]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/kern/imgact_elf.c
MFC r315960: dtrace sched:::preempt should fire only when there is preemption
[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
1045  ret:
1046         free(interp_buf, M_TEMP);
1047         return (error);
1048 }
1049
1050 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1051
1052 int
1053 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1054 {
1055         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1056         Elf_Addr *base;
1057         Elf_Addr *pos;
1058
1059         base = (Elf_Addr *)*stack_base;
1060         pos = base + (imgp->args->argc + imgp->args->envc + 2);
1061
1062         if (args->execfd != -1)
1063                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1064         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1065         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1066         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1067         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1068         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1069         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1070         AUXARGS_ENTRY(pos, AT_BASE, args->base);
1071         if (imgp->execpathp != 0)
1072                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1073         AUXARGS_ENTRY(pos, AT_OSRELDATE,
1074             imgp->proc->p_ucred->cr_prison->pr_osreldate);
1075         if (imgp->canary != 0) {
1076                 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1077                 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1078         }
1079         AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1080         if (imgp->pagesizes != 0) {
1081                 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1082                 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1083         }
1084         if (imgp->sysent->sv_timekeep_base != 0) {
1085                 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1086                     imgp->sysent->sv_timekeep_base);
1087         }
1088         AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1089             != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1090             imgp->sysent->sv_stackprot);
1091         AUXARGS_ENTRY(pos, AT_NULL, 0);
1092
1093         free(imgp->auxargs, M_TEMP);
1094         imgp->auxargs = NULL;
1095
1096         base--;
1097         suword(base, (long)imgp->args->argc);
1098         *stack_base = (register_t *)base;
1099         return (0);
1100 }
1101
1102 /*
1103  * Code for generating ELF core dumps.
1104  */
1105
1106 typedef void (*segment_callback)(vm_map_entry_t, void *);
1107
1108 /* Closure for cb_put_phdr(). */
1109 struct phdr_closure {
1110         Elf_Phdr *phdr;         /* Program header to fill in */
1111         Elf_Off offset;         /* Offset of segment in core file */
1112 };
1113
1114 /* Closure for cb_size_segment(). */
1115 struct sseg_closure {
1116         int count;              /* Count of writable segments. */
1117         size_t size;            /* Total size of all writable segments. */
1118 };
1119
1120 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1121
1122 struct note_info {
1123         int             type;           /* Note type. */
1124         outfunc_t       outfunc;        /* Output function. */
1125         void            *outarg;        /* Argument for the output function. */
1126         size_t          outsize;        /* Output size. */
1127         TAILQ_ENTRY(note_info) link;    /* Link to the next note info. */
1128 };
1129
1130 TAILQ_HEAD(note_info_list, note_info);
1131
1132 static void cb_put_phdr(vm_map_entry_t, void *);
1133 static void cb_size_segment(vm_map_entry_t, void *);
1134 static void each_writable_segment(struct thread *, segment_callback, void *);
1135 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1136     int, void *, size_t, struct note_info_list *, size_t, gzFile);
1137 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1138     size_t *);
1139 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1140 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1141 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1142 static int sbuf_drain_core_output(void *, const char *, int);
1143 static int sbuf_drain_count(void *arg, const char *data, int len);
1144
1145 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1146 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1147 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1148 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1150 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1151 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1152 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1153 static void note_procstat_files(void *, struct sbuf *, size_t *);
1154 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1155 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1156 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1157 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1158 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1159
1160 #ifdef COMPRESS_USER_CORES
1161 extern int compress_user_cores;
1162 extern int compress_user_cores_gzlevel;
1163 #endif
1164
1165 static int
1166 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1167     struct ucred *active_cred, struct ucred *file_cred,
1168     struct thread *td, char *core_buf, gzFile gzfile) {
1169
1170         int error;
1171         if (gzfile) {
1172 #ifdef COMPRESS_USER_CORES
1173                 error = compress_core(gzfile, base, core_buf, len, td);
1174 #else
1175                 panic("shouldn't be here");
1176 #endif
1177         } else {
1178                 /*
1179                  * EFAULT is a non-fatal error that we can get, for example,
1180                  * if the segment is backed by a file but extends beyond its
1181                  * end.
1182                  */
1183                 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1184                     UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1185                     NULL, td);
1186                 if (error == EFAULT) {
1187                         log(LOG_WARNING, "Failed to fully fault in a core file "
1188                             "segment at VA %p with size 0x%zx to be written at "
1189                             "offset 0x%jx for process %s\n", base, len, offset,
1190                             curproc->p_comm);
1191
1192                         /*
1193                          * Write a "real" zero byte at the end of the target
1194                          * region in the case this is the last segment.
1195                          * The intermediate space will be implicitly
1196                          * zero-filled.
1197                          */
1198                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
1199                             __DECONST(void *, zero_region), 1, offset + len - 1,
1200                             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, active_cred,
1201                             file_cred, NULL, td);
1202                 }
1203         }
1204         return (error);
1205 }
1206
1207 /* Coredump output parameters for sbuf drain routine. */
1208 struct sbuf_drain_core_params {
1209         off_t           offset;
1210         struct ucred    *active_cred;
1211         struct ucred    *file_cred;
1212         struct thread   *td;
1213         struct vnode    *vp;
1214 #ifdef COMPRESS_USER_CORES
1215         gzFile          gzfile;
1216 #endif
1217 };
1218
1219 /*
1220  * Drain into a core file.
