FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_ksyms.c
1 /* $NetBSD: kern_ksyms.c,v 1.107 2022/07/15 06:40:24 mrg Exp $ */
2
3 /*-
4 * Copyright (c) 2008 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software developed for The NetBSD Foundation
8 * by Andrew Doran.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 /*
33 * Copyright (c) 2001, 2003 Anders Magnusson (ragge@ludd.luth.se).
34 * All rights reserved.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 * 3. The name of the author may not be used to endorse or promote products
45 * derived from this software without specific prior written permission
46 *
47 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
48 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
49 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
50 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
51 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
52 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
53 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
54 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
55 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
56 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
57 */
58
59 /*
60 * Code to deal with in-kernel symbol table management + /dev/ksyms.
61 *
62 * For each loaded module the symbol table info is kept track of by a
63 * struct, placed in a circular list. The first entry is the kernel
64 * symbol table.
65 */
66
67 /*
68 * TODO:
69 *
70 * Add support for mmap, poll.
71 * Constify tables.
72 * Constify db_symtab and move it to .rodata.
73 */
74
75 #include <sys/cdefs.h>
76 __KERNEL_RCSID(0, "$NetBSD: kern_ksyms.c,v 1.107 2022/07/15 06:40:24 mrg Exp $");
77
78 #if defined(_KERNEL) && defined(_KERNEL_OPT)
79 #include "opt_copy_symtab.h"
80 #include "opt_ddb.h"
81 #include "opt_dtrace.h"
82 #endif
83
84 #define _KSYMS_PRIVATE
85
86 #include <sys/param.h>
87 #include <sys/queue.h>
88 #include <sys/exec.h>
89 #include <sys/file.h>
90 #include <sys/filedesc.h>
91 #include <sys/kauth.h>
92 #include <sys/systm.h>
93 #include <sys/conf.h>
94 #include <sys/kmem.h>
95 #include <sys/proc.h>
96 #include <sys/atomic.h>
97 #include <sys/ksyms.h>
98 #include <sys/kernel.h>
99 #include <sys/intr.h>
100 #include <sys/pserialize.h>
101 #include <sys/stat.h>
102
103 #include <uvm/uvm_extern.h>
104
105 #ifdef DDB
106 #include <ddb/db_output.h>
107 #endif
108
109 #include "ksyms.h"
110 #if NKSYMS > 0
111 #include "ioconf.h"
112 #endif
113
114 struct ksyms_snapshot {
115 uint64_t ks_refcnt;
116 uint64_t ks_gen;
117 struct uvm_object *ks_uobj;
118 size_t ks_size;
119 dev_t ks_dev;
120 int ks_maxlen;
121 };
122
123 #define KSYMS_MAX_ID 98304
124 #ifdef KDTRACE_HOOKS
125 static uint32_t ksyms_nmap[KSYMS_MAX_ID]; /* sorted symbol table map */
126 #else
127 static uint32_t *ksyms_nmap = NULL;
128 #endif
129
130 static int ksyms_maxlen;
131 static bool ksyms_initted;
132 static bool ksyms_loaded;
133 static kmutex_t ksyms_lock __cacheline_aligned;
134 static struct ksyms_symtab kernel_symtab;
135 static kcondvar_t ksyms_cv;
136 static struct lwp *ksyms_snapshotting;
137 static struct ksyms_snapshot *ksyms_snapshot;
138 static uint64_t ksyms_snapshot_gen;
139 static pserialize_t ksyms_psz __read_mostly;
140
141 static void ksyms_hdr_init(const void *);
142 static void ksyms_sizes_calc(void);
143 static struct ksyms_snapshot *ksyms_snapshot_alloc(int, size_t, dev_t,
144 uint64_t);
145 static void ksyms_snapshot_release(struct ksyms_snapshot *);
146
147 #ifdef KSYMS_DEBUG
148 #define FOLLOW_CALLS 1
149 #define FOLLOW_MORE_CALLS 2
150 #define FOLLOW_DEVKSYMS 4
151 static int ksyms_debug;
152 #endif
153
154 #define SYMTAB_FILLER "|This is the symbol table!"
155
156 #ifdef makeoptions_COPY_SYMTAB
157 extern char db_symtab[];
158 extern int db_symtabsize;
159 #endif
160
161 /*
162 * used by savecore(8) so non-static
163 */
164 struct ksyms_hdr ksyms_hdr;
165 int ksyms_symsz;
166 int ksyms_strsz;
167 int ksyms_ctfsz; /* this is not currently used by savecore(8) */
168 TAILQ_HEAD(ksyms_symtab_queue, ksyms_symtab) ksyms_symtabs =
169 TAILQ_HEAD_INITIALIZER(ksyms_symtabs);
170 static struct pslist_head ksyms_symtabs_psz = PSLIST_INITIALIZER;
171
172 static int
173 ksyms_verify(const void *symstart, const void *strstart)
174 {
175 #if defined(DIAGNOSTIC) || defined(DEBUG)
176 if (symstart == NULL)
177 printf("ksyms: Symbol table not found\n");
178 if (strstart == NULL)
179 printf("ksyms: String table not found\n");
180 if (symstart == NULL || strstart == NULL)
181 printf("ksyms: Perhaps the kernel is stripped?\n");
182 #endif
183 if (symstart == NULL || strstart == NULL)
184 return 0;
185 return 1;
186 }
187
188 /*
189 * Finds a certain symbol name in a certain symbol table.
190 */
191 static Elf_Sym *
192 findsym(const char *name, struct ksyms_symtab *table, int type)
193 {
194 Elf_Sym *sym, *maxsym;
195 int low, mid, high, nglob;
196 char *str, *cmp;
197
198 sym = table->sd_symstart;
199 str = table->sd_strstart - table->sd_usroffset;
200 nglob = table->sd_nglob;
201 low = 0;
202 high = nglob;
203
204 /*
205 * Start with a binary search of all global symbols in this table.
206 * Global symbols must have unique names.
207 */
208 while (low < high) {
209 mid = (low + high) >> 1;
210 cmp = sym[mid].st_name + str;
211 if (cmp[0] < name[0] || strcmp(cmp, name) < 0) {
212 low = mid + 1;
213 } else {
214 high = mid;
215 }
216 }
217 KASSERT(low == high);
218 if (__predict_true(low < nglob &&
219 strcmp(sym[low].st_name + str, name) == 0)) {
220 KASSERT(ELF_ST_BIND(sym[low].st_info) == STB_GLOBAL);
221 return &sym[low];
222 }
223
224 /*
225 * Perform a linear search of local symbols (rare). Many local
226 * symbols with the same name can exist so are not included in
227 * the binary search.
228 */
229 if (type != KSYMS_EXTERN) {
230 maxsym = sym + table->sd_symsize / sizeof(Elf_Sym);
231 for (sym += nglob; sym < maxsym; sym++) {
232 if (strcmp(name, sym->st_name + str) == 0) {
233 return sym;
234 }
235 }
236 }
237 return NULL;
238 }
239
240 /*
241 * The "attach" is in reality done in ksyms_init().
242 */
243 #if NKSYMS > 0
244 /*
245 * ksyms can be loaded even if the kernel has a missing "pseudo-device ksyms"
246 * statement because ddb and modules require it. Fixing it properly requires
247 * fixing config to warn about required, but missing preudo-devices. For now,
248 * if we don't have the pseudo-device we don't need the attach function; this
249 * is fine, as it does nothing.
