xref: /freebsd/lib/libpmcstat/libpmcstat_image.c (revision 4731124cace5e7a0224e29784617d2856e5c59ab)
1 /*-
2  * Copyright (c) 2003-2008 Joseph Koshy
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/types.h>
31 #include <sys/cpuset.h>
32 #include <sys/param.h>
33 #include <sys/endian.h>
34 #include <sys/pmc.h>
35 #include <sys/sysctl.h>
36 #include <sys/imgact_aout.h>
37 #include <sys/imgact_elf.h>
38 
39 #include <netinet/in.h>
40 
41 #include <assert.h>
42 #include <err.h>
43 #include <fcntl.h>
44 #include <pmc.h>
45 #include <pmclog.h>
46 #include <stdbool.h>
47 #include <stdio.h>
48 #include <stdlib.h>
49 #include <string.h>
50 #include <sysexits.h>
51 #include <unistd.h>
52 
53 #include "libpmcstat.h"
54 
55 #define	min(A,B)		((A) < (B) ? (A) : (B))
56 #define	max(A,B)		((A) > (B) ? (A) : (B))
57 
58 /*
59  * Add the list of symbols in the given section to the list associated
60  * with the object.
61  */
62 void
63 pmcstat_image_add_symbols(struct pmcstat_image *image, Elf *e,
64     Elf_Scn *scn, GElf_Shdr *sh)
65 {
66 	int firsttime;
67 	size_t n, newsyms, nshsyms, nfuncsyms;
68 	struct pmcstat_symbol *symptr;
69 	char *fnname;
70 	GElf_Sym sym;
71 	Elf_Data *data;
72 
73 	if ((data = elf_getdata(scn, NULL)) == NULL)
74 		return;
75 
76 	/*
77 	 * Determine the number of functions named in this
78 	 * section.
79 	 */
80 
81 	nshsyms = sh->sh_size / sh->sh_entsize;
82 	for (n = nfuncsyms = 0; n < nshsyms; n++) {
83 		if (gelf_getsym(data, (int) n, &sym) != &sym)
84 			return;
85 		if (GELF_ST_TYPE(sym.st_info) == STT_FUNC)
86 			nfuncsyms++;
87 	}
88 
89 	if (nfuncsyms == 0)
90 		return;
91 
92 	/*
93 	 * Allocate space for the new entries.
94 	 */
95 	firsttime = image->pi_symbols == NULL;
96 	symptr = reallocarray(image->pi_symbols,
97 	    image->pi_symcount + nfuncsyms, sizeof(*symptr));
98 	if (symptr == image->pi_symbols) /* realloc() failed. */
99 		return;
100 	image->pi_symbols = symptr;
101 
102 	/*
103 	 * Append new symbols to the end of the current table.
104 	 */
105 	symptr += image->pi_symcount;
106 
107 	for (n = newsyms = 0; n < nshsyms; n++) {
108 		if (gelf_getsym(data, (int) n, &sym) != &sym)
109 			return;
110 		if (GELF_ST_TYPE(sym.st_info) != STT_FUNC)
111 			continue;
112 
113 		if (sym.st_shndx == STN_UNDEF)
114 			continue;
115 
116 		if (!firsttime && pmcstat_symbol_search(image, sym.st_value))
117 			continue; /* We've seen this symbol already. */
118 
119 		if ((fnname = elf_strptr(e, sh->sh_link, sym.st_name))
120 		    == NULL)
121 			continue;
122 #ifdef __arm__
123 		/* Remove spurious ARM function name. */
124 		if (fnname[0] == '$' &&
125 		    (fnname[1] == 'a' || fnname[1] == 't' ||
126 		    fnname[1] == 'd') &&
127 		    fnname[2] == '\0')
128 			continue;
129 #endif
130 
131 		symptr->ps_name  = pmcstat_string_intern(fnname);
132 		symptr->ps_start = sym.st_value - image->pi_vaddr;
133 		symptr->ps_end   = symptr->ps_start + sym.st_size;
134 
135 		symptr++;
136 		newsyms++;
137 	}
138 
139 	image->pi_symcount += newsyms;
140 	if (image->pi_symcount == 0)
141 		return;
142 
143 	assert(newsyms <= nfuncsyms);
144 
145 	/*
146 	 * Return space to the system if there were duplicates.
