap. USDT spec string uniquely defines
179  * everything about how to fetch USDT arguments, so two USDT call sites
180  * sharing USDT spec string can safely share the same USDT spec and spec ID.
181  * Note, this spec string deduplication is happening only during the same USDT
182  * attachment, so each USDT spec shares the same USDT cookie value. This is
183  * not generally true for other USDT attachments within the same BPF object,
184  * as even if USDT spec string is the same, USDT cookie value can be
185  * different. It was deemed excessive to try to deduplicate across independent
186  * USDT attachments by taking into account USDT spec string *and* USDT cookie
187  * value, which would complicated spec ID accounting significantly for little
188  * gain.
189  */
190 
191 #define USDT_BASE_SEC ".stapsdt.base"
192 #define USDT_SEMA_SEC ".probes"
193 #define USDT_NOTE_SEC  ".note.stapsdt"
194 #define USDT_NOTE_TYPE 3
195 #define USDT_NOTE_NAME "stapsdt"
196 
197 /* should match exactly enum __bpf_usdt_arg_type from usdt.bpf.h */
198 enum usdt_arg_type {
199 	USDT_ARG_CONST,
200 	USDT_ARG_REG,
201 	USDT_ARG_REG_DEREF,
202 };
203 
204 /* should match exactly struct __bpf_usdt_arg_spec from usdt.bpf.h */
205 struct usdt_arg_spec {
206 	__u64 val_off;
207 	enum usdt_arg_type arg_type;
208 	short reg_off;
209 	bool arg_signed;
210 	char arg_bitshift;
211 };
212 
213 /* should match BPF_USDT_MAX_ARG_CNT in usdt.bpf.h */
214 #define USDT_MAX_ARG_CNT 12
215 
216 /* should match struct __bpf_usdt_spec from usdt.bpf.h */
217 struct usdt_spec {
218 	struct usdt_arg_spec args[USDT_MAX_ARG_CNT];
219 	__u64 usdt_cookie;
220 	short arg_cnt;
221 };
222 
223 struct usdt_note {
224 	const char *provider;
225 	const char *name;
226 	/* USDT args specification string, e.g.:
227 	 * "-4@%esi -4@-24(%rbp) -4@%ecx 2@%ax 8@%rdx"
228 	 */
229 	const char *args;
230 	long loc_addr;
231 	long base_addr;
232 	long sema_addr;
233 };
234 
235 struct usdt_target {
236 	long abs_ip;
237 	long rel_ip;
238 	long sema_off;
239 	struct usdt_spec spec;
240 	const char *spec_str;
241 };
242 
243 struct usdt_manager {
244 	struct bpf_map *specs_map;
245 	struct bpf_map *ip_to_spec_id_map;
246 
247 	int *free_spec_ids;
248 	size_t free_spec_cnt;
249 	size_t next_free_spec_id;
250 
251 	bool has_bpf_cookie;
252 	bool has_sema_refcnt;
253 };
254 
255 struct usdt_manager *usdt_manager_new(struct bpf_object *obj)
256 {
257 	static const char *ref_ctr_sysfs_path = "/sys/bus/event_source/devices/uprobe/format/ref_ctr_offset";
258 	struct usdt_manager *man;
259 	struct bpf_map *specs_map, *ip_to_spec_id_map;
260 
261 	specs_map = bpf_object__find_map_by_name(obj, "__bpf_usdt_specs");
262 	ip_to_spec_id_map = bpf_object__find_map_by_name(obj, "__bpf_usdt_ip_to_spec_id");
263 	if (!specs_map || !ip_to_spec_id_map) {
264 		pr_warn("usdt: failed to find USDT support BPF maps, did you forget to include bpf/usdt.bpf.h?\n");
265 		return ERR_PTR(-ESRCH);
266 	}
267 
268 	man = calloc(1, sizeof(*man));
269 	if (!man)
270 		return ERR_PTR(-ENOMEM);
271 
272 	man->specs_map = specs_map;
273 	man->ip_to_spec_id_map = ip_to_spec_id_map;
274 
275 	/* Detect if BPF cookie is supported for kprobes.
276 	 * We don't need IP-to-ID mapping if we can use BPF cookies.
277 	 * Added in: 7adfc6c9b315 ("bpf: Add bpf_get_attach_cookie() BPF helper to access bpf_cookie value")
278 	 */
279 	man->has_bpf_cookie = kernel_supports(obj, FEAT_BPF_COOKIE);
280 
281 	/* Detect kernel support for automatic refcounting of USDT semaphore.
282 	 * If this is not supported, USDTs with semaphores will not be supported.
283 	 * Added in: a6ca88b241d5 ("trace_uprobe: support reference counter in fd-based uprobe")
284 	 */
285 	man->has_sema_refcnt = access(ref_ctr_sysfs_path, F_OK) == 0;
286 
287 	return man;
288 }
289 
290 void usdt_manager_free(struct usdt_manager *man)
291 {
292 	if (IS_ERR_OR_NULL(man))
293 		return;
294 
295 	free(man->free_spec_ids);
296 	free(man);
297 }
298 
299 static int sanity_check_usdt_elf(Elf *elf, const char *path)
300 {
301 	GElf_Ehdr ehdr;
302 	int endianness;
303 
304 	if (elf_kind(elf) != ELF_K_ELF) {
305 		pr_warn("usdt: unrecognized ELF kind %d for '%s'\n", elf_kind(elf), path);
306 		return -EBADF;
307 	}
308 
309 	switch (gelf_getclass(elf)) {
310 	case ELFCLASS64:
311 		if (sizeof(void *) != 8) {
312 			pr_warn("usdt: attaching to 64-bit ELF binary '%s' is not supported\n", path);
313 			return -EBADF;
314 		}
315 		break;
316 	case ELFCLASS32:
317 		if (sizeof(void *) != 4) {
318 			pr_warn("usdt: attaching to 32-bit ELF binary '%s' is not supported\n", path);
319 			return -EBADF;
320 		}
321 		break;
322 	default:
323 		pr_warn("usdt: unsupported ELF class for '%s'\n", path);
324 		return -EBADF;
325 	}
326 
327 	if (!gelf_getehdr(elf, &ehdr))
328 		return -EINVAL;
329 
330 	if (ehdr.e_type != ET_EXEC && ehdr.e_type != ET_DYN) {
331 		pr_warn("usdt: unsupported type of ELF binary '%s' (%d), only ET_EXEC and ET_DYN are supported\n",
332 			path, ehdr.e_type);
333 		return -EBADF;
334 	}
335 
336 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
337 	endianness = ELFDATA2LSB;
338 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
339 	endianness = ELFDATA2MSB;
340 #else
341 # error "Unrecognized __BYTE_ORDER__"
342 #endif
343 	if (endianness != ehdr.e_ident[EI_DATA]) {
344 		pr_warn("usdt: ELF endianness mismatch for '%s'\n", path);
345 		return -EBADF;
346 	}
347 
348 	return 0;
349 }
350 
351 static int find_elf_sec_by_name(Elf *elf, const char *sec_name, GElf_Shdr *shdr, Elf_Scn **scn)
352 {
353 	Elf_Scn *sec = NULL;
354 	size_t shstrndx;
355 
356 	if (elf_getshdrstrndx(elf, &shstrndx))
357 		return -EINVAL;
358 
359 	/* check if ELF is corrupted and avoid calling elf_strptr if yes */
