2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
41 #include <sys/fcntl.h>
42 #include <sys/imgact.h>
43 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
50 #include <sys/namei.h>
51 #include <sys/pioctl.h>
53 #include <sys/procfs.h>
54 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/rwlock.h>
58 #include <sys/sf_buf.h>
60 #include <sys/systm.h>
61 #include <sys/signalvar.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>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
85 #define ELF_NOTE_ROUNDSIZE 4
86 #define OLD_EI_BRAND 8
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,
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
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);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108 #ifdef COMPRESS_USER_CORES
109 static int compress_core(gzFile, char *, char *, unsigned int,
112 #define CORE_BUF_SIZE (16 * 1024)
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));
121 static int elf_legacy_coredump = 0;
122 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
123 &elf_legacy_coredump, 0, "");
125 int __elfN(nxstack) =
126 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
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");
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");
143 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
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))
149 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
151 Elf_Brandnote __elfN(freebsd_brandnote) = {
152 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
153 .hdr.n_descsz = sizeof(int32_t),
155 .vendor = FREEBSD_ABI_VENDOR,
156 .flags = BN_TRANSLATE_OSREL,
157 .trans_osrel = __elfN(freebsd_trans_osrel)
161 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
165 p = (uintptr_t)(note + 1);
166 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
167 *osrel = *(const int32_t *)(p);
172 static const char GNU_ABI_VENDOR[] = "GNU";
173 static int GNU_KFREEBSD_ABI_DESC = 3;
175 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
176 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
177 .hdr.n_descsz = 16, /* XXX at least 16 */
179 .vendor = GNU_ABI_VENDOR,
180 .flags = BN_TRANSLATE_OSREL,
181 .trans_osrel = kfreebsd_trans_osrel
185 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
187 const Elf32_Word *desc;
190 p = (uintptr_t)(note + 1);
191 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
193 desc = (const Elf32_Word *)p;
194 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
198 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
199 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
201 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
207 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
211 for (i = 0; i < MAX_BRANDS; i++) {
212 if (elf_brand_list[i] == NULL) {
213 elf_brand_list[i] = entry;
217 if (i == MAX_BRANDS) {
218 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
226 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
230 for (i = 0; i < MAX_BRANDS; i++) {
231 if (elf_brand_list[i] == entry) {
232 elf_brand_list[i] = NULL;
242 __elfN(brand_inuse)(Elf_Brandinfo *entry)
247 sx_slock(&allproc_lock);
248 FOREACH_PROC_IN_SYSTEM(p) {
249 if (p->p_sysent == entry->sysvec) {
254 sx_sunlock(&allproc_lock);
259 static Elf_Brandinfo *
260 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
261 int interp_name_len, int32_t *osrel)
263 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
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.
275 /* Look for an ".note.ABI-tag" ELF section */
276 for (i = 0; i < MAX_BRANDS; i++) {
277 bi = elf_brand_list[i];
280 if (hdr->e_machine == bi->machine && (bi->flags &
281 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
282 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
288 /* If the executable has a brand, search for it in the brand list. */
289 for (i = 0; i < MAX_BRANDS; i++) {
290 bi = elf_brand_list[i];
291 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
293 if (hdr->e_machine == bi->machine &&
294 (hdr->e_ident[EI_OSABI] == bi->brand ||
295 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
296 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
300 /* Lacking a known brand, search for a recognized interpreter. */
301 if (interp != NULL) {
302 for (i = 0; i < MAX_BRANDS; i++) {
303 bi = elf_brand_list[i];
304 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
306 if (hdr->e_machine == bi->machine &&
307 /* ELF image p_filesz includes terminating zero */
308 strlen(bi->interp_path) + 1 == interp_name_len &&
309 strncmp(interp, bi->interp_path, interp_name_len)
315 /* Lacking a recognized interpreter, try the default brand */
316 for (i = 0; i < MAX_BRANDS; i++) {
317 bi = elf_brand_list[i];
318 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
320 if (hdr->e_machine == bi->machine &&
321 __elfN(fallback_brand) == bi->brand)
328 __elfN(check_header)(const Elf_Ehdr *hdr)
334 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
335 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
336 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
337 hdr->e_phentsize != sizeof(Elf_Phdr) ||
338 hdr->e_version != ELF_TARG_VER)
342 * Make sure we have at least one brand for this machine.
345 for (i = 0; i < MAX_BRANDS; i++) {
346 bi = elf_brand_list[i];
347 if (bi != NULL && bi->machine == hdr->e_machine)
357 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
358 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
365 * Create the page if it doesn't exist yet. Ignore errors.
368 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
369 VM_PROT_ALL, VM_PROT_ALL, 0);
373 * Find the page from the underlying object.
376 sf = vm_imgact_map_page(object, offset);
378 return (KERN_FAILURE);
379 off = offset - trunc_page(offset);
380 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
382 vm_imgact_unmap_page(sf);
384 return (KERN_FAILURE);
388 return (KERN_SUCCESS);
392 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
393 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
400 if (start != trunc_page(start)) {
401 rv = __elfN(map_partial)(map, object, offset, start,
402 round_page(start), prot);
405 offset += round_page(start) - start;
406 start = round_page(start);
408 if (end != round_page(end)) {
409 rv = __elfN(map_partial)(map, object, offset +
410 trunc_page(end) - start, trunc_page(end), end, prot);
413 end = trunc_page(end);
416 if (offset & PAGE_MASK) {
418 * The mapping is not page aligned. This means we have
419 * to copy the data. Sigh.
