proc->thread stage 3.5: Add an IO_CORE flag so coda doesn't have to dig
[dragonfly.git] / sys / kern / imgact_elf.c
1 /*-
2  * Copyright (c) 1995-1996 Søren Schmidt
3  * Copyright (c) 1996 Peter Wemm
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer
11  *    in this position and unchanged.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote products
16  *    derived from this software withough specific prior written permission
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  *
29  * $FreeBSD: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $
30  * $DragonFly: src/sys/kern/imgact_elf.c,v 1.3 2003/06/24 02:11:55 dillon Exp $
31  */
32
33 #include <sys/param.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_elf.h>
38 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/mman.h>
41 #include <sys/namei.h>
42 #include <sys/pioctl.h>
43 #include <sys/proc.h>
44 #include <sys/procfs.h>
45 #include <sys/resourcevar.h>
46 #include <sys/signalvar.h>
47 #include <sys/stat.h>
48 #include <sys/syscall.h>
49 #include <sys/sysctl.h>
50 #include <sys/sysent.h>
51 #include <sys/systm.h>
52 #include <sys/vnode.h>
53
54 #include <vm/vm.h>
55 #include <vm/vm_kern.h>
56 #include <vm/vm_param.h>
57 #include <vm/pmap.h>
58 #include <sys/lock.h>
59 #include <vm/vm_map.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_extern.h>
62
63 #include <machine/elf.h>
64 #include <machine/md_var.h>
65
66 #define OLD_EI_BRAND    8
67
68 __ElfType(Brandinfo);
69 __ElfType(Auxargs);
70
71 static int elf_check_header __P((const Elf_Ehdr *hdr));
72 static int elf_freebsd_fixup __P((register_t **stack_base,
73     struct image_params *imgp));
74 static int elf_load_file __P((struct proc *p, const char *file, u_long *addr,
75     u_long *entry));
76 static int elf_load_section __P((struct proc *p,
77     struct vmspace *vmspace, struct vnode *vp,
78     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
79     vm_prot_t prot));
80 static int exec_elf_imgact __P((struct image_params *imgp));
81
82 static int elf_trace = 0;
83 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
84 static int elf_legacy_coredump = 0;
85 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
86     &elf_legacy_coredump, 0, "");
87
88 static struct sysentvec elf_freebsd_sysvec = {
89         SYS_MAXSYSCALL,
90         sysent,
91         0,
92         0,
93         0,
94         0,
95         0,
96         0,
97         elf_freebsd_fixup,
98         sendsig,
99         sigcode,
100         &szsigcode,
101         0,
102         "FreeBSD ELF",
103         elf_coredump,
104         NULL,
105         MINSIGSTKSZ
106 };
107
108 static Elf_Brandinfo freebsd_brand_info = {
109                                                 ELFOSABI_FREEBSD,
110                                                 "FreeBSD",
111                                                 "",
112                                                 "/usr/libexec/ld-elf.so.1",
113                                                 &elf_freebsd_sysvec
114                                           };
115 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
116                                                         &freebsd_brand_info,
117                                                         NULL, NULL, NULL,
118                                                         NULL, NULL, NULL, NULL
119                                                     };
120
121 int
122 elf_insert_brand_entry(Elf_Brandinfo *entry)
123 {
124         int i;
125
126         for (i=1; i<MAX_BRANDS; i++) {
127                 if (elf_brand_list[i] == NULL) {
128                         elf_brand_list[i] = entry;
129                         break;
130                 }
131         }
132         if (i == MAX_BRANDS)
133                 return -1;
134         return 0;
135 }
136
137 int
138 elf_remove_brand_entry(Elf_Brandinfo *entry)
139 {
140         int i;
141
142         for (i=1; i<MAX_BRANDS; i++) {
143                 if (elf_brand_list[i] == entry) {
144                         elf_brand_list[i] = NULL;
145                         break;
146                 }
147         }
148         if (i == MAX_BRANDS)
149                 return -1;
150         return 0;
151 }
152
153 int
154 elf_brand_inuse(Elf_Brandinfo *entry)
155 {
156         struct proc *p;
157         int rval = FALSE;
158
159         LIST_FOREACH(p, &allproc, p_list) {
160                 if (p->p_sysent == entry->sysvec) {
161                         rval = TRUE;
162                         break;
163                 }
164         }
165
166         return (rval);
167 }
168
169 static int
170 elf_check_header(const Elf_Ehdr *hdr)
171 {
172         if (!IS_ELF(*hdr) ||
173             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
174             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
175             hdr->e_ident[EI_VERSION] != EV_CURRENT)
176                 return ENOEXEC;
177
178         if (!ELF_MACHINE_OK(hdr->e_machine))
179                 return ENOEXEC;
180
181         if (hdr->e_version != ELF_TARG_VER)
182                 return ENOEXEC;
183         
184         return 0;
185 }
186
187 static int
188 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
189 {
190         size_t map_len;
191         vm_offset_t map_addr;
192         int error, rv, cow;
193         size_t copy_len;
194         vm_object_t object;
195         vm_offset_t file_addr;
196         vm_offset_t data_buf = 0;
197
198         VOP_GETVOBJECT(vp, &object);
199         error = 0;
200
201         /*
202          * It's necessary to fail if the filsz + offset taken from the
203          * header is greater than the actual file pager object's size.
