e1000: 3 new driver stats for managability testing
[linux-2.6] / fs / binfmt_elf_fdpic.c
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2  *
3  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  * Derived from binfmt_elf.c
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  */
12
13 #include <linux/module.h>
14
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/smp_lock.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/param.h>
40 #include <asm/pgalloc.h>
41
42 typedef char *elf_caddr_t;
43
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55
56 MODULE_LICENSE("GPL");
57
58 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61                               struct mm_struct *, const char *);
62
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64                                    struct elf_fdpic_params *,
65                                    struct elf_fdpic_params *);
66
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
69                                             unsigned long *);
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71                                                    struct file *,
72                                                    struct mm_struct *);
73 #endif
74
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76                                              struct file *, struct mm_struct *);
77
78 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
79 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *);
80 #endif
81
82 static struct linux_binfmt elf_fdpic_format = {
83         .module         = THIS_MODULE,
84         .load_binary    = load_elf_fdpic_binary,
85 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
86         .core_dump      = elf_fdpic_core_dump,
87 #endif
88         .min_coredump   = ELF_EXEC_PAGESIZE,
89 };
90
91 static int __init init_elf_fdpic_binfmt(void)
92 {
93         return register_binfmt(&elf_fdpic_format);
94 }
95
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98         unregister_binfmt(&elf_fdpic_format);
99 }
100
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103
104 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
105 {
106         if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107                 return 0;
108         if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109                 return 0;
110         if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
111                 return 0;
112         if (!file->f_op || !file->f_op->mmap)
113                 return 0;
114         return 1;
115 }
116
117 /*****************************************************************************/
118 /*
119  * read the program headers table into memory
120  */
121 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
122                                  struct file *file)
123 {
124         struct elf32_phdr *phdr;
125         unsigned long size;
126         int retval, loop;
127
128         if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
129                 return -ENOMEM;
130         if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
131                 return -ENOMEM;
132
133         size = params->hdr.e_phnum * sizeof(struct elf_phdr);
134         params->phdrs = kmalloc(size, GFP_KERNEL);
135         if (!params->phdrs)
136                 return -ENOMEM;
137
138         retval = kernel_read(file, params->hdr.e_phoff,
139                              (char *) params->phdrs, size);
140         if (retval < 0)
141                 return retval;
142
143         /* determine stack size for this binary */
144         phdr = params->phdrs;
145         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
146                 if (phdr->p_type != PT_GNU_STACK)
147                         continue;
148
149                 if (phdr->p_flags & PF_X)
150                         params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
151                 else
152                         params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
153
154                 params->stack_size = phdr->p_memsz;
155                 break;
156         }
157
158         return 0;
159 }
160
161 /*****************************************************************************/
162 /*
163  * load an fdpic binary into various bits of memory
164  */
165 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
166                                  struct pt_regs *regs)
167 {
168         struct elf_fdpic_params exec_params, interp_params;
169         struct elf_phdr *phdr;
170         unsigned long stack_size, entryaddr;
171 #ifndef CONFIG_MMU
172         unsigned long fullsize;
173 #endif
174 #ifdef ELF_FDPIC_PLAT_INIT
175         unsigned long dynaddr;
176 #endif
177         struct file *interpreter = NULL; /* to shut gcc up */
178         char *interpreter_name = NULL;
179         int executable_stack;
180         int retval, i;
181
182         memset(&exec_params, 0, sizeof(exec_params));
183         memset(&interp_params, 0, sizeof(interp_params));
184
185         exec_params.hdr = *(struct elfhdr *) bprm->buf;
186         exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
187
188         /* check that this is a binary we know how to deal with */
189         retval = -ENOEXEC;
190         if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
191                 goto error;
192
193         /* read the program header table */
194         retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
195         if (retval < 0)
196                 goto error;
197
198         /* scan for a program header that specifies an interpreter */
199         phdr = exec_params.phdrs;
200
201         for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
202                 switch (phdr->p_type) {
203                 case PT_INTERP:
204                         retval = -ENOMEM;
205                         if (phdr->p_filesz > PATH_MAX)
206                                 goto error;
207                         retval = -ENOENT;