1221  */
1222 static int
1223 sbuf_drain_core_output(void *arg, const char *data, int len)
1224 {
1225         struct sbuf_drain_core_params *p;
1226         int error, locked;
1227
1228         p = (struct sbuf_drain_core_params *)arg;
1229
1230         /*
1231          * Some kern_proc out routines that print to this sbuf may
1232          * call us with the process lock held. Draining with the
1233          * non-sleepable lock held is unsafe. The lock is needed for
1234          * those routines when dumping a live process. In our case we
1235          * can safely release the lock before draining and acquire
1236          * again after.
1237          */
1238         locked = PROC_LOCKED(p->td->td_proc);
1239         if (locked)
1240                 PROC_UNLOCK(p->td->td_proc);
1241 #ifdef COMPRESS_USER_CORES
1242         if (p->gzfile != Z_NULL)
1243                 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1244                     len, p->td);
1245         else
1246 #endif
1247                 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1248                     __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1249                     IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1250                     p->td);
1251         if (locked)
1252                 PROC_LOCK(p->td->td_proc);
1253         if (error != 0)
1254                 return (-error);
1255         p->offset += len;
1256         return (len);
1257 }
1258
1259 /*
1260  * Drain into a counter.
1261  */
1262 static int
1263 sbuf_drain_count(void *arg, const char *data __unused, int len)
1264 {
1265         size_t *sizep;
1266
1267         sizep = (size_t *)arg;
1268         *sizep += len;
1269         return (len);
1270 }
1271
1272 int
1273 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1274 {
1275         struct ucred *cred = td->td_ucred;
1276         int error = 0;
1277         struct sseg_closure seginfo;
1278         struct note_info_list notelst;
1279         struct note_info *ninfo;
1280         void *hdr;
1281         size_t hdrsize, notesz, coresize;
1282
1283         gzFile gzfile = Z_NULL;
1284         char *core_buf = NULL;
1285 #ifdef COMPRESS_USER_CORES
1286         char gzopen_flags[8];
1287         char *p;
1288         int doing_compress = flags & IMGACT_CORE_COMPRESS;
1289 #endif
1290
1291         hdr = NULL;
1292         TAILQ_INIT(&notelst);
1293
1294 #ifdef COMPRESS_USER_CORES
1295         if (doing_compress) {
1296                 p = gzopen_flags;
1297                 *p++ = 'w';
1298                 if (compress_user_cores_gzlevel >= 0 &&
1299                     compress_user_cores_gzlevel <= 9)
1300                         *p++ = '0' + compress_user_cores_gzlevel;
1301                 *p = 0;
1302                 gzfile = gz_open("", gzopen_flags, vp);
1303                 if (gzfile == Z_NULL) {
1304                         error = EFAULT;
1305                         goto done;
1306                 }
1307                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1308                 if (!core_buf) {
1309                         error = ENOMEM;
1310                         goto done;
1311                 }
1312         }
1313 #endif
1314
1315         /* Size the program segments. */
1316         seginfo.count = 0;
1317         seginfo.size = 0;
1318         each_writable_segment(td, cb_size_segment, &seginfo);
1319
1320         /*
1321          * Collect info about the core file header area.
1322          */
1323         hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1324         __elfN(prepare_notes)(td, &notelst, &notesz);
1325         coresize = round_page(hdrsize + notesz) + seginfo.size;
1326
1327 #ifdef RACCT
1328         if (racct_enable) {
1329                 PROC_LOCK(td->td_proc);
1330                 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1331                 PROC_UNLOCK(td->td_proc);
1332                 if (error != 0) {
1333                         error = EFAULT;
1334                         goto done;
1335                 }
1336         }
1337 #endif
1338         if (coresize >= limit) {
1339                 error = EFAULT;
1340                 goto done;
1341         }
1342
1343         /*
1344          * Allocate memory for building the header, fill it up,
1345          * and write it out following the notes.
1346          */
1347         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1348         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1349             &notelst, notesz, gzfile);
1350
1351         /* Write the contents of all of the writable segments. */
1352         if (error == 0) {
1353                 Elf_Phdr *php;
1354                 off_t offset;
1355                 int i;
1356
1357                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1358                 offset = round_page(hdrsize + notesz);
1359                 for (i = 0; i < seginfo.count; i++) {
1360                         error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1361                             php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1362                         if (error != 0)
1363                                 break;
1364                         offset += php->p_filesz;
1365                         php++;
1366                 }
1367         }
1368         if (error) {
1369                 log(LOG_WARNING,
1370                     "Failed to write core file for process %s (error %d)\n",
1371                     curproc->p_comm, error);
1372         }
1373
1374 done:
1375 #ifdef COMPRESS_USER_CORES
1376         if (core_buf)
1377                 free(core_buf, M_TEMP);
1378         if (gzfile)
1379                 gzclose(gzfile);
1380 #endif
1381         while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1382                 TAILQ_REMOVE(&notelst, ninfo, link);
1383                 free(ninfo, M_TEMP);
1384         }
1385         if (hdr != NULL)
1386                 free(hdr, M_TEMP);
1387
1388         return (error);
1389 }
1390
1391 /*
1392  * A callback for each_writable_segment() to write out the segment's
1393  * program header entry.