147 	 */
148 	if (newsyms < nfuncsyms)
149 		image->pi_symbols = reallocarray(image->pi_symbols,
150 		    image->pi_symcount, sizeof(*symptr));
151 
152 	/*
153 	 * Keep the list of symbols sorted.
154 	 */
155 	qsort(image->pi_symbols, image->pi_symcount, sizeof(*symptr),
156 	    pmcstat_symbol_compare);
157 
158 	/*
159 	 * Deal with function symbols that have a size of 'zero' by
160 	 * making them extend to the next higher address.  These
161 	 * symbols are usually defined in assembly code.
162 	 */
163 	for (symptr = image->pi_symbols;
164 	     symptr < image->pi_symbols + (image->pi_symcount - 1);
165 	     symptr++)
166 		if (symptr->ps_start == symptr->ps_end)
167 			symptr->ps_end = (symptr+1)->ps_start;
168 }
169 
170 /*
171  * Record the fact that PC values from 'start' to 'end' come from
172  * image 'image'.
173  */
174 
175 void
176 pmcstat_image_link(struct pmcstat_process *pp, struct pmcstat_image *image,
177     uintfptr_t start)
178 {
179 	struct pmcstat_pcmap *pcm, *pcmnew;
180 	uintfptr_t offset;
181 #ifdef __powerpc__
182 	unsigned long kernbase;
183 	size_t kernbase_len;
184 #endif
185 
186 	assert(image->pi_type != PMCSTAT_IMAGE_UNKNOWN &&
187 	    image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE);
188 
189 	if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
190 		err(EX_OSERR, "ERROR: Cannot create a map entry");
191 
192 	/*
193 	 * PowerPC kernel is of DYN type and it has a base address
194 	 * where it is initially loaded, before being relocated.
195 	 * As the address in 'start' is where the kernel was relocated to,
196 	 * but the symbols always use the original base address, we need to
197 	 * subtract it to get the correct offset.
198 	 */
199 #ifdef __powerpc__
200 	if (pp->pp_pid == -1) {
201 		kernbase = 0;
202 		kernbase_len = sizeof(kernbase);
203 		if (sysctlbyname("kern.base_address", &kernbase, &kernbase_len,
204 		    NULL, 0) == -1)
205 			warnx(
206 			    "WARNING: Could not retrieve kernel base address");
207 		else
208 			start -= kernbase;
209 	}
210 #endif
211 
212 	/*
213 	 * Adjust the map entry to only cover the text portion
214 	 * of the object.
215 	 */
216 
217 	offset = start - image->pi_vaddr;
218 	pcmnew->ppm_lowpc  = image->pi_start + offset;
219 	pcmnew->ppm_highpc = image->pi_end + offset;
220 	pcmnew->ppm_image  = image;
221 
222 	assert(pcmnew->ppm_lowpc < pcmnew->ppm_highpc);
223 
224 	/* Overlapped mmap()'s are assumed to never occur. */
225 	TAILQ_FOREACH(pcm, &pp->pp_map, ppm_next)
226 	    if (pcm->ppm_lowpc >= pcmnew->ppm_highpc)
227 		    break;
228 
229 	if (pcm == NULL)
230 		TAILQ_INSERT_TAIL(&pp->pp_map, pcmnew, ppm_next);
231 	else
232 		TAILQ_INSERT_BEFORE(pcm, pcmnew, ppm_next);
233 }
234 
235 /*
236  * Determine whether a given executable image is an A.OUT object, and
237  * if so, fill in its parameters from the text file.
238  * Sets image->pi_type.