208                         if (phdr->p_filesz < 2)
209                                 goto error;
210
211                         /* read the name of the interpreter into memory */
212                         interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
213                         if (!interpreter_name)
214                                 goto error;
215
216                         retval = kernel_read(bprm->file,
217                                              phdr->p_offset,
218                                              interpreter_name,
219                                              phdr->p_filesz);
220                         if (retval < 0)
221                                 goto error;
222
223                         retval = -ENOENT;
224                         if (interpreter_name[phdr->p_filesz - 1] != '\0')
225                                 goto error;
226
227                         kdebug("Using ELF interpreter %s", interpreter_name);
228
229                         /* replace the program with the interpreter */
230                         interpreter = open_exec(interpreter_name);
231                         retval = PTR_ERR(interpreter);
232                         if (IS_ERR(interpreter)) {
233                                 interpreter = NULL;
234                                 goto error;
235                         }
236
237                         retval = kernel_read(interpreter, 0, bprm->buf,
238                                              BINPRM_BUF_SIZE);
239                         if (retval < 0)
240                                 goto error;
241
242                         interp_params.hdr = *((struct elfhdr *) bprm->buf);
243                         break;
244
245                 case PT_LOAD:
246 #ifdef CONFIG_MMU
247                         if (exec_params.load_addr == 0)
248                                 exec_params.load_addr = phdr->p_vaddr;
249 #endif
250                         break;
251                 }
252
253         }
254
255         if (elf_check_const_displacement(&exec_params.hdr))
256                 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
257
258         /* perform insanity checks on the interpreter */
259         if (interpreter_name) {
260                 retval = -ELIBBAD;
261                 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
262                         goto error;
263
264                 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
265
266                 /* read the interpreter's program header table */
267                 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
268                 if (retval < 0)
269                         goto error;
270         }
271
272         stack_size = exec_params.stack_size;
273         if (stack_size < interp_params.stack_size)
274                 stack_size = interp_params.stack_size;
275
276         if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
277                 executable_stack = EXSTACK_ENABLE_X;
278         else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
279                 executable_stack = EXSTACK_DISABLE_X;
280         else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
281                 executable_stack = EXSTACK_ENABLE_X;
282         else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
283                 executable_stack = EXSTACK_DISABLE_X;
284         else
285                 executable_stack = EXSTACK_DEFAULT;
286
287         retval = -ENOEXEC;
288         if (stack_size == 0)
289                 goto error;
290
291         if (elf_check_const_displacement(&interp_params.hdr))
292                 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
293
294         /* flush all traces of the currently running executable */
295         retval = flush_old_exec(bprm);
296         if (retval)
297                 goto error;
298
299         /* there's now no turning back... the old userspace image is dead,
300          * defunct, deceased, etc. after this point we have to exit via
301          * error_kill */
302         set_personality(PER_LINUX_FDPIC);
303         set_binfmt(&elf_fdpic_format);
304
305         current->mm->start_code = 0;
306         current->mm->end_code = 0;
307         current->mm->start_stack = 0;
308         current->mm->start_data = 0;
309         current->mm->end_data = 0;
310         current->mm->context.exec_fdpic_loadmap = 0;
311         current->mm->context.interp_fdpic_loadmap = 0;
312
313         current->flags &= ~PF_FORKNOEXEC;
314
315 #ifdef CONFIG_MMU
316         elf_fdpic_arch_lay_out_mm(&exec_params,
317                                   &interp_params,
318                                   &current->mm->start_stack,
319                                   &current->mm->start_brk);
320
321         retval = setup_arg_pages(bprm, current->mm->start_stack,
322                                  executable_stack);
323         if (retval < 0) {
324                 send_sig(SIGKILL, current, 0);
325                 goto error_kill;
326         }
327 #endif
328
329         /* load the executable and interpreter into memory */
330         retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
331                                     "executable");
332         if (retval < 0)
333                 goto error_kill;
334
335         if (interpreter_name) {
336                 retval = elf_fdpic_map_file(&interp_params, interpreter,
337                                             current->mm, "interpreter");
338                 if (retval < 0) {
339                         printk(KERN_ERR "Unable to load interpreter\n");
340                         goto error_kill;
341                 }
342
343                 allow_write_access(interpreter);
344                 fput(interpreter);
345                 interpreter = NULL;
346         }
347
348 #ifdef CONFIG_MMU
349         if (!current->mm->start_brk)
350                 current->mm->start_brk = current->mm->end_data;
351
352         current->mm->brk = current->mm->start_brk =