250 */
251 void
252 ksymsattach(int arg)
253 {
254 }
255 #endif
256
257 void
258 ksyms_init(void)
259 {
260
261 #ifdef makeoptions_COPY_SYMTAB
262 if (!ksyms_loaded &&
263 strncmp(db_symtab, SYMTAB_FILLER, sizeof(SYMTAB_FILLER))) {
264 ksyms_addsyms_elf(db_symtabsize, db_symtab,
265 db_symtab + db_symtabsize);
266 }
267 #endif
268
269 if (!ksyms_initted) {
270 mutex_init(&ksyms_lock, MUTEX_DEFAULT, IPL_NONE);
271 cv_init(&ksyms_cv, "ksyms");
272 ksyms_psz = pserialize_create();
273 ksyms_initted = true;
274 }
275 }
276
277 /*
278 * Are any symbols available?
279 */
280 bool
281 ksyms_available(void)
282 {
283
284 return ksyms_loaded;
285 }
286
287 /*
288 * Add a symbol table.
289 * This is intended for use when the symbol table and its corresponding
290 * string table are easily available. If they are embedded in an ELF
291 * image, use addsymtab_elf() instead.
292 *
293 * name - Symbol's table name.
294 * symstart, symsize - Address and size of the symbol table.
295 * strstart, strsize - Address and size of the string table.
296 * tab - Symbol table to be updated with this information.
297 * newstart - Address to which the symbol table has to be copied during
298 * shrinking. If NULL, it is not moved.
299 */
300 static const char *addsymtab_strstart;
301
302 static int
303 addsymtab_compar(const void *a, const void *b)
304 {
305 const Elf_Sym *sa, *sb;
306
307 sa = a;
308 sb = b;
309
310 /*
311 * Split the symbol table into two, with globals at the start
312 * and locals at the end.
313 */
314 if (ELF_ST_BIND(sa->st_info) != ELF_ST_BIND(sb->st_info)) {
315 if (ELF_ST_BIND(sa->st_info) == STB_GLOBAL) {
316 return -1;
317 }
318 if (ELF_ST_BIND(sb->st_info) == STB_GLOBAL) {
319 return 1;
320 }
321 }
322
323 /* Within each band, sort by name. */
324 return strcmp(sa->st_name + addsymtab_strstart,
325 sb->st_name + addsymtab_strstart);
326 }
327
328 static void
329 addsymtab(const char *name, void *symstart, size_t symsize,
330 void *strstart, size_t strsize, struct ksyms_symtab *tab,
331 void *newstart, void *ctfstart, size_t ctfsize, uint32_t *nmap)
332 {
333 Elf_Sym *sym, *nsym, ts;
334 int i, j, n, nglob;
335 char *str;
336 int nsyms = symsize / sizeof(Elf_Sym);
337 int s;
338
339 /* Sanity check for pre-allocated map table used during startup. */
340 if ((nmap == ksyms_nmap) && (nsyms >= KSYMS_MAX_ID)) {
341 printf("kern_ksyms: ERROR %d > %d, increase KSYMS_MAX_ID\n",
342 nsyms, KSYMS_MAX_ID);
343
344 /* truncate for now */
345 nsyms = KSYMS_MAX_ID - 1;
346 }
347
348 tab->sd_symstart = symstart;
349 tab->sd_symsize = symsize;
350 tab->sd_strstart = strstart;
351 tab->sd_strsize = strsize;
352 tab->sd_name = name;
353 tab->sd_minsym = UINTPTR_MAX;
354 tab->sd_maxsym = 0;
355 tab->sd_usroffset = 0;
356 tab->sd_ctfstart = ctfstart;
357 tab->sd_ctfsize = ctfsize;
358 tab->sd_nmap = nmap;
359 tab->sd_nmapsize = nsyms;
360 #ifdef KSYMS_DEBUG
361 printf("newstart %p sym %p ksyms_symsz %zu str %p strsz %zu send %p\n",
362 newstart, symstart, symsize, strstart, strsize,
363 tab->sd_strstart + tab->sd_strsize);
364 #endif
365
366 if (nmap) {
367 memset(nmap, 0, nsyms * sizeof(uint32_t));
368 }
369
370 /* Pack symbol table by removing all file name references. */
371 sym = tab->sd_symstart;
372 nsym = (Elf_Sym *)newstart;
373 str = tab->sd_strstart;
374 nglob = 0;
375 for (i = n = 0; i < nsyms; i++) {
376
377 /*
378 * This breaks CTF mapping, so don't do it when
379 * DTrace is enabled.
380 */
381 #ifndef KDTRACE_HOOKS
382 /*
383 * Remove useless symbols.
384 * Should actually remove all typeless symbols.
385 */
386 if (sym[i].st_name == 0)
387 continue; /* Skip nameless entries */
388 if (sym[i].st_shndx == SHN_UNDEF)
389 continue; /* Skip external references */
390 if (ELF_ST_TYPE(sym[i].st_info) == STT_FILE)
391 continue; /* Skip filenames */
392 if (ELF_ST_TYPE(sym[i].st_info) == STT_NOTYPE &&
393 sym[i].st_value == 0 &&
394 strcmp(str + sym[i].st_name, "*ABS*") == 0)
395 continue; /* XXX */
396 if (ELF_ST_TYPE(sym[i].st_info) == STT_NOTYPE &&
397 strcmp(str + sym[i].st_name, "gcc2_compiled.") == 0)
398 continue; /* XXX */
399 #endif
400
401 /* Save symbol. Set it as an absolute offset */
402 nsym[n] = sym[i];
403
404 #ifdef KDTRACE_HOOKS
405 if (nmap != NULL) {
406 /*
407 * Save the size, replace it with the symbol id so
408 * the mapping can be done after the cleanup and sort.
409 */
410 nmap[i] = nsym[n].st_size;
411 nsym[n].st_size = i + 1; /* zero is reserved */
412 }
413 #endif
414
415 if (sym[i].st_shndx != SHN_ABS) {
416 nsym[n].st_shndx = SHBSS;
417 } else {
418 /* SHN_ABS is a magic value, don't overwrite it */
419 }
420
421 j = strlen(nsym[n].st_name + str) + 1;
422 if (j > ksyms_maxlen)
423 ksyms_maxlen = j;
424 nglob += (ELF_ST_BIND(nsym[n].st_info) == STB_GLOBAL);
425
426 /* Compute min and max symbols. */
427 if (strcmp(str + sym[i].st_name, "*ABS*") != 0
428 && ELF_ST_TYPE(nsym[n].st_info) != STT_NOTYPE) {
429 if (nsym[n].st_value < tab->sd_minsym) {
430 tab->sd_minsym = nsym[n].st_value;
431 }
432 if (nsym[n].st_value > tab->sd_maxsym) {
433 tab->sd_maxsym = nsym[n].st_value;
434 }
435 }
436 n++;
437 }
438
439 /* Fill the rest of the record, and sort the symbols. */
440 tab->sd_symstart = nsym;
441 tab->sd_symsize = n * sizeof(Elf_Sym);
442 tab->sd_nglob = nglob;
443
444 addsymtab_strstart = str;
445 if (kheapsort(nsym, n, sizeof(Elf_Sym), addsymtab_compar, &ts) != 0)
446 panic("addsymtab");
447
448 #ifdef KDTRACE_HOOKS
449 /*
450 * Build the mapping from original symbol id to new symbol table.
451 * Deleted symbols will have a zero map, indices will be one based
452 * instead of zero based.
