]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/kern/imgact_elf.c
MFC 313407,313449: Copy ELF machine/flags from binaries to core dumps.
[FreeBSD/stable/10.git] / sys / kern / imgact_elf.c
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 Søren Schmidt
4  * Copyright (c) 1996 Peter Wemm
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_core.h"
37
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mman.h>
50 #include <sys/namei.h>
51 #include <sys/pioctl.h>
52 #include <sys/proc.h>
53 #include <sys/procfs.h>
54 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/rwlock.h>
57 #include <sys/sbuf.h>
58 #include <sys/sf_buf.h>
59 #include <sys/smp.h>
60 #include <sys/systm.h>
61 #include <sys/signalvar.h>
62 #include <sys/stat.h>
63 #include <sys/sx.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/syslog.h>
69 #include <sys/eventhandler.h>
70 #include <sys/user.h>
71
72 #include <net/zlib.h>
73
74 #include <vm/vm.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
81
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
84
85 #define ELF_NOTE_ROUNDSIZE      4
86 #define OLD_EI_BRAND    8
87
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90     const char *interp, int interp_name_len, int32_t *osrel);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
92     u_long *entry, size_t pagesize);
93 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
94     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95     size_t pagesize);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
106     "");
107
108 #ifdef COMPRESS_USER_CORES
109 static int compress_core(gzFile, char *, char *, unsigned int,
110     struct thread * td);
111 #endif
112 #define CORE_BUF_SIZE   (16 * 1024)
113
114 int __elfN(fallback_brand) = -1;
115 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
116     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
117     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
118 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
119     &__elfN(fallback_brand));
120
121 static int elf_legacy_coredump = 0;
122 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 
123     &elf_legacy_coredump, 0, "");
124
125 int __elfN(nxstack) =
126 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
127         1;
128 #else
129         0;
130 #endif
131 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
132     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
133     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
134
135 #if __ELF_WORD_SIZE == 32
136 #if defined(__amd64__) || defined(__ia64__)
137 int i386_read_exec = 0;
138 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
139     "enable execution from readable segments");
140 #endif
141 #endif
142
143 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
144
145 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
146 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
147 #define aligned(a, t)   (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
148
149 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
150
151 Elf_Brandnote __elfN(freebsd_brandnote) = {
152         .hdr.n_namesz   = sizeof(FREEBSD_ABI_VENDOR),
153         .hdr.n_descsz   = sizeof(int32_t),
154         .hdr.n_type     = 1,
155         .vendor         = FREEBSD_ABI_VENDOR,
156         .flags          = BN_TRANSLATE_OSREL,
157         .trans_osrel    = __elfN(freebsd_trans_osrel)
158 };
159
160 static boolean_t
161 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
162 {
163         uintptr_t p;
164
165         p = (uintptr_t)(note + 1);
166         p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
167         *osrel = *(const int32_t *)(p);
168
169         return (TRUE);
170 }
171
172 static const char GNU_ABI_VENDOR[] = "GNU";
173 static int GNU_KFREEBSD_ABI_DESC = 3;
174
175 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
176         .hdr.n_namesz   = sizeof(GNU_ABI_VENDOR),
177         .hdr.n_descsz   = 16,   /* XXX at least 16 */
178         .hdr.n_type     = 1,
179         .vendor         = GNU_ABI_VENDOR,
180         .flags          = BN_TRANSLATE_OSREL,
181         .trans_osrel    = kfreebsd_trans_osrel
182 };
183
184 static boolean_t
185 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
186 {
187         const Elf32_Word *desc;
188         uintptr_t p;
189
190         p = (uintptr_t)(note + 1);
191         p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
192
193         desc = (const Elf32_Word *)p;
194         if (desc[0] != GNU_KFREEBSD_ABI_DESC)
195                 return (FALSE);
196
197         /*
198          * Debian GNU/kFreeBSD embed the earliest compatible kernel version
199          * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
200          */
201         *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
202
203         return (TRUE);
204 }
205
206 int
207 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
208 {
209         int i;
210
211         for (i = 0; i < MAX_BRANDS; i++) {
212                 if (elf_brand_list[i] == NULL) {
213                         elf_brand_list[i] = entry;
214                         break;
215                 }
216         }
217         if (i == MAX_BRANDS) {
218                 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
219                         __func__, entry);
220                 return (-1);
221         }
222         return (0);
223 }
224
225 int
226 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
227 {
228         int i;
229
230         for (i = 0; i < MAX_BRANDS; i++) {
231                 if (elf_brand_list[i] == entry) {
232                         elf_brand_list[i] = NULL;
233                         break;
234                 }
235         }
236         if (i == MAX_BRANDS)
237                 return (-1);
238         return (0);
239 }
240
241 int
242 __elfN(brand_inuse)(Elf_Brandinfo *entry)
243 {
244         struct proc *p;
245         int rval = FALSE;
246
247         sx_slock(&allproc_lock);
248         FOREACH_PROC_IN_SYSTEM(p) {
249                 if (p->p_sysent == entry->sysvec) {
250                         rval = TRUE;
251                         break;
252                 }
253         }
254         sx_sunlock(&allproc_lock);
255
256         return (rval);
257 }
258
259 static Elf_Brandinfo *
260 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
261     int interp_name_len, int32_t *osrel)
262 {
263         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
264         Elf_Brandinfo *bi, *bi_m;
265         boolean_t ret;
266         int i;
267
268         /*
269          * We support four types of branding -- (1) the ELF EI_OSABI field
270          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
271          * branding w/in the ELF header, (3) path of the `interp_path'
272          * field, and (4) the ".note.ABI-tag" ELF section.
273          */
274
275         /* Look for an ".note.ABI-tag" ELF section */
276         bi_m = NULL;
277         for (i = 0; i < MAX_BRANDS; i++) {
278                 bi = elf_brand_list[i];
279                 if (bi == NULL)
280                         continue;
281                 if (hdr->e_machine == bi->machine && (bi->flags &
282                     (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
283                         ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
284                         /*
285                          * If note checker claimed the binary, but the
286                          * interpreter path in the image does not
287                          * match default one for the brand, try to
288                          * search for other brands with the same
289                          * interpreter.  Either there is better brand
290                          * with the right interpreter, or, failing
291                          * this, we return first brand which accepted
292                          * our note and, optionally, header.
293                          */
294                         if (ret && bi_m == NULL && (strlen(bi->interp_path) +
295                             1 != interp_name_len || strncmp(interp,
296                             bi->interp_path, interp_name_len) != 0)) {
297                                 bi_m = bi;
298                                 ret = 0;
299                         }
300                         if (ret)
301                                 return (bi);
302                 }
303         }
304         if (bi_m != NULL)
305                 return (bi_m);
306
307         /* If the executable has a brand, search for it in the brand list. */
308         for (i = 0; i < MAX_BRANDS; i++) {
309                 bi = elf_brand_list[i];
310                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
311                         continue;
312                 if (hdr->e_machine == bi->machine &&
313                     (hdr->e_ident[EI_OSABI] == bi->brand ||
314                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
315                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
316                         return (bi);
317         }
318
319         /* Lacking a known brand, search for a recognized interpreter. */
320         if (interp != NULL) {
321                 for (i = 0; i < MAX_BRANDS; i++) {
322                         bi = elf_brand_list[i];
323                         if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
324                                 continue;
325                         if (hdr->e_machine == bi->machine &&
326                             /* ELF image p_filesz includes terminating zero */
327                             strlen(bi->interp_path) + 1 == interp_name_len &&
328                             strncmp(interp, bi->interp_path, interp_name_len)
329                             == 0)
330                                 return (bi);
331                 }
332         }
333
334         /* Lacking a recognized interpreter, try the default brand */
335         for (i = 0; i < MAX_BRANDS; i++) {
336                 bi = elf_brand_list[i];
337                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
338                         continue;
339                 if (hdr->e_machine == bi->machine &&
340                     __elfN(fallback_brand) == bi->brand)
341                         return (bi);
342         }
343         return (NULL);
344 }
345
346 static int
347 __elfN(check_header)(const Elf_Ehdr *hdr)
348 {
349         Elf_Brandinfo *bi;
350         int i;
351
352         if (!IS_ELF(*hdr) ||
353             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
354             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
355             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
356             hdr->e_phentsize != sizeof(Elf_Phdr) ||
357             hdr->e_version != ELF_TARG_VER)
358                 return (ENOEXEC);
359
360         /*
361          * Make sure we have at least one brand for this machine.
362          */
363
364         for (i = 0; i < MAX_BRANDS; i++) {
365                 bi = elf_brand_list[i];
366                 if (bi != NULL && bi->machine == hdr->e_machine)
367                         break;
368         }
369         if (i == MAX_BRANDS)
370                 return (ENOEXEC);
371
372         return (0);
373 }
374
375 static int
376 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
377     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
378 {
379         struct sf_buf *sf;
380         int error;
381         vm_offset_t off;
382
383         /*
384          * Create the page if it doesn't exist yet. Ignore errors.