204          * If we were to allow this, then the vm_map_find() below would
205          * walk right off the end of the file object and into the ether.
206          *
207          * While I'm here, might as well check for something else that
208          * is invalid: filsz cannot be greater than memsz.
209          */
210         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
211             filsz > memsz) {
212                 uprintf("elf_load_section: truncated ELF file\n");
213                 return (ENOEXEC);
214         }
215
216         map_addr = trunc_page((vm_offset_t)vmaddr);
217         file_addr = trunc_page(offset);
218
219         /*
220          * We have two choices.  We can either clear the data in the last page
221          * of an oversized mapping, or we can start the anon mapping a page
222          * early and copy the initialized data into that first page.  We
223          * choose the second..
224          */
225         if (memsz > filsz)
226                 map_len = trunc_page(offset+filsz) - file_addr;
227         else
228                 map_len = round_page(offset+filsz) - file_addr;
229
230         if (map_len != 0) {
231                 vm_object_reference(object);
232
233                 /* cow flags: don't dump readonly sections in core */
234                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
235                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
236
237                 vm_map_lock(&vmspace->vm_map);
238                 rv = vm_map_insert(&vmspace->vm_map,
239                                       object,
240                                       file_addr,        /* file offset */
241                                       map_addr,         /* virtual start */
242                                       map_addr + map_len,/* virtual end */
243                                       prot,
244                                       VM_PROT_ALL,
245                                       cow);
246                 vm_map_unlock(&vmspace->vm_map);
247                 if (rv != KERN_SUCCESS) {
248                         vm_object_deallocate(object);
249                         return EINVAL;
250                 }
251
252                 /* we can stop now if we've covered it all */
253                 if (memsz == filsz) {
254                         return 0;
255                 }
256         }
257
258
259         /*
260          * We have to get the remaining bit of the file into the first part
261          * of the oversized map segment.  This is normally because the .data
262          * segment in the file is extended to provide bss.  It's a neat idea
263          * to try and save a page, but it's a pain in the behind to implement.
264          */
265         copy_len = (offset + filsz) - trunc_page(offset + filsz);
266         map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
267         map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
268
269         /* This had damn well better be true! */
270         if (map_len != 0) {
271                 vm_map_lock(&vmspace->vm_map);
272                 rv = vm_map_insert(&vmspace->vm_map, NULL, 0,
273                                         map_addr, map_addr + map_len,
274                                         VM_PROT_ALL, VM_PROT_ALL, 0);
275                 vm_map_unlock(&vmspace->vm_map);
276                 if (rv != KERN_SUCCESS) {
277                         return EINVAL; 
278                 }
279         }
280
281         if (copy_len != 0) {
282                 vm_object_reference(object);
283                 rv = vm_map_find(exec_map,
284                                  object, 
285                                  trunc_page(offset + filsz),
286                                  &data_buf,
287                                  PAGE_SIZE,
288                                  TRUE,
289                                  VM_PROT_READ,
290                                  VM_PROT_ALL,
291                                  MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
292                 if (rv != KERN_SUCCESS) {
293                         vm_object_deallocate(object);
294                         return EINVAL;
295                 }
296
297                 /* send the page fragment to user space */
298                 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
299                 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
300                 if (error) {
301                         return (error);
302                 }
303         }
304
305         /*
306          * set it to the specified protection
307          */
308         vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,  prot,
309                        FALSE);
310
311         return error;
312 }
313
314 /*
315  * Load the file "file" into memory.  It may be either a shared object
316  * or an executable.