1394  */
1395 static void
1396 cb_put_phdr(entry, closure)
1397         vm_map_entry_t entry;
1398         void *closure;
1399 {
1400         struct phdr_closure *phc = (struct phdr_closure *)closure;
1401         Elf_Phdr *phdr = phc->phdr;
1402
1403         phc->offset = round_page(phc->offset);
1404
1405         phdr->p_type = PT_LOAD;
1406         phdr->p_offset = phc->offset;
1407         phdr->p_vaddr = entry->start;
1408         phdr->p_paddr = 0;
1409         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1410         phdr->p_align = PAGE_SIZE;
1411         phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1412
1413         phc->offset += phdr->p_filesz;
1414         phc->phdr++;
1415 }
1416
1417 /*
1418  * A callback for each_writable_segment() to gather information about
1419  * the number of segments and their total size.
1420  */
1421 static void
1422 cb_size_segment(entry, closure)
1423         vm_map_entry_t entry;
1424         void *closure;
1425 {
1426         struct sseg_closure *ssc = (struct sseg_closure *)closure;
1427
1428         ssc->count++;
1429         ssc->size += entry->end - entry->start;
1430 }
1431
1432 /*
1433  * For each writable segment in the process's memory map, call the given
1434  * function with a pointer to the map entry and some arbitrary
1435  * caller-supplied data.
1436  */
1437 static void
1438 each_writable_segment(td, func, closure)
1439         struct thread *td;
1440         segment_callback func;
1441         void *closure;
1442 {
1443         struct proc *p = td->td_proc;
1444         vm_map_t map = &p->p_vmspace->vm_map;
1445         vm_map_entry_t entry;
1446         vm_object_t backing_object, object;
1447         boolean_t ignore_entry;
1448
1449         vm_map_lock_read(map);
1450         for (entry = map->header.next; entry != &map->header;
1451             entry = entry->next) {
1452                 /*
1453                  * Don't dump inaccessible mappings, deal with legacy
1454                  * coredump mode.
1455                  *
1456                  * Note that read-only segments related to the elf binary
1457                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1458                  * need to arbitrarily ignore such segments.
1459                  */
1460                 if (elf_legacy_coredump) {
1461                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1462                                 continue;
1463                 } else {
1464                         if ((entry->protection & VM_PROT_ALL) == 0)
1465                                 continue;
1466                 }
1467
1468                 /*
1469                  * Dont include memory segment in the coredump if
1470                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1471                  * madvise(2).  Do not dump submaps (i.e. parts of the
1472                  * kernel map).
1473                  */
1474                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1475                         continue;
1476
1477                 if ((object = entry->object.vm_object) == NULL)
1478                         continue;
1479
1480                 /* Ignore memory-mapped devices and such things. */
1481                 VM_OBJECT_RLOCK(object);
1482                 while ((backing_object = object->backing_object) != NULL) {
1483                         VM_OBJECT_RLOCK(backing_object);
1484                         VM_OBJECT_RUNLOCK(object);
1485                         object = backing_object;
1486                 }
1487                 ignore_entry = object->type != OBJT_DEFAULT &&
1488                     object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1489                     object->type != OBJT_PHYS;
1490                 VM_OBJECT_RUNLOCK(object);
1491                 if (ignore_entry)
1492                         continue;
1493
1494                 (*func)(entry, closure);
1495         }
1496         vm_map_unlock_read(map);
1497 }
1498
1499 /*
1500  * Write the core file header to the file, including padding up to
1501  * the page boundary.
1502  */
1503 static int
1504 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1505     int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1506     size_t notesz, gzFile gzfile)
1507 {
1508         struct sbuf_drain_core_params params;
1509         struct note_info *ninfo;
1510         struct sbuf *sb;
1511         int error;
1512
1513         /* Fill in the header. */
1514         bzero(hdr, hdrsize);
1515         __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1516
1517         params.offset = 0;
1518         params.active_cred = cred;
1519         params.file_cred = NOCRED;
1520         params.td = td;
1521         params.vp = vp;
1522 #ifdef COMPRESS_USER_CORES
1523         params.gzfile = gzfile;
1524 #endif
1525         sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1526         sbuf_set_drain(sb, sbuf_drain_core_output, &params);
1527         sbuf_start_section(sb, NULL);
1528         sbuf_bcat(sb, hdr, hdrsize);
1529         TAILQ_FOREACH(ninfo, notelst, link)
1530             __elfN(putnote)(ninfo, sb);
1531         /* Align up to a page boundary for the program segments. */
1532         sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1533         error = sbuf_finish(sb);
1534         sbuf_delete(sb);
1535
1536         return (error);
1537 }
1538
1539 static void
1540 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1541     size_t *sizep)
1542 {
1543         struct proc *p;
1544         struct thread *thr;
1545         size_t size;
1546
1547         p = td->td_proc;
1548         size = 0;
1549
1550         size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1551
1552         /*
1553          * To have the debugger select the right thread (LWP) as the initial
1554          * thread, we dump the state of the thread passed to us in td first.
1555          * This is the thread that causes the core dump and thus likely to
1556          * be the right thread one wants to have selected in the debugger.