421 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
422 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
427 return (KERN_SUCCESS);
428 for (; start < end; start += sz) {
429 sf = vm_imgact_map_page(object, offset);
431 return (KERN_FAILURE);
432 off = offset - trunc_page(offset);
434 if (sz > PAGE_SIZE - off)
435 sz = PAGE_SIZE - off;
436 error = copyout((caddr_t)sf_buf_kva(sf) + off,
438 vm_imgact_unmap_page(sf);
440 return (KERN_FAILURE);
446 vm_object_reference(object);
448 rv = vm_map_insert(map, object, offset, start, end,
449 prot, VM_PROT_ALL, cow);
451 if (rv != KERN_SUCCESS)
452 vm_object_deallocate(object);
456 return (KERN_SUCCESS);
461 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
462 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
469 vm_offset_t map_addr;
472 vm_offset_t file_addr;
475 * It's necessary to fail if the filsz + offset taken from the
476 * header is greater than the actual file pager object's size.
477 * If we were to allow this, then the vm_map_find() below would
478 * walk right off the end of the file object and into the ether.
480 * While I'm here, might as well check for something else that
481 * is invalid: filsz cannot be greater than memsz.
483 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
484 uprintf("elf_load_section: truncated ELF file\n");
488 object = imgp->object;
489 map = &imgp->proc->p_vmspace->vm_map;
490 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
491 file_addr = trunc_page_ps(offset, pagesize);
494 * We have two choices. We can either clear the data in the last page
495 * of an oversized mapping, or we can start the anon mapping a page
496 * early and copy the initialized data into that first page. We
497 * choose the second..
500 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
502 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
505 /* cow flags: don't dump readonly sections in core */
506 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
507 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
509 rv = __elfN(map_insert)(map,
511 file_addr, /* file offset */
512 map_addr, /* virtual start */
513 map_addr + map_len,/* virtual end */
516 if (rv != KERN_SUCCESS)
519 /* we can stop now if we've covered it all */
520 if (memsz == filsz) {
527 * We have to get the remaining bit of the file into the first part
528 * of the oversized map segment. This is normally because the .data
529 * segment in the file is extended to provide bss. It's a neat idea
530 * to try and save a page, but it's a pain in the behind to implement.
532 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
533 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
534 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
537 /* This had damn well better be true! */
539 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
540 map_len, VM_PROT_ALL, 0);
541 if (rv != KERN_SUCCESS) {
549 sf = vm_imgact_map_page(object, offset + filsz);
553 /* send the page fragment to user space */
554 off = trunc_page_ps(offset + filsz, pagesize) -
555 trunc_page(offset + filsz);
556 error = copyout((caddr_t)sf_buf_kva(sf) + off,
557 (caddr_t)map_addr, copy_len);
558 vm_imgact_unmap_page(sf);
565 * set it to the specified protection.
566 * XXX had better undo the damage from pasting over the cracks here!
568 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
569 map_len), prot, FALSE);
575 * Load the file "file" into memory. It may be either a shared object
578 * The "addr" reference parameter is in/out. On entry, it specifies
579 * the address where a shared object should be loaded. If the file is
580 * an executable, this value is ignored. On exit, "addr" specifies
581 * where the file was actually loaded.
583 * The "entry" reference parameter is out only. On exit, it specifies
584 * the entry point for the loaded file.
587 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
588 u_long *entry, size_t pagesize)
593 struct image_params image_params;
595 const Elf_Ehdr *hdr = NULL;
596 const Elf_Phdr *phdr = NULL;
597 struct nameidata *nd;
599 struct image_params *imgp;
602 u_long base_addr = 0;
603 int error, i, numsegs;
605 #ifdef CAPABILITY_MODE
607 * XXXJA: This check can go away once we are sufficiently confident
608 * that the checks in namei() are correct.
610 if (IN_CAPABILITY_MODE(curthread))
614 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
616 attr = &tempdata->attr;
617 imgp = &tempdata->image_params;
620 * Initialize part of the common data
624 imgp->firstpage = NULL;
625 imgp->image_header = NULL;
627 imgp->execlabel = NULL;
629 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
630 if ((error = namei(nd)) != 0) {
634 NDFREE(nd, NDF_ONLY_PNBUF);
635 imgp->vp = nd->ni_vp;
638 * Check permissions, modes, uid, etc on the file, and "open" it.
640 error = exec_check_permissions(imgp);
644 error = exec_map_first_page(imgp);
649 * Also make certain that the interpreter stays the same, so set
650 * its VV_TEXT flag, too.