360 	if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL))
361 		return -EINVAL;
362 
363 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
364 		char *name;
365 
366 		if (!gelf_getshdr(sec, shdr))
367 			return -EINVAL;
368 
369 		name = elf_strptr(elf, shstrndx, shdr->sh_name);
370 		if (name && strcmp(sec_name, name) == 0) {
371 			*scn = sec;
372 			return 0;
373 		}
374 	}
375 
376 	return -ENOENT;
377 }
378 
379 struct elf_seg {
380 	long start;
381 	long end;
382 	long offset;
383 	bool is_exec;
384 };
385 
386 static int cmp_elf_segs(const void *_a, const void *_b)
387 {
388 	const struct elf_seg *a = _a;
389 	const struct elf_seg *b = _b;
390 
391 	return a->start < b->start ? -1 : 1;
392 }
393 
394 static int parse_elf_segs(Elf *elf, const char *path, struct elf_seg **segs, size_t *seg_cnt)
395 {
396 	GElf_Phdr phdr;
397 	size_t n;
398 	int i, err;
399 	struct elf_seg *seg;
400 	void *tmp;
401 
402 	*seg_cnt = 0;
403 
404 	if (elf_getphdrnum(elf, &n)) {
405 		err = -errno;
406 		return err;
407 	}
408 
409 	for (i = 0; i < n; i++) {
410 		if (!gelf_getphdr(elf, i, &phdr)) {
411 			err = -errno;
412 			return err;
413 		}
414 
415 		pr_debug("usdt: discovered PHDR #%d in '%s': vaddr 0x%lx memsz 0x%lx offset 0x%lx type 0x%lx flags 0x%lx\n",
416 			 i, path, (long)phdr.p_vaddr, (long)phdr.p_memsz, (long)phdr.p_offset,
417 			 (long)phdr.p_type, (long)phdr.p_flags);
418 		if (phdr.p_type != PT_LOAD)
419 			continue;
420 
421 		tmp = libbpf_reallocarray(*segs, *seg_cnt + 1, sizeof(**segs));
422 		if (!tmp)
423 			return -ENOMEM;
424 
425 		*segs = tmp;
426 		seg = *segs + *seg_cnt;
427 		(*seg_cnt)++;
428 
429 		seg->start = phdr.p_vaddr;
430 		seg->end = phdr.p_vaddr + phdr.p_memsz;
431 		seg->offset = phdr.p_offset;
432 		seg->is_exec = phdr.p_flags & PF_X;
433 	}
434 
435 	if (*seg_cnt == 0) {
436 		pr_warn("usdt: failed to find PT_LOAD program headers in '%s'\n", path);
437 		return -ESRCH;
438 	}
439 
440 	qsort(*segs, *seg_cnt, sizeof(**segs), cmp_elf_segs);
441 	return 0;
442 }
443 
444 static int parse_vma_segs(int pid, const char *lib_path, struct elf_seg **segs, size_t *seg_cnt)
445 {
446 	char path[PATH_MAX], line[PATH_MAX], mode[16];
447 	size_t seg_start, seg_end, seg_off;
448 	struct elf_seg *seg;
449 	int tmp_pid, i, err;
450 	FILE *f;
451 
452 	*seg_cnt = 0;
453 
454 	/* Handle containerized binaries only accessible from
455 	 * /proc/<pid>/root/<path>. They will be reported as just /<path> in
456 	 * /proc/<pid>/maps.
457 	 */
458 	if (sscanf(lib_path, "/proc/%d/root%s", &tmp_pid, path) == 2 && pid == tmp_pid)
459 		goto proceed;
460 
461 	if (!realpath(lib_path, path)) {
462 		pr_warn("usdt: failed to get absolute path of '%s' (err %d), using path as is...\n",
463 			lib_path, -errno);
464 		libbpf_strlcpy(path, lib_path, sizeof(path));
465 	}
466 
467 proceed:
468 	sprintf(line, "/proc/%d/maps", pid);
469 	f = fopen(line, "r");
470 	if (!f) {
471 		err = -errno;
472 		pr_warn("usdt: failed to open '%s' to get base addr of '%s': %d\n",
473 			line, lib_path, err);
474 		return err;
475 	}
476 
477 	/* We need to handle lines with no path at the end:
478 	 *
479 	 * 7f5c6f5d1000-7f5c6f5d3000 rw-p 001c7000 08:04 21238613      /usr/lib64/libc-2.17.so
480 	 * 7f5c6f5d3000-7f5c6f5d8000 rw-p 00000000 00:00 0
481 	 * 7f5c6f5d8000-7f5c6f5d9000 r-xp 00000000 103:01 362990598    /data/users/andriin/linux/tools/bpf/usdt/libhello_usdt.so
482 	 */
483 	while (fscanf(f, "%zx-%zx %s %zx %*s %*d%[^\n]\n",
484 		      &seg_start, &seg_end, mode, &seg_off, line) == 5) {
485 		void *tmp;
486 
487 		/* to handle no path case (see above) we need to capture line
488 		 * without skipping any whitespaces. So we need to strip
489 		 * leading whitespaces manually here
490 		 */
491 		i = 0;
492 		while (isblank(line[i]))
493 			i++;
494 		if (strcmp(line + i, path) != 0)
495 			continue;
496 
497 		pr_debug("usdt: discovered segment for lib '%s': addrs %zx-%zx mode %s offset %zx\n",
498 			 path, seg_start, seg_end, mode, seg_off);
499 
500 		/* ignore non-executable sections for shared libs */
501 		if (mode[2] != 'x')
502 			continue;
503 
504 		tmp = libbpf_reallocarray(*segs, *seg_cnt + 1, sizeof(**segs));
505 		if (!tmp) {
506 			err = -ENOMEM;
507 			goto err_out;
508 		}
509 
510 		*segs = tmp;
511 		seg = *segs + *seg_cnt;
512 		*seg_cnt += 1;
513 
514 		seg->start = seg_start;
515 		seg->end = seg_end;
516 		seg->offset = seg_off;
517 		seg->is_exec = true;
518 	}
519 
520 	if (*seg_cnt == 0) {
521 		pr_warn("usdt: failed to find '%s' (resolved to '%s') within PID %d memory mappings\n",
522 			lib_path, path, pid);
523 		err = -ESRCH;
524 		goto err_out;
525 	}
526 
527 	qsort(*segs, *seg_cnt, sizeof(**segs), cmp_elf_segs);
528 	err = 0;
529 err_out:
530 	fclose(f);
531 	return err;
532 }
533 
534 static struct elf_seg *find_elf_seg(struct elf_seg *segs, size_t seg_cnt, long virtaddr)
535 {
536 	struct elf_seg *seg;
537 	int i;
538 
539 	/* for ELF binaries (both executables and shared libraries), we are
540 	 * given virtual address (absolute for executables, relative for
541 	 * libraries) which should match address range of [seg_start, seg_end)
542 	 */
543 	for (i = 0, seg = segs; i < seg_cnt; i++, seg++) {
544 		if (seg->start <= virtaddr && virtaddr < seg->end)
545 			return seg;
546 	}
547 	return NULL;
548 }
549 
550 static struct elf_seg *find_vma_seg(struct elf_seg *segs, size_t seg_cnt, long offset)
551 {
552 	struct elf_seg *seg;
553 	int i;
554 
555 	/* for VMA segments from /proc/<pid>/maps file, provided "address" is
556 	 * actually a file offset, so should be fall within logical
557 	 * offset-based range of [offset_start, offset_end)
558 	 */
559 	for (i = 0, seg = segs; i < seg_cnt; i++, seg++) {
560 		if (seg->offset <= offset && offset < seg->offset + (seg->end - seg->start))