353                 PAGE_ALIGN(current->mm->start_brk);
354
355 #else
356         /* create a stack and brk area big enough for everyone
357          * - the brk heap starts at the bottom and works up
358          * - the stack starts at the top and works down
359          */
360         stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
361         if (stack_size < PAGE_SIZE * 2)
362                 stack_size = PAGE_SIZE * 2;
363
364         down_write(&current->mm->mmap_sem);
365         current->mm->start_brk = do_mmap(NULL, 0, stack_size,
366                                          PROT_READ | PROT_WRITE | PROT_EXEC,
367                                          MAP_PRIVATE | MAP_ANON | MAP_GROWSDOWN,
368                                          0);
369
370         if (IS_ERR_VALUE(current->mm->start_brk)) {
371                 up_write(&current->mm->mmap_sem);
372                 retval = current->mm->start_brk;
373                 current->mm->start_brk = 0;
374                 goto error_kill;
375         }
376
377         /* expand the stack mapping to use up the entire allocation granule */
378         fullsize = ksize((char *) current->mm->start_brk);
379         if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
380                                     fullsize, 0, 0)))
381                 stack_size = fullsize;
382         up_write(&current->mm->mmap_sem);
383
384         current->mm->brk = current->mm->start_brk;
385         current->mm->context.end_brk = current->mm->start_brk;
386         current->mm->context.end_brk +=
387                 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
388         current->mm->start_stack = current->mm->start_brk + stack_size;
389 #endif
390
391         compute_creds(bprm);
392         current->flags &= ~PF_FORKNOEXEC;
393         if (create_elf_fdpic_tables(bprm, current->mm,
394                                     &exec_params, &interp_params) < 0)
395                 goto error_kill;
396
397         kdebug("- start_code  %lx", current->mm->start_code);
398         kdebug("- end_code    %lx", current->mm->end_code);
399         kdebug("- start_data  %lx", current->mm->start_data);
400         kdebug("- end_data    %lx", current->mm->end_data);
401         kdebug("- start_brk   %lx", current->mm->start_brk);
402         kdebug("- brk         %lx", current->mm->brk);
403         kdebug("- start_stack %lx", current->mm->start_stack);
404
405 #ifdef ELF_FDPIC_PLAT_INIT
406         /*
407          * The ABI may specify that certain registers be set up in special
408          * ways (on i386 %edx is the address of a DT_FINI function, for
409          * example.  This macro performs whatever initialization to
410          * the regs structure is required.
411          */
412         dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
413         ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
414                             dynaddr);
415 #endif
416
417         /* everything is now ready... get the userspace context ready to roll */
418         entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
419         start_thread(regs, entryaddr, current->mm->start_stack);
420
421         if (unlikely(current->ptrace & PT_PTRACED)) {
422                 if (current->ptrace & PT_TRACE_EXEC)
423                         ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
424                 else
425                         send_sig(SIGTRAP, current, 0);
426         }
427
428         retval = 0;
429
430 error:
431         if (interpreter) {
432                 allow_write_access(interpreter);
433                 fput(interpreter);
434         }
435         kfree(interpreter_name);
436         kfree(exec_params.phdrs);
437         kfree(exec_params.loadmap);
438         kfree(interp_params.phdrs);
439         kfree(interp_params.loadmap);
440         return retval;
441
442         /* unrecoverable error - kill the process */
443 error_kill:
444         send_sig(SIGSEGV, current, 0);
445         goto error;
446
447 }
448
449 /*****************************************************************************/
450 /*
451  * present useful information to the program
452  */
453 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
454                                    struct mm_struct *mm,
455                                    struct elf_fdpic_params *exec_params,
456                                    struct elf_fdpic_params *interp_params)
457 {
458         unsigned long sp, csp, nitems;
459         elf_caddr_t __user *argv, *envp;
460         size_t platform_len = 0, len;
461         char *k_platform;
462         char __user *u_platform, *p;
463         long hwcap;
464         int loop;
465
466         /* we're going to shovel a whole load of stuff onto the stack */
467 #ifdef CONFIG_MMU
468         sp = bprm->p;
469 #else
470         sp = mm->start_stack;
471
472         /* stack the program arguments and environment */
473         if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
474                 return -EFAULT;
475 #endif
476
477         /* get hold of platform and hardware capabilities masks for the machine
478          * we are running on.  In some cases (Sparc), this info is impossible
479          * to get, in others (i386) it is merely difficult.
480          */
481         hwcap = ELF_HWCAP;
482         k_platform = ELF_PLATFORM;
483         u_platform = NULL;
484
485         if (k_platform) {
486                 platform_len = strlen(k_platform) + 1;
487                 sp -= platform_len;
488                 u_platform = (char __user *) sp;
489                 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
490                         return -EFAULT;
491         }
492
493 #if defined(__i386__) && defined(CONFIG_SMP)
494         /* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
495          * by the processes running on the same package. One thing we can do is
496          * to shuffle the initial stack for them.