385          */
386         vm_map_lock(map);
387         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
388             VM_PROT_ALL, VM_PROT_ALL, 0);
389         vm_map_unlock(map);
390
391         /*
392          * Find the page from the underlying object.
393          */
394         if (object) {
395                 sf = vm_imgact_map_page(object, offset);
396                 if (sf == NULL)
397                         return (KERN_FAILURE);
398                 off = offset - trunc_page(offset);
399                 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
400                     end - start);
401                 vm_imgact_unmap_page(sf);
402                 if (error != 0)
403                         return (KERN_FAILURE);
404         }
405
406         return (KERN_SUCCESS);
407 }
408
409 static int
410 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
411     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
412 {
413         struct sf_buf *sf;
414         vm_offset_t off;
415         vm_size_t sz;
416         int error, rv;
417
418         if (start != trunc_page(start)) {
419                 rv = __elfN(map_partial)(map, object, offset, start,
420                     round_page(start), prot);
421                 if (rv)
422                         return (rv);
423                 offset += round_page(start) - start;
424                 start = round_page(start);
425         }
426         if (end != round_page(end)) {
427                 rv = __elfN(map_partial)(map, object, offset +
428                     trunc_page(end) - start, trunc_page(end), end, prot);
429                 if (rv)
430                         return (rv);
431                 end = trunc_page(end);
432         }
433         if (end > start) {
434                 if (offset & PAGE_MASK) {
435                         /*
436                          * The mapping is not page aligned. This means we have
437                          * to copy the data. Sigh.
438                          */
439                         rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
440                             VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
441                             0);
442                         if (rv != KERN_SUCCESS)
443                                 return (rv);
444                         if (object == NULL)
445                                 return (KERN_SUCCESS);
446                         for (; start < end; start += sz) {
447                                 sf = vm_imgact_map_page(object, offset);
448                                 if (sf == NULL)
449                                         return (KERN_FAILURE);
450                                 off = offset - trunc_page(offset);
451                                 sz = end - start;
452                                 if (sz > PAGE_SIZE - off)
453                                         sz = PAGE_SIZE - off;
454                                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
455                                     (caddr_t)start, sz);
456                                 vm_imgact_unmap_page(sf);
457                                 if (error != 0)
458                                         return (KERN_FAILURE);
459                                 offset += sz;
460                         }
461                         rv = KERN_SUCCESS;
462                 } else {
463                         vm_object_reference(object);
464                         vm_map_lock(map);
465                         rv = vm_map_insert(map, object, offset, start, end,
466                             prot, VM_PROT_ALL, cow);
467                         vm_map_unlock(map);
468                         if (rv != KERN_SUCCESS)
469                                 vm_object_deallocate(object);
470                 }
471                 return (rv);
472         } else {
473                 return (KERN_SUCCESS);
474         }
475 }
476
477 static int
478 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
479     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
480     size_t pagesize)
481 {
482         struct sf_buf *sf;
483         size_t map_len;
484         vm_map_t map;
485         vm_object_t object;
486         vm_offset_t map_addr;
487         int error, rv, cow;
488         size_t copy_len;
489         vm_offset_t file_addr;
490
491         /*
492          * It's necessary to fail if the filsz + offset taken from the
493          * header is greater than the actual file pager object's size.
494          * If we were to allow this, then the vm_map_find() below would
495          * walk right off the end of the file object and into the ether.
496          *
497          * While I'm here, might as well check for something else that
498          * is invalid: filsz cannot be greater than memsz.
499          */
500         if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
501                 uprintf("elf_load_section: truncated ELF file\n");
502                 return (ENOEXEC);
503         }
504
505         object = imgp->object;
506         map = &imgp->proc->p_vmspace->vm_map;
507         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
508         file_addr = trunc_page_ps(offset, pagesize);
509
510         /*
511          * We have two choices.  We can either clear the data in the last page
512          * of an oversized mapping, or we can start the anon mapping a page
513          * early and copy the initialized data into that first page.  We
514          * choose the second..
515          */
516         if (memsz > filsz)
517                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
518         else
519                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
520
521         if (map_len != 0) {
522                 /* cow flags: don't dump readonly sections in core */
523                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
524                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
525
526                 rv = __elfN(map_insert)(map,
527                                       object,
528                                       file_addr,        /* file offset */
529                                       map_addr,         /* virtual start */
530                                       map_addr + map_len,/* virtual end */
531                                       prot,
532                                       cow);
533                 if (rv != KERN_SUCCESS)
534                         return (EINVAL);
535
536                 /* we can stop now if we've covered it all */
537                 if (memsz == filsz) {
538                         return (0);
539                 }
540         }
541
542
543         /*
544          * We have to get the remaining bit of the file into the first part
545          * of the oversized map segment.  This is normally because the .data
546          * segment in the file is extended to provide bss.  It's a neat idea
547          * to try and save a page, but it's a pain in the behind to implement.
548          */
549         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
550         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
551         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
552             map_addr;
553
554         /* This had damn well better be true! */
555         if (map_len != 0) {
556                 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
557                     map_len, VM_PROT_ALL, 0);
558                 if (rv != KERN_SUCCESS) {
559                         return (EINVAL);
560                 }
561         }
562
563         if (copy_len != 0) {
564                 vm_offset_t off;
565
566                 sf = vm_imgact_map_page(object, offset + filsz);
567                 if (sf == NULL)
568                         return (EIO);
569
570                 /* send the page fragment to user space */
571                 off = trunc_page_ps(offset + filsz, pagesize) -
572                     trunc_page(offset + filsz);
573                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
574                     (caddr_t)map_addr, copy_len);
575                 vm_imgact_unmap_page(sf);
576                 if (error) {
577                         return (error);
578                 }
579         }
580
581         /*
582          * set it to the specified protection.
583          * XXX had better undo the damage from pasting over the cracks here!
584          */
585         vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
586             map_len), prot, FALSE);
587
588         return (0);
589 }
590
591 /*
592  * Load the file "file" into memory.  It may be either a shared object
593  * or an executable.
594  *
595  * The "addr" reference parameter is in/out.  On entry, it specifies
596  * the address where a shared object should be loaded.  If the file is
597  * an executable, this value is ignored.  On exit, "addr" specifies
598  * where the file was actually loaded.
599  *
600  * The "entry" reference parameter is out only.  On exit, it specifies
601  * the entry point for the loaded file.
602  */
603 static int
604 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
605         u_long *entry, size_t pagesize)
606 {
607         struct {
608                 struct nameidata nd;
609                 struct vattr attr;
610                 struct image_params image_params;
611         } *tempdata;
612         const Elf_Ehdr *hdr = NULL;
613         const Elf_Phdr *phdr = NULL;
614         struct nameidata *nd;
615         struct vattr *attr;
616         struct image_params *imgp;
617         vm_prot_t prot;
618         u_long rbase;
619         u_long base_addr = 0;
620         int error, i, numsegs;
621
622 #ifdef CAPABILITY_MODE
623         /*
624          * XXXJA: This check can go away once we are sufficiently confident
625          * that the checks in namei() are correct.
626          */
627         if (IN_CAPABILITY_MODE(curthread))
628                 return (ECAPMODE);
629 #endif
630
631         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
632         nd = &tempdata->nd;
633         attr = &tempdata->attr;
634         imgp = &tempdata->image_params;
635
636         /*
637          * Initialize part of the common data
638          */
639         imgp->proc = p;
640         imgp->attr = attr;
641         imgp->firstpage = NULL;
642         imgp->image_header = NULL;
643         imgp->object = NULL;
644         imgp->execlabel = NULL;
645
646         NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
647         if ((error = namei(nd)) != 0) {
648                 nd->ni_vp = NULL;
649                 goto fail;
650         }
651         NDFREE(nd, NDF_ONLY_PNBUF);
652         imgp->vp = nd->ni_vp;
653
654         /*
655          * Check permissions, modes, uid, etc on the file, and "open" it.
656          */
657         error = exec_check_permissions(imgp);
658         if (error)
659                 goto fail;
660
661         error = exec_map_first_page(imgp);
662         if (error)
663                 goto fail;
664
665         /*
666          * Also make certain that the interpreter stays the same, so set
667          * its VV_TEXT flag, too.
668          */
669         VOP_SET_TEXT(nd->ni_vp);
670
671         imgp->object = nd->ni_vp->v_object;
672
673         hdr = (const Elf_Ehdr *)imgp->image_header;
674         if ((error = __elfN(check_header)(hdr)) != 0)
675                 goto fail;
676         if (hdr->e_type == ET_DYN)
677                 rbase = *addr;
678         else if (hdr->e_type == ET_EXEC)
679                 rbase = 0;
680         else {
681                 error = ENOEXEC;
682                 goto fail;
683         }
684
685         /* Only support headers that fit within first page for now      */
686         if ((hdr->e_phoff > PAGE_SIZE) ||
687             (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
688                 error = ENOEXEC;
689                 goto fail;
690         }
691
692         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
693         if (!aligned(phdr, Elf_Addr)) {
694                 error = ENOEXEC;
695                 goto fail;
696         }
697
698         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
699                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
700                         /* Loadable segment */
701                         prot = __elfN(trans_prot)(phdr[i].p_flags);
702                         error = __elfN(load_section)(imgp, phdr[i].p_offset,
703                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
704                             phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
705                         if (error != 0)
706                                 goto fail;
707                         /*
708                          * Establish the base address if this is the
709                          * first segment.