317  *
318  * The "addr" reference parameter is in/out.  On entry, it specifies
319  * the address where a shared object should be loaded.  If the file is
320  * an executable, this value is ignored.  On exit, "addr" specifies
321  * where the file was actually loaded.
322  *
323  * The "entry" reference parameter is out only.  On exit, it specifies
324  * the entry point for the loaded file.
325  */
326 static int
327 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
328 {
329         struct {
330                 struct nameidata nd;
331                 struct vattr attr;
332                 struct image_params image_params;
333         } *tempdata;
334         const Elf_Ehdr *hdr = NULL;
335         const Elf_Phdr *phdr = NULL;
336         struct nameidata *nd;
337         struct vmspace *vmspace = p->p_vmspace;
338         struct vattr *attr;
339         struct image_params *imgp;
340         vm_prot_t prot;
341         u_long rbase;
342         u_long base_addr = 0;
343         int error, i, numsegs;
344
345         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
346         nd = &tempdata->nd;
347         attr = &tempdata->attr;
348         imgp = &tempdata->image_params;
349
350         /*
351          * Initialize part of the common data
352          */
353         imgp->proc = p;
354         imgp->uap = NULL;
355         imgp->attr = attr;
356         imgp->firstpage = NULL;
357         imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
358
359         if (imgp->image_header == NULL) {
360                 nd->ni_vp = NULL;
361                 error = ENOMEM;
362                 goto fail;
363         }
364
365         NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p);   
366                          
367         if ((error = namei(nd)) != 0) {
368                 nd->ni_vp = NULL;
369                 goto fail;
370         }
371         NDFREE(nd, NDF_ONLY_PNBUF);
372         imgp->vp = nd->ni_vp;
373
374         /*
375          * Check permissions, modes, uid, etc on the file, and "open" it.
376          */
377         error = exec_check_permissions(imgp);
378         if (error) {
379                 VOP_UNLOCK(nd->ni_vp, 0, p);
380                 goto fail;
381         }
382
383         error = exec_map_first_page(imgp);
384         /*
385          * Also make certain that the interpreter stays the same, so set
386          * its VTEXT flag, too.
387          */
388         if (error == 0)
389                 nd->ni_vp->v_flag |= VTEXT;
390         VOP_UNLOCK(nd->ni_vp, 0, p);
391         if (error)
392                 goto fail;
393
394         hdr = (const Elf_Ehdr *)imgp->image_header;
395         if ((error = elf_check_header(hdr)) != 0)
396                 goto fail;
397         if (hdr->e_type == ET_DYN)
398                 rbase = *addr;
399         else if (hdr->e_type == ET_EXEC)
400                 rbase = 0;
401         else {
402                 error = ENOEXEC;
403                 goto fail;
404         }
405
406         /* Only support headers that fit within first page for now */
407         if ((hdr->e_phoff > PAGE_SIZE) ||
408             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
409                 error = ENOEXEC;
410                 goto fail;
411         }
412
413         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
414
415         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
416                 if (phdr[i].p_type == PT_LOAD) {        /* Loadable segment */
417                         prot = 0;
418                         if (phdr[i].p_flags & PF_X)
419                                 prot |= VM_PROT_EXECUTE;
420                         if (phdr[i].p_flags & PF_W)
421                                 prot |= VM_PROT_WRITE;
422                         if (phdr[i].p_flags & PF_R)
423                                 prot |= VM_PROT_READ;
424
425                         if ((error = elf_load_section(p, vmspace, nd->ni_vp,
426                                                      phdr[i].p_offset,
427                                                      (caddr_t)phdr[i].p_vaddr +
428                                                         rbase,
429                                                      phdr[i].p_memsz,
430                                                      phdr[i].p_filesz, prot)) != 0)
431                                 goto fail;
432                         /*
433                          * Establish the base address if this is the
434                          * first segment.