497          *
498          * the conditionals here are unneeded, but kept in to make the code
499          * behaviour the same as pre change unless we have hyperthreaded
500          * processors. This keeps Mr Marcelo Person happier but should be
501          * removed for 2.5
502          */
503         if (smp_num_siblings > 1)
504                 sp = sp - ((current->pid % 64) << 7);
505 #endif
506
507         sp &= ~7UL;
508
509         /* stack the load map(s) */
510         len = sizeof(struct elf32_fdpic_loadmap);
511         len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
512         sp = (sp - len) & ~7UL;
513         exec_params->map_addr = sp;
514
515         if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
516                 return -EFAULT;
517
518         current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
519
520         if (interp_params->loadmap) {
521                 len = sizeof(struct elf32_fdpic_loadmap);
522                 len += sizeof(struct elf32_fdpic_loadseg) *
523                         interp_params->loadmap->nsegs;
524                 sp = (sp - len) & ~7UL;
525                 interp_params->map_addr = sp;
526
527                 if (copy_to_user((void __user *) sp, interp_params->loadmap,
528                                  len) != 0)
529                         return -EFAULT;
530
531                 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
532         }
533
534         /* force 16 byte _final_ alignment here for generality */
535 #define DLINFO_ITEMS 13
536
537         nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0);
538 #ifdef DLINFO_ARCH_ITEMS
539         nitems += DLINFO_ARCH_ITEMS;
540 #endif
541
542         csp = sp;
543         sp -= nitems * 2 * sizeof(unsigned long);
544         sp -= (bprm->envc + 1) * sizeof(char *);        /* envv[] */
545         sp -= (bprm->argc + 1) * sizeof(char *);        /* argv[] */
546         sp -= 1 * sizeof(unsigned long);                /* argc */
547
548         csp -= sp & 15UL;
549         sp -= sp & 15UL;
550
551         /* put the ELF interpreter info on the stack */
552 #define NEW_AUX_ENT(nr, id, val)                                        \
553         do {                                                            \
554                 struct { unsigned long _id, _val; } __user *ent;        \
555                                                                         \
556                 ent = (void __user *) csp;                              \
557                 __put_user((id), &ent[nr]._id);                         \
558                 __put_user((val), &ent[nr]._val);                       \
559         } while (0)
560
561         csp -= 2 * sizeof(unsigned long);
562         NEW_AUX_ENT(0, AT_NULL, 0);
563         if (k_platform) {
564                 csp -= 2 * sizeof(unsigned long);
565                 NEW_AUX_ENT(0, AT_PLATFORM,
566                             (elf_addr_t) (unsigned long) u_platform);
567         }
568
569         csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
570         NEW_AUX_ENT( 0, AT_HWCAP,       hwcap);
571         NEW_AUX_ENT( 1, AT_PAGESZ,      PAGE_SIZE);
572         NEW_AUX_ENT( 2, AT_CLKTCK,      CLOCKS_PER_SEC);
573         NEW_AUX_ENT( 3, AT_PHDR,        exec_params->ph_addr);
574         NEW_AUX_ENT( 4, AT_PHENT,       sizeof(struct elf_phdr));
575         NEW_AUX_ENT( 5, AT_PHNUM,       exec_params->hdr.e_phnum);
576         NEW_AUX_ENT( 6, AT_BASE,        interp_params->elfhdr_addr);
577         NEW_AUX_ENT( 7, AT_FLAGS,       0);
578         NEW_AUX_ENT( 8, AT_ENTRY,       exec_params->entry_addr);
579         NEW_AUX_ENT( 9, AT_UID,         (elf_addr_t) current->uid);
580         NEW_AUX_ENT(10, AT_EUID,        (elf_addr_t) current->euid);
581         NEW_AUX_ENT(11, AT_GID,         (elf_addr_t) current->gid);
582         NEW_AUX_ENT(12, AT_EGID,        (elf_addr_t) current->egid);
583
584 #ifdef ARCH_DLINFO
585         /* ARCH_DLINFO must come last so platform specific code can enforce
586          * special alignment requirements on the AUXV if necessary (eg. PPC).
587          */
588         ARCH_DLINFO;
589 #endif
590 #undef NEW_AUX_ENT
591
592         /* allocate room for argv[] and envv[] */
593         csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
594         envp = (elf_caddr_t __user *) csp;
595         csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
596         argv = (elf_caddr_t __user *) csp;
597
598         /* stack argc */
599         csp -= sizeof(unsigned long);
600         __put_user(bprm->argc, (unsigned long __user *) csp);
601
602         BUG_ON(csp != sp);
603
604         /* fill in the argv[] array */
605 #ifdef CONFIG_MMU
606         current->mm->arg_start = bprm->p;
607 #else
608         current->mm->arg_start = current->mm->start_stack -
609                 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
610 #endif
611
612         p = (char __user *) current->mm->arg_start;
613         for (loop = bprm->argc; loop > 0; loop--) {
614                 __put_user((elf_caddr_t) p, argv++);
615                 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES);
616                 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES)
617                         return -EINVAL;
618                 p += len;
619         }
620         __put_user(NULL, argv);
621         current->mm->arg_end = (unsigned long) p;
622
623         /* fill in the envv[] array */
624         current->mm->env_start = (unsigned long) p;
625         for (loop = bprm->envc; loop > 0; loop--) {
626                 __put_user((elf_caddr_t)(unsigned long) p, envp++);
627                 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES);
628                 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES)
629                         return -EINVAL;
630                 p += len;
631         }
632         __put_user(NULL, envp);
633         current->mm->env_end = (unsigned long) p;
634