453 * Resulting map is sd_nmap[original_index] = new_index + 1
454 */
455 if (nmap != NULL) {
456 int new;
457 for (new = 0; new < n; new++) {
458 uint32_t orig = nsym[new].st_size - 1;
459 uint32_t size = nmap[orig];
460
461 nmap[orig] = new + 1;
462
463 /* restore the size */
464 nsym[new].st_size = size;
465 }
466 }
467 #endif
468
469 KASSERT(strcmp(name, "netbsd") == 0 || mutex_owned(&ksyms_lock));
470 KASSERT(cold || mutex_owned(&ksyms_lock));
471
472 /*
473 * Publish the symtab. Do this at splhigh to ensure ddb never
474 * witnesses an inconsistent state of the queue, unless memory
475 * is so corrupt that we crash in PSLIST_WRITER_INSERT_AFTER or
476 * TAILQ_INSERT_TAIL.
477 */
478 PSLIST_ENTRY_INIT(tab, sd_pslist);
479 s = splhigh();
480 if (TAILQ_EMPTY(&ksyms_symtabs)) {
481 PSLIST_WRITER_INSERT_HEAD(&ksyms_symtabs_psz, tab, sd_pslist);
482 } else {
483 struct ksyms_symtab *last;
484
485 last = TAILQ_LAST(&ksyms_symtabs, ksyms_symtab_queue);
486 PSLIST_WRITER_INSERT_AFTER(last, tab, sd_pslist);
487 }
488 TAILQ_INSERT_TAIL(&ksyms_symtabs, tab, sd_queue);
489 splx(s);
490
491 ksyms_sizes_calc();
492 ksyms_loaded = true;
493 }
494
495 /*
496 * Setup the kernel symbol table stuff.
497 */
498 void
499 ksyms_addsyms_elf(int symsize, void *start, void *end)
500 {
501 int i, j;
502 Elf_Shdr *shdr;
503 char *symstart = NULL, *strstart = NULL;
504 size_t strsize = 0;
505 Elf_Ehdr *ehdr;
506 char *ctfstart = NULL;
507 size_t ctfsize = 0;
508
509 if (symsize <= 0) {
510 printf("[ Kernel symbol table missing! ]\n");
511 return;
512 }
513
514 /* Sanity check */
515 if (ALIGNED_POINTER(start, long) == 0) {
516 printf("[ Kernel symbol table has bad start address %p ]\n",
517 start);
518 return;
519 }
520
521 ehdr = (Elf_Ehdr *)start;
522
523 /* check if this is a valid ELF header */
524 /* No reason to verify arch type, the kernel is actually running! */
525 if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) ||
526 ehdr->e_ident[EI_CLASS] != ELFCLASS ||
527 ehdr->e_version > 1) {
528 printf("[ Kernel symbol table invalid! ]\n");
529 return; /* nothing to do */
530 }
531
532 /* Loaded header will be scratched in addsymtab */
533 ksyms_hdr_init(start);
534
535 /* Find the symbol table and the corresponding string table. */
536 shdr = (Elf_Shdr *)((uint8_t *)start + ehdr->e_shoff);
537 for (i = 1; i < ehdr->e_shnum; i++) {
538 if (shdr[i].sh_type != SHT_SYMTAB)
539 continue;
540 if (shdr[i].sh_offset == 0)
541 continue;
542 symstart = (uint8_t *)start + shdr[i].sh_offset;
543 symsize = shdr[i].sh_size;
544 j = shdr[i].sh_link;
545 if (shdr[j].sh_offset == 0)
546 continue; /* Can this happen? */
547 strstart = (uint8_t *)start + shdr[j].sh_offset;
548 strsize = shdr[j].sh_size;
549 break;
550 }
551
552 #ifdef KDTRACE_HOOKS
553 /* Find the CTF section */
554 shdr = (Elf_Shdr *)((uint8_t *)start + ehdr->e_shoff);
555 if (ehdr->e_shstrndx != 0) {
556 char *shstr = (uint8_t *)start +
557 shdr[ehdr->e_shstrndx].sh_offset;
558 for (i = 1; i < ehdr->e_shnum; i++) {
559 #ifdef KSYMS_DEBUG
560 printf("ksyms: checking %s\n", &shstr[shdr[i].sh_name]);
561 #endif
562 if (shdr[i].sh_type != SHT_PROGBITS)
563 continue;
564 if (strncmp(".SUNW_ctf", &shstr[shdr[i].sh_name], 10)
565 != 0)
566 continue;
567 ctfstart = (uint8_t *)start + shdr[i].sh_offset;
568 ctfsize = shdr[i].sh_size;
569 ksyms_ctfsz = ctfsize;
570 #ifdef DEBUG
571 aprint_normal("Found CTF at %p, size 0x%zx\n",
572 ctfstart, ctfsize);
573 #endif
574 break;
575 }
576 #ifdef DEBUG
577 } else {
578 printf("ksyms: e_shstrndx == 0\n");
579 #endif
580 }
581 #endif
582
583 if (!ksyms_verify(symstart, strstart))
584 return;
585
586 addsymtab("netbsd", symstart, symsize, strstart, strsize,
587 &kernel_symtab, symstart, ctfstart, ctfsize, ksyms_nmap);
588
589 #ifdef DEBUG
590 aprint_normal("Loaded initial symtab at %p, strtab at %p, # entries %ld\n",
591 kernel_symtab.sd_symstart, kernel_symtab.sd_strstart,
592 (long)kernel_symtab.sd_symsize/sizeof(Elf_Sym));
593 #endif
594
595 /* Should be no snapshot to invalidate yet. */
596 KASSERT(ksyms_snapshot == NULL);
597 }
598
599 /*
600 * Setup the kernel symbol table stuff.
601 * Use this when the address of the symbol and string tables are known;
602 * otherwise use ksyms_init with an ELF image.
603 * We need to pass a minimal ELF header which will later be completed by
604 * ksyms_hdr_init and handed off to userland through /dev/ksyms. We use
605 * a void *rather than a pointer to avoid exposing the Elf_Ehdr type.
606 */
607 void
608 ksyms_addsyms_explicit(void *ehdr, void *symstart, size_t symsize,
609 void *strstart, size_t strsize)
610 {
611 if (!ksyms_verify(symstart, strstart))
612 return;
613
614 ksyms_hdr_init(ehdr);
615 addsymtab("netbsd", symstart, symsize, strstart, strsize,
616 &kernel_symtab, symstart, NULL, 0, ksyms_nmap);
617
618 /* Should be no snapshot to invalidate yet. */
619 KASSERT(ksyms_snapshot == NULL);
620 }
621
622 /*
623 * Get the value associated with a symbol.
624 * "mod" is the module name, or null if any module.
625 * "sym" is the symbol name.
626 * "val" is a pointer to the corresponding value, if call succeeded.
627 * Returns 0 if success or ENOENT if no such entry.
628 *
629 * If symp is nonnull, caller must hold ksyms_lock or module_lock, have
630 * ksyms_opencnt nonzero, be in a pserialize read section, be in ddb
631 * with all other CPUs quiescent.