435                          */
436                         if (numsegs == 0)
437                                 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
438                         numsegs++;
439                 }
440         }
441         *addr = base_addr;
442         *entry=(unsigned long)hdr->e_entry + rbase;
443
444 fail:
445         if (imgp->firstpage)
446                 exec_unmap_first_page(imgp);
447         if (imgp->image_header)
448                 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
449                         PAGE_SIZE);
450         if (nd->ni_vp)
451                 vrele(nd->ni_vp);
452
453         free(tempdata, M_TEMP);
454
455         return error;
456 }
457
458 /*
459  * non static, as it can be overridden by start_init()
460  */
461 int fallback_elf_brand = -1;
462 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
463                 &fallback_elf_brand, -1,
464                 "ELF brand of last resort");
465
466 static int
467 exec_elf_imgact(struct image_params *imgp)
468 {
469         const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
470         const Elf_Phdr *phdr;
471         Elf_Auxargs *elf_auxargs = NULL;
472         struct vmspace *vmspace;
473         vm_prot_t prot;
474         u_long text_size = 0, data_size = 0, total_size = 0;
475         u_long text_addr = 0, data_addr = 0;
476         u_long seg_size, seg_addr;
477         u_long addr, entry = 0, proghdr = 0;
478         int error, i;
479         const char *interp = NULL;
480         Elf_Brandinfo *brand_info;
481         char *path;
482
483         /*
484          * Do we have a valid ELF header ?
485          */
486         if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
487                 return -1;
488
489         /*
490          * From here on down, we return an errno, not -1, as we've
491          * detected an ELF file.
492          */
493
494         if ((hdr->e_phoff > PAGE_SIZE) ||
495             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
496                 /* Only support headers in first page for now */
497                 return ENOEXEC;
498         }
499         phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
500         
501         /*
502          * From this point on, we may have resources that need to be freed.
503          */
504
505         if ((error = exec_extract_strings(imgp)) != 0)
506                 goto fail;
507
508         exec_new_vmspace(imgp);
509
510         /*
511          * Yeah, I'm paranoid.  There is every reason in the world to get
512          * VTEXT now since from here on out, there are places we can have
513          * a context switch.  Better safe than sorry; I really don't want
514          * the file to change while it's being loaded.
515          */
516         simple_lock(&imgp->vp->v_interlock);
517         imgp->vp->v_flag |= VTEXT;
518         simple_unlock(&imgp->vp->v_interlock);
519
520         vmspace = imgp->proc->p_vmspace;
521
522         for (i = 0; i < hdr->e_phnum; i++) {
523                 switch(phdr[i].p_type) {
524
525                 case PT_LOAD:   /* Loadable segment */
526                         prot = 0;
527                         if (phdr[i].p_flags & PF_X)
528                                 prot |= VM_PROT_EXECUTE;
529                         if (phdr[i].p_flags & PF_W)
530                                 prot |= VM_PROT_WRITE;
531                         if (phdr[i].p_flags & PF_R)
532                                 prot |= VM_PROT_READ;
533
534                         if ((error = elf_load_section(imgp->proc,
535                                                      vmspace, imgp->vp,
536                                                      phdr[i].p_offset,
537                                                      (caddr_t)phdr[i].p_vaddr,
538                                                      phdr[i].p_memsz,
539                                                      phdr[i].p_filesz, prot)) != 0)
540                                 goto fail;
541
542                         seg_addr = trunc_page(phdr[i].p_vaddr);
543                         seg_size = round_page(phdr[i].p_memsz +
544                                 phdr[i].p_vaddr - seg_addr);
545
546                         /*
547                          * Is this .text or .data?  We can't use
548                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
549                          * alpha terribly and possibly does other bad
550                          * things so we stick to the old way of figuring
551                          * it out:  If the segment contains the program
552                          * entry point, it's a text segment, otherwise it
553                          * is a data segment.
554                          *
555                          * Note that obreak() assumes that data_addr + 
556                          * data_size == end of data load area, and the ELF
557                          * file format expects segments to be sorted by
558                          * address.  If multiple data segments exist, the
559                          * last one will be used.