652 VOP_SET_TEXT(nd->ni_vp);
654 imgp->object = nd->ni_vp->v_object;
656 hdr = (const Elf_Ehdr *)imgp->image_header;
657 if ((error = __elfN(check_header)(hdr)) != 0)
659 if (hdr->e_type == ET_DYN)
661 else if (hdr->e_type == ET_EXEC)
668 /* Only support headers that fit within first page for now */
669 if ((hdr->e_phoff > PAGE_SIZE) ||
670 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
675 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
676 if (!aligned(phdr, Elf_Addr)) {
681 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
682 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
683 /* Loadable segment */
684 prot = __elfN(trans_prot)(phdr[i].p_flags);
685 error = __elfN(load_section)(imgp, phdr[i].p_offset,
686 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
687 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
691 * Establish the base address if this is the
695 base_addr = trunc_page(phdr[i].p_vaddr +
701 *entry = (unsigned long)hdr->e_entry + rbase;
705 exec_unmap_first_page(imgp);
710 free(tempdata, M_TEMP);
716 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
718 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
719 const Elf_Phdr *phdr;
720 Elf_Auxargs *elf_auxargs;
721 struct vmspace *vmspace;
723 u_long text_size = 0, data_size = 0, total_size = 0;
724 u_long text_addr = 0, data_addr = 0;
725 u_long seg_size, seg_addr;
726 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
728 int error = 0, i, n, interp_name_len = 0;
729 const char *interp = NULL, *newinterp = NULL;
730 Elf_Brandinfo *brand_info;
732 struct sysentvec *sv;
735 * Do we have a valid ELF header ?
737 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
738 * if particular brand doesn't support it.
740 if (__elfN(check_header)(hdr) != 0 ||
741 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
745 * From here on down, we return an errno, not -1, as we've
746 * detected an ELF file.
749 if ((hdr->e_phoff > PAGE_SIZE) ||
750 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
751 /* Only support headers in first page for now */
754 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
755 if (!aligned(phdr, Elf_Addr))
759 for (i = 0; i < hdr->e_phnum; i++) {
760 switch (phdr[i].p_type) {
763 baddr = phdr[i].p_vaddr;
767 /* Path to interpreter */
768 if (phdr[i].p_filesz > MAXPATHLEN ||
769 phdr[i].p_offset > PAGE_SIZE ||
770 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
772 interp = imgp->image_header + phdr[i].p_offset;
773 interp_name_len = phdr[i].p_filesz;
778 __elfN(trans_prot)(phdr[i].p_flags);
779 imgp->stack_sz = phdr[i].p_memsz;
784 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
786 if (brand_info == NULL) {
787 uprintf("ELF binary type \"%u\" not known.\n",
788 hdr->e_ident[EI_OSABI]);
791 if (hdr->e_type == ET_DYN) {
792 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
795 * Honour the base load address from the dso if it is
796 * non-zero for some reason.
799 et_dyn_addr = ET_DYN_LOAD_ADDR;
804 sv = brand_info->sysvec;
805 if (interp != NULL && brand_info->interp_newpath != NULL)
806 newinterp = brand_info->interp_newpath;
809 * Avoid a possible deadlock if the current address space is destroyed
810 * and that address space maps the locked vnode. In the common case,
811 * the locked vnode's v_usecount is decremented but remains greater
812 * than zero. Consequently, the vnode lock is not needed by vrele().
813 * However, in cases where the vnode lock is external, such as nullfs,
814 * v_usecount may become zero.
816 * The VV_TEXT flag prevents modifications to the executable while
817 * the vnode is unlocked.
819 VOP_UNLOCK(imgp->vp, 0);
821 error = exec_new_vmspace(imgp, sv);
822 imgp->proc->p_sysent = sv;
824 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
828 for (i = 0; i < hdr->e_phnum; i++) {
829 switch (phdr[i].p_type) {
830 case PT_LOAD: /* Loadable segment */
831 if (phdr[i].p_memsz == 0)
833 prot = __elfN(trans_prot)(phdr[i].p_flags);
834 error = __elfN(load_section)(imgp, phdr[i].p_offset,
835 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
836 phdr[i].p_memsz, phdr[i].p_filesz, prot,
842 * If this segment contains the program headers,
843 * remember their virtual address for the AT_PHDR
844 * aux entry. Static binaries don't usually include
847 if (phdr[i].p_offset == 0 &&
848 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
850 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
853 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
854 seg_size = round_page(phdr[i].p_memsz +
855 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
858 * Make the largest executable segment the official
859 * text segment and all others data.
861 * Note that obreak() assumes that data_addr +
862 * data_size == end of data load area, and the ELF
863 * file format expects segments to be sorted by
864 * address. If multiple data segments exist, the
865 * last one will be used.
868 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
869 text_size = seg_size;
870 text_addr = seg_addr;
872 data_size = seg_size;
873 data_addr = seg_addr;
875 total_size += seg_size;
877 case PT_PHDR: /* Program header table info */
878 proghdr = phdr[i].p_vaddr + et_dyn_addr;
885 if (data_addr == 0 && data_size == 0) {
886 data_addr = text_addr;
887 data_size = text_size;
890 entry = (u_long)hdr->e_entry + et_dyn_addr;
893 * Check limits. It should be safe to check the
894 * limits after loading the segments since we do
895 * not actually fault in all the segments pages.
897 PROC_LOCK(imgp->proc);
898 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
899 text_size > maxtsiz ||
900 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
901 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
902 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
903 PROC_UNLOCK(imgp->proc);
907 vmspace = imgp->proc->p_vmspace;
908 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
909 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
910 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
911 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
914 * We load the dynamic linker where a userland call
915 * to mmap(0, ...) would put it. The rationale behind this
916 * calculation is that it leaves room for the heap to grow to
917 * its maximum allowed size.
