xref: /openbmc/linux/tools/perf/util/dso.c (revision 56d06fa2)
1 #include <asm/bug.h>
2 #include <sys/time.h>
3 #include <sys/resource.h>
4 #include "symbol.h"
5 #include "dso.h"
6 #include "machine.h"
7 #include "auxtrace.h"
8 #include "util.h"
9 #include "debug.h"
10 
11 char dso__symtab_origin(const struct dso *dso)
12 {
13 	static const char origin[] = {
14 		[DSO_BINARY_TYPE__KALLSYMS]			= 'k',
15 		[DSO_BINARY_TYPE__VMLINUX]			= 'v',
16 		[DSO_BINARY_TYPE__JAVA_JIT]			= 'j',
17 		[DSO_BINARY_TYPE__DEBUGLINK]			= 'l',
18 		[DSO_BINARY_TYPE__BUILD_ID_CACHE]		= 'B',
19 		[DSO_BINARY_TYPE__FEDORA_DEBUGINFO]		= 'f',
20 		[DSO_BINARY_TYPE__UBUNTU_DEBUGINFO]		= 'u',
21 		[DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO]	= 'o',
22 		[DSO_BINARY_TYPE__BUILDID_DEBUGINFO]		= 'b',
23 		[DSO_BINARY_TYPE__SYSTEM_PATH_DSO]		= 'd',
24 		[DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE]		= 'K',
25 		[DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP]	= 'm',
26 		[DSO_BINARY_TYPE__GUEST_KALLSYMS]		= 'g',
27 		[DSO_BINARY_TYPE__GUEST_KMODULE]		= 'G',
28 		[DSO_BINARY_TYPE__GUEST_KMODULE_COMP]		= 'M',
29 		[DSO_BINARY_TYPE__GUEST_VMLINUX]		= 'V',
30 	};
31 
32 	if (dso == NULL || dso->symtab_type == DSO_BINARY_TYPE__NOT_FOUND)
33 		return '!';
34 	return origin[dso->symtab_type];
35 }
36 
37 int dso__read_binary_type_filename(const struct dso *dso,
38 				   enum dso_binary_type type,
39 				   char *root_dir, char *filename, size_t size)
40 {
41 	char build_id_hex[BUILD_ID_SIZE * 2 + 1];
42 	int ret = 0;
43 	size_t len;
44 
45 	switch (type) {
46 	case DSO_BINARY_TYPE__DEBUGLINK: {
47 		char *debuglink;
48 
49 		len = __symbol__join_symfs(filename, size, dso->long_name);
50 		debuglink = filename + len;
51 		while (debuglink != filename && *debuglink != '/')
52 			debuglink--;
53 		if (*debuglink == '/')
54 			debuglink++;
55 
56 		ret = -1;
57 		if (!is_regular_file(filename))
58 			break;
59 
60 		ret = filename__read_debuglink(filename, debuglink,
61 					       size - (debuglink - filename));
62 		}
63 		break;
64 	case DSO_BINARY_TYPE__BUILD_ID_CACHE:
65 		/* skip the locally configured cache if a symfs is given */
66 		if (symbol_conf.symfs[0] ||
67 		    (dso__build_id_filename(dso, filename, size) == NULL))
68 			ret = -1;
69 		break;
70 
71 	case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
72 		len = __symbol__join_symfs(filename, size, "/usr/lib/debug");
73 		snprintf(filename + len, size - len, "%s.debug", dso->long_name);
74 		break;
75 
76 	case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
77 		len = __symbol__join_symfs(filename, size, "/usr/lib/debug");
78 		snprintf(filename + len, size - len, "%s", dso->long_name);
79 		break;
80 
81 	case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
82 	{
83 		const char *last_slash;
84 		size_t dir_size;
85 
86 		last_slash = dso->long_name + dso->long_name_len;
87 		while (last_slash != dso->long_name && *last_slash != '/')
88 			last_slash--;
89 
90 		len = __symbol__join_symfs(filename, size, "");
91 		dir_size = last_slash - dso->long_name + 2;
92 		if (dir_size > (size - len)) {
93 			ret = -1;
94 			break;
95 		}
96 		len += scnprintf(filename + len, dir_size, "%s",  dso->long_name);
97 		len += scnprintf(filename + len , size - len, ".debug%s",
98 								last_slash);
99 		break;
100 	}
101 
102 	case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
103 		if (!dso->has_build_id) {
104 			ret = -1;
105 			break;
106 		}
107 
108 		build_id__sprintf(dso->build_id,
109 				  sizeof(dso->build_id),
110 				  build_id_hex);
111 		len = __symbol__join_symfs(filename, size, "/usr/lib/debug/.build-id/");
112 		snprintf(filename + len, size - len, "%.2s/%s.debug",
113 			 build_id_hex, build_id_hex + 2);
114 		break;
115 
116 	case DSO_BINARY_TYPE__VMLINUX:
117 	case DSO_BINARY_TYPE__GUEST_VMLINUX:
118 	case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
119 		__symbol__join_symfs(filename, size, dso->long_name);
120 		break;
121 
122 	case DSO_BINARY_TYPE__GUEST_KMODULE:
123 	case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
124 		path__join3(filename, size, symbol_conf.symfs,
125 			    root_dir, dso->long_name);
126 		break;
127 
128 	case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
129 	case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
130 		__symbol__join_symfs(filename, size, dso->long_name);
131 		break;
132 
133 	case DSO_BINARY_TYPE__KCORE:
134 	case DSO_BINARY_TYPE__GUEST_KCORE:
135 		snprintf(filename, size, "%s", dso->long_name);
136 		break;
137 
138 	default:
139 	case DSO_BINARY_TYPE__KALLSYMS:
140 	case DSO_BINARY_TYPE__GUEST_KALLSYMS:
141 	case DSO_BINARY_TYPE__JAVA_JIT:
142 	case DSO_BINARY_TYPE__NOT_FOUND:
143 		ret = -1;
144 		break;
145 	}
146 
147 	return ret;
148 }
149 
150 static const struct {
151 	const char *fmt;
152 	int (*decompress)(const char *input, int output);
153 } compressions[] = {
154 #ifdef HAVE_ZLIB_SUPPORT
155 	{ "gz", gzip_decompress_to_file },
156 #endif
157 #ifdef HAVE_LZMA_SUPPORT
158 	{ "xz", lzma_decompress_to_file },
159 #endif
160 	{ NULL, NULL },
161 };
162 
163 bool is_supported_compression(const char *ext)
164 {
165 	unsigned i;
166 
167 	for (i = 0; compressions[i].fmt; i++) {
168 		if (!strcmp(ext, compressions[i].fmt))
169 			return true;
170 	}
171 	return false;
172 }
173 
174 bool is_kernel_module(const char *pathname, int cpumode)
175 {
176 	struct kmod_path m;
177 	int mode = cpumode & PERF_RECORD_MISC_CPUMODE_MASK;
178 
179 	WARN_ONCE(mode != cpumode,
180 		  "Internal error: passing unmasked cpumode (%x) to is_kernel_module",
181 		  cpumode);
182 
183 	switch (mode) {
184 	case PERF_RECORD_MISC_USER:
185 	case PERF_RECORD_MISC_HYPERVISOR:
186 	case PERF_RECORD_MISC_GUEST_USER:
187 		return false;
188 	/* Treat PERF_RECORD_MISC_CPUMODE_UNKNOWN as kernel */
189 	default:
190 		if (kmod_path__parse(&m, pathname)) {
191 			pr_err("Failed to check whether %s is a kernel module or not. Assume it is.",
192 					pathname);
193 			return true;
194 		}
195 	}
196 
197 	return m.kmod;
198 }
199 
200 bool decompress_to_file(const char *ext, const char *filename, int output_fd)
201 {
202 	unsigned i;
203 
204 	for (i = 0; compressions[i].fmt; i++) {
205 		if (!strcmp(ext, compressions[i].fmt))
206 			return !compressions[i].decompress(filename,
207 							   output_fd);
208 	}
209 	return false;
210 }
211 
212 bool dso__needs_decompress(struct dso *dso)
213 {
214 	return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP ||
215 		dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE_COMP;
216 }
217 
218 /*
219  * Parses kernel module specified in @path and updates
220  * @m argument like:
221  *
222  *    @comp - true if @path contains supported compression suffix,
223  *            false otherwise
224  *    @kmod - true if @path contains '.ko' suffix in right position,
225  *            false otherwise
226  *    @name - if (@alloc_name && @kmod) is true, it contains strdup-ed base name
227  *            of the kernel module without suffixes, otherwise strudup-ed
228  *            base name of @path
229  *    @ext  - if (@alloc_ext && @comp) is true, it contains strdup-ed string
230  *            the compression suffix
231  *
232  * Returns 0 if there's no strdup error, -ENOMEM otherwise.
233  */
234 int __kmod_path__parse(struct kmod_path *m, const char *path,
235 		       bool alloc_name, bool alloc_ext)
236 {
237 	const char *name = strrchr(path, '/');
238 	const char *ext  = strrchr(path, '.');
239 	bool is_simple_name = false;
240 
241 	memset(m, 0x0, sizeof(*m));
242 	name = name ? name + 1 : path;
243 
244 	/*
245 	 * '.' is also a valid character for module name. For example:
246 	 * [aaa.bbb] is a valid module name. '[' should have higher
247 	 * priority than '.ko' suffix.
248 	 *
249 	 * The kernel names are from machine__mmap_name. Such
250 	 * name should belong to kernel itself, not kernel module.
251 	 */
252 	if (name[0] == '[') {
253 		is_simple_name = true;
254 		if ((strncmp(name, "[kernel.kallsyms]", 17) == 0) ||
255 		    (strncmp(name, "[guest.kernel.kallsyms", 22) == 0) ||
256 		    (strncmp(name, "[vdso]", 6) == 0) ||
257 		    (strncmp(name, "[vsyscall]", 10) == 0)) {
258 			m->kmod = false;
259 
260 		} else
261 			m->kmod = true;
262 	}
263 
264 	/* No extension, just return name. */
265 	if ((ext == NULL) || is_simple_name) {
266 		if (alloc_name) {
267 			m->name = strdup(name);
268 			return m->name ? 0 : -ENOMEM;
269 		}
270 		return 0;
271 	}
272 
273 	if (is_supported_compression(ext + 1)) {
274 		m->comp = true;
275 		ext -= 3;
276 	}
277 
278 	/* Check .ko extension only if there's enough name left. */
279 	if (ext > name)
280 		m->kmod = !strncmp(ext, ".ko", 3);
281 
282 	if (alloc_name) {
283 		if (m->kmod) {
284 			if (asprintf(&m->name, "[%.*s]", (int) (ext - name), name) == -1)
285 				return -ENOMEM;
286 		} else {
287 			if (asprintf(&m->name, "%s", name) == -1)
288 				return -ENOMEM;
289 		}
290 
291 		strxfrchar(m->name, '-', '_');
292 	}
293 
294 	if (alloc_ext && m->comp) {
295 		m->ext = strdup(ext + 4);
296 		if (!m->ext) {
297 			free((void *) m->name);
298 			return -ENOMEM;
299 		}
300 	}
301 
302 	return 0;
303 }
304 
305 /*
306  * Global list of open DSOs and the counter.