561 			return seg;
562 	}
563 	return NULL;
564 }
565 
566 static int parse_usdt_note(Elf *elf, const char *path, GElf_Nhdr *nhdr,
567 			   const char *data, size_t name_off, size_t desc_off,
568 			   struct usdt_note *usdt_note);
569 
570 static int parse_usdt_spec(struct usdt_spec *spec, const struct usdt_note *note, __u64 usdt_cookie);
571 
572 static int collect_usdt_targets(struct usdt_manager *man, Elf *elf, const char *path, pid_t pid,
573 				const char *usdt_provider, const char *usdt_name, __u64 usdt_cookie,
574 				struct usdt_target **out_targets, size_t *out_target_cnt)
575 {
576 	size_t off, name_off, desc_off, seg_cnt = 0, vma_seg_cnt = 0, target_cnt = 0;
577 	struct elf_seg *segs = NULL, *vma_segs = NULL;
578 	struct usdt_target *targets = NULL, *target;
579 	long base_addr = 0;
580 	Elf_Scn *notes_scn, *base_scn;
581 	GElf_Shdr base_shdr, notes_shdr;
582 	GElf_Ehdr ehdr;
583 	GElf_Nhdr nhdr;
584 	Elf_Data *data;
585 	int err;
586 
587 	*out_targets = NULL;
588 	*out_target_cnt = 0;
589 
590 	err = find_elf_sec_by_name(elf, USDT_NOTE_SEC, &notes_shdr, &notes_scn);
591 	if (err) {
592 		pr_warn("usdt: no USDT notes section (%s) found in '%s'\n", USDT_NOTE_SEC, path);
593 		return err;
594 	}
595 
596 	if (notes_shdr.sh_type != SHT_NOTE || !gelf_getehdr(elf, &ehdr)) {
597 		pr_warn("usdt: invalid USDT notes section (%s) in '%s'\n", USDT_NOTE_SEC, path);
598 		return -EINVAL;
599 	}
600 
601 	err = parse_elf_segs(elf, path, &segs, &seg_cnt);
602 	if (err) {
603 		pr_warn("usdt: failed to process ELF program segments for '%s': %d\n", path, err);
604 		goto err_out;
605 	}
606 
607 	/* .stapsdt.base ELF section is optional, but is used for prelink
608 	 * offset compensation (see a big comment further below)
609 	 */
610 	if (find_elf_sec_by_name(elf, USDT_BASE_SEC, &base_shdr, &base_scn) == 0)
611 		base_addr = base_shdr.sh_addr;
612 
613 	data = elf_getdata(notes_scn, 0);
614 	off = 0;
615 	while ((off = gelf_getnote(data, off, &nhdr, &name_off, &desc_off)) > 0) {
616 		long usdt_abs_ip, usdt_rel_ip, usdt_sema_off = 0;
617 		struct usdt_note note;
618 		struct elf_seg *seg = NULL;
619 		void *tmp;
620 
621 		err = parse_usdt_note(elf, path, &nhdr, data->d_buf, name_off, desc_off, &note);
622 		if (err)
623 			goto err_out;
624 
625 		if (strcmp(note.provider, usdt_provider) != 0 || strcmp(note.name, usdt_name) != 0)
626 			continue;
627 
628 		/* We need to compensate "prelink effect". See [0] for details,
629 		 * relevant parts quoted here:
630 		 *
631 		 * Each SDT probe also expands into a non-allocated ELF note. You can
632 		 * find this by looking at SHT_NOTE sections and decoding the format;
633 		 * see below for details. Because the note is non-allocated, it means
634 		 * there is no runtime cost, and also preserved in both stripped files
635 		 * and .debug files.
636 		 *
637 		 * However, this means that prelink won't adjust the note's contents
638 		 * for address offsets. Instead, this is done via the .stapsdt.base
639 		 * section. This is a special section that is added to the text. We
640 		 * will only ever have one of these sections in a final link and it
641 		 * will only ever be one byte long. Nothing about this section itself
642 		 * matters, we just use it as a marker to detect prelink address
643 		 * adjustments.
644 		 *
645 		 * Each probe note records the link-time address of the .stapsdt.base
646 		 * section alongside the probe PC address. The decoder compares the
647 		 * base address stored in the note with the .stapsdt.base section's
648 		 * sh_addr. Initially these are the same, but the section header will
649 		 * be adjusted by prelink. So the decoder applies the difference to
650 		 * the probe PC address to get the correct prelinked PC address; the
651 		 * same adjustment is applied to the semaphore address, if any.
652 		 *
653 		 *   [0] https://sourceware.org/systemtap/wiki/UserSpaceProbeImplementation
654 		 */
655 		usdt_abs_ip = note.loc_addr;
656 		if (base_addr)
657 			usdt_abs_ip += base_addr - note.base_addr;
658 
659 		/* When attaching uprobes (which is what USDTs basically are)
660 		 * kernel expects file offset to be specified, not a relative
661 		 * virtual address, so we need to translate virtual address to
662 		 * file offset, for both ET_EXEC and ET_DYN binaries.
663 		 */
664 		seg = find_elf_seg(segs, seg_cnt, usdt_abs_ip);
665 		if (!seg) {
666 			err = -ESRCH;
667 			pr_warn("usdt: failed to find ELF program segment for '%s:%s' in '%s' at IP 0x%lx\n",
668 				usdt_provider, usdt_name, path, usdt_abs_ip);
669 			goto err_out;
670 		}
671 		if (!seg->is_exec) {
672 			err = -ESRCH;
673 			pr_warn("usdt: matched ELF binary '%s' segment [0x%lx, 0x%lx) for '%s:%s' at IP 0x%lx is not executable\n",
674 				path, seg->start, seg->end, usdt_provider, usdt_name,
675 				usdt_abs_ip);
676 			goto err_out;
677 		}
678 		/* translate from virtual address to file offset */
679 		usdt_rel_ip = usdt_abs_ip - seg->start + seg->offset;
680 
681 		if (ehdr.e_type == ET_DYN && !man->has_bpf_cookie) {
682 			/* If we don't have BPF cookie support but need to
683 			 * attach to a shared library, we'll need to know and
684 			 * record absolute addresses of attach points due to
685 			 * the need to lookup USDT spec by absolute IP of
686 			 * triggered uprobe. Doing this resolution is only
687 			 * possible when we have a specific PID of the process
688 			 * that's using specified shared library. BPF cookie
689 			 * removes the absolute address limitation as we don't
690 			 * need to do this lookup (we just use BPF cookie as
691 			 * an index of USDT spec), so for newer kernels with
692 			 * BPF cookie support libbpf supports USDT attachment
693 			 * to shared libraries with no PID filter.