632 */
633 int
634 ksyms_getval_unlocked(const char *mod, const char *sym, Elf_Sym **symp,
635 unsigned long *val, int type)
636 {
637 struct ksyms_symtab *st;
638 Elf_Sym *es;
639 int s, error = ENOENT;
640
641 #ifdef KSYMS_DEBUG
642 if (ksyms_debug & FOLLOW_CALLS)
643 printf("%s: mod %s sym %s valp %p\n", __func__, mod, sym, val);
644 #endif
645
646 s = pserialize_read_enter();
647 PSLIST_READER_FOREACH(st, &ksyms_symtabs_psz, struct ksyms_symtab,
648 sd_pslist) {
649 if (mod != NULL && strcmp(st->sd_name, mod))
650 continue;
651 if ((es = findsym(sym, st, type)) != NULL) {
652 *val = es->st_value;
653 if (symp)
654 *symp = es;
655 error = 0;
656 break;
657 }
658 }
659 pserialize_read_exit(s);
660 return error;
661 }
662
663 int
664 ksyms_getval(const char *mod, const char *sym, unsigned long *val, int type)
665 {
666
667 if (!ksyms_loaded)
668 return ENOENT;
669
670 /* No locking needed -- we read the table pserialized. */
671 return ksyms_getval_unlocked(mod, sym, NULL, val, type);
672 }
673
674 /*
675 * ksyms_get_mod(mod)
676 *
677 * Return the symtab for the given module name. Caller must ensure
678 * that the module cannot be unloaded until after this returns.
679 */
680 struct ksyms_symtab *
681 ksyms_get_mod(const char *mod)
682 {
683 struct ksyms_symtab *st;
684 int s;
685
686 s = pserialize_read_enter();
687 PSLIST_READER_FOREACH(st, &ksyms_symtabs_psz, struct ksyms_symtab,
688 sd_pslist) {
689 if (mod != NULL && strcmp(st->sd_name, mod))
690 continue;
691 break;
692 }
693 pserialize_read_exit(s);
694
695 return st;
696 }
697
698
699 /*
700 * ksyms_mod_foreach()
701 *
702 * Iterate over the symbol table of the specified module, calling the callback
703 * handler for each symbol. Stop iterating if the handler return is non-zero.
704 *
705 */
706
707 int
708 ksyms_mod_foreach(const char *mod, ksyms_callback_t callback, void *opaque)
709 {
710 struct ksyms_symtab *st;
711 Elf_Sym *sym, *maxsym;
712 char *str;
713 int symindx;
714
715 if (!ksyms_loaded)
716 return ENOENT;
717
718 mutex_enter(&ksyms_lock);
719
720 /* find the module */
721 TAILQ_FOREACH(st, &ksyms_symtabs, sd_queue) {
722 if (mod != NULL && strcmp(st->sd_name, mod))
723 continue;
724
725 sym = st->sd_symstart;
726 str = st->sd_strstart - st->sd_usroffset;
727
728 /* now iterate through the symbols */
729 maxsym = sym + st->sd_symsize / sizeof(Elf_Sym);
730 for (symindx = 0; sym < maxsym; sym++, symindx++) {
731 if (callback(str + sym->st_name, symindx,
732 (void *)sym->st_value,
733 sym->st_size,
734 sym->st_info,
735 opaque) != 0) {
736 break;
737 }
738 }
739 }
740 mutex_exit(&ksyms_lock);
741
742 return 0;
743 }
744
745 /*
746 * Get "mod" and "symbol" associated with an address.
747 * Returns 0 if success or ENOENT if no such entry.
748 *
749 * Caller must hold ksyms_lock or module_lock, have ksyms_opencnt
750 * nonzero, be in a pserialize read section, or be in ddb with all
751 * other CPUs quiescent.
752 */
753 int
754 ksyms_getname(const char **mod, const char **sym, vaddr_t v, int f)
755 {
756 struct ksyms_symtab *st;
757 Elf_Sym *les, *es = NULL;
758 vaddr_t laddr = 0;
759 const char *lmod = NULL;
760 char *stable = NULL;
761 int type, i, sz;
762
763 if (!ksyms_loaded)
764 return ENOENT;
765
766 PSLIST_READER_FOREACH(st, &ksyms_symtabs_psz, struct ksyms_symtab,
767 sd_pslist) {
768 if (v < st->sd_minsym || v > st->sd_maxsym)
769 continue;
770 sz = st->sd_symsize/sizeof(Elf_Sym);
771 for (i = 0; i < sz; i++) {
772 les = st->sd_symstart + i;
773 type = ELF_ST_TYPE(les->st_info);
774
775 if ((f & KSYMS_PROC) && (type != STT_FUNC))
776 continue;
777
778 if (type == STT_NOTYPE)
779 continue;
780
781 if (((f & KSYMS_ANY) == 0) &&
782 (type != STT_FUNC) && (type != STT_OBJECT))
783 continue;
784
785 if ((les->st_value <= v) && (les->st_value > laddr)) {
786 laddr = les->st_value;
787 es = les;
788 lmod = st->sd_name;
789 stable = st->sd_strstart - st->sd_usroffset;
790 }
791 }
792 }
793 if (es == NULL)
794 return ENOENT;
795 if ((f & KSYMS_EXACT) && (v != es->st_value))
796 return ENOENT;
797 if (mod)
798 *mod = lmod;
799 if (sym)
800 *sym = stable + es->st_name;
801 return 0;
802 }
803
804 /*
805 * Add a symbol table from a loadable module.
806 */
807 void
808 ksyms_modload(const char *name, void *symstart, vsize_t symsize,
809 char *strstart, vsize_t strsize)
810 {
811 struct ksyms_symtab *st;
812 struct ksyms_snapshot *ks;
813 void *nmap;
814
815 st = kmem_zalloc(sizeof(*st), KM_SLEEP);
816 nmap = kmem_zalloc(symsize / sizeof(Elf_Sym) * sizeof (uint32_t),
817 KM_SLEEP);
818 mutex_enter(&ksyms_lock);
819 addsymtab(name, symstart, symsize, strstart, strsize, st, symstart,
820 NULL, 0, nmap);
821 ks = ksyms_snapshot;
822 ksyms_snapshot = NULL;
823 mutex_exit(&ksyms_lock);
824
825 if (ks)
826 ksyms_snapshot_release(ks);
827 }
828
829 /*
830 * Remove a symbol table from a loadable module.
831 */
832 void
833 ksyms_modunload(const char *name)
834 {
835 struct ksyms_symtab *st;
836 struct ksyms_snapshot *ks;
837 int s;
838
839 mutex_enter(&ksyms_lock);
840 TAILQ_FOREACH(st, &ksyms_symtabs, sd_queue) {
841 if (strcmp(name, st->sd_name) != 0)
842 continue;
843 break;
844 }
845 KASSERT(st != NULL);
846
847 /* Wait for any snapshot in progress to complete. */
848 while (ksyms_snapshotting)
849 cv_wait(&ksyms_cv, &ksyms_lock);
850
851 /*
852 * Remove the symtab. Do this at splhigh to ensure ddb never
853 * witnesses an inconsistent state of the queue, unless memory
854 * is so corrupt that we crash in TAILQ_REMOVE or
855 * PSLIST_WRITER_REMOVE.
856 */
857 s = splhigh();
858 TAILQ_REMOVE(&ksyms_symtabs, st, sd_queue);
859 PSLIST_WRITER_REMOVE(st, sd_pslist);
860 splx(s);
861
862 /*
863 * And wait a grace period, in case there are any pserialized
864 * readers in flight.
865 */
866 pserialize_perform(ksyms_psz);
867 PSLIST_ENTRY_DESTROY(st, sd_pslist);
868
869 /* Recompute the ksyms sizes now that we've removed st. */
870 ksyms_sizes_calc();
871
872 /* Invalidate the global ksyms snapshot. */
873 ks = ksyms_snapshot;
874 ksyms_snapshot = NULL;
875 mutex_exit(&ksyms_lock);
876
877 /*
878 * No more references are possible. Free the name map and the
879 * symtab itself, which we had allocated in ksyms_modload.