560                          */
561                         if (hdr->e_entry >= phdr[i].p_vaddr &&
562                             hdr->e_entry < (phdr[i].p_vaddr +
563                             phdr[i].p_memsz)) {
564                                 text_size = seg_size;
565                                 text_addr = seg_addr;
566                                 entry = (u_long)hdr->e_entry;
567                         } else {
568                                 data_size = seg_size;
569                                 data_addr = seg_addr;
570                         }
571                         total_size += seg_size;
572
573                         /*
574                          * Check limits.  It should be safe to check the
575                          * limits after loading the segment since we do
576                          * not actually fault in all the segment's pages.
577                          */
578                         if (data_size >
579                             imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
580                             text_size > maxtsiz ||
581                             total_size >
582                             imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
583                                 error = ENOMEM;
584                                 goto fail;
585                         }
586                         break;
587                 case PT_INTERP: /* Path to interpreter */
588                         if (phdr[i].p_filesz > MAXPATHLEN ||
589                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
590                                 error = ENOEXEC;
591                                 goto fail;
592                         }
593                         interp = imgp->image_header + phdr[i].p_offset;
594                         break;
595                 case PT_PHDR:   /* Program header table info */
596                         proghdr = phdr[i].p_vaddr;
597                         break;
598                 default:
599                         break;
600                 }
601         }
602
603         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
604         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
605         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
606         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
607
608         addr = ELF_RTLD_ADDR(vmspace);
609
610         imgp->entry_addr = entry;
611
612         brand_info = NULL;
613
614         /* We support three types of branding -- (1) the ELF EI_OSABI field
615          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
616          * branding w/in the ELF header, and (3) path of the `interp_path'
617          * field.  We should also look for an ".note.ABI-tag" ELF section now
618          * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
619          */
620
621         /* If the executable has a brand, search for it in the brand list. */
622         if (brand_info == NULL) {
623                 for (i = 0;  i < MAX_BRANDS;  i++) {
624                         Elf_Brandinfo *bi = elf_brand_list[i];
625
626                         if (bi != NULL && 
627                             (hdr->e_ident[EI_OSABI] == bi->brand
628                             || 0 == 
629                             strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 
630                             bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
631                                 brand_info = bi;
632                                 break;
633                         }
634                 }
635         }
636
637         /* Lacking a known brand, search for a recognized interpreter. */
638         if (brand_info == NULL && interp != NULL) {
639                 for (i = 0;  i < MAX_BRANDS;  i++) {
640                         Elf_Brandinfo *bi = elf_brand_list[i];
641
642                         if (bi != NULL &&
643                             strcmp(interp, bi->interp_path) == 0) {
644                                 brand_info = bi;
645                                 break;
646                         }
647                 }
648         }
649
650         /* Lacking a recognized interpreter, try the default brand */
651         if (brand_info == NULL) {
652                 for (i = 0; i < MAX_BRANDS; i++) {
653                         Elf_Brandinfo *bi = elf_brand_list[i];
654
655                         if (bi != NULL && fallback_elf_brand == bi->brand) {
656                                 brand_info = bi;
657                                 break;
658                         }
659                 }
660         }
661
662         if (brand_info == NULL) {
663                 uprintf("ELF binary type \"%u\" not known.\n",
664                     hdr->e_ident[EI_OSABI]);
665                 error = ENOEXEC;
666                 goto fail;
667         }
668
669         imgp->proc->p_sysent = brand_info->sysvec;
670         if (interp != NULL) {
671                 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
672                 snprintf(path, MAXPATHLEN, "%s%s",
673                          brand_info->emul_path, interp);
674                 if ((error = elf_load_file(imgp->proc, path, &addr,
675                                            &imgp->entry_addr)) != 0) {
676                         if ((error = elf_load_file(imgp->proc, interp, &addr,
677                                                    &imgp->entry_addr)) != 0) {
678                                 uprintf("ELF interpreter %s not found\n", path);
679                                 free(path, M_TEMP);
680                                 goto fail;
681                         }
682                 }
683                 free(path, M_TEMP);
684         }
685
686         /*
687          * Construct auxargs table (used by the fixup routine)
688          */
689         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
690         elf_auxargs->execfd = -1;
691         elf_auxargs->phdr = proghdr;
692         elf_auxargs->phent = hdr->e_phentsize;
693         elf_auxargs->phnum = hdr->e_phnum;
694         elf_auxargs->pagesz = PAGE_SIZE;
695         elf_auxargs->base = addr;
696         elf_auxargs->flags = 0;
697         elf_auxargs->entry = entry;
698         elf_auxargs->trace = elf_trace;
699
700         imgp->auxargs = elf_auxargs;
701         imgp->interpreted = 0;
702
703 fail:
704         return error;
705 }
706
707 static int
708 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
709 {
710         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
711         register_t *pos;
712
713         pos = *stack_base + (imgp->argc + imgp->envc + 2);
714
715         if (args->trace) {
716                 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
717         }
718         if (args->execfd != -1) {
719                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
720         }
721         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
722         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
723         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
724         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
725         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
726         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
727         AUXARGS_ENTRY(pos, AT_BASE, args->base);
728         AUXARGS_ENTRY(pos, AT_NULL, 0);
729
730         free(imgp->auxargs, M_TEMP);
731         imgp->auxargs = NULL;
732
733         (*stack_base)--;
734         suword(*stack_base, (long) imgp->argc);
735         return 0;
736
737
738 /*
739  * Code for generating ELF core dumps.