1557          */
1558         thr = td;
1559         while (thr != NULL) {
1560                 size += register_note(list, NT_PRSTATUS,
1561                     __elfN(note_prstatus), thr);
1562                 size += register_note(list, NT_FPREGSET,
1563                     __elfN(note_fpregset), thr);
1564                 size += register_note(list, NT_THRMISC,
1565                     __elfN(note_thrmisc), thr);
1566                 size += register_note(list, -1,
1567                     __elfN(note_threadmd), thr);
1568
1569                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1570                     TAILQ_NEXT(thr, td_plist);
1571                 if (thr == td)
1572                         thr = TAILQ_NEXT(thr, td_plist);
1573         }
1574
1575         size += register_note(list, NT_PROCSTAT_PROC,
1576             __elfN(note_procstat_proc), p);
1577         size += register_note(list, NT_PROCSTAT_FILES,
1578             note_procstat_files, p);
1579         size += register_note(list, NT_PROCSTAT_VMMAP,
1580             note_procstat_vmmap, p);
1581         size += register_note(list, NT_PROCSTAT_GROUPS,
1582             note_procstat_groups, p);
1583         size += register_note(list, NT_PROCSTAT_UMASK,
1584             note_procstat_umask, p);
1585         size += register_note(list, NT_PROCSTAT_RLIMIT,
1586             note_procstat_rlimit, p);
1587         size += register_note(list, NT_PROCSTAT_OSREL,
1588             note_procstat_osrel, p);
1589         size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1590             __elfN(note_procstat_psstrings), p);
1591         size += register_note(list, NT_PROCSTAT_AUXV,
1592             __elfN(note_procstat_auxv), p);
1593
1594         *sizep = size;
1595 }
1596
1597 static void
1598 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1599     size_t notesz)
1600 {
1601         Elf_Ehdr *ehdr;
1602         Elf_Phdr *phdr;
1603         struct phdr_closure phc;
1604
1605         ehdr = (Elf_Ehdr *)hdr;
1606         phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1607
1608         ehdr->e_ident[EI_MAG0] = ELFMAG0;
1609         ehdr->e_ident[EI_MAG1] = ELFMAG1;
1610         ehdr->e_ident[EI_MAG2] = ELFMAG2;
1611         ehdr->e_ident[EI_MAG3] = ELFMAG3;
1612         ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1613         ehdr->e_ident[EI_DATA] = ELF_DATA;
1614         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1615         ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1616         ehdr->e_ident[EI_ABIVERSION] = 0;
1617         ehdr->e_ident[EI_PAD] = 0;
1618         ehdr->e_type = ET_CORE;
1619 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1620         ehdr->e_machine = ELF_ARCH32;
1621 #else
1622         ehdr->e_machine = ELF_ARCH;
1623 #endif
1624         ehdr->e_version = EV_CURRENT;
1625         ehdr->e_entry = 0;
1626         ehdr->e_phoff = sizeof(Elf_Ehdr);
1627         ehdr->e_flags = 0;
1628         ehdr->e_ehsize = sizeof(Elf_Ehdr);
1629         ehdr->e_phentsize = sizeof(Elf_Phdr);
1630         ehdr->e_phnum = numsegs + 1;
1631         ehdr->e_shentsize = sizeof(Elf_Shdr);
1632         ehdr->e_shnum = 0;
1633         ehdr->e_shstrndx = SHN_UNDEF;
1634
1635         /*
1636          * Fill in the program header entries.
1637          */
1638
1639         /* The note segement. */
1640         phdr->p_type = PT_NOTE;
1641         phdr->p_offset = hdrsize;
1642         phdr->p_vaddr = 0;
1643         phdr->p_paddr = 0;
1644         phdr->p_filesz = notesz;
1645         phdr->p_memsz = 0;
1646         phdr->p_flags = PF_R;
1647         phdr->p_align = ELF_NOTE_ROUNDSIZE;
1648         phdr++;
1649
1650         /* All the writable segments from the program. */
1651         phc.phdr = phdr;
1652         phc.offset = round_page(hdrsize + notesz);
1653         each_writable_segment(td, cb_put_phdr, &phc);
1654 }
1655
1656 static size_t
1657 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1658 {
1659         struct note_info *ninfo;
1660         size_t size, notesize;
1661
1662         size = 0;
1663         out(arg, NULL, &size);
1664         ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1665         ninfo->type = type;
1666         ninfo->outfunc = out;
1667         ninfo->outarg = arg;
1668         ninfo->outsize = size;
1669         TAILQ_INSERT_TAIL(list, ninfo, link);
1670
1671         if (type == -1)
1672                 return (size);
1673
1674         notesize = sizeof(Elf_Note) +           /* note header */
1675             roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1676                                                 /* note name */
1677             roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1678
1679         return (notesize);
1680 }
1681
1682 static size_t
1683 append_note_data(const void *src, void *dst, size_t len)
1684 {
1685         size_t padded_len;
1686
1687         padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1688         if (dst != NULL) {
1689                 bcopy(src, dst, len);
1690                 bzero((char *)dst + len, padded_len - len);
1691         }
1692         return (padded_len);
1693 }
1694
1695 size_t
1696 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1697 {
1698         Elf_Note *note;
1699         char *buf;
1700         size_t notesize;
1701
1702         buf = dst;
1703         if (buf != NULL) {
1704                 note = (Elf_Note *)buf;
1705                 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1706                 note->n_descsz = size;
1707                 note->n_type = type;
1708                 buf += sizeof(*note);
1709                 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1710                     sizeof(FREEBSD_ABI_VENDOR));
1711                 append_note_data(src, buf, size);
1712                 if (descp != NULL)
1713                         *descp = buf;
1714         }
1715
1716         notesize = sizeof(Elf_Note) +           /* note header */
1717             roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1718                                                 /* note name */
1719             roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1720
1721         return (notesize);
1722 }
1723
1724 static void
1725 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1726 {
1727         Elf_Note note;
1728         ssize_t old_len, sect_len;
1729         size_t new_len, descsz, i;
1730
1731         if (ninfo->type == -1) {
1732                 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1733                 return;
1734         }
1735
1736         note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1737         note.n_descsz = ninfo->outsize;
1738         note.n_type = ninfo->type;
1739
1740         sbuf_bcat(sb, &note, sizeof(note));
1741         sbuf_start_section(sb, &old_len);
1742         sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1743         sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1744         if (note.n_descsz == 0)
1745                 return;
1746         sbuf_start_section(sb, &old_len);
1747         ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1748         sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1749         if (sect_len < 0)
1750                 return;
1751
1752         new_len = (size_t)sect_len;
1753         descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1754         if (new_len < descsz) {
1755                 /*
1756                  * It is expected that individual note emitters will correctly
1757                  * predict their expected output size and fill up to that size
1758                  * themselves, padding in a format-specific way if needed.