710                          */
711                         if (numsegs == 0)
712                                 base_addr = trunc_page(phdr[i].p_vaddr +
713                                     rbase);
714                         numsegs++;
715                 }
716         }
717         *addr = base_addr;
718         *entry = (unsigned long)hdr->e_entry + rbase;
719
720 fail:
721         if (imgp->firstpage)
722                 exec_unmap_first_page(imgp);
723
724         if (nd->ni_vp)
725                 vput(nd->ni_vp);
726
727         free(tempdata, M_TEMP);
728
729         return (error);
730 }
731
732 static int
733 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
734 {
735         struct thread *td;
736         const Elf_Ehdr *hdr;
737         const Elf_Phdr *phdr;
738         Elf_Auxargs *elf_auxargs;
739         struct vmspace *vmspace;
740         const char *err_str, *newinterp;
741         char *interp, *interp_buf, *path;
742         Elf_Brandinfo *brand_info;
743         struct sysentvec *sv;
744         vm_prot_t prot;
745         u_long text_size, data_size, total_size, text_addr, data_addr;
746         u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
747         int32_t osrel;
748         int error, i, n, interp_name_len, have_interp;
749
750         hdr = (const Elf_Ehdr *)imgp->image_header;
751
752         /*
753          * Do we have a valid ELF header ?
754          *
755          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
756          * if particular brand doesn't support it.
757          */
758         if (__elfN(check_header)(hdr) != 0 ||
759             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
760                 return (-1);
761
762         /*
763          * From here on down, we return an errno, not -1, as we've
764          * detected an ELF file.
765          */
766
767         if ((hdr->e_phoff > PAGE_SIZE) ||
768             (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
769                 /* Only support headers in first page for now */
770                 uprintf("Program headers not in the first page\n");
771                 return (ENOEXEC);
772         }
773         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 
774         if (!aligned(phdr, Elf_Addr)) {
775                 uprintf("Unaligned program headers\n");
776                 return (ENOEXEC);
777         }
778
779         n = error = 0;
780         baddr = 0;
781         osrel = 0;
782         text_size = data_size = total_size = text_addr = data_addr = 0;
783         entry = proghdr = 0;
784         interp_name_len = 0;
785         err_str = newinterp = NULL;
786         interp = interp_buf = NULL;
787         td = curthread;
788
789         for (i = 0; i < hdr->e_phnum; i++) {
790                 switch (phdr[i].p_type) {
791                 case PT_LOAD:
792                         if (n == 0)
793                                 baddr = phdr[i].p_vaddr;
794                         n++;
795                         break;
796                 case PT_INTERP:
797                         /* Path to interpreter */
798                         if (phdr[i].p_filesz > MAXPATHLEN) {
799                                 uprintf("Invalid PT_INTERP\n");
800                                 error = ENOEXEC;
801                                 goto ret;
802                         }
803                         if (interp != NULL) {
804                                 uprintf("Multiple PT_INTERP headers\n");
805                                 error = ENOEXEC;
806                                 goto ret;
807                         }
808                         interp_name_len = phdr[i].p_filesz;
809                         if (phdr[i].p_offset > PAGE_SIZE ||
810                             interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
811                                 VOP_UNLOCK(imgp->vp, 0);
812                                 interp_buf = malloc(interp_name_len + 1, M_TEMP,
813                                     M_WAITOK);
814                                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
815                                 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
816                                     interp_name_len, phdr[i].p_offset,
817                                     UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
818                                     NOCRED, NULL, td);
819                                 if (error != 0) {
820                                         uprintf("i/o error PT_INTERP\n");
821                                         goto ret;
822                                 }
823                                 interp_buf[interp_name_len] = '\0';
824                                 interp = interp_buf;
825                         } else {
826                                 interp = __DECONST(char *, imgp->image_header) +
827                                     phdr[i].p_offset;
828                         }
829                         break;
830                 case PT_GNU_STACK:
831                         if (__elfN(nxstack))
832                                 imgp->stack_prot =
833                                     __elfN(trans_prot)(phdr[i].p_flags);
834                         imgp->stack_sz = phdr[i].p_memsz;
835                         break;
836                 }
837         }
838
839         brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
840             &osrel);
841         if (brand_info == NULL) {
842                 uprintf("ELF binary type \"%u\" not known.\n",
843                     hdr->e_ident[EI_OSABI]);
844                 error = ENOEXEC;
845                 goto ret;
846         }
847         if (hdr->e_type == ET_DYN) {
848                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
849                         uprintf("Cannot execute shared object\n");
850                         error = ENOEXEC;
851                         goto ret;
852                 }
853                 /*
854                  * Honour the base load address from the dso if it is
855                  * non-zero for some reason.
856                  */
857                 if (baddr == 0)
858                         et_dyn_addr = ET_DYN_LOAD_ADDR;
859                 else
860                         et_dyn_addr = 0;
861         } else
862                 et_dyn_addr = 0;
863         sv = brand_info->sysvec;
864         if (interp != NULL && brand_info->interp_newpath != NULL)
865                 newinterp = brand_info->interp_newpath;
866
867         /*
868          * Avoid a possible deadlock if the current address space is destroyed
869          * and that address space maps the locked vnode.  In the common case,
870          * the locked vnode's v_usecount is decremented but remains greater
871          * than zero.  Consequently, the vnode lock is not needed by vrele().
872          * However, in cases where the vnode lock is external, such as nullfs,
873          * v_usecount may become zero.
874          *
875          * The VV_TEXT flag prevents modifications to the executable while
876          * the vnode is unlocked.
877          */
878         VOP_UNLOCK(imgp->vp, 0);
879
880         error = exec_new_vmspace(imgp, sv);
881         imgp->proc->p_sysent = sv;
882
883         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
884         if (error != 0)
885                 goto ret;
886
887         for (i = 0; i < hdr->e_phnum; i++) {
888                 switch (phdr[i].p_type) {
889                 case PT_LOAD:   /* Loadable segment */
890                         if (phdr[i].p_memsz == 0)
891                                 break;
892                         prot = __elfN(trans_prot)(phdr[i].p_flags);
893                         error = __elfN(load_section)(imgp, phdr[i].p_offset,
894                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
895                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
896                             sv->sv_pagesize);
897                         if (error != 0)
898                                 goto ret;
899
900                         /*
901                          * If this segment contains the program headers,
902                          * remember their virtual address for the AT_PHDR
903                          * aux entry. Static binaries don't usually include
904                          * a PT_PHDR entry.
905                          */
906                         if (phdr[i].p_offset == 0 &&
907                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
908                                 <= phdr[i].p_filesz)
909                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
910                                     et_dyn_addr;
911
912                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
913                         seg_size = round_page(phdr[i].p_memsz +
914                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
915
916                         /*
917                          * Make the largest executable segment the official
918                          * text segment and all others data.
919                          *
920                          * Note that obreak() assumes that data_addr + 
921                          * data_size == end of data load area, and the ELF
922                          * file format expects segments to be sorted by
923                          * address.  If multiple data segments exist, the
924                          * last one will be used.
925                          */
926
927                         if (phdr[i].p_flags & PF_X && text_size < seg_size) {
928                                 text_size = seg_size;
929                                 text_addr = seg_addr;
930                         } else {
931                                 data_size = seg_size;
932                                 data_addr = seg_addr;
933                         }
934                         total_size += seg_size;
935                         break;
936                 case PT_PHDR:   /* Program header table info */
937                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
938                         break;
939                 default:
940                         break;
941                 }
942         }
943         
944         if (data_addr == 0 && data_size == 0) {
945                 data_addr = text_addr;
946                 data_size = text_size;
947         }
948
949         entry = (u_long)hdr->e_entry + et_dyn_addr;
950
951         /*
952          * Check limits.  It should be safe to check the
953          * limits after loading the segments since we do
954          * not actually fault in all the segments pages.
955          */
956         PROC_LOCK(imgp->proc);
957         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA))
958                 err_str = "Data segment size exceeds process limit";
959         else if (text_size > maxtsiz)
960                 err_str = "Text segment size exceeds system limit";
961         else if (total_size > lim_cur(imgp->proc, RLIMIT_VMEM))
962                 err_str = "Total segment size exceeds process limit";
963         else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
964                 err_str = "Data segment size exceeds resource limit";
965         else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
966                 err_str = "Total segment size exceeds resource limit";
967         if (err_str != NULL) {
968                 PROC_UNLOCK(imgp->proc);
969                 uprintf("%s\n", err_str);
970                 error = ENOMEM;
971                 goto ret;
972         }
973
974         vmspace = imgp->proc->p_vmspace;
975         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
976         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
977         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
978         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
979
980         /*
981          * We load the dynamic linker where a userland call
982          * to mmap(0, ...) would put it.  The rationale behind this
983          * calculation is that it leaves room for the heap to grow to
984          * its maximum allowed size.