635         mm->start_stack = (unsigned long) sp;
636         return 0;
637 }
638
639 /*****************************************************************************/
640 /*
641  * transfer the program arguments and environment from the holding pages onto
642  * the stack
643  */
644 #ifndef CONFIG_MMU
645 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
646                                             unsigned long *_sp)
647 {
648         unsigned long index, stop, sp;
649         char *src;
650         int ret = 0;
651
652         stop = bprm->p >> PAGE_SHIFT;
653         sp = *_sp;
654
655         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
656                 src = kmap(bprm->page[index]);
657                 sp -= PAGE_SIZE;
658                 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
659                         ret = -EFAULT;
660                 kunmap(bprm->page[index]);
661                 if (ret < 0)
662                         goto out;
663         }
664
665         *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
666
667 out:
668         return ret;
669 }
670 #endif
671
672 /*****************************************************************************/
673 /*
674  * load the appropriate binary image (executable or interpreter) into memory
675  * - we assume no MMU is available
676  * - if no other PIC bits are set in params->hdr->e_flags
677  *   - we assume that the LOADable segments in the binary are independently relocatable
678  *   - we assume R/O executable segments are shareable
679  * - else
680  *   - we assume the loadable parts of the image to require fixed displacement
681  *   - the image is not shareable
682  */
683 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
684                               struct file *file,
685                               struct mm_struct *mm,
686                               const char *what)
687 {
688         struct elf32_fdpic_loadmap *loadmap;
689 #ifdef CONFIG_MMU
690         struct elf32_fdpic_loadseg *mseg;
691 #endif
692         struct elf32_fdpic_loadseg *seg;
693         struct elf32_phdr *phdr;
694         unsigned long load_addr, stop;
695         unsigned nloads, tmp;
696         size_t size;
697         int loop, ret;
698
699         /* allocate a load map table */
700         nloads = 0;
701         for (loop = 0; loop < params->hdr.e_phnum; loop++)
702                 if (params->phdrs[loop].p_type == PT_LOAD)
703                         nloads++;
704
705         if (nloads == 0)
706                 return -ELIBBAD;
707
708         size = sizeof(*loadmap) + nloads * sizeof(*seg);
709         loadmap = kzalloc(size, GFP_KERNEL);
710         if (!loadmap)
711                 return -ENOMEM;
712
713         params->loadmap = loadmap;
714
715         loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
716         loadmap->nsegs = nloads;
717
718         load_addr = params->load_addr;
719         seg = loadmap->segs;
720
721         /* map the requested LOADs into the memory space */
722         switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
723         case ELF_FDPIC_FLAG_CONSTDISP:
724         case ELF_FDPIC_FLAG_CONTIGUOUS:
725 #ifndef CONFIG_MMU
726                 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
727                 if (ret < 0)
728                         return ret;
729                 break;
730 #endif
731         default:
732                 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
733                 if (ret < 0)
734                         return ret;
735                 break;
736         }
737
738         /* map the entry point */
739         if (params->hdr.e_entry) {
740                 seg = loadmap->segs;
741                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
742                         if (params->hdr.e_entry >= seg->p_vaddr &&
743                             params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
744                                 params->entry_addr =
745                                         (params->hdr.e_entry - seg->p_vaddr) +
746                                         seg->addr;
747                                 break;
748                         }
749                 }
750         }
751
752         /* determine where the program header table has wound up if mapped */
753         stop = params->hdr.e_phoff;
754         stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
755         phdr = params->phdrs;
756
757         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
758                 if (phdr->p_type != PT_LOAD)
759                         continue;
760
761                 if (phdr->p_offset > params->hdr.e_phoff ||
762                     phdr->p_offset + phdr->p_filesz < stop)
763                         continue;
764
765                 seg = loadmap->segs;
766                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
767                         if (phdr->p_vaddr >= seg->p_vaddr &&
768                             phdr->p_vaddr + phdr->p_filesz <=
769                             seg->p_vaddr + seg->p_memsz) {
770                                 params->ph_addr =
771                                         (phdr->p_vaddr - seg->p_vaddr) +
772                                         seg->addr +
773                                         params->hdr.e_phoff - phdr->p_offset;
774                                 break;
775                         }
776                 }
777                 break;
778         }
779
780         /* determine where the dynamic section has wound up if there is one */
781         phdr = params->phdrs;
782         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
783                 if (phdr->p_type != PT_DYNAMIC)
784                         continue;
785
786                 seg = loadmap->segs;
787                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
788                         if (phdr->p_vaddr >= seg->p_vaddr &&
789                             phdr->p_vaddr + phdr->p_memsz <=