307  */
308 static LIST_HEAD(dso__data_open);
309 static long dso__data_open_cnt;
310 static pthread_mutex_t dso__data_open_lock = PTHREAD_MUTEX_INITIALIZER;
311 
312 static void dso__list_add(struct dso *dso)
313 {
314 	list_add_tail(&dso->data.open_entry, &dso__data_open);
315 	dso__data_open_cnt++;
316 }
317 
318 static void dso__list_del(struct dso *dso)
319 {
320 	list_del(&dso->data.open_entry);
321 	WARN_ONCE(dso__data_open_cnt <= 0,
322 		  "DSO data fd counter out of bounds.");
323 	dso__data_open_cnt--;
324 }
325 
326 static void close_first_dso(void);
327 
328 static int do_open(char *name)
329 {
330 	int fd;
331 	char sbuf[STRERR_BUFSIZE];
332 
333 	do {
334 		fd = open(name, O_RDONLY);
335 		if (fd >= 0)
336 			return fd;
337 
338 		pr_debug("dso open failed: %s\n",
339 			 strerror_r(errno, sbuf, sizeof(sbuf)));
340 		if (!dso__data_open_cnt || errno != EMFILE)
341 			break;
342 
343 		close_first_dso();
344 	} while (1);
345 
346 	return -1;
347 }
348 
349 static int __open_dso(struct dso *dso, struct machine *machine)
350 {
351 	int fd;
352 	char *root_dir = (char *)"";
353 	char *name = malloc(PATH_MAX);
354 
355 	if (!name)
356 		return -ENOMEM;
357 
358 	if (machine)
359 		root_dir = machine->root_dir;
360 
361 	if (dso__read_binary_type_filename(dso, dso->binary_type,
362 					    root_dir, name, PATH_MAX)) {
363 		free(name);
364 		return -EINVAL;
365 	}
366 
367 	fd = do_open(name);
368 	free(name);
369 	return fd;
370 }
371 
372 static void check_data_close(void);
373 
374 /**
375  * dso_close - Open DSO data file
376  * @dso: dso object
377  *
378  * Open @dso's data file descriptor and updates
379  * list/count of open DSO objects.
380  */
381 static int open_dso(struct dso *dso, struct machine *machine)
382 {
383 	int fd = __open_dso(dso, machine);
384 
385 	if (fd >= 0) {
386 		dso__list_add(dso);
387 		/*
388 		 * Check if we crossed the allowed number
389 		 * of opened DSOs and close one if needed.
390 		 */
391 		check_data_close();
392 	}
393 
394 	return fd;
395 }
396 
397 static void close_data_fd(struct dso *dso)
398 {
399 	if (dso->data.fd >= 0) {
400 		close(dso->data.fd);
401 		dso->data.fd = -1;
402 		dso->data.file_size = 0;
403 		dso__list_del(dso);
404 	}
405 }
406 
407 /**
408  * dso_close - Close DSO data file
409  * @dso: dso object
410  *
411  * Close @dso's data file descriptor and updates
412  * list/count of open DSO objects.
413  */
414 static void close_dso(struct dso *dso)
415 {
416 	close_data_fd(dso);
417 }
418 
419 static void close_first_dso(void)
420 {
421 	struct dso *dso;
422 
423 	dso = list_first_entry(&dso__data_open, struct dso, data.open_entry);
424 	close_dso(dso);
425 }
426 
427 static rlim_t get_fd_limit(void)
428 {
429 	struct rlimit l;
430 	rlim_t limit = 0;
431 
432 	/* Allow half of the current open fd limit. */
433 	if (getrlimit(RLIMIT_NOFILE, &l) == 0) {
434 		if (l.rlim_cur == RLIM_INFINITY)
435 			limit = l.rlim_cur;
436 		else
437 			limit = l.rlim_cur / 2;
438 	} else {
439 		pr_err("failed to get fd limit\n");
440 		limit = 1;
441 	}
442 
443 	return limit;
444 }
445 
446 static bool may_cache_fd(void)
447 {
448 	static rlim_t limit;
449 
450 	if (!limit)
451 		limit = get_fd_limit();
452 
453 	if (limit == RLIM_INFINITY)
454 		return true;
455 
456 	return limit > (rlim_t) dso__data_open_cnt;
457 }
458 
459 /*
460  * Check and close LRU dso if we crossed allowed limit
461  * for opened dso file descriptors. The limit is half
462  * of the RLIMIT_NOFILE files opened.
463 */
464 static void check_data_close(void)
465 {
466 	bool cache_fd = may_cache_fd();
467 
468 	if (!cache_fd)
469 		close_first_dso();
470 }
471 
472 /**
473  * dso__data_close - Close DSO data file
474  * @dso: dso object
475  *
476  * External interface to close @dso's data file descriptor.
477  */
478 void dso__data_close(struct dso *dso)
479 {
480 	pthread_mutex_lock(&dso__data_open_lock);
481 	close_dso(dso);
482 	pthread_mutex_unlock(&dso__data_open_lock);
483 }
484 
485 static void try_to_open_dso(struct dso *dso, struct machine *machine)
486 {
487 	enum dso_binary_type binary_type_data[] = {
488 		DSO_BINARY_TYPE__BUILD_ID_CACHE,
489 		DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
490 		DSO_BINARY_TYPE__NOT_FOUND,
491 	};
492 	int i = 0;
493 
494 	if (dso->data.fd >= 0)
495 		return;
496 
497 	if (dso->binary_type != DSO_BINARY_TYPE__NOT_FOUND) {
498 		dso->data.fd = open_dso(dso, machine);
499 		goto out;
500 	}
501 
502 	do {
503 		dso->binary_type = binary_type_data[i++];
504 
505 		dso->data.fd = open_dso(dso, machine);
506 		if (dso->data.fd >= 0)
507 			goto out;
508 
509 	} while (dso->binary_type != DSO_BINARY_TYPE__NOT_FOUND);
510 out:
511 	if (dso->data.fd >= 0)
512 		dso->data.status = DSO_DATA_STATUS_OK;
513 	else
514 		dso->data.status = DSO_DATA_STATUS_ERROR;
515 }
516 
517 /**
518  * dso__data_get_fd - Get dso's data file descriptor
519  * @dso: dso object
520  * @machine: machine object
521  *
522  * External interface to find dso's file, open it and
523  * returns file descriptor.  It should be paired with
524  * dso__data_put_fd() if it returns non-negative value.