985          */
986         addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
987             RLIMIT_DATA));
988         PROC_UNLOCK(imgp->proc);
989
990         imgp->entry_addr = entry;
991
992         if (interp != NULL) {
993                 have_interp = FALSE;
994                 VOP_UNLOCK(imgp->vp, 0);
995                 if (brand_info->emul_path != NULL &&
996                     brand_info->emul_path[0] != '\0') {
997                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
998                         snprintf(path, MAXPATHLEN, "%s%s",
999                             brand_info->emul_path, interp);
1000                         error = __elfN(load_file)(imgp->proc, path, &addr,
1001                             &imgp->entry_addr, sv->sv_pagesize);
1002                         free(path, M_TEMP);
1003                         if (error == 0)
1004                                 have_interp = TRUE;
1005                 }
1006                 if (!have_interp && newinterp != NULL &&
1007                     (brand_info->interp_path == NULL ||
1008                     strcmp(interp, brand_info->interp_path) == 0)) {
1009                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1010                             &imgp->entry_addr, sv->sv_pagesize);
1011                         if (error == 0)
1012                                 have_interp = TRUE;
1013                 }
1014                 if (!have_interp) {
1015                         error = __elfN(load_file)(imgp->proc, interp, &addr,
1016                             &imgp->entry_addr, sv->sv_pagesize);
1017                 }
1018                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1019                 if (error != 0) {
1020                         uprintf("ELF interpreter %s not found, error %d\n",
1021                             interp, error);
1022                         goto ret;
1023                 }
1024         } else
1025                 addr = et_dyn_addr;
1026
1027         /*
1028          * Construct auxargs table (used by the fixup routine)
1029          */
1030         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1031         elf_auxargs->execfd = -1;
1032         elf_auxargs->phdr = proghdr;
1033         elf_auxargs->phent = hdr->e_phentsize;
1034         elf_auxargs->phnum = hdr->e_phnum;
1035         elf_auxargs->pagesz = PAGE_SIZE;
1036         elf_auxargs->base = addr;
1037         elf_auxargs->flags = 0;
1038         elf_auxargs->entry = entry;
1039
1040         imgp->auxargs = elf_auxargs;
1041         imgp->interpreted = 0;
1042         imgp->reloc_base = addr;
1043         imgp->proc->p_osrel = osrel;
1044         imgp->proc->p_elf_machine = hdr->e_machine;
1045         imgp->proc->p_elf_flags = hdr->e_flags;
1046
1047  ret:
1048         free(interp_buf, M_TEMP);
1049         return (error);
1050 }
1051
1052 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1053
1054 int
1055 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1056 {
1057         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1058         Elf_Addr *base;
1059         Elf_Addr *pos;
1060
1061         base = (Elf_Addr *)*stack_base;
1062         pos = base + (imgp->args->argc + imgp->args->envc + 2);
1063
1064         if (args->execfd != -1)
1065                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1066         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1067         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1068         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1069         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1070         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1071         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1072         AUXARGS_ENTRY(pos, AT_BASE, args->base);
1073         if (imgp->execpathp != 0)
1074                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1075         AUXARGS_ENTRY(pos, AT_OSRELDATE,
1076             imgp->proc->p_ucred->cr_prison->pr_osreldate);
1077         if (imgp->canary != 0) {
1078                 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1079                 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1080         }
1081         AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1082         if (imgp->pagesizes != 0) {
1083                 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1084                 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1085         }
1086         if (imgp->sysent->sv_timekeep_base != 0) {
1087                 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1088                     imgp->sysent->sv_timekeep_base);
1089         }
1090         AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1091             != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1092             imgp->sysent->sv_stackprot);
1093         AUXARGS_ENTRY(pos, AT_NULL, 0);
1094
1095         free(imgp->auxargs, M_TEMP);
1096         imgp->auxargs = NULL;
1097
1098         base--;
1099         suword(base, (long)imgp->args->argc);
1100         *stack_base = (register_t *)base;
1101         return (0);
1102 }
1103
1104 /*
1105  * Code for generating ELF core dumps.
1106  */
1107
1108 typedef void (*segment_callback)(vm_map_entry_t, void *);
1109
1110 /* Closure for cb_put_phdr(). */
1111 struct phdr_closure {
1112         Elf_Phdr *phdr;         /* Program header to fill in */
1113         Elf_Off offset;         /* Offset of segment in core file */
1114 };
1115
1116 /* Closure for cb_size_segment(). */
1117 struct sseg_closure {
1118         int count;              /* Count of writable segments. */
1119         size_t size;            /* Total size of all writable segments. */
1120 };
1121
1122 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1123
1124 struct note_info {
1125         int             type;           /* Note type. */
1126         outfunc_t       outfunc;        /* Output function. */
1127         void            *outarg;        /* Argument for the output function. */
1128         size_t          outsize;        /* Output size. */
1129         TAILQ_ENTRY(note_info) link;    /* Link to the next note info. */
1130 };
1131
1132 TAILQ_HEAD(note_info_list, note_info);
1133
1134 static void cb_put_phdr(vm_map_entry_t, void *);
1135 static void cb_size_segment(vm_map_entry_t, void *);
1136 static void each_writable_segment(struct thread *, segment_callback, void *);
1137 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1138     int, void *, size_t, struct note_info_list *, size_t, gzFile);
1139 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1140     size_t *);
1141 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1142 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1143 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1144 static int sbuf_drain_core_output(void *, const char *, int);
1145 static int sbuf_drain_count(void *arg, const char *data, int len);
1146
1147 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1148 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1150 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1151 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1152 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1153 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1154 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1155 static void note_procstat_files(void *, struct sbuf *, size_t *);
1156 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1157 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1158 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1159 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1160 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1161
1162 #ifdef COMPRESS_USER_CORES
1163 extern int compress_user_cores;
1164 extern int compress_user_cores_gzlevel;
1165 #endif
1166
1167 static int
1168 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1169     struct ucred *active_cred, struct ucred *file_cred,
1170     struct thread *td, char *core_buf, gzFile gzfile) {
1171
1172         int error;
1173         if (gzfile) {
1174 #ifdef COMPRESS_USER_CORES
1175                 error = compress_core(gzfile, base, core_buf, len, td);
1176 #else
1177                 panic("shouldn't be here");
1178 #endif
1179         } else {
1180                 /*
1181                  * EFAULT is a non-fatal error that we can get, for example,
1182                  * if the segment is backed by a file but extends beyond its
1183                  * end.
1184                  */
1185                 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1186                     UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1187                     NULL, td);
1188                 if (error == EFAULT) {
1189                         log(LOG_WARNING, "Failed to fully fault in a core file "
1190                             "segment at VA %p with size 0x%zx to be written at "
1191                             "offset 0x%jx for process %s\n", base, len, offset,
1192                             curproc->p_comm);
1193
1194                         /*
1195                          * Write a "real" zero byte at the end of the target
1196                          * region in the case this is the last segment.
1197                          * The intermediate space will be implicitly
1198                          * zero-filled.
1199                          */
1200                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
1201                             __DECONST(void *, zero_region), 1, offset + len - 1,
1202                             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, active_cred,
1203                             file_cred, NULL, td);
1204                 }
1205         }
1206         return (error);
1207 }
1208
1209 /* Coredump output parameters for sbuf drain routine. */
1210 struct sbuf_drain_core_params {
1211         off_t           offset;
1212         struct ucred    *active_cred;
1213         struct ucred    *file_cred;
1214         struct thread   *td;
1215         struct vnode    *vp;
1216 #ifdef COMPRESS_USER_CORES
1217         gzFile          gzfile;
1218 #endif
1219 };
1220
1221 /*
1222  * Drain into a core file.
1223  */
1224 static int
1225 sbuf_drain_core_output(void *arg, const char *data, int len)
1226 {
1227         struct sbuf_drain_core_params *p;
1228         int error, locked;
1229
1230         p = (struct sbuf_drain_core_params *)arg;
1231
1232         /*
1233          * Some kern_proc out routines that print to this sbuf may
1234          * call us with the process lock held. Draining with the
1235          * non-sleepable lock held is unsafe. The lock is needed for
1236          * those routines when dumping a live process. In our case we
1237          * can safely release the lock before draining and acquire
1238          * again after.
1239          */
1240         locked = PROC_LOCKED(p->td->td_proc);
1241         if (locked)
1242                 PROC_UNLOCK(p->td->td_proc);
1243 #ifdef COMPRESS_USER_CORES
1244         if (p->gzfile != Z_NULL)
1245                 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1246                     len, p->td);
1247         else
1248 #endif
1249                 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1250                     __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1251                     IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1252                     p->td);
1253         if (locked)
1254                 PROC_LOCK(p->td->td_proc);
1255         if (error != 0)
1256                 return (-error);
1257         p->offset += len;
1258         return (len);
1259 }
1260
1261 /*
1262  * Drain into a counter.