790                             seg->p_vaddr + seg->p_memsz) {
791                                 params->dynamic_addr =
792                                         (phdr->p_vaddr - seg->p_vaddr) +
793                                         seg->addr;
794
795                                 /* check the dynamic section contains at least
796                                  * one item, and that the last item is a NULL
797                                  * entry */
798                                 if (phdr->p_memsz == 0 ||
799                                     phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
800                                         goto dynamic_error;
801
802                                 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
803                                 if (((Elf32_Dyn *)
804                                      params->dynamic_addr)[tmp - 1].d_tag != 0)
805                                         goto dynamic_error;
806                                 break;
807                         }
808                 }
809                 break;
810         }
811
812         /* now elide adjacent segments in the load map on MMU linux
813          * - on uClinux the holes between may actually be filled with system
814          *   stuff or stuff from other processes
815          */
816 #ifdef CONFIG_MMU
817         nloads = loadmap->nsegs;
818         mseg = loadmap->segs;
819         seg = mseg + 1;
820         for (loop = 1; loop < nloads; loop++) {
821                 /* see if we have a candidate for merging */
822                 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
823                         load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
824                         if (load_addr == (seg->addr & PAGE_MASK)) {
825                                 mseg->p_memsz +=
826                                         load_addr -
827                                         (mseg->addr + mseg->p_memsz);
828                                 mseg->p_memsz += seg->addr & ~PAGE_MASK;
829                                 mseg->p_memsz += seg->p_memsz;
830                                 loadmap->nsegs--;
831                                 continue;
832                         }
833                 }
834
835                 mseg++;
836                 if (mseg != seg)
837                         *mseg = *seg;
838         }
839 #endif
840
841         kdebug("Mapped Object [%s]:", what);
842         kdebug("- elfhdr   : %lx", params->elfhdr_addr);
843         kdebug("- entry    : %lx", params->entry_addr);
844         kdebug("- PHDR[]   : %lx", params->ph_addr);
845         kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
846         seg = loadmap->segs;
847         for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
848                 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
849                        loop,
850                        seg->addr, seg->addr + seg->p_memsz - 1,
851                        seg->p_vaddr, seg->p_memsz);
852
853         return 0;
854
855 dynamic_error:
856         printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
857                what, file->f_path.dentry->d_inode->i_ino);
858         return -ELIBBAD;
859 }
860
861 /*****************************************************************************/
862 /*
863  * map a file with constant displacement under uClinux
864  */
865 #ifndef CONFIG_MMU
866 static int elf_fdpic_map_file_constdisp_on_uclinux(
867         struct elf_fdpic_params *params,
868         struct file *file,
869         struct mm_struct *mm)
870 {
871         struct elf32_fdpic_loadseg *seg;
872         struct elf32_phdr *phdr;
873         unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
874         loff_t fpos;
875         int loop, ret;
876
877         load_addr = params->load_addr;
878         seg = params->loadmap->segs;
879
880         /* determine the bounds of the contiguous overall allocation we must
881          * make */
882         phdr = params->phdrs;
883         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
884                 if (params->phdrs[loop].p_type != PT_LOAD)
885                         continue;
886
887                 if (base > phdr->p_vaddr)
888                         base = phdr->p_vaddr;
889                 if (top < phdr->p_vaddr + phdr->p_memsz)
890                         top = phdr->p_vaddr + phdr->p_memsz;
891         }
892
893         /* allocate one big anon block for everything */
894         mflags = MAP_PRIVATE;
895         if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
896                 mflags |= MAP_EXECUTABLE;
897
898         down_write(&mm->mmap_sem);
899         maddr = do_mmap(NULL, load_addr, top - base,
900                         PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
901         up_write(&mm->mmap_sem);
902         if (IS_ERR_VALUE(maddr))
903                 return (int) maddr;
904
905         if (load_addr != 0)
906                 load_addr += PAGE_ALIGN(top - base);
907
908         /* and then load the file segments into it */
909         phdr = params->phdrs;
910         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
911                 if (params->phdrs[loop].p_type != PT_LOAD)
912                         continue;
913
914                 fpos = phdr->p_offset;
915
916                 seg->addr = maddr + (phdr->p_vaddr - base);
917                 seg->p_vaddr = phdr->p_vaddr;
918                 seg->p_memsz = phdr->p_memsz;
919
920                 ret = file->f_op->read(file, (void *) seg->addr,
921                                        phdr->p_filesz, &fpos);
922                 if (ret < 0)
923                         return ret;
924
925                 /* map the ELF header address if in this segment */
926                 if (phdr->p_offset == 0)
927                         params->elfhdr_addr = seg->addr;
928
929                 /* clear any space allocated but not loaded */