919 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
921 PROC_UNLOCK(imgp->proc);
923 imgp->entry_addr = entry;
925 if (interp != NULL) {
926 int have_interp = FALSE;
927 VOP_UNLOCK(imgp->vp, 0);
928 if (brand_info->emul_path != NULL &&
929 brand_info->emul_path[0] != '\0') {
930 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
931 snprintf(path, MAXPATHLEN, "%s%s",
932 brand_info->emul_path, interp);
933 error = __elfN(load_file)(imgp->proc, path, &addr,
934 &imgp->entry_addr, sv->sv_pagesize);
939 if (!have_interp && newinterp != NULL) {
940 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
941 &imgp->entry_addr, sv->sv_pagesize);
946 error = __elfN(load_file)(imgp->proc, interp, &addr,
947 &imgp->entry_addr, sv->sv_pagesize);
949 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
951 uprintf("ELF interpreter %s not found\n", interp);
958 * Construct auxargs table (used by the fixup routine)
960 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
961 elf_auxargs->execfd = -1;
962 elf_auxargs->phdr = proghdr;
963 elf_auxargs->phent = hdr->e_phentsize;
964 elf_auxargs->phnum = hdr->e_phnum;
965 elf_auxargs->pagesz = PAGE_SIZE;
966 elf_auxargs->base = addr;
967 elf_auxargs->flags = 0;
968 elf_auxargs->entry = entry;
970 imgp->auxargs = elf_auxargs;
971 imgp->interpreted = 0;
972 imgp->reloc_base = addr;
973 imgp->proc->p_osrel = osrel;
978 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
981 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
983 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
987 base = (Elf_Addr *)*stack_base;
988 pos = base + (imgp->args->argc + imgp->args->envc + 2);
990 if (args->execfd != -1)
991 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
992 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
993 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
994 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
995 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
996 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
997 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
998 AUXARGS_ENTRY(pos, AT_BASE, args->base);
999 if (imgp->execpathp != 0)
1000 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1001 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1002 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1003 if (imgp->canary != 0) {
1004 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1005 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1007 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1008 if (imgp->pagesizes != 0) {
1009 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1010 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1012 if (imgp->sysent->sv_timekeep_base != 0) {
1013 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1014 imgp->sysent->sv_timekeep_base);
1016 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1017 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1018 imgp->sysent->sv_stackprot);
1019 AUXARGS_ENTRY(pos, AT_NULL, 0);
1021 free(imgp->auxargs, M_TEMP);
1022 imgp->auxargs = NULL;
1025 suword(base, (long)imgp->args->argc);
1026 *stack_base = (register_t *)base;
1031 * Code for generating ELF core dumps.
1034 typedef void (*segment_callback)(vm_map_entry_t, void *);
1036 /* Closure for cb_put_phdr(). */
1037 struct phdr_closure {
1038 Elf_Phdr *phdr; /* Program header to fill in */
1039 Elf_Off offset; /* Offset of segment in core file */
1042 /* Closure for cb_size_segment(). */
1043 struct sseg_closure {
1044 int count; /* Count of writable segments. */
1045 size_t size; /* Total size of all writable segments. */
1048 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1051 int type; /* Note type. */
1052 outfunc_t outfunc; /* Output function. */
1053 void *outarg; /* Argument for the output function. */
1054 size_t outsize; /* Output size. */
1055 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1058 TAILQ_HEAD(note_info_list, note_info);
1060 static void cb_put_phdr(vm_map_entry_t, void *);
1061 static void cb_size_segment(vm_map_entry_t, void *);
1062 static void each_writable_segment(struct thread *, segment_callback, void *);
1063 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1064 int, void *, size_t, struct note_info_list *, size_t, gzFile);
1065 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1067 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1068 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1069 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1070 static int sbuf_drain_core_output(void *, const char *, int);
1071 static int sbuf_drain_count(void *arg, const char *data, int len);
1073 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1074 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1075 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1076 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1077 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1078 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1079 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1080 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1081 static void note_procstat_files(void *, struct sbuf *, size_t *);
1082 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1083 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1084 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1085 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1086 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1088 #ifdef COMPRESS_USER_CORES
1089 extern int compress_user_cores;
1090 extern int compress_user_cores_gzlevel;
1094 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1095 struct ucred *active_cred, struct ucred *file_cred,
1096 struct thread *td, char *core_buf, gzFile gzfile) {
1100 #ifdef COMPRESS_USER_CORES
1101 error = compress_core(gzfile, base, core_buf, len, td);
1103 panic("shouldn't be here");
1106 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1107 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1113 /* Coredump output parameters for sbuf drain routine. */
1114 struct sbuf_drain_core_params {
1116 struct ucred *active_cred;
1117 struct ucred *file_cred;
1120 #ifdef COMPRESS_USER_CORES
1126 * Drain into a core file.
1129 sbuf_drain_core_output(void *arg, const char *data, int len)
1131 struct sbuf_drain_core_params *p;
1134 p = (struct sbuf_drain_core_params *)arg;
1137 * Some kern_proc out routines that print to this sbuf may
1138 * call us with the process lock held. Draining with the
1139 * non-sleepable lock held is unsafe. The lock is needed for
1140 * those routines when dumping a live process. In our case we
1141 * can safely release the lock before draining and acquire
1144 locked = PROC_LOCKED(p->td->td_proc);
1146 PROC_UNLOCK(p->td->td_proc);
1147 #ifdef COMPRESS_USER_CORES
1148 if (p->gzfile != Z_NULL)
1149 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1153 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1154 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1155 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1158 PROC_LOCK(p->td->td_proc);
1166 * Drain into a counter.