239  */
240 
241 void
242 pmcstat_image_get_aout_params(struct pmcstat_image *image,
243     struct pmcstat_args *args)
244 {
245 	int fd;
246 	ssize_t nbytes;
247 	struct exec ex;
248 	const char *path;
249 	char buffer[PATH_MAX];
250 
251 	path = pmcstat_string_unintern(image->pi_execpath);
252 	assert(path != NULL);
253 
254 	if (image->pi_iskernelmodule)
255 		errx(EX_SOFTWARE,
256 		    "ERROR: a.out kernel modules are unsupported \"%s\"", path);
257 
258 	(void) snprintf(buffer, sizeof(buffer), "%s%s",
259 	    args->pa_fsroot, path);
260 
261 	if ((fd = open(buffer, O_RDONLY, 0)) < 0 ||
262 	    (nbytes = read(fd, &ex, sizeof(ex))) < 0) {
263 		if (args->pa_verbosity >= 2)
264 			warn("WARNING: Cannot determine type of \"%s\"",
265 			    path);
266 		image->pi_type = PMCSTAT_IMAGE_INDETERMINABLE;
267 		if (fd != -1)
268 			(void) close(fd);
269 		return;
270 	}
271 
272 	(void) close(fd);
273 
274 	if ((unsigned) nbytes != sizeof(ex) ||
275 	    N_BADMAG(ex))
276 		return;
277 
278 	image->pi_type = PMCSTAT_IMAGE_AOUT;
279 
280 	/* TODO: the rest of a.out processing */
281 
282 	return;
283 }
284 
285 /*
286  * Examine an ELF file to determine the size of its text segment.
287  * Sets image->pi_type if anything conclusive can be determined about
288  * this image.
289  */
290 
291 void
292 pmcstat_image_get_elf_params(struct pmcstat_image *image,
293     struct pmcstat_args *args)
294 {
295 	int fd;
296 	size_t i, nph, nsh;
297 	const char *path, *elfbase;
298 	char *p, *endp;
299 	bool first_exec_segment;
300 	uintfptr_t minva, maxva;
301 	Elf *e;
302 	Elf_Scn *scn;
303 	GElf_Ehdr eh;
304 	GElf_Phdr ph;
305 	GElf_Shdr sh;
306 	enum pmcstat_image_type image_type;
307 	char buffer[PATH_MAX];
308 	char buffer_modules[PATH_MAX];
309 
310 	assert(image->pi_type == PMCSTAT_IMAGE_UNKNOWN);
311 
312 	image->pi_start = minva = ~(uintfptr_t) 0;
313 	image->pi_end = maxva = (uintfptr_t) 0;
314 	image->pi_type = image_type = PMCSTAT_IMAGE_INDETERMINABLE;
315 	image->pi_isdynamic = 0;
316 	image->pi_dynlinkerpath = NULL;
317 	image->pi_vaddr = 0;
318 
319 	path = pmcstat_string_unintern(image->pi_execpath);
320 	assert(path != NULL);
321 
322 	/*
323 	 * Look for kernel modules under FSROOT/KERNELPATH/NAME and
324 	 * FSROOT/boot/modules/NAME, and user mode executable objects
325 	 * under FSROOT/PATHNAME.