930                 if (phdr->p_filesz < phdr->p_memsz)
931                         clear_user((void *) (seg->addr + phdr->p_filesz),
932                                    phdr->p_memsz - phdr->p_filesz);
933
934                 if (mm) {
935                         if (phdr->p_flags & PF_X) {
936                                 mm->start_code = seg->addr;
937                                 mm->end_code = seg->addr + phdr->p_memsz;
938                         } else if (!mm->start_data) {
939                                 mm->start_data = seg->addr;
940 #ifndef CONFIG_MMU
941                                 mm->end_data = seg->addr + phdr->p_memsz;
942 #endif
943                         }
944
945 #ifdef CONFIG_MMU
946                         if (seg->addr + phdr->p_memsz > mm->end_data)
947                                 mm->end_data = seg->addr + phdr->p_memsz;
948 #endif
949                 }
950
951                 seg++;
952         }
953
954         return 0;
955 }
956 #endif
957
958 /*****************************************************************************/
959 /*
960  * map a binary by direct mmap() of the individual PT_LOAD segments
961  */
962 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
963                                              struct file *file,
964                                              struct mm_struct *mm)
965 {
966         struct elf32_fdpic_loadseg *seg;
967         struct elf32_phdr *phdr;
968         unsigned long load_addr, delta_vaddr;
969         int loop, dvset;
970
971         load_addr = params->load_addr;
972         delta_vaddr = 0;
973         dvset = 0;
974
975         seg = params->loadmap->segs;
976
977         /* deal with each load segment separately */
978         phdr = params->phdrs;
979         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
980                 unsigned long maddr, disp, excess, excess1;
981                 int prot = 0, flags;
982
983                 if (phdr->p_type != PT_LOAD)
984                         continue;
985
986                 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
987                        (unsigned long) phdr->p_vaddr,
988                        (unsigned long) phdr->p_offset,
989                        (unsigned long) phdr->p_filesz,
990                        (unsigned long) phdr->p_memsz);
991
992                 /* determine the mapping parameters */
993                 if (phdr->p_flags & PF_R) prot |= PROT_READ;
994                 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
995                 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
996
997                 flags = MAP_PRIVATE | MAP_DENYWRITE;
998                 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
999                         flags |= MAP_EXECUTABLE;
1000
1001                 maddr = 0;
1002
1003                 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1004                 case ELF_FDPIC_FLAG_INDEPENDENT:
1005                         /* PT_LOADs are independently locatable */
1006                         break;
1007
1008                 case ELF_FDPIC_FLAG_HONOURVADDR:
1009                         /* the specified virtual address must be honoured */
1010                         maddr = phdr->p_vaddr;
1011                         flags |= MAP_FIXED;
1012                         break;
1013
1014                 case ELF_FDPIC_FLAG_CONSTDISP:
1015                         /* constant displacement
1016                          * - can be mapped anywhere, but must be mapped as a
1017                          *   unit
1018                          */
1019                         if (!dvset) {
1020                                 maddr = load_addr;
1021                                 delta_vaddr = phdr->p_vaddr;
1022                                 dvset = 1;
1023                         } else {
1024                                 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1025                                 flags |= MAP_FIXED;
1026                         }
1027                         break;
1028
1029                 case ELF_FDPIC_FLAG_CONTIGUOUS:
1030                         /* contiguity handled later */
1031                         break;
1032
1033                 default:
1034                         BUG();
1035                 }
1036
1037                 maddr &= PAGE_MASK;
1038
1039                 /* create the mapping */
1040                 disp = phdr->p_vaddr & ~PAGE_MASK;
1041                 down_write(&mm->mmap_sem);
1042                 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1043                                 phdr->p_offset - disp);
1044                 up_write(&mm->mmap_sem);
1045
1046                 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1047                        loop, phdr->p_memsz + disp, prot, flags,
1048                        phdr->p_offset - disp, maddr);
1049
1050                 if (IS_ERR_VALUE(maddr))
1051                         return (int) maddr;
1052
1053                 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1054                     ELF_FDPIC_FLAG_CONTIGUOUS)
1055                         load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1056
1057                 seg->addr = maddr + disp;
1058                 seg->p_vaddr = phdr->p_vaddr;
1059                 seg->p_memsz = phdr->p_memsz;
1060
1061                 /* map the ELF header address if in this segment */
1062                 if (phdr->p_offset == 0)
1063                         params->elfhdr_addr = seg->addr;
1064
1065                 /* clear the bit between beginning of mapping and beginning of
1066                  * PT_LOAD */
1067                 if (prot & PROT_WRITE && disp > 0) {
1068                         kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1069                         clear_user((void __user *) maddr, disp);
1070                         maddr += disp;
1071                 }
1072
1073                 /* clear any space allocated but not loaded