694 			 */
695 			if (pid < 0) {
696 				pr_warn("usdt: attaching to shared libraries without specific PID is not supported on current kernel\n");
697 				err = -ENOTSUP;
698 				goto err_out;
699 			}
700 
701 			/* vma_segs are lazily initialized only if necessary */
702 			if (vma_seg_cnt == 0) {
703 				err = parse_vma_segs(pid, path, &vma_segs, &vma_seg_cnt);
704 				if (err) {
705 					pr_warn("usdt: failed to get memory segments in PID %d for shared library '%s': %d\n",
706 						pid, path, err);
707 					goto err_out;
708 				}
709 			}
710 
711 			seg = find_vma_seg(vma_segs, vma_seg_cnt, usdt_rel_ip);
712 			if (!seg) {
713 				err = -ESRCH;
714 				pr_warn("usdt: failed to find shared lib memory segment for '%s:%s' in '%s' at relative IP 0x%lx\n",
715 					usdt_provider, usdt_name, path, usdt_rel_ip);
716 				goto err_out;
717 			}
718 
719 			usdt_abs_ip = seg->start - seg->offset + usdt_rel_ip;
720 		}
721 
722 		pr_debug("usdt: probe for '%s:%s' in %s '%s': addr 0x%lx base 0x%lx (resolved abs_ip 0x%lx rel_ip 0x%lx) args '%s' in segment [0x%lx, 0x%lx) at offset 0x%lx\n",
723 			 usdt_provider, usdt_name, ehdr.e_type == ET_EXEC ? "exec" : "lib ", path,
724 			 note.loc_addr, note.base_addr, usdt_abs_ip, usdt_rel_ip, note.args,
725 			 seg ? seg->start : 0, seg ? seg->end : 0, seg ? seg->offset : 0);
726 
727 		/* Adjust semaphore address to be a file offset */
728 		if (note.sema_addr) {
729 			if (!man->has_sema_refcnt) {
730 				pr_warn("usdt: kernel doesn't support USDT semaphore refcounting for '%s:%s' in '%s'\n",
731 					usdt_provider, usdt_name, path);
732 				err = -ENOTSUP;
733 				goto err_out;
734 			}
735 
736 			seg = find_elf_seg(segs, seg_cnt, note.sema_addr);
737 			if (!seg) {
738 				err = -ESRCH;
739 				pr_warn("usdt: failed to find ELF loadable segment with semaphore of '%s:%s' in '%s' at 0x%lx\n",
740 					usdt_provider, usdt_name, path, note.sema_addr);
741 				goto err_out;
742 			}
743 			if (seg->is_exec) {
744 				err = -ESRCH;
745 				pr_warn("usdt: matched ELF binary '%s' segment [0x%lx, 0x%lx] for semaphore of '%s:%s' at 0x%lx is executable\n",
746 					path, seg->start, seg->end, usdt_provider, usdt_name,
747 					note.sema_addr);
748 				goto err_out;
749 			}
750 
751 			usdt_sema_off = note.sema_addr - seg->start + seg->offset;
752 
753 			pr_debug("usdt: sema  for '%s:%s' in %s '%s': addr 0x%lx base 0x%lx (resolved 0x%lx) in segment [0x%lx, 0x%lx] at offset 0x%lx\n",
754 				 usdt_provider, usdt_name, ehdr.e_type == ET_EXEC ? "exec" : "lib ",
755 				 path, note.sema_addr, note.base_addr, usdt_sema_off,
756 				 seg->start, seg->end, seg->offset);
757 		}
758 
759 		/* Record adjusted addresses and offsets and parse USDT spec */
760 		tmp = libbpf_reallocarray(targets, target_cnt + 1, sizeof(*targets));
761 		if (!tmp) {
762 			err = -ENOMEM;
763 			goto err_out;
764 		}
765 		targets = tmp;
766 
767 		target = &targets[target_cnt];
768 		memset(target, 0, sizeof(*target));
769 
770 		target->abs_ip = usdt_abs_ip;
771 		target->rel_ip = usdt_rel_ip;
772 		target->sema_off = usdt_sema_off;
773 
774 		/* notes.args references strings from Elf itself, so they can
775 		 * be referenced safely until elf_end() call
776 		 */
777 		target->spec_str = note.args;
778 
779 		err = parse_usdt_spec(&target->spec, &note, usdt_cookie);
780 		if (err)
781 			goto err_out;
782 
783 		target_cnt++;
784 	}
785 
786 	*out_targets = targets;
787 	*out_target_cnt = target_cnt;
788 	err = target_cnt;
789 
790 err_out:
791 	free(segs);
792 	free(vma_segs);
793 	if (err < 0)
794 		free(targets);
795 	return err;
796 }
797 
798 struct bpf_link_usdt {
799 	struct bpf_link link;
800 
801 	struct usdt_manager *usdt_man;
802 
803 	size_t spec_cnt;
804 	int *spec_ids;
805 
806 	size_t uprobe_cnt;
807 	struct {
808 		long abs_ip;
809 		struct bpf_link *link;
810 	} *uprobes;
811 };
812 
813 static int bpf_link_usdt_detach(struct bpf_link *link)
814 {
815 	struct bpf_link_usdt *usdt_link = container_of(link, struct bpf_link_usdt, link);
816 	struct usdt_manager *man = usdt_link->usdt_man;
817 	int i;
818 
819 	for (i = 0; i < usdt_link->uprobe_cnt; i++) {
820 		/* detach underlying uprobe link */
821 		bpf_link__destroy(usdt_link->uprobes[i].link);
822 		/* there is no need to update specs map because it will be
823 		 * unconditionally overwritten on subsequent USDT attaches,
824 		 * but if BPF cookies are not used we need to remove entry
825 		 * from ip_to_spec_id map, otherwise we'll run into false
826 		 * conflicting IP errors
827 		 */
828 		if (!man->has_bpf_cookie) {
829 			/* not much we can do about errors here */
830 			(void)bpf_map_delete_elem(bpf_map__fd(man->ip_to_spec_id_map),
831 						  &usdt_link->uprobes[i].abs_ip);
832 		}
833 	}
834 
835 	/* try to return the list of previously used spec IDs to usdt_manager
836 	 * for future reuse for subsequent USDT attaches
837 	 */
838 	if (!man->free_spec_ids) {
839 		/* if there were no free spec IDs yet, just transfer our IDs */
840 		man->free_spec_ids = usdt_link->spec_ids;
841 		man->free_spec_cnt = usdt_link->spec_cnt;
842 		usdt_link->spec_ids = NULL;
843 	} else {
844 		/* otherwise concat IDs */
845 		size_t new_cnt = man->free_spec_cnt + usdt_link->spec_cnt;
846 		int *new_free_ids;
847 
848 		new_free_ids = libbpf_reallocarray(man->free_spec_ids, new_cnt,
849 						   sizeof(*new_free_ids));
850 		/* If we couldn't resize free_spec_ids, we'll just leak
851 		 * a bunch of free IDs; this is very unlikely to happen and if
852 		 * system is so exhausted on memory, it's the least of user's
853 		 * concerns, probably.
854 		 * So just do our best here to return those IDs to usdt_manager.