1759                  * However, in case they don't, just do it here with zeros.
1760                  */
1761                 for (i = 0; i < descsz - new_len; i++)
1762                         sbuf_putc(sb, 0);
1763         } else if (new_len > descsz) {
1764                 /*
1765                  * We can't always truncate sb -- we may have drained some
1766                  * of it already.
1767                  */
1768                 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1769                     "read it (%zu > %zu).  Since it is longer than "
1770                     "expected, this coredump's notes are corrupt.  THIS "
1771                     "IS A BUG in the note_procstat routine for type %u.\n",
1772                     __func__, (unsigned)note.n_type, new_len, descsz,
1773                     (unsigned)note.n_type));
1774         }
1775 }
1776
1777 /*
1778  * Miscellaneous note out functions.
1779  */
1780
1781 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1782 #include <compat/freebsd32/freebsd32.h>
1783
1784 typedef struct prstatus32 elf_prstatus_t;
1785 typedef struct prpsinfo32 elf_prpsinfo_t;
1786 typedef struct fpreg32 elf_prfpregset_t;
1787 typedef struct fpreg32 elf_fpregset_t;
1788 typedef struct reg32 elf_gregset_t;
1789 typedef struct thrmisc32 elf_thrmisc_t;
1790 #define ELF_KERN_PROC_MASK      KERN_PROC_MASK32
1791 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1792 typedef uint32_t elf_ps_strings_t;
1793 #else
1794 typedef prstatus_t elf_prstatus_t;
1795 typedef prpsinfo_t elf_prpsinfo_t;
1796 typedef prfpregset_t elf_prfpregset_t;
1797 typedef prfpregset_t elf_fpregset_t;
1798 typedef gregset_t elf_gregset_t;
1799 typedef thrmisc_t elf_thrmisc_t;
1800 #define ELF_KERN_PROC_MASK      0
1801 typedef struct kinfo_proc elf_kinfo_proc_t;
1802 typedef vm_offset_t elf_ps_strings_t;
1803 #endif
1804
1805 static void
1806 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1807 {
1808         struct sbuf sbarg;
1809         size_t len;
1810         char *cp, *end;
1811         struct proc *p;
1812         elf_prpsinfo_t *psinfo;
1813         int error;
1814
1815         p = (struct proc *)arg;
1816         if (sb != NULL) {
1817                 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1818                 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1819                 psinfo->pr_version = PRPSINFO_VERSION;
1820                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1821                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1822                 PROC_LOCK(p);
1823                 if (p->p_args != NULL) {
1824                         len = sizeof(psinfo->pr_psargs) - 1;
1825                         if (len > p->p_args->ar_length)
1826                                 len = p->p_args->ar_length;
1827                         memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1828                         PROC_UNLOCK(p);
1829                         error = 0;
1830                 } else {
1831                         _PHOLD(p);
1832                         PROC_UNLOCK(p);
1833                         sbuf_new(&sbarg, psinfo->pr_psargs,
1834                             sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1835                         error = proc_getargv(curthread, p, &sbarg);
1836                         PRELE(p);
1837                         if (sbuf_finish(&sbarg) == 0)
1838                                 len = sbuf_len(&sbarg) - 1;
1839                         else
1840                                 len = sizeof(psinfo->pr_psargs) - 1;
1841                         sbuf_delete(&sbarg);
1842                 }
1843                 if (error || len == 0)
1844                         strlcpy(psinfo->pr_psargs, p->p_comm,
1845                             sizeof(psinfo->pr_psargs));
1846                 else {
1847                         KASSERT(len < sizeof(psinfo->pr_psargs),
1848                             ("len is too long: %zu vs %zu", len,
1849                             sizeof(psinfo->pr_psargs)));
1850                         cp = psinfo->pr_psargs;
1851                         end = cp + len - 1;
1852                         for (;;) {
1853                                 cp = memchr(cp, '\0', end - cp);
1854                                 if (cp == NULL)
1855                                         break;
1856                                 *cp = ' ';
1857                         }
1858                 }
1859                 psinfo->pr_pid = p->p_pid;
1860                 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1861                 free(psinfo, M_TEMP);
1862         }
1863         *sizep = sizeof(*psinfo);
1864 }
1865
1866 static void
1867 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1868 {
1869         struct thread *td;
1870         elf_prstatus_t *status;
1871
1872         td = (struct thread *)arg;
1873         if (sb != NULL) {
1874                 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1875                 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1876                 status->pr_version = PRSTATUS_VERSION;
1877                 status->pr_statussz = sizeof(elf_prstatus_t);
1878                 status->pr_gregsetsz = sizeof(elf_gregset_t);
1879                 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1880                 status->pr_osreldate = osreldate;
1881                 status->pr_cursig = td->td_proc->p_sig;
1882                 status->pr_pid = td->td_tid;
1883 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1884                 fill_regs32(td, &status->pr_reg);
1885 #else
1886                 fill_regs(td, &status->pr_reg);
1887 #endif
1888                 sbuf_bcat(sb, status, sizeof(*status));
1889                 free(status, M_TEMP);
1890         }
1891         *sizep = sizeof(*status);
1892 }
1893
1894 static void
1895 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1896 {
1897         struct thread *td;
1898         elf_prfpregset_t *fpregset;
1899
1900         td = (struct thread *)arg;
1901         if (sb != NULL) {
1902                 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1903                 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1904 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1905                 fill_fpregs32(td, fpregset);
1906 #else
1907                 fill_fpregs(td, fpregset);
1908 #endif
1909                 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1910                 free(fpregset, M_TEMP);