880 */
881 kmem_free(st->sd_nmap, st->sd_nmapsize * sizeof(uint32_t));
882 kmem_free(st, sizeof(*st));
883
884 /* Release the formerly global ksyms snapshot, if any. */
885 if (ks)
886 ksyms_snapshot_release(ks);
887 }
888
889 #ifdef DDB
890 /*
891 * Keep sifting stuff here, to avoid export of ksyms internals.
892 *
893 * Systems is expected to be quiescent, so no locking done.
894 */
895 int
896 ksyms_sift(char *mod, char *sym, int mode)
897 {
898 struct ksyms_symtab *st;
899 char *sb;
900 int i, sz;
901
902 if (!ksyms_loaded)
903 return ENOENT;
904
905 TAILQ_FOREACH(st, &ksyms_symtabs, sd_queue) {
906 if (mod && strcmp(mod, st->sd_name))
907 continue;
908 sb = st->sd_strstart - st->sd_usroffset;
909
910 sz = st->sd_symsize/sizeof(Elf_Sym);
911 for (i = 0; i < sz; i++) {
912 Elf_Sym *les = st->sd_symstart + i;
913 char c;
914
915 if (strstr(sb + les->st_name, sym) == NULL)
916 continue;
917
918 if (mode == 'F') {
919 switch (ELF_ST_TYPE(les->st_info)) {
920 case STT_OBJECT:
921 c = '+';
922 break;
923 case STT_FUNC:
924 c = '*';
925 break;
926 case STT_SECTION:
927 c = '&';
928 break;
929 case STT_FILE:
930 c = '/';
931 break;
932 default:
933 c = ' ';
934 break;
935 }
936 db_printf("%s%c ", sb + les->st_name, c);
937 } else
938 db_printf("%s ", sb + les->st_name);
939 }
940 }
941 return ENOENT;
942 }
943 #endif /* DDB */
944
945 /*
946 * In case we exposing the symbol table to the userland using the pseudo-
947 * device /dev/ksyms, it is easier to provide all the tables as one.
948 * However, it means we have to change all the st_name fields for the
949 * symbols so they match the ELF image that the userland will read
950 * through the device.
951 *
952 * The actual (correct) value of st_name is preserved through a global
953 * offset stored in the symbol table structure.
954 *
955 * Call with ksyms_lock held.
956 */
957 static void
958 ksyms_sizes_calc(void)
959 {
960 struct ksyms_symtab *st;
961 int i, delta;
962
963 KASSERT(cold || mutex_owned(&ksyms_lock));
964
965 ksyms_symsz = ksyms_strsz = 0;
966 TAILQ_FOREACH(st, &ksyms_symtabs, sd_queue) {
967 delta = ksyms_strsz - st->sd_usroffset;
968 if (delta != 0) {
969 for (i = 0; i < st->sd_symsize/sizeof(Elf_Sym); i++)
970 st->sd_symstart[i].st_name += delta;
971 st->sd_usroffset = ksyms_strsz;
972 }
973 ksyms_symsz += st->sd_symsize;
974 ksyms_strsz += st->sd_strsize;
975 }
976 }
977
978 static void
979 ksyms_fill_note(void)
980 {
981 int32_t *note = ksyms_hdr.kh_note;
982 note[0] = ELF_NOTE_NETBSD_NAMESZ;
983 note[1] = ELF_NOTE_NETBSD_DESCSZ;
984 note[2] = ELF_NOTE_TYPE_NETBSD_TAG;
985 memcpy(¬e[3], "NetBSD\0", 8);
986 note[5] = __NetBSD_Version__;
987 }
988
989 static void
990 ksyms_hdr_init(const void *hdraddr)
991 {
992 /* Copy the loaded elf exec header */
993 memcpy(&ksyms_hdr.kh_ehdr, hdraddr, sizeof(Elf_Ehdr));
994
995 /* Set correct program/section header sizes, offsets and numbers */
996 ksyms_hdr.kh_ehdr.e_phoff = offsetof(struct ksyms_hdr, kh_phdr[0]);
997 ksyms_hdr.kh_ehdr.e_phentsize = sizeof(Elf_Phdr);
998 ksyms_hdr.kh_ehdr.e_phnum = NPRGHDR;
999 ksyms_hdr.kh_ehdr.e_shoff = offsetof(struct ksyms_hdr, kh_shdr[0]);
1000 ksyms_hdr.kh_ehdr.e_shentsize = sizeof(Elf_Shdr);
1001 ksyms_hdr.kh_ehdr.e_shnum = NSECHDR;
1002 ksyms_hdr.kh_ehdr.e_shstrndx = SHSTRTAB;
1003
1004 /* Text/data - fake */
1005 ksyms_hdr.kh_phdr[0].p_type = PT_LOAD;
1006 ksyms_hdr.kh_phdr[0].p_memsz = (unsigned long)-1L;
1007 ksyms_hdr.kh_phdr[0].p_flags = PF_R | PF_X | PF_W;
1008
1009 #define SHTCOPY(name) strlcpy(&ksyms_hdr.kh_strtab[offs], (name), \
1010 sizeof(ksyms_hdr.kh_strtab) - offs), offs += sizeof(name)
1011
1012 uint32_t offs = 1;
1013 /* First section header ".note.netbsd.ident" */
1014 ksyms_hdr.kh_shdr[SHNOTE].sh_name = offs;
1015 ksyms_hdr.kh_shdr[SHNOTE].sh_type = SHT_NOTE;
1016 ksyms_hdr.kh_shdr[SHNOTE].sh_offset =
1017 offsetof(struct ksyms_hdr, kh_note[0]);
1018 ksyms_hdr.kh_shdr[SHNOTE].sh_size = sizeof(ksyms_hdr.kh_note);
1019 ksyms_hdr.kh_shdr[SHNOTE].sh_addralign = sizeof(int);
1020 SHTCOPY(".note.netbsd.ident");
1021 ksyms_fill_note();
1022
1023 /* Second section header; ".symtab" */
1024 ksyms_hdr.kh_shdr[SYMTAB].sh_name = offs;
1025 ksyms_hdr.kh_shdr[SYMTAB].sh_type = SHT_SYMTAB;
1026 ksyms_hdr.kh_shdr[SYMTAB].sh_offset = sizeof(struct ksyms_hdr);
1027 /* ksyms_hdr.kh_shdr[SYMTAB].sh_size = filled in at open */
1028 ksyms_hdr.kh_shdr[SYMTAB].sh_link = STRTAB; /* Corresponding strtab */
1029 ksyms_hdr.kh_shdr[SYMTAB].sh_addralign = sizeof(long);
1030 ksyms_hdr.kh_shdr[SYMTAB].sh_entsize = sizeof(Elf_Sym);
1031 SHTCOPY(".symtab");
1032
1033 /* Third section header; ".strtab" */
1034 ksyms_hdr.kh_shdr[STRTAB].sh_name = offs;
1035 ksyms_hdr.kh_shdr[STRTAB].sh_type = SHT_STRTAB;
1036 /* ksyms_hdr.kh_shdr[STRTAB].sh_offset = filled in at open */
1037 /* ksyms_hdr.kh_shdr[STRTAB].sh_size = filled in at open */
1038 ksyms_hdr.kh_shdr[STRTAB].sh_addralign = sizeof(char);
1039 SHTCOPY(".strtab");
1040
1041 /* Fourth section, ".shstrtab" */
1042 ksyms_hdr.kh_shdr[SHSTRTAB].sh_name = offs;
1043 ksyms_hdr.kh_shdr[SHSTRTAB].sh_type = SHT_STRTAB;
1044 ksyms_hdr.kh_shdr[SHSTRTAB].sh_offset =
1045 offsetof(struct ksyms_hdr, kh_strtab);