525  */
526 int dso__data_get_fd(struct dso *dso, struct machine *machine)
527 {
528 	if (dso->data.status == DSO_DATA_STATUS_ERROR)
529 		return -1;
530 
531 	if (pthread_mutex_lock(&dso__data_open_lock) < 0)
532 		return -1;
533 
534 	try_to_open_dso(dso, machine);
535 
536 	if (dso->data.fd < 0)
537 		pthread_mutex_unlock(&dso__data_open_lock);
538 
539 	return dso->data.fd;
540 }
541 
542 void dso__data_put_fd(struct dso *dso __maybe_unused)
543 {
544 	pthread_mutex_unlock(&dso__data_open_lock);
545 }
546 
547 bool dso__data_status_seen(struct dso *dso, enum dso_data_status_seen by)
548 {
549 	u32 flag = 1 << by;
550 
551 	if (dso->data.status_seen & flag)
552 		return true;
553 
554 	dso->data.status_seen |= flag;
555 
556 	return false;
557 }
558 
559 static void
560 dso_cache__free(struct dso *dso)
561 {
562 	struct rb_root *root = &dso->data.cache;
563 	struct rb_node *next = rb_first(root);
564 
565 	pthread_mutex_lock(&dso->lock);
566 	while (next) {
567 		struct dso_cache *cache;
568 
569 		cache = rb_entry(next, struct dso_cache, rb_node);
570 		next = rb_next(&cache->rb_node);
571 		rb_erase(&cache->rb_node, root);
572 		free(cache);
573 	}
574 	pthread_mutex_unlock(&dso->lock);
575 }
576 
577 static struct dso_cache *dso_cache__find(struct dso *dso, u64 offset)
578 {
579 	const struct rb_root *root = &dso->data.cache;
580 	struct rb_node * const *p = &root->rb_node;
581 	const struct rb_node *parent = NULL;
582 	struct dso_cache *cache;
583 
584 	while (*p != NULL) {
585 		u64 end;
586 
587 		parent = *p;
588 		cache = rb_entry(parent, struct dso_cache, rb_node);
589 		end = cache->offset + DSO__DATA_CACHE_SIZE;
590 
591 		if (offset < cache->offset)
592 			p = &(*p)->rb_left;
593 		else if (offset >= end)
594 			p = &(*p)->rb_right;
595 		else
596 			return cache;
597 	}
598 
599 	return NULL;
600 }
601 
602 static struct dso_cache *
603 dso_cache__insert(struct dso *dso, struct dso_cache *new)
604 {
605 	struct rb_root *root = &dso->data.cache;
606 	struct rb_node **p = &root->rb_node;
607 	struct rb_node *parent = NULL;
608 	struct dso_cache *cache;
609 	u64 offset = new->offset;
610 
611 	pthread_mutex_lock(&dso->lock);
612 	while (*p != NULL) {
613 		u64 end;
614 
615 		parent = *p;
616 		cache = rb_entry(parent, struct dso_cache, rb_node);
617 		end = cache->offset + DSO__DATA_CACHE_SIZE;
618 
619 		if (offset < cache->offset)
620 			p = &(*p)->rb_left;
621 		else if (offset >= end)
622 			p = &(*p)->rb_right;
623 		else
624 			goto out;
625 	}
626 
627 	rb_link_node(&new->rb_node, parent, p);
628 	rb_insert_color(&new->rb_node, root);
629 
630 	cache = NULL;
631 out:
632 	pthread_mutex_unlock(&dso->lock);
633 	return cache;
634 }
635 
636 static ssize_t
637 dso_cache__memcpy(struct dso_cache *cache, u64 offset,
638 		  u8 *data, u64 size)
639 {
640 	u64 cache_offset = offset - cache->offset;
641 	u64 cache_size   = min(cache->size - cache_offset, size);
642 
643 	memcpy(data, cache->data + cache_offset, cache_size);
644 	return cache_size;
645 }
646 
647 static ssize_t
648 dso_cache__read(struct dso *dso, struct machine *machine,
649 		u64 offset, u8 *data, ssize_t size)
650 {
651 	struct dso_cache *cache;
652 	struct dso_cache *old;
653 	ssize_t ret;
654 
655 	do {
656 		u64 cache_offset;
657 
658 		cache = zalloc(sizeof(*cache) + DSO__DATA_CACHE_SIZE);
659 		if (!cache)
660 			return -ENOMEM;
661 
662 		pthread_mutex_lock(&dso__data_open_lock);
663 
664 		/*
665 		 * dso->data.fd might be closed if other thread opened another
666 		 * file (dso) due to open file limit (RLIMIT_NOFILE).
667 		 */
668 		try_to_open_dso(dso, machine);
669 
670 		if (dso->data.fd < 0) {
671 			ret = -errno;
672 			dso->data.status = DSO_DATA_STATUS_ERROR;
673 			break;
674 		}
675 
676 		cache_offset = offset & DSO__DATA_CACHE_MASK;
677 
678 		ret = pread(dso->data.fd, cache->data, DSO__DATA_CACHE_SIZE, cache_offset);
679 		if (ret <= 0)
680 			break;
681 
682 		cache->offset = cache_offset;
683 		cache->size   = ret;
684 	} while (0);
685 
686 	pthread_mutex_unlock(&dso__data_open_lock);
687 
688 	if (ret > 0) {
689 		old = dso_cache__insert(dso, cache);
690 		if (old) {
691 			/* we lose the race */
692 			free(cache);
693 			cache = old;
694 		}
695 
696 		ret = dso_cache__memcpy(cache, offset, data, size);
697 	}
698 
699 	if (ret <= 0)
700 		free(cache);
701 
702 	return ret;
703 }
704 
705 static ssize_t dso_cache_read(struct dso *dso, struct machine *machine,
706 			      u64 offset, u8 *data, ssize_t size)
707 {
708 	struct dso_cache *cache;
709 
710 	cache = dso_cache__find(dso, offset);
711 	if (cache)
712 		return dso_cache__memcpy(cache, offset, data, size);
713 	else
714 		return dso_cache__read(dso, machine, offset, data, size);
715 }
716 
717 /*
718  * Reads and caches dso data DSO__DATA_CACHE_SIZE size chunks
719  * in the rb_tree. Any read to already cached data is served
720  * by cached data.