1263  */
1264 static int
1265 sbuf_drain_count(void *arg, const char *data __unused, int len)
1266 {
1267         size_t *sizep;
1268
1269         sizep = (size_t *)arg;
1270         *sizep += len;
1271         return (len);
1272 }
1273
1274 int
1275 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1276 {
1277         struct ucred *cred = td->td_ucred;
1278         int error = 0;
1279         struct sseg_closure seginfo;
1280         struct note_info_list notelst;
1281         struct note_info *ninfo;
1282         void *hdr;
1283         size_t hdrsize, notesz, coresize;
1284
1285         gzFile gzfile = Z_NULL;
1286         char *core_buf = NULL;
1287 #ifdef COMPRESS_USER_CORES
1288         char gzopen_flags[8];
1289         char *p;
1290         int doing_compress = flags & IMGACT_CORE_COMPRESS;
1291 #endif
1292
1293         hdr = NULL;
1294         TAILQ_INIT(&notelst);
1295
1296 #ifdef COMPRESS_USER_CORES
1297         if (doing_compress) {
1298                 p = gzopen_flags;
1299                 *p++ = 'w';
1300                 if (compress_user_cores_gzlevel >= 0 &&
1301                     compress_user_cores_gzlevel <= 9)
1302                         *p++ = '0' + compress_user_cores_gzlevel;
1303                 *p = 0;
1304                 gzfile = gz_open("", gzopen_flags, vp);
1305                 if (gzfile == Z_NULL) {
1306                         error = EFAULT;
1307                         goto done;
1308                 }
1309                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1310                 if (!core_buf) {
1311                         error = ENOMEM;
1312                         goto done;
1313                 }
1314         }
1315 #endif
1316
1317         /* Size the program segments. */
1318         seginfo.count = 0;
1319         seginfo.size = 0;
1320         each_writable_segment(td, cb_size_segment, &seginfo);
1321
1322         /*
1323          * Collect info about the core file header area.
1324          */
1325         hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1326         __elfN(prepare_notes)(td, &notelst, &notesz);
1327         coresize = round_page(hdrsize + notesz) + seginfo.size;
1328
1329 #ifdef RACCT
1330         if (racct_enable) {
1331                 PROC_LOCK(td->td_proc);
1332                 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1333                 PROC_UNLOCK(td->td_proc);
1334                 if (error != 0) {
1335                         error = EFAULT;
1336                         goto done;
1337                 }
1338         }
1339 #endif
1340         if (coresize >= limit) {
1341                 error = EFAULT;
1342                 goto done;
1343         }
1344
1345         /*
1346          * Allocate memory for building the header, fill it up,
1347          * and write it out following the notes.
1348          */
1349         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1350         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1351             &notelst, notesz, gzfile);
1352
1353         /* Write the contents of all of the writable segments. */
1354         if (error == 0) {
1355                 Elf_Phdr *php;
1356                 off_t offset;
1357                 int i;
1358
1359                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1360                 offset = round_page(hdrsize + notesz);
1361                 for (i = 0; i < seginfo.count; i++) {
1362                         error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1363                             php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1364                         if (error != 0)
1365                                 break;
1366                         offset += php->p_filesz;
1367                         php++;
1368                 }
1369         }
1370         if (error) {
1371                 log(LOG_WARNING,
1372                     "Failed to write core file for process %s (error %d)\n",
1373                     curproc->p_comm, error);
1374         }
1375
1376 done:
1377 #ifdef COMPRESS_USER_CORES
1378         if (core_buf)
1379                 free(core_buf, M_TEMP);
1380         if (gzfile)
1381                 gzclose(gzfile);
1382 #endif
1383         while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1384                 TAILQ_REMOVE(&notelst, ninfo, link);
1385                 free(ninfo, M_TEMP);
1386         }
1387         if (hdr != NULL)
1388                 free(hdr, M_TEMP);
1389
1390         return (error);
1391 }
1392
1393 /*
1394  * A callback for each_writable_segment() to write out the segment's
1395  * program header entry.
1396  */
1397 static void
1398 cb_put_phdr(entry, closure)
1399         vm_map_entry_t entry;
1400         void *closure;
1401 {
1402         struct phdr_closure *phc = (struct phdr_closure *)closure;
1403         Elf_Phdr *phdr = phc->phdr;
1404
1405         phc->offset = round_page(phc->offset);
1406
1407         phdr->p_type = PT_LOAD;
1408         phdr->p_offset = phc->offset;
1409         phdr->p_vaddr = entry->start;
1410         phdr->p_paddr = 0;
1411         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1412         phdr->p_align = PAGE_SIZE;
1413         phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1414
1415         phc->offset += phdr->p_filesz;
1416         phc->phdr++;
1417 }
1418
1419 /*
1420  * A callback for each_writable_segment() to gather information about
1421  * the number of segments and their total size.
1422  */
1423 static void
1424 cb_size_segment(entry, closure)
1425         vm_map_entry_t entry;
1426         void *closure;
1427 {
1428         struct sseg_closure *ssc = (struct sseg_closure *)closure;
1429
1430         ssc->count++;
1431         ssc->size += entry->end - entry->start;
1432 }
1433
1434 /*
1435  * For each writable segment in the process's memory map, call the given
1436  * function with a pointer to the map entry and some arbitrary
1437  * caller-supplied data.
1438  */
1439 static void
1440 each_writable_segment(td, func, closure)
1441         struct thread *td;
1442         segment_callback func;
1443         void *closure;
1444 {
1445         struct proc *p = td->td_proc;
1446         vm_map_t map = &p->p_vmspace->vm_map;
1447         vm_map_entry_t entry;
1448         vm_object_t backing_object, object;
1449         boolean_t ignore_entry;
1450
1451         vm_map_lock_read(map);
1452         for (entry = map->header.next; entry != &map->header;
1453             entry = entry->next) {
1454                 /*
1455                  * Don't dump inaccessible mappings, deal with legacy
1456                  * coredump mode.
1457                  *
1458                  * Note that read-only segments related to the elf binary
1459                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1460                  * need to arbitrarily ignore such segments.
1461                  */
1462                 if (elf_legacy_coredump) {
1463                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1464                                 continue;
1465                 } else {
1466                         if ((entry->protection & VM_PROT_ALL) == 0)
1467                                 continue;
1468                 }
1469
1470                 /*
1471                  * Dont include memory segment in the coredump if
1472                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1473                  * madvise(2).  Do not dump submaps (i.e. parts of the
1474                  * kernel map).
1475                  */
1476                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1477                         continue;
1478
1479                 if ((object = entry->object.vm_object) == NULL)
1480                         continue;
1481
1482                 /* Ignore memory-mapped devices and such things. */
1483                 VM_OBJECT_RLOCK(object);
1484                 while ((backing_object = object->backing_object) != NULL) {
1485                         VM_OBJECT_RLOCK(backing_object);
1486                         VM_OBJECT_RUNLOCK(object);
1487                         object = backing_object;
1488                 }
1489                 ignore_entry = object->type != OBJT_DEFAULT &&
1490                     object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1491                     object->type != OBJT_PHYS;
1492                 VM_OBJECT_RUNLOCK(object);
1493                 if (ignore_entry)
1494                         continue;
1495
1496                 (*func)(entry, closure);
1497         }
1498         vm_map_unlock_read(map);
1499 }
1500
1501 /*
1502  * Write the core file header to the file, including padding up to
1503  * the page boundary.
1504  */
1505 static int
1506 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1507     int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1508     size_t notesz, gzFile gzfile)
1509 {
1510         struct sbuf_drain_core_params params;
1511         struct note_info *ninfo;
1512         struct sbuf *sb;
1513         int error;
1514
1515         /* Fill in the header. */
1516         bzero(hdr, hdrsize);
1517         __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1518
1519         params.offset = 0;
1520         params.active_cred = cred;
1521         params.file_cred = NOCRED;
1522         params.td = td;
1523         params.vp = vp;
1524 #ifdef COMPRESS_USER_CORES
1525         params.gzfile = gzfile;
1526 #endif
1527         sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1528         sbuf_set_drain(sb, sbuf_drain_core_output, &params);
1529         sbuf_start_section(sb, NULL);
1530         sbuf_bcat(sb, hdr, hdrsize);
1531         TAILQ_FOREACH(ninfo, notelst, link)
1532             __elfN(putnote)(ninfo, sb);
1533         /* Align up to a page boundary for the program segments. */
1534         sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1535         error = sbuf_finish(sb);
1536         sbuf_delete(sb);
1537
1538         return (error);
1539 }
1540
1541 static void
1542 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1543     size_t *sizep)
1544 {
1545         struct proc *p;
1546         struct thread *thr;
1547         size_t size;
1548
1549         p = td->td_proc;
1550         size = 0;
1551
1552         size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1553
1554         /*
1555          * To have the debugger select the right thread (LWP) as the initial
1556          * thread, we dump the state of the thread passed to us in td first.