1169 sbuf_drain_count(void *arg, const char *data __unused, int len)
1173 sizep = (size_t *)arg;
1179 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1181 struct ucred *cred = td->td_ucred;
1183 struct sseg_closure seginfo;
1184 struct note_info_list notelst;
1185 struct note_info *ninfo;
1187 size_t hdrsize, notesz, coresize;
1189 gzFile gzfile = Z_NULL;
1190 char *core_buf = NULL;
1191 #ifdef COMPRESS_USER_CORES
1192 char gzopen_flags[8];
1194 int doing_compress = flags & IMGACT_CORE_COMPRESS;
1198 TAILQ_INIT(¬elst);
1200 #ifdef COMPRESS_USER_CORES
1201 if (doing_compress) {
1204 if (compress_user_cores_gzlevel >= 0 &&
1205 compress_user_cores_gzlevel <= 9)
1206 *p++ = '0' + compress_user_cores_gzlevel;
1208 gzfile = gz_open("", gzopen_flags, vp);
1209 if (gzfile == Z_NULL) {
1213 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1221 /* Size the program segments. */
1224 each_writable_segment(td, cb_size_segment, &seginfo);
1227 * Collect info about the core file header area.
1229 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1230 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1231 coresize = round_page(hdrsize + notesz) + seginfo.size;
1235 PROC_LOCK(td->td_proc);
1236 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1237 PROC_UNLOCK(td->td_proc);
1244 if (coresize >= limit) {
1250 * Allocate memory for building the header, fill it up,
1251 * and write it out following the notes.
1253 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1258 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1259 ¬elst, notesz, gzfile);
1261 /* Write the contents of all of the writable segments. */
1267 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1268 offset = round_page(hdrsize + notesz);
1269 for (i = 0; i < seginfo.count; i++) {
1270 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1271 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1274 offset += php->p_filesz;
1280 "Failed to write core file for process %s (error %d)\n",
1281 curproc->p_comm, error);
1285 #ifdef COMPRESS_USER_CORES
1287 free(core_buf, M_TEMP);
1291 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1292 TAILQ_REMOVE(¬elst, ninfo, link);
1293 free(ninfo, M_TEMP);
1302 * A callback for each_writable_segment() to write out the segment's
1303 * program header entry.
1306 cb_put_phdr(entry, closure)
1307 vm_map_entry_t entry;
1310 struct phdr_closure *phc = (struct phdr_closure *)closure;
1311 Elf_Phdr *phdr = phc->phdr;
1313 phc->offset = round_page(phc->offset);
1315 phdr->p_type = PT_LOAD;
1316 phdr->p_offset = phc->offset;
1317 phdr->p_vaddr = entry->start;
1319 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1320 phdr->p_align = PAGE_SIZE;
1321 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1323 phc->offset += phdr->p_filesz;
1328 * A callback for each_writable_segment() to gather information about
1329 * the number of segments and their total size.
1332 cb_size_segment(entry, closure)
1333 vm_map_entry_t entry;
1336 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1339 ssc->size += entry->end - entry->start;
1343 * For each writable segment in the process's memory map, call the given
1344 * function with a pointer to the map entry and some arbitrary
1345 * caller-supplied data.
1348 each_writable_segment(td, func, closure)
1350 segment_callback func;
1353 struct proc *p = td->td_proc;
1354 vm_map_t map = &p->p_vmspace->vm_map;
1355 vm_map_entry_t entry;
1356 vm_object_t backing_object, object;
1357 boolean_t ignore_entry;
1359 vm_map_lock_read(map);
1360 for (entry = map->header.next; entry != &map->header;
1361 entry = entry->next) {
1363 * Don't dump inaccessible mappings, deal with legacy
1366 * Note that read-only segments related to the elf binary
1367 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1368 * need to arbitrarily ignore such segments.
1370 if (elf_legacy_coredump) {
1371 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1374 if ((entry->protection & VM_PROT_ALL) == 0)
1379 * Dont include memory segment in the coredump if
1380 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1381 * madvise(2). Do not dump submaps (i.e. parts of the
1384 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1387 if ((object = entry->object.vm_object) == NULL)
1390 /* Ignore memory-mapped devices and such things. */
1391 VM_OBJECT_RLOCK(object);
1392 while ((backing_object = object->backing_object) != NULL) {
1393 VM_OBJECT_RLOCK(backing_object);
1394 VM_OBJECT_RUNLOCK(object);
1395 object = backing_object;
1397 ignore_entry = object->type != OBJT_DEFAULT &&
1398 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1399 object->type != OBJT_PHYS;
1400 VM_OBJECT_RUNLOCK(object);
1404 (*func)(entry, closure);
1406 vm_map_unlock_read(map);
1410 * Write the core file header to the file, including padding up to
1411 * the page boundary.