1911         }
1912         *sizep = sizeof(*fpregset);
1913 }
1914
1915 static void
1916 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1917 {
1918         struct thread *td;
1919         elf_thrmisc_t thrmisc;
1920
1921         td = (struct thread *)arg;
1922         if (sb != NULL) {
1923                 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1924                 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1925                 strcpy(thrmisc.pr_tname, td->td_name);
1926                 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1927         }
1928         *sizep = sizeof(thrmisc);
1929 }
1930
1931 /*
1932  * Allow for MD specific notes, as well as any MD
1933  * specific preparations for writing MI notes.
1934  */
1935 static void
1936 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1937 {
1938         struct thread *td;
1939         void *buf;
1940         size_t size;
1941
1942         td = (struct thread *)arg;
1943         size = *sizep;
1944         if (size != 0 && sb != NULL)
1945                 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1946         else
1947                 buf = NULL;
1948         size = 0;
1949         __elfN(dump_thread)(td, buf, &size);
1950         KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1951         if (size != 0 && sb != NULL)
1952                 sbuf_bcat(sb, buf, size);
1953         free(buf, M_TEMP);
1954         *sizep = size;
1955 }
1956
1957 #ifdef KINFO_PROC_SIZE
1958 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1959 #endif
1960
1961 static void
1962 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1963 {
1964         struct proc *p;
1965         size_t size;
1966         int structsize;
1967
1968         p = (struct proc *)arg;
1969         size = sizeof(structsize) + p->p_numthreads *
1970             sizeof(elf_kinfo_proc_t);
1971
1972         if (sb != NULL) {
1973                 KASSERT(*sizep == size, ("invalid size"));
1974                 structsize = sizeof(elf_kinfo_proc_t);
1975                 sbuf_bcat(sb, &structsize, sizeof(structsize));
1976                 PROC_LOCK(p);
1977                 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1978         }
1979         *sizep = size;
1980 }
1981
1982 #ifdef KINFO_FILE_SIZE
1983 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1984 #endif
1985
1986 static void
1987 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1988 {
1989         struct proc *p;
1990         size_t size, sect_sz, i;
1991         ssize_t start_len, sect_len;
1992         int structsize, filedesc_flags;
1993
1994         if (coredump_pack_fileinfo)
1995                 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1996         else
1997                 filedesc_flags = 0;
1998
1999         p = (struct proc *)arg;
2000         structsize = sizeof(struct kinfo_file);
2001         if (sb == NULL) {
2002                 size = 0;
2003                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2004                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2005                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2006                 PROC_LOCK(p);
2007                 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2008                 sbuf_finish(sb);
2009                 sbuf_delete(sb);
2010                 *sizep = size;
2011         } else {
2012                 sbuf_start_section(sb, &start_len);
2013
2014                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2015                 PROC_LOCK(p);
2016                 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2017                     filedesc_flags);
2018
2019                 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2020                 if (sect_len < 0)
2021                         return;
2022                 sect_sz = sect_len;
2023
2024                 KASSERT(sect_sz <= *sizep,
2025                     ("kern_proc_filedesc_out did not respect maxlen; "
2026                      "requested %zu, got %zu", *sizep - sizeof(structsize),
2027                      sect_sz - sizeof(structsize)));
2028
2029                 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2030                         sbuf_putc(sb, 0);
2031         }
2032 }
2033
2034 #ifdef KINFO_VMENTRY_SIZE
2035 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2036 #endif
2037
2038 static void
2039 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2040 {
2041         struct proc *p;
2042         size_t size;
2043         int structsize, vmmap_flags;
2044
2045         if (coredump_pack_vmmapinfo)
2046                 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2047         else
2048                 vmmap_flags = 0;
2049
2050         p = (struct proc *)arg;
2051         structsize = sizeof(struct kinfo_vmentry);
2052         if (sb == NULL) {
2053                 size = 0;
2054                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2055                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2056                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2057                 PROC_LOCK(p);
2058                 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2059                 sbuf_finish(sb);
2060                 sbuf_delete(sb);
2061                 *sizep = size;
2062         } else {
2063                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2064                 PROC_LOCK(p);
2065                 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2066                     vmmap_flags);
2067         }
2068 }
2069
2070 static void
2071 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2072 {
2073         struct proc *p;
2074         size_t size;
2075         int structsize;
2076
2077         p = (struct proc *)arg;
2078         size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2079         if (sb != NULL) {
2080                 KASSERT(*sizep == size, ("invalid size"));
2081                 structsize = sizeof(gid_t);
2082                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2083                 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2084                     sizeof(gid_t));
2085         }
2086         *sizep = size;
2087 }
2088
2089 static void
2090 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2091 {
2092         struct proc *p;
2093         size_t size;
2094         int structsize;
2095