1557          * This is the thread that causes the core dump and thus likely to
1558          * be the right thread one wants to have selected in the debugger.
1559          */
1560         thr = td;
1561         while (thr != NULL) {
1562                 size += register_note(list, NT_PRSTATUS,
1563                     __elfN(note_prstatus), thr);
1564                 size += register_note(list, NT_FPREGSET,
1565                     __elfN(note_fpregset), thr);
1566                 size += register_note(list, NT_THRMISC,
1567                     __elfN(note_thrmisc), thr);
1568                 size += register_note(list, -1,
1569                     __elfN(note_threadmd), thr);
1570
1571                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1572                     TAILQ_NEXT(thr, td_plist);
1573                 if (thr == td)
1574                         thr = TAILQ_NEXT(thr, td_plist);
1575         }
1576
1577         size += register_note(list, NT_PROCSTAT_PROC,
1578             __elfN(note_procstat_proc), p);
1579         size += register_note(list, NT_PROCSTAT_FILES,
1580             note_procstat_files, p);
1581         size += register_note(list, NT_PROCSTAT_VMMAP,
1582             note_procstat_vmmap, p);
1583         size += register_note(list, NT_PROCSTAT_GROUPS,
1584             note_procstat_groups, p);
1585         size += register_note(list, NT_PROCSTAT_UMASK,
1586             note_procstat_umask, p);
1587         size += register_note(list, NT_PROCSTAT_RLIMIT,
1588             note_procstat_rlimit, p);
1589         size += register_note(list, NT_PROCSTAT_OSREL,
1590             note_procstat_osrel, p);
1591         size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1592             __elfN(note_procstat_psstrings), p);
1593         size += register_note(list, NT_PROCSTAT_AUXV,
1594             __elfN(note_procstat_auxv), p);
1595
1596         *sizep = size;
1597 }
1598
1599 static void
1600 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1601     size_t notesz)
1602 {
1603         Elf_Ehdr *ehdr;
1604         Elf_Phdr *phdr;
1605         struct phdr_closure phc;
1606
1607         ehdr = (Elf_Ehdr *)hdr;
1608         phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1609
1610         ehdr->e_ident[EI_MAG0] = ELFMAG0;
1611         ehdr->e_ident[EI_MAG1] = ELFMAG1;
1612         ehdr->e_ident[EI_MAG2] = ELFMAG2;
1613         ehdr->e_ident[EI_MAG3] = ELFMAG3;
1614         ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1615         ehdr->e_ident[EI_DATA] = ELF_DATA;
1616         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1617         ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1618         ehdr->e_ident[EI_ABIVERSION] = 0;
1619         ehdr->e_ident[EI_PAD] = 0;
1620         ehdr->e_type = ET_CORE;
1621         ehdr->e_machine = td->td_proc->p_elf_machine;
1622         ehdr->e_version = EV_CURRENT;
1623         ehdr->e_entry = 0;
1624         ehdr->e_phoff = sizeof(Elf_Ehdr);
1625         ehdr->e_flags = td->td_proc->p_elf_flags;
1626         ehdr->e_ehsize = sizeof(Elf_Ehdr);
1627         ehdr->e_phentsize = sizeof(Elf_Phdr);
1628         ehdr->e_phnum = numsegs + 1;
1629         ehdr->e_shentsize = sizeof(Elf_Shdr);
1630         ehdr->e_shnum = 0;
1631         ehdr->e_shstrndx = SHN_UNDEF;
1632
1633         /*
1634          * Fill in the program header entries.
1635          */
1636
1637         /* The note segement. */
1638         phdr->p_type = PT_NOTE;
1639         phdr->p_offset = hdrsize;
1640         phdr->p_vaddr = 0;
1641         phdr->p_paddr = 0;
1642         phdr->p_filesz = notesz;
1643         phdr->p_memsz = 0;
1644         phdr->p_flags = PF_R;
1645         phdr->p_align = ELF_NOTE_ROUNDSIZE;
1646         phdr++;
1647
1648         /* All the writable segments from the program. */
1649         phc.phdr = phdr;
1650         phc.offset = round_page(hdrsize + notesz);
1651         each_writable_segment(td, cb_put_phdr, &phc);
1652 }
1653
1654 static size_t
1655 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1656 {
1657         struct note_info *ninfo;
1658         size_t size, notesize;
1659
1660         size = 0;
1661         out(arg, NULL, &size);
1662         ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1663         ninfo->type = type;
1664         ninfo->outfunc = out;
1665         ninfo->outarg = arg;
1666         ninfo->outsize = size;
1667         TAILQ_INSERT_TAIL(list, ninfo, link);
1668
1669         if (type == -1)
1670                 return (size);
1671
1672         notesize = sizeof(Elf_Note) +           /* note header */
1673             roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1674                                                 /* note name */
1675             roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1676
1677         return (notesize);
1678 }
1679
1680 static size_t
1681 append_note_data(const void *src, void *dst, size_t len)
1682 {
1683         size_t padded_len;
1684
1685         padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1686         if (dst != NULL) {
1687                 bcopy(src, dst, len);
1688                 bzero((char *)dst + len, padded_len - len);
1689         }
1690         return (padded_len);
1691 }
1692
1693 size_t
1694 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1695 {
1696         Elf_Note *note;
1697         char *buf;
1698         size_t notesize;
1699
1700         buf = dst;
1701         if (buf != NULL) {
1702                 note = (Elf_Note *)buf;
1703                 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1704                 note->n_descsz = size;
1705                 note->n_type = type;
1706                 buf += sizeof(*note);
1707                 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1708                     sizeof(FREEBSD_ABI_VENDOR));
1709                 append_note_data(src, buf, size);
1710                 if (descp != NULL)
1711                         *descp = buf;
1712         }
1713
1714         notesize = sizeof(Elf_Note) +           /* note header */
1715             roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1716                                                 /* note name */
1717             roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1718
1719         return (notesize);
1720 }
1721
1722 static void
1723 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1724 {
1725         Elf_Note note;
1726         ssize_t old_len, sect_len;
1727         size_t new_len, descsz, i;
1728
1729         if (ninfo->type == -1) {
1730                 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1731                 return;
1732         }
1733
1734         note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1735         note.n_descsz = ninfo->outsize;
1736         note.n_type = ninfo->type;
1737
1738         sbuf_bcat(sb, &note, sizeof(note));
1739         sbuf_start_section(sb, &old_len);
1740         sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1741         sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1742         if (note.n_descsz == 0)
1743                 return;
1744         sbuf_start_section(sb, &old_len);
1745         ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1746         sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1747         if (sect_len < 0)
1748                 return;
1749
1750         new_len = (size_t)sect_len;
1751         descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1752         if (new_len < descsz) {
1753                 /*
1754                  * It is expected that individual note emitters will correctly
1755                  * predict their expected output size and fill up to that size
1756                  * themselves, padding in a format-specific way if needed.
1757                  * However, in case they don't, just do it here with zeros.
1758                  */
1759                 for (i = 0; i < descsz - new_len; i++)
1760                         sbuf_putc(sb, 0);
1761         } else if (new_len > descsz) {
1762                 /*
1763                  * We can't always truncate sb -- we may have drained some
1764                  * of it already.
1765                  */
1766                 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1767                     "read it (%zu > %zu).  Since it is longer than "
1768                     "expected, this coredump's notes are corrupt.  THIS "
1769                     "IS A BUG in the note_procstat routine for type %u.\n",
1770                     __func__, (unsigned)note.n_type, new_len, descsz,
1771                     (unsigned)note.n_type));
1772         }
1773 }
1774
1775 /*
1776  * Miscellaneous note out functions.