1046 ksyms_hdr.kh_shdr[SHSTRTAB].sh_size = SHSTRSIZ;
1047 ksyms_hdr.kh_shdr[SHSTRTAB].sh_addralign = sizeof(char);
1048 SHTCOPY(".shstrtab");
1049
1050 /* Fifth section, ".bss". All symbols reside here. */
1051 ksyms_hdr.kh_shdr[SHBSS].sh_name = offs;
1052 ksyms_hdr.kh_shdr[SHBSS].sh_type = SHT_NOBITS;
1053 ksyms_hdr.kh_shdr[SHBSS].sh_offset = 0;
1054 ksyms_hdr.kh_shdr[SHBSS].sh_size = (unsigned long)-1L;
1055 ksyms_hdr.kh_shdr[SHBSS].sh_addralign = PAGE_SIZE;
1056 ksyms_hdr.kh_shdr[SHBSS].sh_flags = SHF_ALLOC | SHF_EXECINSTR;
1057 SHTCOPY(".bss");
1058
1059 /* Sixth section header; ".SUNW_ctf" */
1060 ksyms_hdr.kh_shdr[SHCTF].sh_name = offs;
1061 ksyms_hdr.kh_shdr[SHCTF].sh_type = SHT_PROGBITS;
1062 /* ksyms_hdr.kh_shdr[SHCTF].sh_offset = filled in at open */
1063 /* ksyms_hdr.kh_shdr[SHCTF].sh_size = filled in at open */
1064 ksyms_hdr.kh_shdr[SHCTF].sh_link = SYMTAB; /* Corresponding symtab */
1065 ksyms_hdr.kh_shdr[SHCTF].sh_addralign = sizeof(char);
1066 SHTCOPY(".SUNW_ctf");
1067 }
1068
1069 static struct ksyms_snapshot *
1070 ksyms_snapshot_alloc(int maxlen, size_t size, dev_t dev, uint64_t gen)
1071 {
1072 struct ksyms_snapshot *ks;
1073
1074 ks = kmem_zalloc(sizeof(*ks), KM_SLEEP);
1075 ks->ks_refcnt = 1;
1076 ks->ks_gen = gen;
1077 ks->ks_uobj = uao_create(size, 0);
1078 ks->ks_size = size;
1079 ks->ks_dev = dev;
1080 ks->ks_maxlen = maxlen;
1081
1082 return ks;
1083 }
1084
1085 static void
1086 ksyms_snapshot_release(struct ksyms_snapshot *ks)
1087 {
1088 uint64_t refcnt;
1089
1090 mutex_enter(&ksyms_lock);
1091 refcnt = --ks->ks_refcnt;
1092 mutex_exit(&ksyms_lock);
1093
1094 if (refcnt)
1095 return;
1096
1097 uao_detach(ks->ks_uobj);
1098 kmem_free(ks, sizeof(*ks));
1099 }
1100
1101 static int
1102 ubc_copyfrombuf(struct uvm_object *uobj, struct uio *uio, const void *buf,
1103 size_t n)
1104 {
1105 struct iovec iov = { .iov_base = __UNCONST(buf), .iov_len = n };
1106
1107 uio->uio_iov = &iov;
1108 uio->uio_iovcnt = 1;
1109 uio->uio_resid = n;
1110
1111 return ubc_uiomove(uobj, uio, n, UVM_ADV_SEQUENTIAL, UBC_WRITE);
1112 }
1113
1114 static int
1115 ksyms_take_snapshot(struct ksyms_snapshot *ks, struct ksyms_symtab *last)
1116 {
1117 struct uvm_object *uobj = ks->ks_uobj;
1118 struct uio uio;
1119 struct ksyms_symtab *st;
1120 int error;
1121
1122 /* Caller must have initiated snapshotting. */
1123 KASSERT(ksyms_snapshotting == curlwp);
1124
1125 /* Start a uio transfer to reuse incrementally. */
1126 uio.uio_offset = 0;
1127 uio.uio_rw = UIO_WRITE; /* write from buffer to uobj */
1128 UIO_SETUP_SYSSPACE(&uio);
1129
1130 /*
1131 * First: Copy out the ELF header.
1132 */
1133 error = ubc_copyfrombuf(uobj, &uio, &ksyms_hdr, sizeof(ksyms_hdr));
1134 if (error)
1135 return error;
1136
1137 /*
1138 * Copy out the symbol table. The list of symtabs is
1139 * guaranteed to be nonempty because we always have an entry
1140 * for the main kernel. We stop at last, not at the end of the
1141 * tailq or NULL, because entries beyond last are not included
1142 * in this snapshot (and may not be fully initialized memory as
1143 * we witness it).
1144 */
1145 KASSERT(uio.uio_offset == sizeof(struct ksyms_hdr));
1146 for (st = TAILQ_FIRST(&ksyms_symtabs);
1147 ;
1148 st = TAILQ_NEXT(st, sd_queue)) {
1149 error = ubc_copyfrombuf(uobj, &uio, st->sd_symstart,
1150 st->sd_symsize);
1151 if (error)
1152 return error;
1153 if (st == last)
1154 break;
1155 }
1156
1157 /*
1158 * Copy out the string table
1159 */
1160 KASSERT(uio.uio_offset == sizeof(struct ksyms_hdr) +
1161 ksyms_hdr.kh_shdr[SYMTAB].sh_size);
1162 for (st = TAILQ_FIRST(&ksyms_symtabs);
1163 ;
1164 st = TAILQ_NEXT(st, sd_queue)) {
1165 error = ubc_copyfrombuf(uobj, &uio, st->sd_strstart,
1166 st->sd_strsize);
1167 if (error)
1168 return error;
1169 if (st == last)
1170 break;
1171 }
1172
1173 /*
1174 * Copy out the CTF table.
1175 */
1176 KASSERT(uio.uio_offset == sizeof(struct ksyms_hdr) +
1177 ksyms_hdr.kh_shdr[SYMTAB].sh_size +
1178 ksyms_hdr.kh_shdr[STRTAB].sh_size);
1179 st = TAILQ_FIRST(&ksyms_symtabs);
1180 if (st->sd_ctfstart != NULL) {
1181 error = ubc_copyfrombuf(uobj, &uio, st->sd_ctfstart,
1182 st->sd_ctfsize);
1183 if (error)
1184 return error;
1185 }
1186
1187 KASSERT(uio.uio_offset == sizeof(struct ksyms_hdr) +
1188 ksyms_hdr.kh_shdr[SYMTAB].sh_size +
1189 ksyms_hdr.kh_shdr[STRTAB].sh_size +
1190 ksyms_hdr.kh_shdr[SHCTF].sh_size);
1191 KASSERT(uio.uio_offset == ks->ks_size);
1192
1193 return 0;
1194 }
1195
1196 static const struct fileops ksyms_fileops;
1197
1198 static int
1199 ksymsopen(dev_t dev, int flags, int devtype, struct lwp *l)
1200 {
1201 struct file *fp = NULL;
1202 int fd = -1;
1203 struct ksyms_snapshot *ks = NULL;
1204 size_t size;
1205 struct ksyms_symtab *last;
1206 int maxlen;
1207 uint64_t gen;
1208 int error;
1209
1210 if (minor(dev) != 0 || !ksyms_loaded)
1211 return ENXIO;
1212
1213 /* Allocate a private file. */
1214 error = fd_allocfile(&fp, &fd);
1215 if (error)
1216 return error;
1217
1218 mutex_enter(&ksyms_lock);
1219
1220 /*
1221 * Wait until we have a snapshot, or until there is no snapshot
1222 * being taken right now so we can take one.