740  */
741
742 typedef void (*segment_callback) __P((vm_map_entry_t, void *));
743
744 /* Closure for cb_put_phdr(). */
745 struct phdr_closure {
746         Elf_Phdr *phdr;         /* Program header to fill in */
747         Elf_Off offset;         /* Offset of segment in core file */
748 };
749
750 /* Closure for cb_size_segment(). */
751 struct sseg_closure {
752         int count;              /* Count of writable segments. */
753         size_t size;            /* Total size of all writable segments. */
754 };
755
756 static void cb_put_phdr __P((vm_map_entry_t, void *));
757 static void cb_size_segment __P((vm_map_entry_t, void *));
758 static void each_writable_segment __P((struct proc *, segment_callback,
759     void *));
760 static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *,
761     int, void *, size_t));
762 static void elf_puthdr __P((struct proc *, void *, size_t *,
763     const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int));
764 static void elf_putnote __P((void *, size_t *, const char *, int,
765     const void *, size_t));
766
767 extern int osreldate;
768
769 int
770 elf_coredump(p, vp, limit)
771         register struct proc *p;
772         register struct vnode *vp;
773         off_t limit;
774 {
775         register struct ucred *cred = p->p_ucred;
776         int error = 0;
777         struct sseg_closure seginfo;
778         void *hdr;
779         size_t hdrsize;
780
781         /* Size the program segments. */
782         seginfo.count = 0;
783         seginfo.size = 0;
784         each_writable_segment(p, cb_size_segment, &seginfo);
785
786         /*
787          * Calculate the size of the core file header area by making
788          * a dry run of generating it.  Nothing is written, but the
789          * size is calculated.
790          */
791         hdrsize = 0;
792         elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize,
793             (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
794             (const prpsinfo_t *)NULL, seginfo.count);
795
796         if (hdrsize + seginfo.size >= limit)
797                 return (EFAULT);
798
799         /*
800          * Allocate memory for building the header, fill it up,
801          * and write it out.
802          */
803         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
804         if (hdr == NULL) {
805                 return EINVAL;
806         }
807         error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize);
808
809         /* Write the contents of all of the writable segments. */
810         if (error == 0) {
811                 Elf_Phdr *php;
812                 off_t offset;
813                 int i;
814
815                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
816                 offset = hdrsize;
817                 for (i = 0;  i < seginfo.count;  i++) {
818                         error = vn_rdwr_inchunks(UIO_WRITE, vp, 
819                             (caddr_t)php->p_vaddr,
820                             php->p_filesz, offset, UIO_USERSPACE,
821                             IO_UNIT | IO_DIRECT | IO_CORE, cred, (int *)NULL, p);
822                         if (error != 0)
823                                 break;
824                         offset += php->p_filesz;
825                         php++;
826                 }
827         }
828         free(hdr, M_TEMP);
829         
830         return error;
831 }
832
833 /*
834  * A callback for each_writable_segment() to write out the segment's
835  * program header entry.
