1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License, Version 1.0 only
6 * (the "License"). You may not use this file except in compliance
7 * with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or http://www.opensolaris.org/os/licensing.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22 /*
23 * Copyright 2004 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 /*
28 * Copyright 2012 DEY Storage Systems, Inc. All rights reserved.
29 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
30 */
31
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/thread.h>
35 #include <sys/sysmacros.h>
36 #include <sys/signal.h>
37 #include <sys/cred.h>
38 #include <sys/priv.h>
39 #include <sys/user.h>
40 #include <sys/file.h>
41 #include <sys/errno.h>
42 #include <sys/vnode.h>
43 #include <sys/mode.h>
44 #include <sys/vfs.h>
45 #include <sys/mman.h>
46 #include <sys/kmem.h>
47 #include <sys/proc.h>
48 #include <sys/pathname.h>
49 #include <sys/cmn_err.h>
50 #include <sys/systm.h>
51 #include <sys/elf.h>
52 #include <sys/vmsystm.h>
53 #include <sys/debug.h>
54 #include <sys/procfs.h>
55 #include <sys/regset.h>
56 #include <sys/auxv.h>
57 #include <sys/exec.h>
58 #include <sys/prsystm.h>
59 #include <sys/utsname.h>
60 #include <sys/zone.h>
61 #include <vm/as.h>
62 #include <vm/rm.h>
63 #include <sys/modctl.h>
64 #include <sys/systeminfo.h>
65 #include <sys/machelf.h>
66 #include <sys/sunddi.h>
67 #include "elf_impl.h"
68 #if defined(__i386) || defined(__i386_COMPAT)
69 #include <sys/sysi86.h>
70 #endif
71
72 void
73 setup_note_header(Phdr *v, proc_t *p)
74 {
75 int nlwp = p->p_lwpcnt;
76 int nzomb = p->p_zombcnt;
77 int nfd;
78 size_t size;
79 prcred_t *pcrp;
80 uf_info_t *fip;
81 uf_entry_t *ufp;
82 int fd;
83
84 fip = P_FINFO(p);
85 nfd = 0;
86 mutex_enter(&fip->fi_lock);
87 for (fd = 0; fd < fip->fi_nfiles; fd++) {
88 UF_ENTER(ufp, fip, fd);
89 if ((ufp->uf_file != NULL) && (ufp->uf_file->f_count > 0))
90 nfd++;
91 UF_EXIT(ufp);
92 }
93 mutex_exit(&fip->fi_lock);
94
95 v[0].p_type = PT_NOTE;
96 v[0].p_flags = PF_R;
97 v[0].p_filesz = (sizeof (Note) * (10 + 2 * nlwp + nzomb + nfd))
98 + roundup(sizeof (psinfo_t), sizeof (Word))
99 + roundup(sizeof (pstatus_t), sizeof (Word))
100 + roundup(prgetprivsize(), sizeof (Word))
101 + roundup(priv_get_implinfo_size(), sizeof (Word))
102 + roundup(strlen(platform) + 1, sizeof (Word))
103 + roundup(strlen(p->p_zone->zone_name) + 1, sizeof (Word))
104 + roundup(__KERN_NAUXV_IMPL * sizeof (aux_entry_t), sizeof (Word))
105 + roundup(sizeof (utsname), sizeof (Word))
106 + roundup(sizeof (core_content_t), sizeof (Word))
107 + roundup(sizeof (prsecflags_t), sizeof (Word))
108 + (nlwp + nzomb) * roundup(sizeof (lwpsinfo_t), sizeof (Word))
109 + nlwp * roundup(sizeof (lwpstatus_t), sizeof (Word))
110 + nfd * roundup(sizeof (prfdinfo_t), sizeof (Word));
111
112 if (curproc->p_agenttp != NULL) {
113 v[0].p_filesz += sizeof (Note) +
114 roundup(sizeof (psinfo_t), sizeof (Word));
115 }
116
117 size = sizeof (prcred_t) + sizeof (gid_t) * (ngroups_max - 1);
118 pcrp = kmem_alloc(size, KM_SLEEP);
119 prgetcred(p, pcrp);
120 if (pcrp->pr_ngroups != 0) {
121 v[0].p_filesz += sizeof (Note) + roundup(sizeof (prcred_t) +
122 sizeof (gid_t) * (pcrp->pr_ngroups - 1), sizeof (Word));
123 } else {
124 v[0].p_filesz += sizeof (Note) +
125 roundup(sizeof (prcred_t), sizeof (Word));
126 }
127 kmem_free(pcrp, size);
128
129
130 #if defined(__i386) || defined(__i386_COMPAT)
131 mutex_enter(&p->p_ldtlock);
132 size = prnldt(p) * sizeof (struct ssd);
133 mutex_exit(&p->p_ldtlock);
134 if (size != 0)
135 v[0].p_filesz += sizeof (Note) + roundup(size, sizeof (Word));
136 #endif /* __i386 || __i386_COMPAT */
137
138 if ((size = prhasx(p)? prgetprxregsize(p) : 0) != 0)
139 v[0].p_filesz += nlwp * sizeof (Note)
140 + nlwp * roundup(size, sizeof (Word));
141
142 #if defined(__sparc)
143 /*
144 * Figure out the number and sizes of register windows.