1223 */
1224 while ((ks = ksyms_snapshot) == NULL && ksyms_snapshotting) {
1225 error = cv_wait_sig(&ksyms_cv, &ksyms_lock);
1226 if (error)
1227 goto out;
1228 }
1229
1230 /*
1231 * If there's a usable snapshot, increment its reference count
1232 * (can't overflow, 64-bit) and just reuse it.
1233 */
1234 if (ks) {
1235 ks->ks_refcnt++;
1236 goto out;
1237 }
1238
1239 /* Find the current length of the symtab object. */
1240 size = sizeof(struct ksyms_hdr);
1241 size += ksyms_strsz;
1242 size += ksyms_symsz;
1243 size += ksyms_ctfsz;
1244
1245 /* Start a new snapshot. */
1246 ksyms_hdr.kh_shdr[SYMTAB].sh_size = ksyms_symsz;
1247 ksyms_hdr.kh_shdr[SYMTAB].sh_info = ksyms_symsz / sizeof(Elf_Sym);
1248 ksyms_hdr.kh_shdr[STRTAB].sh_offset = ksyms_symsz +
1249 ksyms_hdr.kh_shdr[SYMTAB].sh_offset;
1250 ksyms_hdr.kh_shdr[STRTAB].sh_size = ksyms_strsz;
1251 ksyms_hdr.kh_shdr[SHCTF].sh_offset = ksyms_strsz +
1252 ksyms_hdr.kh_shdr[STRTAB].sh_offset;
1253 ksyms_hdr.kh_shdr[SHCTF].sh_size = ksyms_ctfsz;
1254 last = TAILQ_LAST(&ksyms_symtabs, ksyms_symtab_queue);
1255 maxlen = ksyms_maxlen;
1256 gen = ksyms_snapshot_gen++;
1257
1258 /*
1259 * Prevent ksyms entries from being removed while we take the
1260 * snapshot.
1261 */
1262 KASSERT(ksyms_snapshotting == NULL);
1263 ksyms_snapshotting = curlwp;
1264 mutex_exit(&ksyms_lock);
1265
1266 /* Create a snapshot and write the symtab to it. */
1267 ks = ksyms_snapshot_alloc(maxlen, size, dev, gen);
1268 error = ksyms_take_snapshot(ks, last);
1269
1270 /*
1271 * Snapshot creation is done. Wake up anyone waiting to remove
1272 * entries (module unload).
1273 */
1274 mutex_enter(&ksyms_lock);
1275 KASSERTMSG(ksyms_snapshotting == curlwp, "lwp %p stole snapshot",
1276 ksyms_snapshotting);
1277 ksyms_snapshotting = NULL;
1278 cv_broadcast(&ksyms_cv);
1279
1280 /* If we failed, give up. */
1281 if (error)
1282 goto out;
1283
1284 /* Cache the snapshot for the next reader. */
1285 KASSERT(ksyms_snapshot == NULL);
1286 ksyms_snapshot = ks;
1287 ks->ks_refcnt++;
1288 KASSERT(ks->ks_refcnt == 2);
1289
1290 out: mutex_exit(&ksyms_lock);
1291 if (error) {
1292 if (fp)
1293 fd_abort(curproc, fp, fd);
1294 if (ks)
1295 ksyms_snapshot_release(ks);
1296 } else {
1297 KASSERT(fp);
1298 KASSERT(ks);
1299 error = fd_clone(fp, fd, flags, &ksyms_fileops, ks);
1300 KASSERTMSG(error == EMOVEFD, "error=%d", error);
1301 }
1302 return error;
1303 }
1304
1305 static int
1306 ksymsclose(struct file *fp)
1307 {
1308 struct ksyms_snapshot *ks = fp->f_data;
1309
1310 ksyms_snapshot_release(ks);
1311
1312 return 0;
1313 }
1314
1315 static int
1316 ksymsread(struct file *fp, off_t *offp, struct uio *uio, kauth_cred_t cred,
1317 int flags)
1318 {
1319 const struct ksyms_snapshot *ks = fp->f_data;
1320 size_t count;
1321 int error;
1322
1323 /*
1324 * Since we don't have a per-object lock, we might as well use
1325 * the struct file lock to serialize access to fp->f_offset --
1326 * but if the caller isn't relying on or updating fp->f_offset,
1327 * there's no need to do even that. We could use ksyms_lock,
1328 * but why bother with a global lock if not needed? Either
1329 * way, the lock we use here must agree with what ksymsseek
1330 * takes (nothing else in ksyms uses fp->f_offset).
1331 */
1332 if (offp == &fp->f_offset)
1333 mutex_enter(&fp->f_lock);
1334
1335 /* Refuse negative offsets. */
1336 if (*offp < 0) {
1337 error = EINVAL;
1338 goto out;
1339 }
1340
1341 /* Return nothing at or past end of file. */
1342 if (*offp >= ks->ks_size) {
1343 error = 0;
1344 goto out;
1345 }
1346
1347 /*
1348 * 1. Set up the uio to transfer from offset *offp.
1349 * 2. Transfer as many bytes as we can (at most uio->uio_resid
1350 * or what's left in the ksyms).
1351 * 3. If requested, update *offp to reflect the number of bytes
1352 * transferred.