855 		 */
856 		if (new_free_ids) {
857 			memcpy(new_free_ids + man->free_spec_cnt, usdt_link->spec_ids,
858 			       usdt_link->spec_cnt * sizeof(*usdt_link->spec_ids));
859 			man->free_spec_ids = new_free_ids;
860 			man->free_spec_cnt = new_cnt;
861 		}
862 	}
863 
864 	return 0;
865 }
866 
867 static void bpf_link_usdt_dealloc(struct bpf_link *link)
868 {
869 	struct bpf_link_usdt *usdt_link = container_of(link, struct bpf_link_usdt, link);
870 
871 	free(usdt_link->spec_ids);
872 	free(usdt_link->uprobes);
873 	free(usdt_link);
874 }
875 
876 static size_t specs_hash_fn(const void *key, void *ctx)
877 {
878 	const char *s = key;
879 
880 	return str_hash(s);
881 }
882 
883 static bool specs_equal_fn(const void *key1, const void *key2, void *ctx)
884 {
885 	const char *s1 = key1;
886 	const char *s2 = key2;
887 
888 	return strcmp(s1, s2) == 0;
889 }
890 
891 static int allocate_spec_id(struct usdt_manager *man, struct hashmap *specs_hash,
892 			    struct bpf_link_usdt *link, struct usdt_target *target,
893 			    int *spec_id, bool *is_new)
894 {
895 	void *tmp;
896 	int err;
897 
898 	/* check if we already allocated spec ID for this spec string */
899 	if (hashmap__find(specs_hash, target->spec_str, &tmp)) {
900 		*spec_id = (long)tmp;
901 		*is_new = false;
902 		return 0;
903 	}
904 
905 	/* otherwise it's a new ID that needs to be set up in specs map and
906 	 * returned back to usdt_manager when USDT link is detached
907 	 */
908 	tmp = libbpf_reallocarray(link->spec_ids, link->spec_cnt + 1, sizeof(*link->spec_ids));
909 	if (!tmp)
910 		return -ENOMEM;
911 	link->spec_ids = tmp;
912 
913 	/* get next free spec ID, giving preference to free list, if not empty */
914 	if (man->free_spec_cnt) {
915 		*spec_id = man->free_spec_ids[man->free_spec_cnt - 1];
916 
917 		/* cache spec ID for current spec string for future lookups */
918 		err = hashmap__add(specs_hash, target->spec_str, (void *)(long)*spec_id);
919 		if (err)
920 			 return err;
921 
922 		man->free_spec_cnt--;
923 	} else {
924 		/* don't allocate spec ID bigger than what fits in specs map */
925 		if (man->next_free_spec_id >= bpf_map__max_entries(man->specs_map))
926 			return -E2BIG;
927 
928 		*spec_id = man->next_free_spec_id;
929 
930 		/* cache spec ID for current spec string for future lookups */
931 		err = hashmap__add(specs_hash, target->spec_str, (void *)(long)*spec_id);
932 		if (err)
933 			 return err;
934 
935 		man->next_free_spec_id++;
936 	}
937 
938 	/* remember new spec ID in the link for later return back to free list on detach */
939 	link->spec_ids[link->spec_cnt] = *spec_id;
940 	link->spec_cnt++;
941 	*is_new = true;
942 	return 0;
943 }
944 
945 struct bpf_link *usdt_manager_attach_usdt(struct usdt_manager *man, const struct bpf_program *prog,
946 					  pid_t pid, const char *path,
947 					  const char *usdt_provider, const char *usdt_name,
948 					  __u64 usdt_cookie)
949 {
950 	int i, fd, err, spec_map_fd, ip_map_fd;
951 	LIBBPF_OPTS(bpf_uprobe_opts, opts);
952 	struct hashmap *specs_hash = NULL;
953 	struct bpf_link_usdt *link = NULL;
954 	struct usdt_target *targets = NULL;
955 	size_t target_cnt;
956 	Elf *elf;
957 
958 	spec_map_fd = bpf_map__fd(man->specs_map);
959 	ip_map_fd = bpf_map__fd(man->ip_to_spec_id_map);
960 
961 	/* TODO: perform path resolution similar to uprobe's */
962 	fd = open(path, O_RDONLY);
963 	if (fd < 0) {
964 		err = -errno;
965 		pr_warn("usdt: failed to open ELF binary '%s': %d\n", path, err);
966 		return libbpf_err_ptr(err);
967 	}
968 
969 	elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
970 	if (!elf) {
971 		err = -EBADF;
972 		pr_warn("usdt: failed to parse ELF binary '%s': %s\n", path, elf_errmsg(-1));
973 		goto err_out;
974 	}
975 
976 	err = sanity_check_usdt_elf(elf, path);
977 	if (err)
978 		goto err_out;
979 
980 	/* normalize PID filter */
981 	if (pid < 0)
982 		pid = -1;
983 	else if (pid == 0)
984 		pid = getpid();
985 
986 	/* discover USDT in given binary, optionally limiting
987 	 * activations to a given PID, if pid > 0
988 	 */
989 	err = collect_usdt_targets(man, elf, path, pid, usdt_provider, usdt_name,
990 				   usdt_cookie, &targets, &target_cnt);
991 	if (err <= 0) {
992 		err = (err == 0) ? -ENOENT : err;
993 		goto err_out;
994 	}
995 
996 	specs_hash = hashmap__new(specs_hash_fn, specs_equal_fn, NULL);
997 	if (IS_ERR(specs_hash)) {
998 		err = PTR_ERR(specs_hash);
999 		goto err_out;
1000 	}
1001 
1002 	link = calloc(1, sizeof(*link));
1003 	if (!link) {
1004 		err = -ENOMEM;
1005 		goto err_out;
1006 	}
1007 
1008 	link->usdt_man = man;
1009 	link->link.detach = &bpf_link_usdt_detach;
1010 	link->link.dealloc = &bpf_link_usdt_dealloc;
1011 
1012 	link->uprobes = calloc(target_cnt, sizeof(*link->uprobes));
1013 	if (!link->uprobes) {
1014 		err = -ENOMEM;
1015 		goto err_out;
1016 	}
1017 
1018 	for (i = 0; i < target_cnt; i++) {
1019 		struct usdt_target *target = &targets[i];
1020 		struct bpf_link *uprobe_link;
1021 		bool is_new;
1022 		int spec_id;
1023 
1024 		/* Spec ID can be either reused or newly allocated. If it is
1025 		 * newly allocated, we'll need to fill out spec map, otherwise
1026 		 * entire spec should be valid and can be just used by a new
1027 		 * uprobe. We reuse spec when USDT arg spec is identical. We
1028 		 * also never share specs between two different USDT
1029 		 * attachments ("links"), so all the reused specs already
1030 		 * share USDT cookie value implicitly.