326 	 */
327 	if (image->pi_iskernelmodule) {
328 		(void) snprintf(buffer, sizeof(buffer), "%s%s/%s",
329 		    args->pa_fsroot, args->pa_kernel, path);
330 		(void) snprintf(buffer_modules, sizeof(buffer_modules),
331 		    "%s/boot/modules/%s", args->pa_fsroot, path);
332 	} else {
333 		(void) snprintf(buffer, sizeof(buffer), "%s%s",
334 		    args->pa_fsroot, path);
335 	}
336 
337 	e = NULL;
338 	fd = open(buffer, O_RDONLY, 0);
339 	if (fd < 0 && !image->pi_iskernelmodule) {
340 		warnx("WARNING: Cannot open \"%s\".",
341 		    buffer);
342 		goto done;
343 	}
344 	if (fd < 0 && (fd = open(buffer_modules, O_RDONLY, 0)) < 0) {
345 		warnx("WARNING: Cannot open \"%s\" or \"%s\".",
346 		    buffer, buffer_modules);
347 		goto done;
348 	}
349 	if (elf_version(EV_CURRENT) == EV_NONE) {
350 		warnx("WARNING: failed to init elf\n");
351 		goto done;
352 	}
353 
354 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
355 		warnx("WARNING: Cannot read \"%s\".",
356 		    buffer);
357 		goto done;
358 	}
359 
360 	if (elf_kind(e) != ELF_K_ELF) {
361 		if (args->pa_verbosity >= 2)
362 			warnx("WARNING: Cannot determine the type of \"%s\".",
363 			    buffer);
364 		goto done;
365 	}
366 
367 	if (gelf_getehdr(e, &eh) != &eh) {
368 		warnx(
369 		    "WARNING: Cannot retrieve the ELF Header for \"%s\": %s.",
370 		    buffer, elf_errmsg(-1));
371 		goto done;
372 	}
373 
374 	if (eh.e_type != ET_EXEC && eh.e_type != ET_DYN &&
375 	    !(image->pi_iskernelmodule && eh.e_type == ET_REL)) {
376 		warnx("WARNING: \"%s\" is of an unsupported ELF type.",
377 		    buffer);
378 		goto done;
379 	}
380 
381 	image_type = eh.e_ident[EI_CLASS] == ELFCLASS32 ?
382 	    PMCSTAT_IMAGE_ELF32 : PMCSTAT_IMAGE_ELF64;
383 
384 	/*
385 	 * Determine the virtual address where an executable would be
386 	 * loaded.  Additionally, for dynamically linked executables,
387 	 * save the pathname to the runtime linker.
388 	 */
389 	if (eh.e_type != ET_REL) {
390 		if (elf_getphnum(e, &nph) == 0) {
391 			warnx(
392 "WARNING: Could not determine the number of program headers in \"%s\": %s.",
393 			    buffer,
394 			    elf_errmsg(-1));
395 			goto done;
396 		}
397 		first_exec_segment = true;
398 		for (i = 0; i < eh.e_phnum; i++) {
399 			if (gelf_getphdr(e, i, &ph) != &ph) {
400 				warnx(
401 "WARNING: Retrieval of PHDR entry #%ju in \"%s\" failed: %s.",
402 				    (uintmax_t) i, buffer, elf_errmsg(-1));
403 				goto done;
404 			}
405 			switch (ph.p_type) {
406 			case PT_DYNAMIC:
407 				image->pi_isdynamic = 1;
408 				break;
409 			case PT_INTERP:
410 				if ((elfbase = elf_rawfile(e, NULL)) == NULL) {
411 					warnx(
412 "WARNING: Cannot retrieve the interpreter for \"%s\": %s.",
413 					    buffer, elf_errmsg(-1));
414 					goto done;
415 				}
416 				image->pi_dynlinkerpath =
417 				    pmcstat_string_intern(elfbase +
418 				        ph.p_offset);
419 				break;
420 			case PT_LOAD:
421 				if ((ph.p_flags & PF_X) != 0 &&
422 				    first_exec_segment) {
423 					image->pi_vaddr = ph.p_vaddr & (-ph.p_align);
424 					first_exec_segment = false;
425 				}
426 				break;
427 			}
428 		}
429 	}
430 
431 	/*
432 	 * Get the min and max VA associated with this ELF object.