145 */
146 {
147 kthread_t *t = p->p_tlist;
148 do {
149 if ((size = prnwindows(ttolwp(t))) != 0) {
150 size = sizeof (gwindows_t) -
151 (SPARC_MAXREGWINDOW - size) *
152 sizeof (struct rwindow);
153 v[0].p_filesz += sizeof (Note) +
154 roundup(size, sizeof (Word));
155 }
156 } while ((t = t->t_forw) != p->p_tlist);
157 }
158 /*
159 * Space for the Ancillary State Registers.
160 */
161 if (p->p_model == DATAMODEL_LP64)
162 v[0].p_filesz += nlwp * sizeof (Note)
163 + nlwp * roundup(sizeof (asrset_t), sizeof (Word));
164 #endif /* __sparc */
165 }
166
167 int
168 write_elfnotes(proc_t *p, int sig, vnode_t *vp, offset_t offset,
169 rlim64_t rlimit, cred_t *credp, core_content_t content)
170 {
171 union {
172 psinfo_t psinfo;
173 pstatus_t pstatus;
174 lwpsinfo_t lwpsinfo;
175 lwpstatus_t lwpstatus;
176 #if defined(__sparc)
177 gwindows_t gwindows;
178 asrset_t asrset;
179 #endif /* __sparc */
180 char xregs[1];
181 aux_entry_t auxv[__KERN_NAUXV_IMPL];
182 prcred_t pcred;
183 prpriv_t ppriv;
184 priv_impl_info_t prinfo;
185 struct utsname uts;
186 prsecflags_t psecflags;
187 } *bigwad;
188
189 size_t xregsize = prhasx(p)? prgetprxregsize(p) : 0;
190 size_t crsize = sizeof (prcred_t) + sizeof (gid_t) * (ngroups_max - 1);
191 size_t psize = prgetprivsize();
192 size_t bigsize = MAX(psize, MAX(sizeof (*bigwad),
193 MAX(xregsize, crsize)));
194
195 priv_impl_info_t *prii;
196
197 lwpdir_t *ldp;
198 lwpent_t *lep;
199 kthread_t *t;
200 klwp_t *lwp;
201 user_t *up;
202 int i;
203 int nlwp;
204 int nzomb;
205 int error;
206 uchar_t oldsig;
207 uf_info_t *fip;
208 int fd;
209 vnode_t *vroot;
210
211 #if defined(__i386) || defined(__i386_COMPAT)
212 struct ssd *ssd;
213 size_t ssdsize;
214 #endif /* __i386 || __i386_COMPAT */
215
216 bigsize = MAX(bigsize, priv_get_implinfo_size());
217
218 bigwad = kmem_alloc(bigsize, KM_SLEEP);
219
220 /*
221 * The order of the elfnote entries should be same here
222 * and in the gcore(1) command. Synchronization is
223 * needed between the kernel and gcore(1).
224 */
225
226 /*
227 * Get the psinfo, and set the wait status to indicate that a core was
228 * dumped. We have to forge this since p->p_wcode is not set yet.
229 */
230 mutex_enter(&p->p_lock);
231 prgetpsinfo(p, &bigwad->psinfo);
232 mutex_exit(&p->p_lock);
233 bigwad->psinfo.pr_wstat = wstat(CLD_DUMPED, sig);
234
235 error = elfnote(vp, &offset, NT_PSINFO, sizeof (bigwad->psinfo),
236 (caddr_t)&bigwad->psinfo, rlimit, credp);
237 if (error)
238 goto done;
239
240 /*
241 * Modify t_whystop and lwp_cursig so it appears that the current LWP
242 * is stopped after faulting on the signal that caused the core dump.
243 * As a result, prgetstatus() will record that signal, the saved
244 * lwp_siginfo, and its signal handler in the core file status. We
245 * restore lwp_cursig in case a subsequent signal was received while
246 * dumping core.