1414 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1415 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1416 size_t notesz, gzFile gzfile)
1418 struct sbuf_drain_core_params params;
1419 struct note_info *ninfo;
1423 /* Fill in the header. */
1424 bzero(hdr, hdrsize);
1425 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1428 params.active_cred = cred;
1429 params.file_cred = NOCRED;
1432 #ifdef COMPRESS_USER_CORES
1433 params.gzfile = gzfile;
1435 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1436 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms);
1437 sbuf_start_section(sb, NULL);
1438 sbuf_bcat(sb, hdr, hdrsize);
1439 TAILQ_FOREACH(ninfo, notelst, link)
1440 __elfN(putnote)(ninfo, sb);
1441 /* Align up to a page boundary for the program segments. */
1442 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1443 error = sbuf_finish(sb);
1450 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1460 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1463 * To have the debugger select the right thread (LWP) as the initial
1464 * thread, we dump the state of the thread passed to us in td first.
1465 * This is the thread that causes the core dump and thus likely to
1466 * be the right thread one wants to have selected in the debugger.
1469 while (thr != NULL) {
1470 size += register_note(list, NT_PRSTATUS,
1471 __elfN(note_prstatus), thr);
1472 size += register_note(list, NT_FPREGSET,
1473 __elfN(note_fpregset), thr);
1474 size += register_note(list, NT_THRMISC,
1475 __elfN(note_thrmisc), thr);
1476 size += register_note(list, -1,
1477 __elfN(note_threadmd), thr);
1479 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1480 TAILQ_NEXT(thr, td_plist);
1482 thr = TAILQ_NEXT(thr, td_plist);
1485 size += register_note(list, NT_PROCSTAT_PROC,
1486 __elfN(note_procstat_proc), p);
1487 size += register_note(list, NT_PROCSTAT_FILES,
1488 note_procstat_files, p);
1489 size += register_note(list, NT_PROCSTAT_VMMAP,
1490 note_procstat_vmmap, p);
1491 size += register_note(list, NT_PROCSTAT_GROUPS,
1492 note_procstat_groups, p);
1493 size += register_note(list, NT_PROCSTAT_UMASK,
1494 note_procstat_umask, p);
1495 size += register_note(list, NT_PROCSTAT_RLIMIT,
1496 note_procstat_rlimit, p);
1497 size += register_note(list, NT_PROCSTAT_OSREL,
1498 note_procstat_osrel, p);
1499 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1500 __elfN(note_procstat_psstrings), p);
1501 size += register_note(list, NT_PROCSTAT_AUXV,
1502 __elfN(note_procstat_auxv), p);
1508 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1513 struct phdr_closure phc;
1515 ehdr = (Elf_Ehdr *)hdr;
1516 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1518 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1519 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1520 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1521 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1522 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1523 ehdr->e_ident[EI_DATA] = ELF_DATA;
1524 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1525 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1526 ehdr->e_ident[EI_ABIVERSION] = 0;
1527 ehdr->e_ident[EI_PAD] = 0;
1528 ehdr->e_type = ET_CORE;
1529 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1530 ehdr->e_machine = ELF_ARCH32;
1532 ehdr->e_machine = ELF_ARCH;
1534 ehdr->e_version = EV_CURRENT;
1536 ehdr->e_phoff = sizeof(Elf_Ehdr);
1538 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1539 ehdr->e_phentsize = sizeof(Elf_Phdr);
1540 ehdr->e_phnum = numsegs + 1;
1541 ehdr->e_shentsize = sizeof(Elf_Shdr);
1543 ehdr->e_shstrndx = SHN_UNDEF;
1546 * Fill in the program header entries.
1549 /* The note segement. */
1550 phdr->p_type = PT_NOTE;
1551 phdr->p_offset = hdrsize;
1554 phdr->p_filesz = notesz;
1556 phdr->p_flags = PF_R;
1557 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1560 /* All the writable segments from the program. */
1562 phc.offset = round_page(hdrsize + notesz);
1563 each_writable_segment(td, cb_put_phdr, &phc);
1567 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1569 struct note_info *ninfo;
1570 size_t size, notesize;
1573 out(arg, NULL, &size);
1574 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1576 ninfo->outfunc = out;
1577 ninfo->outarg = arg;
1578 ninfo->outsize = size;
1579 TAILQ_INSERT_TAIL(list, ninfo, link);
1584 notesize = sizeof(Elf_Note) + /* note header */
1585 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1587 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1593 append_note_data(const void *src, void *dst, size_t len)
1597 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1599 bcopy(src, dst, len);
1600 bzero((char *)dst + len, padded_len - len);
1602 return (padded_len);
1606 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1614 note = (Elf_Note *)buf;
1615 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1616 note->n_descsz = size;
1617 note->n_type = type;
1618 buf += sizeof(*note);
1619 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1620 sizeof(FREEBSD_ABI_VENDOR));
1621 append_note_data(src, buf, size);
1626 notesize = sizeof(Elf_Note) + /* note header */
1627 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1629 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1635 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1640 if (ninfo->type == -1) {
1641 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1645 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1646 note.n_descsz = ninfo->outsize;
1647 note.n_type = ninfo->type;
1649 sbuf_bcat(sb, ¬e, sizeof(note));
1650 sbuf_start_section(sb, &old_len);
1651 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1652 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1653 if (note.n_descsz == 0)
1655 sbuf_start_section(sb, &old_len);
1656 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1657 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1661 * Miscellaneous note out functions.