721  */
722 static ssize_t cached_read(struct dso *dso, struct machine *machine,
723 			   u64 offset, u8 *data, ssize_t size)
724 {
725 	ssize_t r = 0;
726 	u8 *p = data;
727 
728 	do {
729 		ssize_t ret;
730 
731 		ret = dso_cache_read(dso, machine, offset, p, size);
732 		if (ret < 0)
733 			return ret;
734 
735 		/* Reached EOF, return what we have. */
736 		if (!ret)
737 			break;
738 
739 		BUG_ON(ret > size);
740 
741 		r      += ret;
742 		p      += ret;
743 		offset += ret;
744 		size   -= ret;
745 
746 	} while (size);
747 
748 	return r;
749 }
750 
751 static int data_file_size(struct dso *dso, struct machine *machine)
752 {
753 	int ret = 0;
754 	struct stat st;
755 	char sbuf[STRERR_BUFSIZE];
756 
757 	if (dso->data.file_size)
758 		return 0;
759 
760 	if (dso->data.status == DSO_DATA_STATUS_ERROR)
761 		return -1;
762 
763 	pthread_mutex_lock(&dso__data_open_lock);
764 
765 	/*
766 	 * dso->data.fd might be closed if other thread opened another
767 	 * file (dso) due to open file limit (RLIMIT_NOFILE).
768 	 */
769 	try_to_open_dso(dso, machine);
770 
771 	if (dso->data.fd < 0) {
772 		ret = -errno;
773 		dso->data.status = DSO_DATA_STATUS_ERROR;
774 		goto out;
775 	}
776 
777 	if (fstat(dso->data.fd, &st) < 0) {
778 		ret = -errno;
779 		pr_err("dso cache fstat failed: %s\n",
780 		       strerror_r(errno, sbuf, sizeof(sbuf)));
781 		dso->data.status = DSO_DATA_STATUS_ERROR;
782 		goto out;
783 	}
784 	dso->data.file_size = st.st_size;
785 
786 out:
787 	pthread_mutex_unlock(&dso__data_open_lock);
788 	return ret;
789 }
790 
791 /**
792  * dso__data_size - Return dso data size
793  * @dso: dso object
794  * @machine: machine object
795  *
796  * Return: dso data size
797  */
798 off_t dso__data_size(struct dso *dso, struct machine *machine)
799 {
800 	if (data_file_size(dso, machine))
801 		return -1;
802 
803 	/* For now just estimate dso data size is close to file size */
804 	return dso->data.file_size;
805 }
806 
807 static ssize_t data_read_offset(struct dso *dso, struct machine *machine,
808 				u64 offset, u8 *data, ssize_t size)
809 {
810 	if (data_file_size(dso, machine))
811 		return -1;
812 
813 	/* Check the offset sanity. */
814 	if (offset > dso->data.file_size)
815 		return -1;
816 
817 	if (offset + size < offset)
818 		return -1;
819 
820 	return cached_read(dso, machine, offset, data, size);
821 }
822 
823 /**
824  * dso__data_read_offset - Read data from dso file offset
825  * @dso: dso object
826  * @machine: machine object
827  * @offset: file offset
828  * @data: buffer to store data
829  * @size: size of the @data buffer
830  *
831  * External interface to read data from dso file offset. Open
832  * dso data file and use cached_read to get the data.
833  */
834 ssize_t dso__data_read_offset(struct dso *dso, struct machine *machine,
835 			      u64 offset, u8 *data, ssize_t size)
836 {
837 	if (dso->data.status == DSO_DATA_STATUS_ERROR)
838 		return -1;
839 
840 	return data_read_offset(dso, machine, offset, data, size);
841 }
842 
843 /**
844  * dso__data_read_addr - Read data from dso address
845  * @dso: dso object
846  * @machine: machine object
847  * @add: virtual memory address
848  * @data: buffer to store data
849  * @size: size of the @data buffer
850  *
851  * External interface to read data from dso address.
852  */
853 ssize_t dso__data_read_addr(struct dso *dso, struct map *map,
854 			    struct machine *machine, u64 addr,
855 			    u8 *data, ssize_t size)
856 {
857 	u64 offset = map->map_ip(map, addr);
858 	return dso__data_read_offset(dso, machine, offset, data, size);
859 }
860 
861 struct map *dso__new_map(const char *name)
862 {
863 	struct map *map = NULL;
864 	struct dso *dso = dso__new(name);
865 
866 	if (dso)
867 		map = map__new2(0, dso, MAP__FUNCTION);
868 
869 	return map;
870 }
871 
872 struct dso *machine__findnew_kernel(struct machine *machine, const char *name,
873 				    const char *short_name, int dso_type)
874 {
875 	/*
876 	 * The kernel dso could be created by build_id processing.
877 	 */
878 	struct dso *dso = machine__findnew_dso(machine, name);
879 
880 	/*
881 	 * We need to run this in all cases, since during the build_id
882 	 * processing we had no idea this was the kernel dso.
883 	 */
884 	if (dso != NULL) {
885 		dso__set_short_name(dso, short_name, false);
886 		dso->kernel = dso_type;
887 	}
888 
889 	return dso;
890 }
891 
892 /*
893  * Find a matching entry and/or link current entry to RB tree.
894  * Either one of the dso or name parameter must be non-NULL or the
895  * function will not work.