247 */
248 mutex_enter(&p->p_lock);
249 lwp = ttolwp(curthread);
250
251 oldsig = lwp->lwp_cursig;
252 lwp->lwp_cursig = (uchar_t)sig;
253 curthread->t_whystop = PR_FAULTED;
254
255 prgetstatus(p, &bigwad->pstatus, p->p_zone);
256 bigwad->pstatus.pr_lwp.pr_why = 0;
257
258 curthread->t_whystop = 0;
259 lwp->lwp_cursig = oldsig;
260 mutex_exit(&p->p_lock);
261
262 error = elfnote(vp, &offset, NT_PSTATUS, sizeof (bigwad->pstatus),
263 (caddr_t)&bigwad->pstatus, rlimit, credp);
264 if (error)
265 goto done;
266
267 error = elfnote(vp, &offset, NT_PLATFORM, strlen(platform) + 1,
268 platform, rlimit, credp);
269 if (error)
270 goto done;
271
272 up = PTOU(p);
273 for (i = 0; i < __KERN_NAUXV_IMPL; i++) {
274 bigwad->auxv[i].a_type = up->u_auxv[i].a_type;
275 bigwad->auxv[i].a_un.a_val = up->u_auxv[i].a_un.a_val;
276 }
277 error = elfnote(vp, &offset, NT_AUXV, sizeof (bigwad->auxv),
278 (caddr_t)bigwad->auxv, rlimit, credp);
279 if (error)
280 goto done;
281
282 bcopy(&utsname, &bigwad->uts, sizeof (struct utsname));
283 if (!INGLOBALZONE(p)) {
284 bcopy(p->p_zone->zone_nodename, &bigwad->uts.nodename,
285 _SYS_NMLN);
286 }
287 error = elfnote(vp, &offset, NT_UTSNAME, sizeof (struct utsname),
288 (caddr_t)&bigwad->uts, rlimit, credp);
289 if (error)
290 goto done;
291
292 prgetsecflags(p, &bigwad->psecflags);
293 error = elfnote(vp, &offset, NT_SECFLAGS, sizeof (prsecflags_t),
294 (caddr_t)&bigwad->psecflags, rlimit, credp);
295 if (error)
296 goto done;
297
298 prgetcred(p, &bigwad->pcred);
299
300 if (bigwad->pcred.pr_ngroups != 0) {
301 crsize = sizeof (prcred_t) +
302 sizeof (gid_t) * (bigwad->pcred.pr_ngroups - 1);
303 } else
304 crsize = sizeof (prcred_t);
305
306 error = elfnote(vp, &offset, NT_PRCRED, crsize,
307 (caddr_t)&bigwad->pcred, rlimit, credp);
308 if (error)
309 goto done;
310
311 error = elfnote(vp, &offset, NT_CONTENT, sizeof (core_content_t),
312 (caddr_t)&content, rlimit, credp);
313 if (error)
314 goto done;
315
316 prgetpriv(p, &bigwad->ppriv);
317
318 error = elfnote(vp, &offset, NT_PRPRIV, psize,
319 (caddr_t)&bigwad->ppriv, rlimit, credp);
320 if (error)
321 goto done;
322
323 prii = priv_hold_implinfo();
324 error = elfnote(vp, &offset, NT_PRPRIVINFO, priv_get_implinfo_size(),
325 (caddr_t)prii, rlimit, credp);
326 priv_release_implinfo();
327 if (error)
328 goto done;
329
330 /* zone can't go away as long as process exists */
331 error = elfnote(vp, &offset, NT_ZONENAME,
332 strlen(p->p_zone->zone_name) + 1, p->p_zone->zone_name,
333 rlimit, credp);
334 if (error)
335 goto done;
336
337
338 /* open file table */
339 vroot = PTOU(p)->u_rdir;
340 if (vroot == NULL)
341 vroot = rootdir;
342
343 VN_HOLD(vroot);
344
345 fip = P_FINFO(p);
346
347 for (fd = 0; fd < fip->fi_nfiles; fd++) {
348 uf_entry_t *ufp;
349 vnode_t *fvp;
350 struct file *fp;
351 vattr_t vattr;
352 prfdinfo_t fdinfo;
353
354 bzero(&fdinfo, sizeof (fdinfo));
355
356 mutex_enter(&fip->fi_lock);
357 UF_ENTER(ufp, fip, fd);
358 if (((fp = ufp->uf_file) == NULL) || (fp->f_count < 1)) {
359 UF_EXIT(ufp);
360 mutex_exit(&fip->fi_lock);
361 continue;
362 }
363
364 fdinfo.pr_fd = fd;
365 fdinfo.pr_fdflags = ufp->uf_flag;
366 fdinfo.pr_fileflags = fp->f_flag2;
367 fdinfo.pr_fileflags <<= 16;
368 fdinfo.pr_fileflags |= fp->f_flag;
369 if ((fdinfo.pr_fileflags & (FSEARCH | FEXEC)) == 0)
370 fdinfo.pr_fileflags += FOPEN;
371 fdinfo.pr_offset = fp->f_offset;
372
373
374 fvp = fp->f_vnode;
375 VN_HOLD(fvp);
376 UF_EXIT(ufp);
377 mutex_exit(&fip->fi_lock);
378
379 /*
380 * There are some vnodes that have no corresponding
381 * path. Its reasonable for this to fail, in which
382 * case the path will remain an empty string.