2096         p = (struct proc *)arg;
2097         size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2098         if (sb != NULL) {
2099                 KASSERT(*sizep == size, ("invalid size"));
2100                 structsize = sizeof(p->p_fd->fd_cmask);
2101                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2102                 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2103         }
2104         *sizep = size;
2105 }
2106
2107 static void
2108 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2109 {
2110         struct proc *p;
2111         struct rlimit rlim[RLIM_NLIMITS];
2112         size_t size;
2113         int structsize, i;
2114
2115         p = (struct proc *)arg;
2116         size = sizeof(structsize) + sizeof(rlim);
2117         if (sb != NULL) {
2118                 KASSERT(*sizep == size, ("invalid size"));
2119                 structsize = sizeof(rlim);
2120                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2121                 PROC_LOCK(p);
2122                 for (i = 0; i < RLIM_NLIMITS; i++)
2123                         lim_rlimit(p, i, &rlim[i]);
2124                 PROC_UNLOCK(p);
2125                 sbuf_bcat(sb, rlim, sizeof(rlim));
2126         }
2127         *sizep = size;
2128 }
2129
2130 static void
2131 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2132 {
2133         struct proc *p;
2134         size_t size;
2135         int structsize;
2136
2137         p = (struct proc *)arg;
2138         size = sizeof(structsize) + sizeof(p->p_osrel);
2139         if (sb != NULL) {
2140                 KASSERT(*sizep == size, ("invalid size"));
2141                 structsize = sizeof(p->p_osrel);
2142                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2143                 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2144         }
2145         *sizep = size;
2146 }
2147
2148 static void
2149 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2150 {
2151         struct proc *p;
2152         elf_ps_strings_t ps_strings;
2153         size_t size;
2154         int structsize;
2155
2156         p = (struct proc *)arg;
2157         size = sizeof(structsize) + sizeof(ps_strings);
2158         if (sb != NULL) {
2159                 KASSERT(*sizep == size, ("invalid size"));
2160                 structsize = sizeof(ps_strings);
2161 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2162                 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2163 #else
2164                 ps_strings = p->p_sysent->sv_psstrings;
2165 #endif
2166                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2167                 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2168         }
2169         *sizep = size;
2170 }
2171
2172 static void
2173 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2174 {
2175         struct proc *p;
2176         size_t size;
2177         int structsize;
2178
2179         p = (struct proc *)arg;
2180         if (sb == NULL) {
2181                 size = 0;
2182                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2183                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2184                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2185                 PHOLD(p);
2186                 proc_getauxv(curthread, p, sb);
2187                 PRELE(p);
2188                 sbuf_finish(sb);
2189                 sbuf_delete(sb);
2190                 *sizep = size;
2191         } else {
2192                 structsize = sizeof(Elf_Auxinfo);
2193                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2194                 PHOLD(p);
2195                 proc_getauxv(curthread, p, sb);
2196                 PRELE(p);
2197         }
2198 }
2199
2200 static boolean_t
2201 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2202     int32_t *osrel, const Elf_Phdr *pnote)
2203 {
2204         const Elf_Note *note, *note0, *note_end;
2205         const char *note_name;
2206         char *buf;
2207         int i, error;
2208         boolean_t res;
2209
2210         /* We need some limit, might as well use PAGE_SIZE. */
2211         if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2212                 return (FALSE);
2213         ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2214         if (pnote->p_offset > PAGE_SIZE ||
2215             pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2216                 VOP_UNLOCK(imgp->vp, 0);
2217                 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2218                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2219                 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2220                     pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2221                     curthread->td_ucred, NOCRED, NULL, curthread);
2222                 if (error != 0) {
2223                         uprintf("i/o error PT_NOTE\n");
2224                         res = FALSE;
2225                         goto ret;
2226                 }
2227                 note = note0 = (const Elf_Note *)buf;
2228                 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2229         } else {
2230                 note = note0 = (const Elf_Note *)(imgp->image_header +
2231                     pnote->p_offset);
2232                 note_end = (const Elf_Note *)(imgp->image_header +
2233                     pnote->p_offset + pnote->p_filesz);
2234                 buf = NULL;
2235         }
2236         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2237                 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2238                     (const char *)note < sizeof(Elf_Note)) {
2239                         res = FALSE;
2240                         goto ret;
2241                 }
2242                 if (note->n_namesz != checknote->hdr.n_namesz ||
2243                     note->n_descsz != checknote->hdr.n_descsz ||
2244                     note->n_type != checknote->hdr.n_type)
2245                         goto nextnote;
2246                 note_name = (const char *)(note + 1);
2247                 if (note_name + checknote->hdr.n_namesz >=
2248                     (const char *)note_end || strncmp(checknote->vendor,
2249                     note_name, checknote->hdr.n_namesz) != 0)
2250                         goto nextnote;
2251
2252                 /*
2253                  * Fetch the osreldate for binary
2254                  * from the ELF OSABI-note if necessary.