896  */
897 static struct dso *__dso__findlink_by_longname(struct rb_root *root,
898 					       struct dso *dso, const char *name)
899 {
900 	struct rb_node **p = &root->rb_node;
901 	struct rb_node  *parent = NULL;
902 
903 	if (!name)
904 		name = dso->long_name;
905 	/*
906 	 * Find node with the matching name
907 	 */
908 	while (*p) {
909 		struct dso *this = rb_entry(*p, struct dso, rb_node);
910 		int rc = strcmp(name, this->long_name);
911 
912 		parent = *p;
913 		if (rc == 0) {
914 			/*
915 			 * In case the new DSO is a duplicate of an existing
916 			 * one, print an one-time warning & put the new entry
917 			 * at the end of the list of duplicates.
918 			 */
919 			if (!dso || (dso == this))
920 				return this;	/* Find matching dso */
921 			/*
922 			 * The core kernel DSOs may have duplicated long name.
923 			 * In this case, the short name should be different.
924 			 * Comparing the short names to differentiate the DSOs.
925 			 */
926 			rc = strcmp(dso->short_name, this->short_name);
927 			if (rc == 0) {
928 				pr_err("Duplicated dso name: %s\n", name);
929 				return NULL;
930 			}
931 		}
932 		if (rc < 0)
933 			p = &parent->rb_left;
934 		else
935 			p = &parent->rb_right;
936 	}
937 	if (dso) {
938 		/* Add new node and rebalance tree */
939 		rb_link_node(&dso->rb_node, parent, p);
940 		rb_insert_color(&dso->rb_node, root);
941 		dso->root = root;
942 	}
943 	return NULL;
944 }
945 
946 static inline struct dso *__dso__find_by_longname(struct rb_root *root,
947 						  const char *name)
948 {
949 	return __dso__findlink_by_longname(root, NULL, name);
950 }
951 
952 void dso__set_long_name(struct dso *dso, const char *name, bool name_allocated)
953 {
954 	struct rb_root *root = dso->root;
955 
956 	if (name == NULL)
957 		return;
958 
959 	if (dso->long_name_allocated)
960 		free((char *)dso->long_name);
961 
962 	if (root) {
963 		rb_erase(&dso->rb_node, root);
964 		/*
965 		 * __dso__findlink_by_longname() isn't guaranteed to add it
966 		 * back, so a clean removal is required here.
967 		 */
968 		RB_CLEAR_NODE(&dso->rb_node);
969 		dso->root = NULL;
970 	}
971 
972 	dso->long_name		 = name;
973 	dso->long_name_len	 = strlen(name);
974 	dso->long_name_allocated = name_allocated;
975 
976 	if (root)
977 		__dso__findlink_by_longname(root, dso, NULL);
978 }
979 
980 void dso__set_short_name(struct dso *dso, const char *name, bool name_allocated)
981 {
982 	if (name == NULL)
983 		return;
984 
985 	if (dso->short_name_allocated)
986 		free((char *)dso->short_name);
987 
988 	dso->short_name		  = name;
989 	dso->short_name_len	  = strlen(name);
990 	dso->short_name_allocated = name_allocated;
991 }
992 
993 static void dso__set_basename(struct dso *dso)
994 {
995        /*
996         * basename() may modify path buffer, so we must pass
997         * a copy.
998         */
999        char *base, *lname = strdup(dso->long_name);
1000 
1001        if (!lname)
1002                return;
1003 
1004        /*
1005         * basename() may return a pointer to internal
1006         * storage which is reused in subsequent calls
1007         * so copy the result.
1008         */
1009        base = strdup(basename(lname));
1010 
1011        free(lname);
1012 
1013        if (!base)
1014                return;
1015 
1016        dso__set_short_name(dso, base, true);
1017 }
1018 
1019 int dso__name_len(const struct dso *dso)
1020 {
1021 	if (!dso)
1022 		return strlen("[unknown]");
1023 	if (verbose)
1024 		return dso->long_name_len;
1025 
1026 	return dso->short_name_len;
1027 }
1028 
1029 bool dso__loaded(const struct dso *dso, enum map_type type)
1030 {
1031 	return dso->loaded & (1 << type);
1032 }
1033 
1034 bool dso__sorted_by_name(const struct dso *dso, enum map_type type)
1035 {
1036 	return dso->sorted_by_name & (1 << type);
1037 }
1038 
1039 void dso__set_sorted_by_name(struct dso *dso, enum map_type type)
1040 {
1041 	dso->sorted_by_name |= (1 << type);
1042 }
1043 
1044 struct dso *dso__new(const char *name)
1045 {
1046 	struct dso *dso = calloc(1, sizeof(*dso) + strlen(name) + 1);
1047 
1048 	if (dso != NULL) {
1049 		int i;
1050 		strcpy(dso->name, name);
1051 		dso__set_long_name(dso, dso->name, false);
1052 		dso__set_short_name(dso, dso->name, false);
1053 		for (i = 0; i < MAP__NR_TYPES; ++i)
1054 			dso->symbols[i] = dso->symbol_names[i] = RB_ROOT;
1055 		dso->data.cache = RB_ROOT;
1056 		dso->data.fd = -1;
1057 		dso->data.status = DSO_DATA_STATUS_UNKNOWN;
1058 		dso->symtab_type = DSO_BINARY_TYPE__NOT_FOUND;
1059 		dso->binary_type = DSO_BINARY_TYPE__NOT_FOUND;
1060 		dso->is_64_bit = (sizeof(void *) == 8);
1061 		dso->loaded = 0;
1062 		dso->rel = 0;