1777  */
1778
1779 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1780 #include <compat/freebsd32/freebsd32.h>
1781
1782 typedef struct prstatus32 elf_prstatus_t;
1783 typedef struct prpsinfo32 elf_prpsinfo_t;
1784 typedef struct fpreg32 elf_prfpregset_t;
1785 typedef struct fpreg32 elf_fpregset_t;
1786 typedef struct reg32 elf_gregset_t;
1787 typedef struct thrmisc32 elf_thrmisc_t;
1788 #define ELF_KERN_PROC_MASK      KERN_PROC_MASK32
1789 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1790 typedef uint32_t elf_ps_strings_t;
1791 #else
1792 typedef prstatus_t elf_prstatus_t;
1793 typedef prpsinfo_t elf_prpsinfo_t;
1794 typedef prfpregset_t elf_prfpregset_t;
1795 typedef prfpregset_t elf_fpregset_t;
1796 typedef gregset_t elf_gregset_t;
1797 typedef thrmisc_t elf_thrmisc_t;
1798 #define ELF_KERN_PROC_MASK      0
1799 typedef struct kinfo_proc elf_kinfo_proc_t;
1800 typedef vm_offset_t elf_ps_strings_t;
1801 #endif
1802
1803 static void
1804 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1805 {
1806         struct sbuf sbarg;
1807         size_t len;
1808         char *cp, *end;
1809         struct proc *p;
1810         elf_prpsinfo_t *psinfo;
1811         int error;
1812
1813         p = (struct proc *)arg;
1814         if (sb != NULL) {
1815                 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1816                 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1817                 psinfo->pr_version = PRPSINFO_VERSION;
1818                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1819                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1820                 PROC_LOCK(p);
1821                 if (p->p_args != NULL) {
1822                         len = sizeof(psinfo->pr_psargs) - 1;
1823                         if (len > p->p_args->ar_length)
1824                                 len = p->p_args->ar_length;
1825                         memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1826                         PROC_UNLOCK(p);
1827                         error = 0;
1828                 } else {
1829                         _PHOLD(p);
1830                         PROC_UNLOCK(p);
1831                         sbuf_new(&sbarg, psinfo->pr_psargs,
1832                             sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1833                         error = proc_getargv(curthread, p, &sbarg);
1834                         PRELE(p);
1835                         if (sbuf_finish(&sbarg) == 0)
1836                                 len = sbuf_len(&sbarg) - 1;
1837                         else
1838                                 len = sizeof(psinfo->pr_psargs) - 1;
1839                         sbuf_delete(&sbarg);
1840                 }
1841                 if (error || len == 0)
1842                         strlcpy(psinfo->pr_psargs, p->p_comm,
1843                             sizeof(psinfo->pr_psargs));
1844                 else {
1845                         KASSERT(len < sizeof(psinfo->pr_psargs),
1846                             ("len is too long: %zu vs %zu", len,
1847                             sizeof(psinfo->pr_psargs)));
1848                         cp = psinfo->pr_psargs;
1849                         end = cp + len - 1;
1850                         for (;;) {
1851                                 cp = memchr(cp, '\0', end - cp);
1852                                 if (cp == NULL)
1853                                         break;
1854                                 *cp = ' ';
1855                         }
1856                 }
1857                 psinfo->pr_pid = p->p_pid;
1858                 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1859                 free(psinfo, M_TEMP);
1860         }
1861         *sizep = sizeof(*psinfo);
1862 }
1863
1864 static void
1865 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1866 {
1867         struct thread *td;
1868         elf_prstatus_t *status;
1869
1870         td = (struct thread *)arg;
1871         if (sb != NULL) {
1872                 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1873                 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1874                 status->pr_version = PRSTATUS_VERSION;
1875                 status->pr_statussz = sizeof(elf_prstatus_t);
1876                 status->pr_gregsetsz = sizeof(elf_gregset_t);
1877                 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1878                 status->pr_osreldate = osreldate;
1879                 status->pr_cursig = td->td_proc->p_sig;
1880                 status->pr_pid = td->td_tid;
1881 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1882                 fill_regs32(td, &status->pr_reg);
1883 #else
1884                 fill_regs(td, &status->pr_reg);
1885 #endif
1886                 sbuf_bcat(sb, status, sizeof(*status));
1887                 free(status, M_TEMP);
1888         }
1889         *sizep = sizeof(*status);
1890 }
1891
1892 static void
1893 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1894 {
1895         struct thread *td;
1896         elf_prfpregset_t *fpregset;
1897
1898         td = (struct thread *)arg;
1899         if (sb != NULL) {
1900                 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1901                 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1902 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1903                 fill_fpregs32(td, fpregset);
1904 #else
1905                 fill_fpregs(td, fpregset);
1906 #endif
1907                 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1908                 free(fpregset, M_TEMP);
1909         }
1910         *sizep = sizeof(*fpregset);
1911 }
1912
1913 static void
1914 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1915 {
1916         struct thread *td;
1917         elf_thrmisc_t thrmisc;
1918
1919         td = (struct thread *)arg;
1920         if (sb != NULL) {
1921                 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1922                 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1923                 strcpy(thrmisc.pr_tname, td->td_name);
1924                 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1925         }
1926         *sizep = sizeof(thrmisc);
1927 }
1928
1929 /*
1930  * Allow for MD specific notes, as well as any MD
1931  * specific preparations for writing MI notes.
1932  */
1933 static void
1934 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1935 {
1936         struct thread *td;
1937         void *buf;
1938         size_t size;
1939
1940         td = (struct thread *)arg;
1941         size = *sizep;
1942         if (size != 0 && sb != NULL)
1943                 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1944         else
1945                 buf = NULL;
1946         size = 0;
1947         __elfN(dump_thread)(td, buf, &size);
1948         KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1949         if (size != 0 && sb != NULL)
1950                 sbuf_bcat(sb, buf, size);
1951         free(buf, M_TEMP);
1952         *sizep = size;
1953 }
1954
1955 #ifdef KINFO_PROC_SIZE
1956 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1957 #endif
1958
1959 static void
1960 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1961 {
1962         struct proc *p;
1963         size_t size;
1964         int structsize;
1965
1966         p = (struct proc *)arg;
1967         size = sizeof(structsize) + p->p_numthreads *
1968             sizeof(elf_kinfo_proc_t);
1969
1970         if (sb != NULL) {
1971                 KASSERT(*sizep == size, ("invalid size"));
1972                 structsize = sizeof(elf_kinfo_proc_t);
1973                 sbuf_bcat(sb, &structsize, sizeof(structsize));
1974                 PROC_LOCK(p);
1975                 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1976         }
1977         *sizep = size;
1978 }
1979
1980 #ifdef KINFO_FILE_SIZE
1981 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1982 #endif
1983
1984 static void
1985 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1986 {
1987         struct proc *p;
1988         size_t size, sect_sz, i;
1989         ssize_t start_len, sect_len;
1990         int structsize, filedesc_flags;
1991
1992         if (coredump_pack_fileinfo)
1993                 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1994         else
1995                 filedesc_flags = 0;
1996
1997         p = (struct proc *)arg;
1998         structsize = sizeof(struct kinfo_file);
1999         if (sb == NULL) {
2000                 size = 0;
2001                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2002                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2003                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2004                 PROC_LOCK(p);
2005                 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2006                 sbuf_finish(sb);
2007                 sbuf_delete(sb);
2008                 *sizep = size;
2009         } else {
2010                 sbuf_start_section(sb, &start_len);
2011
2012                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2013                 PROC_LOCK(p);
2014                 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2015                     filedesc_flags);
2016
2017                 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2018                 if (sect_len < 0)
2019                         return;
2020                 sect_sz = sect_len;
2021
2022                 KASSERT(sect_sz <= *sizep,
2023                     ("kern_proc_filedesc_out did not respect maxlen; "
2024                      "requested %zu, got %zu", *sizep - sizeof(structsize),
2025                      sect_sz - sizeof(structsize)));
2026
2027                 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2028                         sbuf_putc(sb, 0);
2029         }
2030 }
2031
2032 #ifdef KINFO_VMENTRY_SIZE
2033 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2034 #endif
2035
2036 static void
2037 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2038 {
2039         struct proc *p;
2040         size_t size;
2041         int structsize, vmmap_flags;
2042
2043         if (coredump_pack_vmmapinfo)
2044                 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2045         else
2046                 vmmap_flags = 0;
2047
2048         p = (struct proc *)arg;
2049         structsize = sizeof(struct kinfo_vmentry);
2050         if (sb == NULL) {
2051                 size = 0;
2052                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2053                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2054                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2055                 PROC_LOCK(p);
2056                 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2057                 sbuf_finish(sb);
2058                 sbuf_delete(sb);
2059                 *sizep = size;
2060         } else {
2061                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2062                 PROC_LOCK(p);
2063                 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2064                     vmmap_flags);
2065         }
2066 }
2067
2068 static void
2069 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2070 {
2071         struct proc *p;
2072         size_t size;
2073         int structsize;
2074
2075         p = (struct proc *)arg;
2076         size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2077         if (sb != NULL) {
2078                 KASSERT(*sizep == size, ("invalid size"));
2079                 structsize = sizeof(gid_t);
2080                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2081                 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2082                     sizeof(gid_t));
2083         }
2084         *sizep = size;
2085 }
2086
2087 static void
2088 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2089 {
2090         struct proc *p;
2091         size_t size;