1031 		 */
1032 		err = allocate_spec_id(man, specs_hash, link, target, &spec_id, &is_new);
1033 		if (err)
1034 			goto err_out;
1035 
1036 		if (is_new && bpf_map_update_elem(spec_map_fd, &spec_id, &target->spec, BPF_ANY)) {
1037 			err = -errno;
1038 			pr_warn("usdt: failed to set USDT spec #%d for '%s:%s' in '%s': %d\n",
1039 				spec_id, usdt_provider, usdt_name, path, err);
1040 			goto err_out;
1041 		}
1042 		if (!man->has_bpf_cookie &&
1043 		    bpf_map_update_elem(ip_map_fd, &target->abs_ip, &spec_id, BPF_NOEXIST)) {
1044 			err = -errno;
1045 			if (err == -EEXIST) {
1046 				pr_warn("usdt: IP collision detected for spec #%d for '%s:%s' in '%s'\n",
1047 				        spec_id, usdt_provider, usdt_name, path);
1048 			} else {
1049 				pr_warn("usdt: failed to map IP 0x%lx to spec #%d for '%s:%s' in '%s': %d\n",
1050 					target->abs_ip, spec_id, usdt_provider, usdt_name,
1051 					path, err);
1052 			}
1053 			goto err_out;
1054 		}
1055 
1056 		opts.ref_ctr_offset = target->sema_off;
1057 		opts.bpf_cookie = man->has_bpf_cookie ? spec_id : 0;
1058 		uprobe_link = bpf_program__attach_uprobe_opts(prog, pid, path,
1059 							      target->rel_ip, &opts);
1060 		err = libbpf_get_error(uprobe_link);
1061 		if (err) {
1062 			pr_warn("usdt: failed to attach uprobe #%d for '%s:%s' in '%s': %d\n",
1063 				i, usdt_provider, usdt_name, path, err);
1064 			goto err_out;
1065 		}
1066 
1067 		link->uprobes[i].link = uprobe_link;
1068 		link->uprobes[i].abs_ip = target->abs_ip;
1069 		link->uprobe_cnt++;
1070 	}
1071 
1072 	free(targets);
1073 	hashmap__free(specs_hash);
1074 	elf_end(elf);
1075 	close(fd);
1076 
1077 	return &link->link;
1078 
1079 err_out:
1080 	if (link)
1081 		bpf_link__destroy(&link->link);
1082 	free(targets);
1083 	hashmap__free(specs_hash);
1084 	if (elf)
1085 		elf_end(elf);
1086 	close(fd);
1087 	return libbpf_err_ptr(err);
1088 }
1089 
1090 /* Parse out USDT ELF note from '.note.stapsdt' section.
1091  * Logic inspired by perf's code.
1092  */
1093 static int parse_usdt_note(Elf *elf, const char *path, GElf_Nhdr *nhdr,
1094 			   const char *data, size_t name_off, size_t desc_off,
1095 			   struct usdt_note *note)
1096 {
1097 	const char *provider, *name, *args;
1098 	long addrs[3];
1099 	size_t len;
1100 
1101 	/* sanity check USDT note name and type first */
1102 	if (strncmp(data + name_off, USDT_NOTE_NAME, nhdr->n_namesz) != 0)
1103 		return -EINVAL;
1104 	if (nhdr->n_type != USDT_NOTE_TYPE)
1105 		return -EINVAL;
1106 
1107 	/* sanity check USDT note contents ("description" in ELF terminology) */
1108 	len = nhdr->n_descsz;
1109 	data = data + desc_off;
1110 
1111 	/* +3 is the very minimum required to store three empty strings */
1112 	if (len < sizeof(addrs) + 3)
1113 		return -EINVAL;
1114 
1115 	/* get location, base, and semaphore addrs */
1116 	memcpy(&addrs, data, sizeof(addrs));
1117 
1118 	/* parse string fields: provider, name, args */
1119 	provider = data + sizeof(addrs);
1120 
1121 	name = (const char *)memchr(provider, '\0', data + len - provider);
1122 	if (!name) /* non-zero-terminated provider */
1123 		return -EINVAL;
1124 	name++;
1125 	if (name >= data + len || *name == '\0') /* missing or empty name */
1126 		return -EINVAL;
1127 
1128 	args = memchr(name, '\0', data + len - name);
1129 	if (!args) /* non-zero-terminated name */
1130 		return -EINVAL;
1131 	++args;
1132 	if (args >= data + len) /* missing arguments spec */
1133 		return -EINVAL;
1134 
1135 	note->provider = provider;
1136 	note->name = name;
1137 	if (*args == '\0' || *args == ':')
1138 		note->args = "";
1139 	else
1140 		note->args = args;
1141 	note->loc_addr = addrs[0];
1142 	note->base_addr = addrs[1];
1143 	note->sema_addr = addrs[2];
1144 
1145 	return 0;
1146 }
1147 
1148 static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg);
1149 
1150 static int parse_usdt_spec(struct usdt_spec *spec, const struct usdt_note *note, __u64 usdt_cookie)
1151 {
1152 	const char *s;
1153 	int len;
1154 
1155 	spec->usdt_cookie = usdt_cookie;
1156 	spec->arg_cnt = 0;
1157 
1158 	s = note->args;
1159 	while (s[0]) {
1160 		if (spec->arg_cnt >= USDT_MAX_ARG_CNT) {
1161 			pr_warn("usdt: too many USDT arguments (> %d) for '%s:%s' with args spec '%s'\n",
1162 				USDT_MAX_ARG_CNT, note->provider, note->name, note->args);
1163 			return -E2BIG;
1164 		}
1165 
1166 		len = parse_usdt_arg(s, spec->arg_cnt, &spec->args[spec->arg_cnt]);
1167 		if (len < 0)
1168 			return len;
1169 
1170 		s += len;
1171 		spec->arg_cnt++;
1172 	}
1173 
1174 	return 0;
1175 }
1176 
1177 /* Architecture-specific logic for parsing USDT argument location specs */
1178 
1179 #if defined(__x86_64__) || defined(__i386__)
1180 
1181 static int calc_pt_regs_off(const char *reg_name)
1182 {
1183 	static struct {
1184 		const char *names[4];
1185 		size_t pt_regs_off;
1186 	} reg_map[] = {
1187 #ifdef __x86_64__
1188 #define reg_off(reg64, reg32) offsetof(struct pt_regs, reg64)
1189 #else
1190 #define reg_off(reg64, reg32) offsetof(struct pt_regs, reg32)
1191 #endif
1192 		{ {"rip", "eip", "", ""}, reg_off(rip, eip) },
1193 		{ {"rax", "eax", "ax", "al"}, reg_off(rax, eax) },
1194 		{ {"rbx", "ebx", "bx", "bl"}, reg_off(rbx, ebx) },
1195 		{ {"rcx", "ecx", "cx", "cl"}, reg_off(rcx, ecx) },
1196 		{ {"rdx", "edx", "dx", "dl"}, reg_off(rdx, edx) },
1197 		{ {"rsi", "esi", "si", "sil"}, reg_off(rsi, esi) },
1198 		{ {"rdi", "edi", "di", "dil"}, reg_off(rdi, edi) },
1199 		{ {"rbp", "ebp", "bp", "bpl"}, reg_off(rbp, ebp) },