1063 		dso->sorted_by_name = 0;
1064 		dso->has_build_id = 0;
1065 		dso->has_srcline = 1;
1066 		dso->a2l_fails = 1;
1067 		dso->kernel = DSO_TYPE_USER;
1068 		dso->needs_swap = DSO_SWAP__UNSET;
1069 		RB_CLEAR_NODE(&dso->rb_node);
1070 		dso->root = NULL;
1071 		INIT_LIST_HEAD(&dso->node);
1072 		INIT_LIST_HEAD(&dso->data.open_entry);
1073 		pthread_mutex_init(&dso->lock, NULL);
1074 		atomic_set(&dso->refcnt, 1);
1075 	}
1076 
1077 	return dso;
1078 }
1079 
1080 void dso__delete(struct dso *dso)
1081 {
1082 	int i;
1083 
1084 	if (!RB_EMPTY_NODE(&dso->rb_node))
1085 		pr_err("DSO %s is still in rbtree when being deleted!\n",
1086 		       dso->long_name);
1087 	for (i = 0; i < MAP__NR_TYPES; ++i)
1088 		symbols__delete(&dso->symbols[i]);
1089 
1090 	if (dso->short_name_allocated) {
1091 		zfree((char **)&dso->short_name);
1092 		dso->short_name_allocated = false;
1093 	}
1094 
1095 	if (dso->long_name_allocated) {
1096 		zfree((char **)&dso->long_name);
1097 		dso->long_name_allocated = false;
1098 	}
1099 
1100 	dso__data_close(dso);
1101 	auxtrace_cache__free(dso->auxtrace_cache);
1102 	dso_cache__free(dso);
1103 	dso__free_a2l(dso);
1104 	zfree(&dso->symsrc_filename);
1105 	pthread_mutex_destroy(&dso->lock);
1106 	free(dso);
1107 }
1108 
1109 struct dso *dso__get(struct dso *dso)
1110 {
1111 	if (dso)
1112 		atomic_inc(&dso->refcnt);
1113 	return dso;
1114 }
1115 
1116 void dso__put(struct dso *dso)
1117 {
1118 	if (dso && atomic_dec_and_test(&dso->refcnt))
1119 		dso__delete(dso);
1120 }
1121 
1122 void dso__set_build_id(struct dso *dso, void *build_id)
1123 {
1124 	memcpy(dso->build_id, build_id, sizeof(dso->build_id));
1125 	dso->has_build_id = 1;
1126 }
1127 
1128 bool dso__build_id_equal(const struct dso *dso, u8 *build_id)
1129 {
1130 	return memcmp(dso->build_id, build_id, sizeof(dso->build_id)) == 0;
1131 }
1132 
1133 void dso__read_running_kernel_build_id(struct dso *dso, struct machine *machine)
1134 {
1135 	char path[PATH_MAX];
1136 
1137 	if (machine__is_default_guest(machine))
1138 		return;
1139 	sprintf(path, "%s/sys/kernel/notes", machine->root_dir);
1140 	if (sysfs__read_build_id(path, dso->build_id,
1141 				 sizeof(dso->build_id)) == 0)
1142 		dso->has_build_id = true;
1143 }
1144 
1145 int dso__kernel_module_get_build_id(struct dso *dso,
1146 				    const char *root_dir)
1147 {
1148 	char filename[PATH_MAX];
1149 	/*
1150 	 * kernel module short names are of the form "[module]" and
1151 	 * we need just "module" here.
1152 	 */
1153 	const char *name = dso->short_name + 1;
1154 
1155 	snprintf(filename, sizeof(filename),
1156 		 "%s/sys/module/%.*s/notes/.note.gnu.build-id",
1157 		 root_dir, (int)strlen(name) - 1, name);
1158 
1159 	if (sysfs__read_build_id(filename, dso->build_id,
1160 				 sizeof(dso->build_id)) == 0)
1161 		dso->has_build_id = true;
1162 
1163 	return 0;
1164 }
1165 
1166 bool __dsos__read_build_ids(struct list_head *head, bool with_hits)
1167 {
1168 	bool have_build_id = false;
1169 	struct dso *pos;
1170 
1171 	list_for_each_entry(pos, head, node) {
1172 		if (with_hits && !pos->hit)
1173 			continue;
1174 		if (pos->has_build_id) {
1175 			have_build_id = true;
1176 			continue;
1177 		}
1178 		if (filename__read_build_id(pos->long_name, pos->build_id,
1179 					    sizeof(pos->build_id)) > 0) {
1180 			have_build_id	  = true;
1181 			pos->has_build_id = true;
1182 		}
1183 	}
1184 
1185 	return have_build_id;
1186 }
1187 
1188 void __dsos__add(struct dsos *dsos, struct dso *dso)
1189 {
1190 	list_add_tail(&dso->node, &dsos->head);
1191 	__dso__findlink_by_longname(&dsos->root, dso, NULL);
1192 	/*
1193 	 * It is now in the linked list, grab a reference, then garbage collect
1194 	 * this when needing memory, by looking at LRU dso instances in the
1195 	 * list with atomic_read(&dso->refcnt) == 1, i.e. no references
1196 	 * anywhere besides the one for the list, do, under a lock for the
1197 	 * list: remove it from the list, then a dso__put(), that probably will
1198 	 * be the last and will then call dso__delete(), end of life.
1199 	 *
1200 	 * That, or at the end of the 'struct machine' lifetime, when all
1201 	 * 'struct dso' instances will be removed from the list, in
1202 	 * dsos__exit(), if they have no other reference from some other data
1203 	 * structure.
1204 	 *
1205 	 * E.g.: after processing a 'perf.data' file and storing references
1206 	 * to objects instantiated while processing events, we will have
1207 	 * references to the 'thread', 'map', 'dso' structs all from 'struct
1208 	 * hist_entry' instances, but we may not need anything not referenced,
1209 	 * so we might as well call machines__exit()/machines__delete() and
1210 	 * garbage collect it.