836  */
837 static void
838 cb_put_phdr(entry, closure)
839         vm_map_entry_t entry;
840         void *closure;
841 {
842         struct phdr_closure *phc = (struct phdr_closure *)closure;
843         Elf_Phdr *phdr = phc->phdr;
844
845         phc->offset = round_page(phc->offset);
846
847         phdr->p_type = PT_LOAD;
848         phdr->p_offset = phc->offset;
849         phdr->p_vaddr = entry->start;
850         phdr->p_paddr = 0;
851         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
852         phdr->p_align = PAGE_SIZE;
853         phdr->p_flags = 0;
854         if (entry->protection & VM_PROT_READ)
855                 phdr->p_flags |= PF_R;
856         if (entry->protection & VM_PROT_WRITE)
857                 phdr->p_flags |= PF_W;
858         if (entry->protection & VM_PROT_EXECUTE)
859                 phdr->p_flags |= PF_X;
860
861         phc->offset += phdr->p_filesz;
862         phc->phdr++;
863 }
864
865 /*
866  * A callback for each_writable_segment() to gather information about
867  * the number of segments and their total size.
868  */
869 static void
870 cb_size_segment(entry, closure)
871         vm_map_entry_t entry;
872         void *closure;
873 {
874         struct sseg_closure *ssc = (struct sseg_closure *)closure;
875
876         ssc->count++;
877         ssc->size += entry->end - entry->start;
878 }
879
880 /*
881  * For each writable segment in the process's memory map, call the given
882  * function with a pointer to the map entry and some arbitrary
883  * caller-supplied data.
884  */
885 static void
886 each_writable_segment(p, func, closure)
887         struct proc *p;
888         segment_callback func;
889         void *closure;
890 {
891         vm_map_t map = &p->p_vmspace->vm_map;
892         vm_map_entry_t entry;
893
894         for (entry = map->header.next;  entry != &map->header;
895             entry = entry->next) {
896                 vm_object_t obj;
897
898                 /*
899                  * Don't dump inaccessible mappings, deal with legacy
900                  * coredump mode.
901                  *
902                  * Note that read-only segments related to the elf binary
903                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
904                  * need to arbitrarily ignore such segments.
905                  */
906                 if (elf_legacy_coredump) {
907                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
908                                 continue;
909                 } else {
910                         if ((entry->protection & VM_PROT_ALL) == 0)
911                                 continue;
912                 }
913
914                 /*
915                  * Dont include memory segment in the coredump if
916                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
917                  * madvise(2).  Do not dump submaps (i.e. parts of the
918                  * kernel map).
919                  */
920                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
921                         continue;
922
923                 if ((obj = entry->object.vm_object) == NULL)
924                         continue;
925
926                 /* Find the deepest backing object. */
927                 while (obj->backing_object != NULL)
928                         obj = obj->backing_object;
929
930                 /* Ignore memory-mapped devices and such things. */
931                 if (obj->type != OBJT_DEFAULT &&
932                     obj->type != OBJT_SWAP &&
933                     obj->type != OBJT_VNODE)
934                         continue;
935
936                 (*func)(entry, closure);
937         }
938 }
939
940 /*
941  * Write the core file header to the file, including padding up to
942  * the page boundary.