1200 		{ {"rsp", "esp", "sp", "spl"}, reg_off(rsp, esp) },
1201 #undef reg_off
1202 #ifdef __x86_64__
1203 		{ {"r8", "r8d", "r8w", "r8b"}, offsetof(struct pt_regs, r8) },
1204 		{ {"r9", "r9d", "r9w", "r9b"}, offsetof(struct pt_regs, r9) },
1205 		{ {"r10", "r10d", "r10w", "r10b"}, offsetof(struct pt_regs, r10) },
1206 		{ {"r11", "r11d", "r11w", "r11b"}, offsetof(struct pt_regs, r11) },
1207 		{ {"r12", "r12d", "r12w", "r12b"}, offsetof(struct pt_regs, r12) },
1208 		{ {"r13", "r13d", "r13w", "r13b"}, offsetof(struct pt_regs, r13) },
1209 		{ {"r14", "r14d", "r14w", "r14b"}, offsetof(struct pt_regs, r14) },
1210 		{ {"r15", "r15d", "r15w", "r15b"}, offsetof(struct pt_regs, r15) },
1211 #endif
1212 	};
1213 	int i, j;
1214 
1215 	for (i = 0; i < ARRAY_SIZE(reg_map); i++) {
1216 		for (j = 0; j < ARRAY_SIZE(reg_map[i].names); j++) {
1217 			if (strcmp(reg_name, reg_map[i].names[j]) == 0)
1218 				return reg_map[i].pt_regs_off;
1219 		}
1220 	}
1221 
1222 	pr_warn("usdt: unrecognized register '%s'\n", reg_name);
1223 	return -ENOENT;
1224 }
1225 
1226 static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg)
1227 {
1228 	char *reg_name = NULL;
1229 	int arg_sz, len, reg_off;
1230 	long off;
1231 
1232 	if (sscanf(arg_str, " %d @ %ld ( %%%m[^)] ) %n", &arg_sz, &off, &reg_name, &len) == 3) {
1233 		/* Memory dereference case, e.g., -4@-20(%rbp) */
1234 		arg->arg_type = USDT_ARG_REG_DEREF;
1235 		arg->val_off = off;
1236 		reg_off = calc_pt_regs_off(reg_name);
1237 		free(reg_name);
1238 		if (reg_off < 0)
1239 			return reg_off;
1240 		arg->reg_off = reg_off;
1241 	} else if (sscanf(arg_str, " %d @ %%%ms %n", &arg_sz, &reg_name, &len) == 2) {
1242 		/* Register read case, e.g., -4@%eax */
1243 		arg->arg_type = USDT_ARG_REG;
1244 		arg->val_off = 0;
1245 
1246 		reg_off = calc_pt_regs_off(reg_name);
1247 		free(reg_name);
1248 		if (reg_off < 0)
1249 			return reg_off;
1250 		arg->reg_off = reg_off;
1251 	} else if (sscanf(arg_str, " %d @ $%ld %n", &arg_sz, &off, &len) == 2) {
1252 		/* Constant value case, e.g., 4@$71 */
1253 		arg->arg_type = USDT_ARG_CONST;
1254 		arg->val_off = off;
1255 		arg->reg_off = 0;
1256 	} else {
1257 		pr_warn("usdt: unrecognized arg #%d spec '%s'\n", arg_num, arg_str);
1258 		return -EINVAL;
1259 	}
1260 
1261 	arg->arg_signed = arg_sz < 0;
1262 	if (arg_sz < 0)
1263 		arg_sz = -arg_sz;
1264 
1265 	switch (arg_sz) {
1266 	case 1: case 2: case 4: case 8:
1267 		arg->arg_bitshift = 64 - arg_sz * 8;
1268 		break;
1269 	default:
1270 		pr_warn("usdt: unsupported arg #%d (spec '%s') size: %d\n",
1271 			arg_num, arg_str, arg_sz);
1272 		return -EINVAL;
1273 	}
1274 
1275 	return len;
1276 }
1277 
1278 #elif defined(__s390x__)
1279 
1280 /* Do not support __s390__ for now, since user_pt_regs is broken with -m31. */
1281 
1282 static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg)
1283 {
1284 	unsigned int reg;
1285 	int arg_sz, len;
1286 	long off;
1287 
1288 	if (sscanf(arg_str, " %d @ %ld ( %%r%u ) %n", &arg_sz, &off, &reg, &len) == 3) {
1289 		/* Memory dereference case, e.g., -2@-28(%r15) */
1290 		arg->arg_type = USDT_ARG_REG_DEREF;
1291 		arg->val_off = off;
1292 		if (reg > 15) {
1293 			pr_warn("usdt: unrecognized register '%%r%u'\n", reg);
1294 			return -EINVAL;
1295 		}
1296 		arg->reg_off = offsetof(user_pt_regs, gprs[reg]);
1297 	} else if (sscanf(arg_str, " %d @ %%r%u %n", &arg_sz, &reg, &len) == 2) {
1298 		/* Register read case, e.g., -8@%r0 */
1299 		arg->arg_type = USDT_ARG_REG;
1300 		arg->val_off = 0;
1301 		if (reg > 15) {
1302 			pr_warn("usdt: unrecognized register '%%r%u'\n", reg);
1303 			return -EINVAL;
1304 		}
1305 		arg->reg_off = offsetof(user_pt_regs, gprs[reg]);
1306 	} else if (sscanf(arg_str, " %d @ %ld %n", &arg_sz, &off, &len) == 2) {
1307 		/* Constant value case, e.g., 4@71 */
1308 		arg->arg_type = USDT_ARG_CONST;
1309 		arg->val_off = off;
1310 		arg->reg_off = 0;
1311 	} else {
1312 		pr_warn("usdt: unrecognized arg #%d spec '%s'\n", arg_num, arg_str);
1313 		return -EINVAL;
1314 	}
1315 
1316 	arg->arg_signed = arg_sz < 0;
1317 	if (arg_sz < 0)
1318 		arg_sz = -arg_sz;
1319 
1320 	switch (arg_sz) {
1321 	case 1: case 2: case 4: case 8:
1322 		arg->arg_bitshift = 64 - arg_sz * 8;
1323 		break;
1324 	default:
1325 		pr_warn("usdt: unsupported arg #%d (spec '%s') size: %d\n",
1326 			arg_num, arg_str, arg_sz);
1327 		return -EINVAL;
1328 	}
1329 
1330 	return len;
1331 }
1332 
1333 #elif defined(__aarch64__)
1334 
1335 static int calc_pt_regs_off(const char *reg_name)
1336 {
1337 	int reg_num;
1338 
1339 	if (sscanf(reg_name, "x%d", &reg_num) == 1) {
1340 		if (reg_num >= 0 && reg_num < 31)
1341 			return offsetof(struct user_pt_regs, regs[reg_num]);
1342 	} else if (strcmp(reg_name, "sp") == 0) {
1343 		return offsetof(struct user_pt_regs, sp);
1344 	}
1345 	pr_warn("usdt: unrecognized register '%s'\n", reg_name);
1346 	return -ENOENT;
1347 }
1348 
1349 static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg)
1350 {
1351 	char *reg_name = NULL;
1352 	int arg_sz, len, reg_off;
1353 	long off;
1354 
1355 	if (sscanf(arg_str, " %d @ \[ %m[a-z0-9], %ld ] %n", &arg_sz, &reg_name, &off, &len) == 3) {
1356 		/* Memory dereference case, e.g., -4@[sp, 96] */
1357 		arg->arg_type = USDT_ARG_REG_DEREF;
1358 		arg->val_off = off;
1359 		reg_off = calc_pt_regs_off(reg_name);
1360 		free(reg_name);
1361 		if (reg_off < 0)
1362 			return reg_off;
1363 		arg->reg_off = reg_off;