1211 	 */
1212 	dso__get(dso);
1213 }
1214 
1215 void dsos__add(struct dsos *dsos, struct dso *dso)
1216 {
1217 	pthread_rwlock_wrlock(&dsos->lock);
1218 	__dsos__add(dsos, dso);
1219 	pthread_rwlock_unlock(&dsos->lock);
1220 }
1221 
1222 struct dso *__dsos__find(struct dsos *dsos, const char *name, bool cmp_short)
1223 {
1224 	struct dso *pos;
1225 
1226 	if (cmp_short) {
1227 		list_for_each_entry(pos, &dsos->head, node)
1228 			if (strcmp(pos->short_name, name) == 0)
1229 				return pos;
1230 		return NULL;
1231 	}
1232 	return __dso__find_by_longname(&dsos->root, name);
1233 }
1234 
1235 struct dso *dsos__find(struct dsos *dsos, const char *name, bool cmp_short)
1236 {
1237 	struct dso *dso;
1238 	pthread_rwlock_rdlock(&dsos->lock);
1239 	dso = __dsos__find(dsos, name, cmp_short);
1240 	pthread_rwlock_unlock(&dsos->lock);
1241 	return dso;
1242 }
1243 
1244 struct dso *__dsos__addnew(struct dsos *dsos, const char *name)
1245 {
1246 	struct dso *dso = dso__new(name);
1247 
1248 	if (dso != NULL) {
1249 		__dsos__add(dsos, dso);
1250 		dso__set_basename(dso);
1251 		/* Put dso here because __dsos_add already got it */
1252 		dso__put(dso);
1253 	}
1254 	return dso;
1255 }
1256 
1257 struct dso *__dsos__findnew(struct dsos *dsos, const char *name)
1258 {
1259 	struct dso *dso = __dsos__find(dsos, name, false);
1260 
1261 	return dso ? dso : __dsos__addnew(dsos, name);
1262 }
1263 
1264 struct dso *dsos__findnew(struct dsos *dsos, const char *name)
1265 {
1266 	struct dso *dso;
1267 	pthread_rwlock_wrlock(&dsos->lock);
1268 	dso = dso__get(__dsos__findnew(dsos, name));
1269 	pthread_rwlock_unlock(&dsos->lock);
1270 	return dso;
1271 }
1272 
1273 size_t __dsos__fprintf_buildid(struct list_head *head, FILE *fp,
1274 			       bool (skip)(struct dso *dso, int parm), int parm)
1275 {
1276 	struct dso *pos;
1277 	size_t ret = 0;
1278 
1279 	list_for_each_entry(pos, head, node) {
1280 		if (skip && skip(pos, parm))
1281 			continue;
1282 		ret += dso__fprintf_buildid(pos, fp);
1283 		ret += fprintf(fp, " %s\n", pos->long_name);
1284 	}
1285 	return ret;
1286 }
1287 
1288 size_t __dsos__fprintf(struct list_head *head, FILE *fp)
1289 {
1290 	struct dso *pos;
1291 	size_t ret = 0;
1292 
1293 	list_for_each_entry(pos, head, node) {
1294 		int i;
1295 		for (i = 0; i < MAP__NR_TYPES; ++i)
1296 			ret += dso__fprintf(pos, i, fp);
1297 	}
1298 
1299 	return ret;
1300 }
1301 
1302 size_t dso__fprintf_buildid(struct dso *dso, FILE *fp)
1303 {
1304 	char sbuild_id[BUILD_ID_SIZE * 2 + 1];
1305 
1306 	build_id__sprintf(dso->build_id, sizeof(dso->build_id), sbuild_id);
1307 	return fprintf(fp, "%s", sbuild_id);
1308 }
1309 
1310 size_t dso__fprintf(struct dso *dso, enum map_type type, FILE *fp)
1311 {
1312 	struct rb_node *nd;
1313 	size_t ret = fprintf(fp, "dso: %s (", dso->short_name);
1314 
1315 	if (dso->short_name != dso->long_name)
1316 		ret += fprintf(fp, "%s, ", dso->long_name);
1317 	ret += fprintf(fp, "%s, %sloaded, ", map_type__name[type],
1318 		       dso__loaded(dso, type) ? "" : "NOT ");
1319 	ret += dso__fprintf_buildid(dso, fp);
1320 	ret += fprintf(fp, ")\n");
1321 	for (nd = rb_first(&dso->symbols[type]); nd; nd = rb_next(nd)) {
1322 		struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
1323 		ret += symbol__fprintf(pos, fp);
1324 	}
1325 
1326 	return ret;
1327 }
1328 
1329 enum dso_type dso__type(struct dso *dso, struct machine *machine)
1330 {
1331 	int fd;
1332 	enum dso_type type = DSO__TYPE_UNKNOWN;
1333 
1334 	fd = dso__data_get_fd(dso, machine);
1335 	if (fd >= 0) {
1336 		type = dso__type_fd(fd);
1337 		dso__data_put_fd(dso);
1338 	}
1339 
1340 	return type;
1341 }
1342 
1343 int dso__strerror_load(struct dso *dso, char *buf, size_t buflen)
1344 {
1345 	int idx, errnum = dso->load_errno;
1346 	/*
1347 	 * This must have a same ordering as the enum dso_load_errno.
1348 	 */
1349 	static const char *dso_load__error_str[] = {
1350 	"Internal tools/perf/ library error",
1351 	"Invalid ELF file",
1352 	"Can not read build id",
1353 	"Mismatching build id",
1354 	"Decompression failure",
1355 	};
1356 
1357 	BUG_ON(buflen == 0);
1358 
1359 	if (errnum >= 0) {
1360 		const char *err = strerror_r(errnum, buf, buflen);
1361 
1362 		if (err != buf)
1363 			scnprintf(buf, buflen, "%s", err);
1364 
1365 		return 0;
1366 	}
1367 
1368 	if (errnum <  __DSO_LOAD_ERRNO__START || errnum >= __DSO_LOAD_ERRNO__END)
1369 		return -1;
1370 
1371 	idx = errnum - __DSO_LOAD_ERRNO__START;
1372 	scnprintf(buf, buflen, "%s", dso_load__error_str[idx]);
1373 	return 0;
1374 }
1375