2255                  */
2256                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2257                     checknote->trans_osrel != NULL) {
2258                         res = checknote->trans_osrel(note, osrel);
2259                         goto ret;
2260                 }
2261                 res = TRUE;
2262                 goto ret;
2263 nextnote:
2264                 note = (const Elf_Note *)((const char *)(note + 1) +
2265                     roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2266                     roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2267         }
2268         res = FALSE;
2269 ret:
2270         free(buf, M_TEMP);
2271         return (res);
2272 }
2273
2274 /*
2275  * Try to find the appropriate ABI-note section for checknote,
2276  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2277  * first page of the image is searched, the same as for headers.
2278  */
2279 static boolean_t
2280 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2281     int32_t *osrel)
2282 {
2283         const Elf_Phdr *phdr;
2284         const Elf_Ehdr *hdr;
2285         int i;
2286
2287         hdr = (const Elf_Ehdr *)imgp->image_header;
2288         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2289
2290         for (i = 0; i < hdr->e_phnum; i++) {
2291                 if (phdr[i].p_type == PT_NOTE &&
2292                     __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2293                         return (TRUE);
2294         }
2295         return (FALSE);
2296
2297 }
2298
2299 /*
2300  * Tell kern_execve.c about it, with a little help from the linker.
2301  */
2302 static struct execsw __elfN(execsw) = {
2303         __CONCAT(exec_, __elfN(imgact)),
2304         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2305 };
2306 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2307
2308 #ifdef COMPRESS_USER_CORES
2309 /*
2310  * Compress and write out a core segment for a user process.
2311  *
2312  * 'inbuf' is the starting address of a VM segment in the process' address
2313  * space that is to be compressed and written out to the core file.  'dest_buf'
2314  * is a buffer in the kernel's address space.  The segment is copied from 
2315  * 'inbuf' to 'dest_buf' first before being processed by the compression
2316  * routine gzwrite().  This copying is necessary because the content of the VM
2317  * segment may change between the compression pass and the crc-computation pass
2318  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
2319  *
2320  * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2321  */
2322 static int
2323 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2324     struct thread *td)
2325 {
2326         int len_compressed;
2327         int error = 0;
2328         unsigned int chunk_len;
2329
2330         while (len) {
2331                 if (inbuf != NULL) {
2332                         chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2333
2334                         /*
2335                          * We can get EFAULT error here.  In that case zero out
2336                          * the current chunk of the segment.
2337                          */
2338                         error = copyin(inbuf, dest_buf, chunk_len);
2339                         if (error != 0) {
2340                                 bzero(dest_buf, chunk_len);
2341                                 error = 0;
2342                         }
2343                         inbuf += chunk_len;
2344                 } else {
2345                         chunk_len = len;
2346                 }
2347                 len_compressed = gzwrite(file, dest_buf, chunk_len);
2348
2349                 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2350
2351                 if ((unsigned int)len_compressed != chunk_len) {
2352                         log(LOG_WARNING,
2353                             "compress_core: length mismatch (0x%x returned, "
2354                             "0x%x expected)\n", len_compressed, chunk_len);
2355                         EVENTHANDLER_INVOKE(app_coredump_error, td,
2356                             "compress_core: length mismatch %x -> %x",
2357                             chunk_len, len_compressed);
2358                         error = EFAULT;
2359                         break;
2360                 }
2361                 len -= chunk_len;
2362                 maybe_yield();
2363         }
2364
2365         return (error);
2366 }
2367 #endif /* COMPRESS_USER_CORES */
2368
2369 static vm_prot_t
2370 __elfN(trans_prot)(Elf_Word flags)
2371 {
2372         vm_prot_t prot;
2373
2374         prot = 0;
2375         if (flags & PF_X)
2376                 prot |= VM_PROT_EXECUTE;
2377         if (flags & PF_W)
2378                 prot |= VM_PROT_WRITE;
2379         if (flags & PF_R)
2380                 prot |= VM_PROT_READ;
2381 #if __ELF_WORD_SIZE == 32
2382 #if defined(__amd64__) || defined(__ia64__)
2383         if (i386_read_exec && (flags & PF_R))
2384                 prot |= VM_PROT_EXECUTE;
2385 #endif
2386 #endif
2387         return (prot);
2388 }
2389
2390 static Elf_Word
2391 __elfN(untrans_prot)(vm_prot_t prot)
2392 {
2393         Elf_Word flags;
2394
2395         flags = 0;
2396         if (prot & VM_PROT_EXECUTE)
2397                 flags |= PF_X;
2398         if (prot & VM_PROT_READ)
2399                 flags |= PF_R;
2400         if (prot & VM_PROT_WRITE)
2401                 flags |= PF_W;
2402         return (flags);
2403 }