433 	 */
434 	if (elf_getshnum(e, &nsh) == 0) {
435 		warnx(
436 "WARNING: Could not determine the number of sections for \"%s\": %s.",
437 		    buffer, elf_errmsg(-1));
438 		goto done;
439 	}
440 
441 	for (i = 0; i < nsh; i++) {
442 		if ((scn = elf_getscn(e, i)) == NULL ||
443 		    gelf_getshdr(scn, &sh) != &sh) {
444 			warnx(
445 "WARNING: Could not retrieve section header #%ju in \"%s\": %s.",
446 			    (uintmax_t) i, buffer, elf_errmsg(-1));
447 			goto done;
448 		}
449 		if (sh.sh_flags & SHF_EXECINSTR) {
450 			minva = min(minva, sh.sh_addr);
451 			maxva = max(maxva, sh.sh_addr + sh.sh_size);
452 		}
453 		if (sh.sh_type == SHT_SYMTAB || sh.sh_type == SHT_DYNSYM)
454 			pmcstat_image_add_symbols(image, e, scn, &sh);
455 	}
456 
457 	image->pi_start = minva;
458 	image->pi_end   = maxva;
459 	image->pi_type  = image_type;
460 	image->pi_fullpath = pmcstat_string_intern(buffer);
461 
462 	/* Build display name
463 	 */
464 	endp = buffer;
465 	for (p = buffer; *p; p++)
466 		if (*p == '/')
467 			endp = p+1;
468 	image->pi_name = pmcstat_string_intern(endp);
469 
470  done:
471 	(void) elf_end(e);
472 	if (fd >= 0)
473 		(void) close(fd);
474 	return;
475 }
476 
477 /*
478  * Given an image descriptor, determine whether it is an ELF, or AOUT.
479  * If no handler claims the image, set its type to 'INDETERMINABLE'.
480  */
481 
482 void
483 pmcstat_image_determine_type(struct pmcstat_image *image,
484     struct pmcstat_args *args)
485 {
486 	assert(image->pi_type == PMCSTAT_IMAGE_UNKNOWN);
487 
488 	/* Try each kind of handler in turn */
489 	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
490 		pmcstat_image_get_elf_params(image, args);
491 	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
492 		pmcstat_image_get_aout_params(image, args);
493 
494 	/*
495 	 * Otherwise, remember that we tried to determine
496 	 * the object's type and had failed.
497 	 */
498 	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
499 		image->pi_type = PMCSTAT_IMAGE_INDETERMINABLE;
500 }
501 
502 /*
503  * Locate an image descriptor given an interned path, adding a fresh
504  * descriptor to the cache if necessary.  This function also finds a
505  * suitable name for this image's sample file.
506  *
507  * We defer filling in the file format specific parts of the image
508  * structure till the time we actually see a sample that would fall
509  * into this image.
510  */
511 
512 struct pmcstat_image *
513 pmcstat_image_from_path(pmcstat_interned_string internedpath,
514     int iskernelmodule, struct pmcstat_args *args,
515     struct pmc_plugins *plugins)
516 {
517 	int hash;
518 	struct pmcstat_image *pi;
519 
520 	hash = pmcstat_string_lookup_hash(internedpath);
521 
522 	/* First, look for an existing entry. */
523 	LIST_FOREACH(pi, &pmcstat_image_hash[hash], pi_next)
524 	    if (pi->pi_execpath == internedpath &&
525 		  pi->pi_iskernelmodule == iskernelmodule)
526 		    return (pi);
527 
528 	/*
529 	 * Allocate a new entry and place it at the head of the hash
530 	 * and LRU lists.
531 	 */
532 	pi = malloc(sizeof(*pi));
533 	if (pi == NULL)
534 		return (NULL);
535 
536 	pi->pi_type = PMCSTAT_IMAGE_UNKNOWN;
537 	pi->pi_execpath = internedpath;
538 	pi->pi_start = ~0;
539 	pi->pi_end = 0;
540 	pi->pi_entry = 0;
541 	pi->pi_vaddr = 0;
542 	pi->pi_isdynamic = 0;
543 	pi->pi_iskernelmodule = iskernelmodule;
544 	pi->pi_dynlinkerpath = NULL;
545 	pi->pi_symbols = NULL;
546 	pi->pi_symcount = 0;
547 	pi->pi_addr2line = NULL;
548 
549 	if (plugins[args->pa_pplugin].pl_initimage != NULL)
550 		plugins[args->pa_pplugin].pl_initimage(pi);
551 	if (plugins[args->pa_plugin].pl_initimage != NULL)
552 		plugins[args->pa_plugin].pl_initimage(pi);
553 
554 	LIST_INSERT_HEAD(&pmcstat_image_hash[hash], pi, pi_next);
555 
556 	return (pi);
557 }
558