1664 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1665 #include <compat/freebsd32/freebsd32.h>
1667 typedef struct prstatus32 elf_prstatus_t;
1668 typedef struct prpsinfo32 elf_prpsinfo_t;
1669 typedef struct fpreg32 elf_prfpregset_t;
1670 typedef struct fpreg32 elf_fpregset_t;
1671 typedef struct reg32 elf_gregset_t;
1672 typedef struct thrmisc32 elf_thrmisc_t;
1673 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1674 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1675 typedef uint32_t elf_ps_strings_t;
1677 typedef prstatus_t elf_prstatus_t;
1678 typedef prpsinfo_t elf_prpsinfo_t;
1679 typedef prfpregset_t elf_prfpregset_t;
1680 typedef prfpregset_t elf_fpregset_t;
1681 typedef gregset_t elf_gregset_t;
1682 typedef thrmisc_t elf_thrmisc_t;
1683 #define ELF_KERN_PROC_MASK 0
1684 typedef struct kinfo_proc elf_kinfo_proc_t;
1685 typedef vm_offset_t elf_ps_strings_t;
1689 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1692 elf_prpsinfo_t *psinfo;
1694 p = (struct proc *)arg;
1696 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1697 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1698 psinfo->pr_version = PRPSINFO_VERSION;
1699 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1700 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1702 * XXX - We don't fill in the command line arguments properly
1705 strlcpy(psinfo->pr_psargs, p->p_comm,
1706 sizeof(psinfo->pr_psargs));
1708 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1709 free(psinfo, M_TEMP);
1711 *sizep = sizeof(*psinfo);
1715 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1718 elf_prstatus_t *status;
1720 td = (struct thread *)arg;
1722 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1723 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1724 status->pr_version = PRSTATUS_VERSION;
1725 status->pr_statussz = sizeof(elf_prstatus_t);
1726 status->pr_gregsetsz = sizeof(elf_gregset_t);
1727 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1728 status->pr_osreldate = osreldate;
1729 status->pr_cursig = td->td_proc->p_sig;
1730 status->pr_pid = td->td_tid;
1731 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1732 fill_regs32(td, &status->pr_reg);
1734 fill_regs(td, &status->pr_reg);
1736 sbuf_bcat(sb, status, sizeof(*status));
1737 free(status, M_TEMP);
1739 *sizep = sizeof(*status);
1743 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1746 elf_prfpregset_t *fpregset;
1748 td = (struct thread *)arg;
1750 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1751 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1752 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1753 fill_fpregs32(td, fpregset);
1755 fill_fpregs(td, fpregset);
1757 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1758 free(fpregset, M_TEMP);
1760 *sizep = sizeof(*fpregset);
1764 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1767 elf_thrmisc_t thrmisc;
1769 td = (struct thread *)arg;
1771 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1772 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1773 strcpy(thrmisc.pr_tname, td->td_name);
1774 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1776 *sizep = sizeof(thrmisc);
1780 * Allow for MD specific notes, as well as any MD
1781 * specific preparations for writing MI notes.
1784 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1790 td = (struct thread *)arg;
1792 if (size != 0 && sb != NULL)
1793 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1797 __elfN(dump_thread)(td, buf, &size);
1798 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1799 if (size != 0 && sb != NULL)
1800 sbuf_bcat(sb, buf, size);
1805 #ifdef KINFO_PROC_SIZE
1806 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1810 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1816 p = (struct proc *)arg;
1817 size = sizeof(structsize) + p->p_numthreads *
1818 sizeof(elf_kinfo_proc_t);
1821 KASSERT(*sizep == size, ("invalid size"));
1822 structsize = sizeof(elf_kinfo_proc_t);
1823 sbuf_bcat(sb, &structsize, sizeof(structsize));
1825 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1830 #ifdef KINFO_FILE_SIZE
1831 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1835 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1841 p = (struct proc *)arg;
1844 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1845 sbuf_set_drain(sb, sbuf_drain_count, &size);
1846 sbuf_bcat(sb, &structsize, sizeof(structsize));
1848 kern_proc_filedesc_out(p, sb, -1);
1853 structsize = sizeof(struct kinfo_file);
1854 sbuf_bcat(sb, &structsize, sizeof(structsize));
1856 kern_proc_filedesc_out(p, sb, -1);
1860 #ifdef KINFO_VMENTRY_SIZE
1861 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1865 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1871 p = (struct proc *)arg;
1874 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1875 sbuf_set_drain(sb, sbuf_drain_count, &size);
1876 sbuf_bcat(sb, &structsize, sizeof(structsize));
1878 kern_proc_vmmap_out(p, sb);
1883 structsize = sizeof(struct kinfo_vmentry);
1884 sbuf_bcat(sb, &structsize, sizeof(structsize));
1886 kern_proc_vmmap_out(p, sb);
1891 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1897 p = (struct proc *)arg;
1898 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1900 KASSERT(*sizep == size, ("invalid size"));
1901 structsize = sizeof(gid_t);
1902 sbuf_bcat(sb, &structsize, sizeof(structsize));
1903 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1910 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1916 p = (struct proc *)arg;
1917 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1919 KASSERT(*sizep == size, ("invalid size"));
1920 structsize = sizeof(p->p_fd->fd_cmask);
1921 sbuf_bcat(sb, &structsize, sizeof(structsize));