2092         int structsize;
2093
2094         p = (struct proc *)arg;
2095         size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2096         if (sb != NULL) {
2097                 KASSERT(*sizep == size, ("invalid size"));
2098                 structsize = sizeof(p->p_fd->fd_cmask);
2099                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2100                 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2101         }
2102         *sizep = size;
2103 }
2104
2105 static void
2106 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2107 {
2108         struct proc *p;
2109         struct rlimit rlim[RLIM_NLIMITS];
2110         size_t size;
2111         int structsize, i;
2112
2113         p = (struct proc *)arg;
2114         size = sizeof(structsize) + sizeof(rlim);
2115         if (sb != NULL) {
2116                 KASSERT(*sizep == size, ("invalid size"));
2117                 structsize = sizeof(rlim);
2118                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2119                 PROC_LOCK(p);
2120                 for (i = 0; i < RLIM_NLIMITS; i++)
2121                         lim_rlimit(p, i, &rlim[i]);
2122                 PROC_UNLOCK(p);
2123                 sbuf_bcat(sb, rlim, sizeof(rlim));
2124         }
2125         *sizep = size;
2126 }
2127
2128 static void
2129 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2130 {
2131         struct proc *p;
2132         size_t size;
2133         int structsize;
2134
2135         p = (struct proc *)arg;
2136         size = sizeof(structsize) + sizeof(p->p_osrel);
2137         if (sb != NULL) {
2138                 KASSERT(*sizep == size, ("invalid size"));
2139                 structsize = sizeof(p->p_osrel);
2140                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2141                 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2142         }
2143         *sizep = size;
2144 }
2145
2146 static void
2147 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2148 {
2149         struct proc *p;
2150         elf_ps_strings_t ps_strings;
2151         size_t size;
2152         int structsize;
2153
2154         p = (struct proc *)arg;
2155         size = sizeof(structsize) + sizeof(ps_strings);
2156         if (sb != NULL) {
2157                 KASSERT(*sizep == size, ("invalid size"));
2158                 structsize = sizeof(ps_strings);
2159 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2160                 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2161 #else
2162                 ps_strings = p->p_sysent->sv_psstrings;
2163 #endif
2164                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2165                 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2166         }
2167         *sizep = size;
2168 }
2169
2170 static void
2171 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2172 {
2173         struct proc *p;
2174         size_t size;
2175         int structsize;
2176
2177         p = (struct proc *)arg;
2178         if (sb == NULL) {
2179                 size = 0;
2180                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2181                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2182                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2183                 PHOLD(p);
2184                 proc_getauxv(curthread, p, sb);
2185                 PRELE(p);
2186                 sbuf_finish(sb);
2187                 sbuf_delete(sb);
2188                 *sizep = size;
2189         } else {
2190                 structsize = sizeof(Elf_Auxinfo);
2191                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2192                 PHOLD(p);
2193                 proc_getauxv(curthread, p, sb);
2194                 PRELE(p);
2195         }
2196 }
2197
2198 static boolean_t
2199 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2200     int32_t *osrel, const Elf_Phdr *pnote)
2201 {
2202         const Elf_Note *note, *note0, *note_end;
2203         const char *note_name;
2204         char *buf;
2205         int i, error;
2206         boolean_t res;
2207
2208         /* We need some limit, might as well use PAGE_SIZE. */
2209         if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2210                 return (FALSE);
2211         ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2212         if (pnote->p_offset > PAGE_SIZE ||
2213             pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2214                 VOP_UNLOCK(imgp->vp, 0);
2215                 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2216                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2217                 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2218                     pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2219                     curthread->td_ucred, NOCRED, NULL, curthread);
2220                 if (error != 0) {
2221                         uprintf("i/o error PT_NOTE\n");
2222                         res = FALSE;
2223                         goto ret;
2224                 }
2225                 note = note0 = (const Elf_Note *)buf;
2226                 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2227         } else {
2228                 note = note0 = (const Elf_Note *)(imgp->image_header +
2229                     pnote->p_offset);
2230                 note_end = (const Elf_Note *)(imgp->image_header +
2231                     pnote->p_offset + pnote->p_filesz);
2232                 buf = NULL;
2233         }
2234         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2235                 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2236                     (const char *)note < sizeof(Elf_Note)) {
2237                         res = FALSE;
2238                         goto ret;
2239                 }
2240                 if (note->n_namesz != checknote->hdr.n_namesz ||
2241                     note->n_descsz != checknote->hdr.n_descsz ||
2242                     note->n_type != checknote->hdr.n_type)
2243                         goto nextnote;
2244                 note_name = (const char *)(note + 1);
2245                 if (note_name + checknote->hdr.n_namesz >=
2246                     (const char *)note_end || strncmp(checknote->vendor,
2247                     note_name, checknote->hdr.n_namesz) != 0)
2248                         goto nextnote;
2249
2250                 /*
2251                  * Fetch the osreldate for binary
2252                  * from the ELF OSABI-note if necessary.
2253                  */
2254                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2255                     checknote->trans_osrel != NULL) {
2256                         res = checknote->trans_osrel(note, osrel);
2257                         goto ret;
2258                 }
2259                 res = TRUE;
2260                 goto ret;
2261 nextnote:
2262                 note = (const Elf_Note *)((const char *)(note + 1) +
2263                     roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2264                     roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2265         }
2266         res = FALSE;
2267 ret:
2268         free(buf, M_TEMP);
2269         return (res);
2270 }
2271
2272 /*
2273  * Try to find the appropriate ABI-note section for checknote,
2274  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2275  * first page of the image is searched, the same as for headers.
2276  */
2277 static boolean_t
2278 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2279     int32_t *osrel)
2280 {
2281         const Elf_Phdr *phdr;
2282         const Elf_Ehdr *hdr;
2283         int i;
2284
2285         hdr = (const Elf_Ehdr *)imgp->image_header;
2286         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2287
2288         for (i = 0; i < hdr->e_phnum; i++) {
2289                 if (phdr[i].p_type == PT_NOTE &&
2290                     __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2291                         return (TRUE);
2292         }
2293         return (FALSE);
2294
2295 }
2296
2297 /*
2298  * Tell kern_execve.c about it, with a little help from the linker.
2299  */
2300 static struct execsw __elfN(execsw) = {
2301         __CONCAT(exec_, __elfN(imgact)),
2302         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2303 };
2304 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2305
2306 #ifdef COMPRESS_USER_CORES
2307 /*
2308  * Compress and write out a core segment for a user process.
2309  *
2310  * 'inbuf' is the starting address of a VM segment in the process' address
2311  * space that is to be compressed and written out to the core file.  'dest_buf'
2312  * is a buffer in the kernel's address space.  The segment is copied from 
2313  * 'inbuf' to 'dest_buf' first before being processed by the compression
2314  * routine gzwrite().  This copying is necessary because the content of the VM
2315  * segment may change between the compression pass and the crc-computation pass
2316  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
2317  *
2318  * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2319  */
2320 static int
2321 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2322     struct thread *td)
2323 {
2324         int len_compressed;
2325         int error = 0;
2326         unsigned int chunk_len;
2327
2328         while (len) {
2329                 if (inbuf != NULL) {
2330                         chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2331
2332                         /*
2333                          * We can get EFAULT error here.  In that case zero out
2334                          * the current chunk of the segment.
2335                          */
2336                         error = copyin(inbuf, dest_buf, chunk_len);
2337                         if (error != 0) {
2338                                 bzero(dest_buf, chunk_len);
2339                                 error = 0;
2340                         }
2341                         inbuf += chunk_len;
2342                 } else {
2343                         chunk_len = len;
2344                 }
2345                 len_compressed = gzwrite(file, dest_buf, chunk_len);
2346
2347                 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2348
2349                 if ((unsigned int)len_compressed != chunk_len) {
2350                         log(LOG_WARNING,
2351                             "compress_core: length mismatch (0x%x returned, "
2352                             "0x%x expected)\n", len_compressed, chunk_len);
2353                         EVENTHANDLER_INVOKE(app_coredump_error, td,
2354                             "compress_core: length mismatch %x -> %x",
2355                             chunk_len, len_compressed);
2356                         error = EFAULT;
2357                         break;
2358                 }
2359                 len -= chunk_len;
2360                 maybe_yield();
2361         }
2362
2363         return (error);
2364 }
2365 #endif /* COMPRESS_USER_CORES */
2366
2367 static vm_prot_t
2368 __elfN(trans_prot)(Elf_Word flags)
2369 {
2370         vm_prot_t prot;
2371
2372         prot = 0;
2373         if (flags & PF_X)
2374                 prot |= VM_PROT_EXECUTE;
2375         if (flags & PF_W)
2376                 prot |= VM_PROT_WRITE;
2377         if (flags & PF_R)
2378                 prot |= VM_PROT_READ;
2379 #if __ELF_WORD_SIZE == 32
2380 #if defined(__amd64__) || defined(__ia64__)
2381         if (i386_read_exec && (flags & PF_R))
2382                 prot |= VM_PROT_EXECUTE;
2383 #endif
2384 #endif
2385         return (prot);
2386 }
2387
2388 static Elf_Word
2389 __elfN(untrans_prot)(vm_prot_t prot)
2390 {
2391         Elf_Word flags;
2392
2393         flags = 0;
2394         if (prot & VM_PROT_EXECUTE)
2395                 flags |= PF_X;
2396         if (prot & VM_PROT_READ)
2397                 flags |= PF_R;
2398         if (prot & VM_PROT_WRITE)
2399                 flags |= PF_W;
2400         return (flags);
2401 }