383 */
384 (void) vnodetopath(vroot, fvp, fdinfo.pr_path,
385 sizeof (fdinfo.pr_path), credp);
386
387 if (VOP_GETATTR(fvp, &vattr, 0, credp, NULL) != 0) {
388 /*
389 * Try to write at least a subset of information
390 */
391 fdinfo.pr_major = 0;
392 fdinfo.pr_minor = 0;
393 fdinfo.pr_ino = 0;
394 fdinfo.pr_mode = 0;
395 fdinfo.pr_uid = (uid_t)-1;
396 fdinfo.pr_gid = (gid_t)-1;
397 fdinfo.pr_rmajor = 0;
398 fdinfo.pr_rminor = 0;
399 fdinfo.pr_size = -1;
400
401 error = elfnote(vp, &offset, NT_FDINFO,
402 sizeof (fdinfo), &fdinfo, rlimit, credp);
403 VN_RELE(fvp);
404 if (error) {
405 VN_RELE(vroot);
406 goto done;
407 }
408 continue;
409 }
410
411 if (fvp->v_type == VSOCK)
412 fdinfo.pr_fileflags |= sock_getfasync(fvp);
413
414 VN_RELE(fvp);
415
416 /*
417 * This logic mirrors fstat(), which we cannot use
418 * directly, as it calls copyout().
419 */
420 fdinfo.pr_major = getmajor(vattr.va_fsid);
421 fdinfo.pr_minor = getminor(vattr.va_fsid);
422 fdinfo.pr_ino = (ino64_t)vattr.va_nodeid;
423 fdinfo.pr_mode = VTTOIF(vattr.va_type) | vattr.va_mode;
424 fdinfo.pr_uid = vattr.va_uid;
425 fdinfo.pr_gid = vattr.va_gid;
426 fdinfo.pr_rmajor = getmajor(vattr.va_rdev);
427 fdinfo.pr_rminor = getminor(vattr.va_rdev);
428 fdinfo.pr_size = (off64_t)vattr.va_size;
429
430 error = elfnote(vp, &offset, NT_FDINFO,
431 sizeof (fdinfo), &fdinfo, rlimit, credp);
432 if (error) {
433 VN_RELE(vroot);
434 goto done;
435 }
436 }
437
438 VN_RELE(vroot);
439
440 #if defined(__i386) || defined(__i386_COMPAT)
441 mutex_enter(&p->p_ldtlock);
442 ssdsize = prnldt(p) * sizeof (struct ssd);
443 if (ssdsize != 0) {
444 ssd = kmem_alloc(ssdsize, KM_SLEEP);
445 prgetldt(p, ssd);
446 error = elfnote(vp, &offset, NT_LDT, ssdsize,
447 (caddr_t)ssd, rlimit, credp);
448 kmem_free(ssd, ssdsize);
449 }
450 mutex_exit(&p->p_ldtlock);
451 if (error)
452 goto done;
453 #endif /* __i386 || defined(__i386_COMPAT) */
454
455 nlwp = p->p_lwpcnt;
456 nzomb = p->p_zombcnt;
457 /* for each entry in the lwp directory ... */
458 for (ldp = p->p_lwpdir; nlwp + nzomb != 0; ldp++) {
459
460 if ((lep = ldp->ld_entry) == NULL) /* empty slot */
461 continue;
462
463 if ((t = lep->le_thread) != NULL) { /* active lwp */
464 ASSERT(nlwp != 0);
465 nlwp--;
466 lwp = ttolwp(t);
467 mutex_enter(&p->p_lock);
468 prgetlwpsinfo(t, &bigwad->lwpsinfo);
469 mutex_exit(&p->p_lock);
470 } else { /* zombie lwp */
471 ASSERT(nzomb != 0);
472 nzomb--;
473 bzero(&bigwad->lwpsinfo, sizeof (bigwad->lwpsinfo));
474 bigwad->lwpsinfo.pr_lwpid = lep->le_lwpid;
475 bigwad->lwpsinfo.pr_state = SZOMB;
476 bigwad->lwpsinfo.pr_sname = 'Z';
477 bigwad->lwpsinfo.pr_start.tv_sec = lep->le_start;
478 }
479 error = elfnote(vp, &offset, NT_LWPSINFO,
480 sizeof (bigwad->lwpsinfo), (caddr_t)&bigwad->lwpsinfo,
481 rlimit, credp);
482 if (error)
483 goto done;
484 if (t == NULL) /* nothing more to do for a zombie */
485 continue;
486
487 mutex_enter(&p->p_lock);
488 if (t == curthread) {
489 /*
490 * Modify t_whystop and lwp_cursig so it appears that
491 * the current LWP is stopped after faulting on the
492 * signal that caused the core dump. As a result,
493 * prgetlwpstatus() will record that signal, the saved
494 * lwp_siginfo, and its signal handler in the core file
495 * status. We restore lwp_cursig in case a subsequent
496 * signal was received while dumping core.