1353 */
1354 uio->uio_offset = *offp;
1355 count = uio->uio_resid;
1356 error = ubc_uiomove(ks->ks_uobj, uio, MIN(count, ks->ks_size - *offp),
1357 UVM_ADV_SEQUENTIAL, UBC_READ|UBC_PARTIALOK);
1358 if (flags & FOF_UPDATE_OFFSET)
1359 *offp += count - uio->uio_resid;
1360
1361 out: if (offp == &fp->f_offset)
1362 mutex_exit(&fp->f_lock);
1363 return error;
1364 }
1365
1366 static int
1367 ksymsstat(struct file *fp, struct stat *st)
1368 {
1369 const struct ksyms_snapshot *ks = fp->f_data;
1370
1371 memset(st, 0, sizeof(*st));
1372
1373 st->st_dev = NODEV;
1374 st->st_ino = 0;
1375 st->st_mode = S_IFCHR;
1376 st->st_nlink = 1;
1377 st->st_uid = kauth_cred_geteuid(fp->f_cred);
1378 st->st_gid = kauth_cred_getegid(fp->f_cred);
1379 st->st_rdev = ks->ks_dev;
1380 st->st_size = ks->ks_size;
1381 /* zero time */
1382 st->st_blksize = MAXPHYS; /* XXX arbitrary */
1383 st->st_blocks = 0;
1384 st->st_gen = ks->ks_gen;
1385
1386 return 0;
1387 }
1388
1389 static int
1390 ksymsmmap(struct file *fp, off_t *offp, size_t nbytes, int prot, int *flagsp,
1391 int *advicep, struct uvm_object **uobjp, int *maxprotp)
1392 {
1393 const struct ksyms_snapshot *ks = fp->f_data;
1394
1395 /* uvm_mmap guarantees page-aligned offset and size. */
1396 KASSERT(*offp == round_page(*offp));
1397 KASSERT(nbytes == round_page(nbytes));
1398 KASSERT(nbytes > 0);
1399
1400 /* Refuse negative offsets. */
1401 if (*offp < 0)
1402 return EINVAL;
1403
1404 /* Refuse mappings that pass the end of file. */
1405 if (nbytes > round_page(ks->ks_size) ||
1406 *offp > round_page(ks->ks_size) - nbytes)
1407 return EINVAL; /* XXX ??? */
1408
1409 /* Success! */
1410 uao_reference(ks->ks_uobj);
1411 *advicep = UVM_ADV_SEQUENTIAL;
1412 *uobjp = ks->ks_uobj;
1413 *maxprotp = prot & VM_PROT_READ;
1414 return 0;
1415 }
1416
1417 static int
1418 ksymsseek(struct file *fp, off_t delta, int whence, off_t *newoffp, int flags)
1419 {
1420 struct ksyms_snapshot *ks = fp->f_data;
1421 off_t base, newoff;
1422 int error;
1423
1424 mutex_enter(&fp->f_lock);
1425
1426 switch (whence) {
1427 case SEEK_CUR:
1428 base = fp->f_offset;
1429 break;
1430 case SEEK_END:
1431 base = ks->ks_size;
1432 break;
1433 case SEEK_SET:
1434 base = 0;
1435 break;
1436 default:
1437 error = EINVAL;
1438 goto out;
1439 }
1440
1441 /* Compute the new offset and validate it. */
1442 newoff = base + delta; /* XXX arithmetic overflow */
1443 if (newoff < 0) {
1444 error = EINVAL;
1445 goto out;
1446 }
1447
1448 /* Success! */
1449 if (newoffp)
1450 *newoffp = newoff;
1451 if (flags & FOF_UPDATE_OFFSET)
1452 fp->f_offset = newoff;
1453 error = 0;
1454
1455 out: mutex_exit(&fp->f_lock);
1456 return error;
1457 }
1458
1459 __CTASSERT(offsetof(struct ksyms_ogsymbol, kg_name) == offsetof(struct ksyms_gsymbol, kg_name));
1460 __CTASSERT(offsetof(struct ksyms_gvalue, kv_name) == offsetof(struct ksyms_gsymbol, kg_name));
1461
1462 static int
1463 ksymsioctl(struct file *fp, u_long cmd, void *data)
1464 {
1465 struct ksyms_snapshot *ks = fp->f_data;
1466 struct ksyms_ogsymbol *okg = (struct ksyms_ogsymbol *)data;
1467 struct ksyms_gsymbol *kg = (struct ksyms_gsymbol *)data;
1468 struct ksyms_gvalue *kv = (struct ksyms_gvalue *)data;
1469 struct ksyms_symtab *st;
1470 Elf_Sym *sym = NULL, copy;
1471 unsigned long val;
1472 int error = 0;
1473 char *str = NULL;
1474 int len, s;
1475
1476 /* Read cached ksyms_maxlen. */
1477 len = ks->ks_maxlen;
1478
1479 if (cmd == OKIOCGVALUE || cmd == OKIOCGSYMBOL ||
1480 cmd == KIOCGVALUE || cmd == KIOCGSYMBOL) {
1481 str = kmem_alloc(len, KM_SLEEP);
1482 if ((error = copyinstr(kg->kg_name, str, len, NULL)) != 0) {
1483 kmem_free(str, len);
1484 return error;
1485 }
1486 }
1487
1488 switch (cmd) {
1489 case OKIOCGVALUE:
1490 /*
1491 * Use the in-kernel symbol lookup code for fast
1492 * retreival of a value.
1493 */
1494 error = ksyms_getval(NULL, str, &val, KSYMS_EXTERN);
1495 if (error == 0)
1496 error = copyout(&val, okg->kg_value, sizeof(long));
1497 kmem_free(str, len);
1498 break;
1499
1500 case OKIOCGSYMBOL:
1501 /*
1502 * Use the in-kernel symbol lookup code for fast
1503 * retreival of a symbol.
1504 */
1505 s = pserialize_read_enter();
1506 PSLIST_READER_FOREACH(st, &ksyms_symtabs_psz,
1507 struct ksyms_symtab, sd_pslist) {
1508 if ((sym = findsym(str, st, KSYMS_ANY)) == NULL)
1509 continue;
1510 #ifdef notdef
1511 /* Skip if bad binding */
1512 if (ELF_ST_BIND(sym->st_info) != STB_GLOBAL) {
1513 sym = NULL;
1514 continue;
1515 }
1516 #endif
1517 break;
1518 }
1519 if (sym != NULL) {
1520 memcpy(©, sym, sizeof(copy));
1521 pserialize_read_exit(s);
1522 error = copyout(©, okg->kg_sym, sizeof(Elf_Sym));
1523 } else {
1524 pserialize_read_exit(s);
1525 error = ENOENT;
1526 }
1527 kmem_free(str, len);
1528 break;
1529
1530 case KIOCGVALUE:
1531 /*
1532 * Use the in-kernel symbol lookup code for fast
1533 * retreival of a value.
1534 */
1535 error = ksyms_getval(NULL, str, &val, KSYMS_EXTERN);
1536 if (error == 0)
1537 kv->kv_value = val;
1538 kmem_free(str, len);
1539 break;
1540
1541 case KIOCGSYMBOL:
1542 /*
1543 * Use the in-kernel symbol lookup code for fast
1544 * retreival of a symbol.
1545 */
1546 s = pserialize_read_enter();
1547 PSLIST_READER_FOREACH(st, &ksyms_symtabs_psz,
1548 struct ksyms_symtab, sd_pslist) {
1549 if ((sym = findsym(str, st, KSYMS_ANY)) == NULL)
1550 continue;
1551 #ifdef notdef
1552 /* Skip if bad binding */
1553 if (ELF_ST_BIND(sym->st_info) != STB_GLOBAL) {
1554 sym = NULL;
1555 continue;
1556 }
1557 #endif
1558 break;
1559 }
1560 if (sym != NULL) {
1561 kg->kg_sym = *sym;
1562 } else {
1563 error = ENOENT;
1564 }
1565 pserialize_read_exit(s);
1566 kmem_free(str, len);
1567 break;
1568
1569 case KIOCGSIZE:
1570 /*
1571 * Get total size of symbol table.
1572 */
1573 *(int *)data = ks->ks_size;
1574 break;
1575
1576 default:
1577 error = ENOTTY;
1578 break;
1579 }
1580
1581 return error;
1582 }
1583
1584 const struct cdevsw ksyms_cdevsw = {
1585 .d_open = ksymsopen,
1586 .d_close = noclose,
1587 .d_read = noread,
1588 .d_write = nowrite,
1589 .d_ioctl = noioctl,
1590 .d_stop = nostop,
1591 .d_tty = notty,
1592 .d_poll = nopoll,
1593 .d_mmap = nommap,
1594 .d_kqfilter = nokqfilter,
1595 .d_discard = nodiscard,
1596 .d_flag = D_OTHER | D_MPSAFE
1597 };
1598
1599 static const struct fileops ksyms_fileops = {
1600 .fo_name = "ksyms",
1601 .fo_read = ksymsread,
1602 .fo_write = fbadop_write,
1603 .fo_ioctl = ksymsioctl,
1604 .fo_fcntl = fnullop_fcntl,
1605 .fo_poll = fnullop_poll,
1606 .fo_stat = ksymsstat,
1607 .fo_close = ksymsclose,
1608 .fo_kqfilter = fnullop_kqfilter,
1609 .fo_restart = fnullop_restart,
1610 .fo_mmap = ksymsmmap,
1611 .fo_seek = ksymsseek,
1612 };
Cache object: 6d17662fa45baca1dabc177ce2726cd1
|