1364 	} else if (sscanf(arg_str, " %d @ \[ %m[a-z0-9] ] %n", &arg_sz, &reg_name, &len) == 2) {
1365 		/* Memory dereference case, e.g., -4@[sp] */
1366 		arg->arg_type = USDT_ARG_REG_DEREF;
1367 		arg->val_off = 0;
1368 		reg_off = calc_pt_regs_off(reg_name);
1369 		free(reg_name);
1370 		if (reg_off < 0)
1371 			return reg_off;
1372 		arg->reg_off = reg_off;
1373 	} else if (sscanf(arg_str, " %d @ %ld %n", &arg_sz, &off, &len) == 2) {
1374 		/* Constant value case, e.g., 4@5 */
1375 		arg->arg_type = USDT_ARG_CONST;
1376 		arg->val_off = off;
1377 		arg->reg_off = 0;
1378 	} else if (sscanf(arg_str, " %d @ %m[a-z0-9] %n", &arg_sz, &reg_name, &len) == 2) {
1379 		/* Register read case, e.g., -8@x4 */
1380 		arg->arg_type = USDT_ARG_REG;
1381 		arg->val_off = 0;
1382 		reg_off = calc_pt_regs_off(reg_name);
1383 		free(reg_name);
1384 		if (reg_off < 0)
1385 			return reg_off;
1386 		arg->reg_off = reg_off;
1387 	} else {
1388 		pr_warn("usdt: unrecognized arg #%d spec '%s'\n", arg_num, arg_str);
1389 		return -EINVAL;
1390 	}
1391 
1392 	arg->arg_signed = arg_sz < 0;
1393 	if (arg_sz < 0)
1394 		arg_sz = -arg_sz;
1395 
1396 	switch (arg_sz) {
1397 	case 1: case 2: case 4: case 8:
1398 		arg->arg_bitshift = 64 - arg_sz * 8;
1399 		break;
1400 	default:
1401 		pr_warn("usdt: unsupported arg #%d (spec '%s') size: %d\n",
1402 			arg_num, arg_str, arg_sz);
1403 		return -EINVAL;
1404 	}
1405 
1406 	return len;
1407 }
1408 
1409 #elif defined(__riscv)
1410 
1411 static int calc_pt_regs_off(const char *reg_name)
1412 {
1413 	static struct {
1414 		const char *name;
1415 		size_t pt_regs_off;
1416 	} reg_map[] = {
1417 		{ "ra", offsetof(struct user_regs_struct, ra) },
1418 		{ "sp", offsetof(struct user_regs_struct, sp) },
1419 		{ "gp", offsetof(struct user_regs_struct, gp) },
1420 		{ "tp", offsetof(struct user_regs_struct, tp) },
1421 		{ "a0", offsetof(struct user_regs_struct, a0) },
1422 		{ "a1", offsetof(struct user_regs_struct, a1) },
1423 		{ "a2", offsetof(struct user_regs_struct, a2) },
1424 		{ "a3", offsetof(struct user_regs_struct, a3) },
1425 		{ "a4", offsetof(struct user_regs_struct, a4) },
1426 		{ "a5", offsetof(struct user_regs_struct, a5) },
1427 		{ "a6", offsetof(struct user_regs_struct, a6) },
1428 		{ "a7", offsetof(struct user_regs_struct, a7) },
1429 		{ "s0", offsetof(struct user_regs_struct, s0) },
1430 		{ "s1", offsetof(struct user_regs_struct, s1) },
1431 		{ "s2", offsetof(struct user_regs_struct, s2) },
1432 		{ "s3", offsetof(struct user_regs_struct, s3) },
1433 		{ "s4", offsetof(struct user_regs_struct, s4) },
1434 		{ "s5", offsetof(struct user_regs_struct, s5) },
1435 		{ "s6", offsetof(struct user_regs_struct, s6) },
1436 		{ "s7", offsetof(struct user_regs_struct, s7) },
1437 		{ "s8", offsetof(struct user_regs_struct, rv_s8) },
1438 		{ "s9", offsetof(struct user_regs_struct, s9) },
1439 		{ "s10", offsetof(struct user_regs_struct, s10) },
1440 		{ "s11", offsetof(struct user_regs_struct, s11) },
1441 		{ "t0", offsetof(struct user_regs_struct, t0) },
1442 		{ "t1", offsetof(struct user_regs_struct, t1) },
1443 		{ "t2", offsetof(struct user_regs_struct, t2) },
1444 		{ "t3", offsetof(struct user_regs_struct, t3) },
1445 		{ "t4", offsetof(struct user_regs_struct, t4) },
1446 		{ "t5", offsetof(struct user_regs_struct, t5) },
1447 		{ "t6", offsetof(struct user_regs_struct, t6) },
1448 	};
1449 	int i;
1450 
1451 	for (i = 0; i < ARRAY_SIZE(reg_map); i++) {
1452 		if (strcmp(reg_name, reg_map[i].name) == 0)
1453 			return reg_map[i].pt_regs_off;
1454 	}
1455 
1456 	pr_warn("usdt: unrecognized register '%s'\n", reg_name);
1457 	return -ENOENT;
1458 }
1459 
1460 static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg)
1461 {
1462 	char *reg_name = NULL;
1463 	int arg_sz, len, reg_off;
1464 	long off;
1465 
1466 	if (sscanf(arg_str, " %d @ %ld ( %m[a-z0-9] ) %n", &arg_sz, &off, &reg_name, &len) == 3) {
1467 		/* Memory dereference case, e.g., -8@-88(s0) */
1468 		arg->arg_type = USDT_ARG_REG_DEREF;
1469 		arg->val_off = off;
1470 		reg_off = calc_pt_regs_off(reg_name);
1471 		free(reg_name);
1472 		if (reg_off < 0)
1473 			return reg_off;
1474 		arg->reg_off = reg_off;
1475 	} else if (sscanf(arg_str, " %d @ %ld %n", &arg_sz, &off, &len) == 2) {
1476 		/* Constant value case, e.g., 4@5 */
1477 		arg->arg_type = USDT_ARG_CONST;
1478 		arg->val_off = off;
1479 		arg->reg_off = 0;
1480 	} else if (sscanf(arg_str, " %d @ %m[a-z0-9] %n", &arg_sz, &reg_name, &len) == 2) {
1481 		/* Register read case, e.g., -8@a1 */
1482 		arg->arg_type = USDT_ARG_REG;
1483 		arg->val_off = 0;
1484 		reg_off = calc_pt_regs_off(reg_name);
1485 		free(reg_name);
1486 		if (reg_off < 0)
1487 			return reg_off;
1488 		arg->reg_off = reg_off;
1489 	} else {
1490 		pr_warn("usdt: unrecognized arg #%d spec '%s'\n", arg_num, arg_str);
1491 		return -EINVAL;
1492 	}
1493 
1494 	arg->arg_signed = arg_sz < 0;
1495 	if (arg_sz < 0)
1496 		arg_sz = -arg_sz;
1497 
1498 	switch (arg_sz) {
1499 	case 1: case 2: case 4: case 8:
1500 		arg->arg_bitshift = 64 - arg_sz * 8;
1501 		break;
1502 	default:
1503 		pr_warn("usdt: unsupported arg #%d (spec '%s') size: %d\n",
1504 			arg_num, arg_str, arg_sz);
1505 		return -EINVAL;
1506 	}
1507 
1508 	return len;
1509 }
1510 
1511 #else
1512 
1513 static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg)
1514 {
1515 	pr_warn("usdt: libbpf doesn't support USDTs on current architecture\n");
1516 	return -ENOTSUP;
1517 }
1518 
1519 #endif
1520