1922 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1928 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1931 struct rlimit rlim[RLIM_NLIMITS];
1935 p = (struct proc *)arg;
1936 size = sizeof(structsize) + sizeof(rlim);
1938 KASSERT(*sizep == size, ("invalid size"));
1939 structsize = sizeof(rlim);
1940 sbuf_bcat(sb, &structsize, sizeof(structsize));
1942 for (i = 0; i < RLIM_NLIMITS; i++)
1943 lim_rlimit(p, i, &rlim[i]);
1945 sbuf_bcat(sb, rlim, sizeof(rlim));
1951 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1957 p = (struct proc *)arg;
1958 size = sizeof(structsize) + sizeof(p->p_osrel);
1960 KASSERT(*sizep == size, ("invalid size"));
1961 structsize = sizeof(p->p_osrel);
1962 sbuf_bcat(sb, &structsize, sizeof(structsize));
1963 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1969 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1972 elf_ps_strings_t ps_strings;
1976 p = (struct proc *)arg;
1977 size = sizeof(structsize) + sizeof(ps_strings);
1979 KASSERT(*sizep == size, ("invalid size"));
1980 structsize = sizeof(ps_strings);
1981 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1982 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1984 ps_strings = p->p_sysent->sv_psstrings;
1986 sbuf_bcat(sb, &structsize, sizeof(structsize));
1987 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1993 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
1999 p = (struct proc *)arg;
2002 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2003 sbuf_set_drain(sb, sbuf_drain_count, &size);
2004 sbuf_bcat(sb, &structsize, sizeof(structsize));
2006 proc_getauxv(curthread, p, sb);
2012 structsize = sizeof(Elf_Auxinfo);
2013 sbuf_bcat(sb, &structsize, sizeof(structsize));
2015 proc_getauxv(curthread, p, sb);
2021 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2022 int32_t *osrel, const Elf_Phdr *pnote)
2024 const Elf_Note *note, *note0, *note_end;
2025 const char *note_name;
2028 if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
2029 pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
2032 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
2033 note_end = (const Elf_Note *)(imgp->image_header +
2034 pnote->p_offset + pnote->p_filesz);
2035 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2036 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2037 (const char *)note < sizeof(Elf_Note))
2039 if (note->n_namesz != checknote->hdr.n_namesz ||
2040 note->n_descsz != checknote->hdr.n_descsz ||
2041 note->n_type != checknote->hdr.n_type)
2043 note_name = (const char *)(note + 1);
2044 if (note_name + checknote->hdr.n_namesz >=
2045 (const char *)note_end || strncmp(checknote->vendor,
2046 note_name, checknote->hdr.n_namesz) != 0)
2050 * Fetch the osreldate for binary
2051 * from the ELF OSABI-note if necessary.
2053 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2054 checknote->trans_osrel != NULL)
2055 return (checknote->trans_osrel(note, osrel));
2059 note = (const Elf_Note *)((const char *)(note + 1) +
2060 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2061 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2068 * Try to find the appropriate ABI-note section for checknote,
2069 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2070 * first page of the image is searched, the same as for headers.
2073 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2076 const Elf_Phdr *phdr;
2077 const Elf_Ehdr *hdr;
2080 hdr = (const Elf_Ehdr *)imgp->image_header;
2081 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2083 for (i = 0; i < hdr->e_phnum; i++) {
2084 if (phdr[i].p_type == PT_NOTE &&
2085 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2093 * Tell kern_execve.c about it, with a little help from the linker.
2095 static struct execsw __elfN(execsw) = {
2096 __CONCAT(exec_, __elfN(imgact)),
2097 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2099 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2101 #ifdef COMPRESS_USER_CORES
2103 * Compress and write out a core segment for a user process.
2105 * 'inbuf' is the starting address of a VM segment in the process' address
2106 * space that is to be compressed and written out to the core file. 'dest_buf'
2107 * is a buffer in the kernel's address space. The segment is copied from
2108 * 'inbuf' to 'dest_buf' first before being processed by the compression
2109 * routine gzwrite(). This copying is necessary because the content of the VM
2110 * segment may change between the compression pass and the crc-computation pass
2111 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel.
2113 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2116 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2121 unsigned int chunk_len;
2124 if (inbuf != NULL) {
2125 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2126 copyin(inbuf, dest_buf, chunk_len);
2131 len_compressed = gzwrite(file, dest_buf, chunk_len);
2133 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2135 if ((unsigned int)len_compressed != chunk_len) {
2137 "compress_core: length mismatch (0x%x returned, "
2138 "0x%x expected)\n", len_compressed, chunk_len);
2139 EVENTHANDLER_INVOKE(app_coredump_error, td,
2140 "compress_core: length mismatch %x -> %x",
2141 chunk_len, len_compressed);
2151 #endif /* COMPRESS_USER_CORES */
2154 __elfN(trans_prot)(Elf_Word flags)
2160 prot |= VM_PROT_EXECUTE;
2162 prot |= VM_PROT_WRITE;
2164 prot |= VM_PROT_READ;
2165 #if __ELF_WORD_SIZE == 32
2166 #if defined(__amd64__) || defined(__ia64__)
2167 if (i386_read_exec && (flags & PF_R))
2168 prot |= VM_PROT_EXECUTE;
2175 __elfN(untrans_prot)(vm_prot_t prot)
2180 if (prot & VM_PROT_EXECUTE)
2182 if (prot & VM_PROT_READ)
2184 if (prot & VM_PROT_WRITE)