497 */
498 oldsig = lwp->lwp_cursig;
499 lwp->lwp_cursig = (uchar_t)sig;
500 t->t_whystop = PR_FAULTED;
501
502 prgetlwpstatus(t, &bigwad->lwpstatus, p->p_zone);
503 bigwad->lwpstatus.pr_why = 0;
504
505 t->t_whystop = 0;
506 lwp->lwp_cursig = oldsig;
507 } else {
508 prgetlwpstatus(t, &bigwad->lwpstatus, p->p_zone);
509 }
510 mutex_exit(&p->p_lock);
511 error = elfnote(vp, &offset, NT_LWPSTATUS,
512 sizeof (bigwad->lwpstatus), (caddr_t)&bigwad->lwpstatus,
513 rlimit, credp);
514 if (error)
515 goto done;
516
517 #if defined(__sparc)
518 /*
519 * Unspilled SPARC register windows.
520 */
521 {
522 size_t size = prnwindows(lwp);
523
524 if (size != 0) {
525 size = sizeof (gwindows_t) -
526 (SPARC_MAXREGWINDOW - size) *
527 sizeof (struct rwindow);
528 prgetwindows(lwp, &bigwad->gwindows);
529 error = elfnote(vp, &offset, NT_GWINDOWS,
530 size, (caddr_t)&bigwad->gwindows,
531 rlimit, credp);
532 if (error)
533 goto done;
534 }
535 }
536 /*
537 * Ancillary State Registers.
538 */
539 if (p->p_model == DATAMODEL_LP64) {
540 prgetasregs(lwp, bigwad->asrset);
541 error = elfnote(vp, &offset, NT_ASRS,
542 sizeof (asrset_t), (caddr_t)bigwad->asrset,
543 rlimit, credp);
544 if (error)
545 goto done;
546 }
547 #endif /* __sparc */
548
549 if (xregsize) {
550 prgetprxregs(lwp, bigwad->xregs);
551 error = elfnote(vp, &offset, NT_PRXREG,
552 xregsize, bigwad->xregs, rlimit, credp);
553 if (error)
554 goto done;
555 }
556
557 if (t->t_lwp->lwp_spymaster != NULL) {
558 void *psaddr = t->t_lwp->lwp_spymaster;
559 #ifdef _ELF32_COMPAT
560 /*
561 * On a 64-bit kernel with 32-bit ELF compatibility,
562 * this file is compiled into two different objects:
563 * one is compiled normally, and the other is compiled
564 * with _ELF32_COMPAT set -- and therefore with a
565 * psinfo_t defined to be a psinfo32_t. However, the
566 * psinfo_t denoting our spymaster is always of the
567 * native type; if we are in the _ELF32_COMPAT case,
568 * we need to explicitly convert it.
569 */
570 if (p->p_model == DATAMODEL_ILP32) {
571 psinfo_kto32(psaddr, &bigwad->psinfo);
572 psaddr = &bigwad->psinfo;
573 }
574 #endif
575
576 error = elfnote(vp, &offset, NT_SPYMASTER,
577 sizeof (psinfo_t), psaddr, rlimit, credp);
578 if (error)
579 goto done;
580 }
581 }
582 ASSERT(nlwp == 0);
583
584 done:
585 kmem_free(bigwad, bigsize);
586 return (error);
587 }