943  */
944 static int
945 elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize)
946         struct proc *p;
947         struct vnode *vp;
948         struct ucred *cred;
949         int numsegs;
950         size_t hdrsize;
951         void *hdr;
952 {
953         struct {
954                 prstatus_t status;
955                 prfpregset_t fpregset;
956                 prpsinfo_t psinfo;
957         } *tempdata;
958         size_t off;
959         prstatus_t *status;
960         prfpregset_t *fpregset;
961         prpsinfo_t *psinfo;
962
963         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
964         status = &tempdata->status;
965         fpregset = &tempdata->fpregset;
966         psinfo = &tempdata->psinfo;
967
968         /* Gather the information for the header. */
969         status->pr_version = PRSTATUS_VERSION;
970         status->pr_statussz = sizeof(prstatus_t);
971         status->pr_gregsetsz = sizeof(gregset_t);
972         status->pr_fpregsetsz = sizeof(fpregset_t);
973         status->pr_osreldate = osreldate;
974         status->pr_cursig = p->p_sig;
975         status->pr_pid = p->p_pid;
976         fill_regs(p, &status->pr_reg);
977
978         fill_fpregs(p, fpregset);
979
980         psinfo->pr_version = PRPSINFO_VERSION;
981         psinfo->pr_psinfosz = sizeof(prpsinfo_t);
982         strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1);
983
984         /* XXX - We don't fill in the command line arguments properly yet. */
985         strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
986
987         /* Fill in the header. */
988         bzero(hdr, hdrsize);
989         off = 0;
990         elf_puthdr(p, hdr, &off, status, fpregset, psinfo, numsegs);
991
992         free(tempdata, M_TEMP);
993
994         /* Write it to the core file. */
995         return vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
996             UIO_SYSSPACE, IO_UNIT | IO_DIRECT | IO_CORE, cred, NULL, p);
997 }
998
999 static void
1000 elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
1001     const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
1002 {
1003         size_t ehoff;
1004         size_t phoff;
1005         size_t noteoff;
1006         size_t notesz;
1007
1008         ehoff = *off;
1009         *off += sizeof(Elf_Ehdr);
1010
1011         phoff = *off;
1012         *off += (numsegs + 1) * sizeof(Elf_Phdr);
1013
1014         noteoff = *off;
1015         elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
1016             sizeof *status);
1017         elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1018             sizeof *fpregset);
1019         elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1020             sizeof *psinfo);
1021         notesz = *off - noteoff;
1022
1023         /* Align up to a page boundary for the program segments. */
1024         *off = round_page(*off);
1025
1026         if (dst != NULL) {
1027                 Elf_Ehdr *ehdr;
1028                 Elf_Phdr *phdr;
1029                 struct phdr_closure phc;
1030
1031                 /*
1032                  * Fill in the ELF header.
1033                  */
1034                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1035                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1036                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1037                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1038                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1039                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1040                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1041                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1042                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1043                 ehdr->e_ident[EI_ABIVERSION] = 0;
1044                 ehdr->e_ident[EI_PAD] = 0;
1045                 ehdr->e_type = ET_CORE;
1046                 ehdr->e_machine = ELF_ARCH;
1047                 ehdr->e_version = EV_CURRENT;
1048                 ehdr->e_entry = 0;
1049                 ehdr->e_phoff = phoff;
1050                 ehdr->e_flags = 0;
1051                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1052                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1053                 ehdr->e_phnum = numsegs + 1;
1054                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1055                 ehdr->e_shnum = 0;
1056                 ehdr->e_shstrndx = SHN_UNDEF;
1057
1058                 /*
1059                  * Fill in the program header entries.
1060                  */
1061                 phdr = (Elf_Phdr *)((char *)dst + phoff);
1062
1063                 /* The note segement. */
1064                 phdr->p_type = PT_NOTE;
1065                 phdr->p_offset = noteoff;
1066                 phdr->p_vaddr = 0;
1067                 phdr->p_paddr = 0;
1068                 phdr->p_filesz = notesz;
1069                 phdr->p_memsz = 0;
1070                 phdr->p_flags = 0;
1071                 phdr->p_align = 0;
1072                 phdr++;
1073
1074                 /* All the writable segments from the program. */
1075                 phc.phdr = phdr;
1076                 phc.offset = *off;
1077                 each_writable_segment(p, cb_put_phdr, &phc);
1078         }
1079 }
1080
1081 static void
1082 elf_putnote(void *dst, size_t *off, const char *name, int type,
1083     const void *desc, size_t descsz)
1084 {
1085         Elf_Note note;
1086
1087         note.n_namesz = strlen(name) + 1;
1088         note.n_descsz = descsz;
1089         note.n_type = type;
1090         if (dst != NULL)
1091                 bcopy(&note, (char *)dst + *off, sizeof note);
1092         *off += sizeof note;
1093         if (dst != NULL)
1094                 bcopy(name, (char *)dst + *off, note.n_namesz);
1095         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1096         if (dst != NULL)
1097                 bcopy(desc, (char *)dst + *off, note.n_descsz);
1098         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1099 }
1100
1101 /*
1102  * Tell kern_execve.c about it, with a little help from the linker.
1103  */
1104 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1105 EXEC_SET(elf, elf_execsw);