1074                  * - on uClinux we can just clear the lot
1075                  * - on MMU linux we'll get a SIGBUS beyond the last page
1076                  *   extant in the file
1077                  */
1078                 excess = phdr->p_memsz - phdr->p_filesz;
1079                 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1080
1081 #ifdef CONFIG_MMU
1082                 if (excess > excess1) {
1083                         unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1084                         unsigned long xmaddr;
1085
1086                         flags |= MAP_FIXED | MAP_ANONYMOUS;
1087                         down_write(&mm->mmap_sem);
1088                         xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1089                                          prot, flags, 0);
1090                         up_write(&mm->mmap_sem);
1091
1092                         kdebug("mmap[%d] <anon>"
1093                                " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1094                                loop, xaddr, excess - excess1, prot, flags,
1095                                xmaddr);
1096
1097                         if (xmaddr != xaddr)
1098                                 return -ENOMEM;
1099                 }
1100
1101                 if (prot & PROT_WRITE && excess1 > 0) {
1102                         kdebug("clear[%d] ad=%lx sz=%lx",
1103                                loop, maddr + phdr->p_filesz, excess1);
1104                         clear_user((void __user *) maddr + phdr->p_filesz,
1105                                    excess1);
1106                 }
1107
1108 #else
1109                 if (excess > 0) {
1110                         kdebug("clear[%d] ad=%lx sz=%lx",
1111                                loop, maddr + phdr->p_filesz, excess);
1112                         clear_user((void *) maddr + phdr->p_filesz, excess);
1113                 }
1114 #endif
1115
1116                 if (mm) {
1117                         if (phdr->p_flags & PF_X) {
1118                                 mm->start_code = maddr;
1119                                 mm->end_code = maddr + phdr->p_memsz;
1120                         } else if (!mm->start_data) {
1121                                 mm->start_data = maddr;
1122                                 mm->end_data = maddr + phdr->p_memsz;
1123                         }
1124                 }
1125
1126                 seg++;
1127         }
1128
1129         return 0;
1130 }
1131
1132 /*****************************************************************************/
1133 /*
1134  * ELF-FDPIC core dumper
1135  *
1136  * Modelled on fs/exec.c:aout_core_dump()
1137  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1138  *
1139  * Modelled on fs/binfmt_elf.c core dumper
1140  */
1141 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1142
1143 /*
1144  * These are the only things you should do on a core-file: use only these
1145  * functions to write out all the necessary info.
1146  */
1147 static int dump_write(struct file *file, const void *addr, int nr)
1148 {
1149         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1150 }
1151
1152 static int dump_seek(struct file *file, loff_t off)
1153 {
1154         if (file->f_op->llseek) {
1155                 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1156                         return 0;
1157         } else {
1158                 file->f_pos = off;
1159         }
1160         return 1;
1161 }
1162
1163 /*
1164  * Decide whether a segment is worth dumping; default is yes to be
1165  * sure (missing info is worse than too much; etc).
1166  * Personally I'd include everything, and use the coredump limit...
1167  *
1168  * I think we should skip something. But I am not sure how. H.J.
1169  */
1170 static int maydump(struct vm_area_struct *vma)
1171 {
1172         /* Do not dump I/O mapped devices or special mappings */
1173         if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1174                 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1175                 return 0;
1176         }
1177
1178         /* If we may not read the contents, don't allow us to dump
1179          * them either. "dump_write()" can't handle it anyway.
1180          */
1181         if (!(vma->vm_flags & VM_READ)) {
1182                 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1183                 return 0;
1184         }
1185
1186         /* Dump shared memory only if mapped from an anonymous file. */
1187         if (vma->vm_flags & VM_SHARED) {
1188                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1189                         kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags);
1190                         return 1;
1191                 }
1192
1193                 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags);
1194                 return 0;
1195         }
1196
1197 #ifdef CONFIG_MMU
1198         /* If it hasn't been written to, don't write it out */
1199         if (!vma->anon_vma) {
1200                 kdcore("%08lx: %08lx: no (!anon)", vma->vm_start, vma->vm_flags);
1201                 return 0;
1202         }
1203 #endif
1204
1205         kdcore("%08lx: %08lx: yes", vma->vm_start, vma->vm_flags);
1206         return 1;
1207 }
1208
1209 /* An ELF note in memory */
1210 struct memelfnote
1211 {
1212         const char *name;
1213         int type;
1214         unsigned int datasz;
1215         void *data;
1216 };
1217
1218 static int notesize(struct memelfnote *en)
1219 {
1220         int sz;
1221
1222         sz = sizeof(struct elf_note);
1223         sz += roundup(strlen(en->name) + 1, 4);
1224         sz += roundup(en->datasz, 4);
1225
1226         return sz;
1227 }
1228
1229 /* #define DEBUG */
1230
1231 #define DUMP_WRITE(addr, nr)    \
1232         do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1233 #define DUMP_SEEK(off)  \
1234         do { if (!dump_seek(file, (off))) return 0; } while(0)
1235
1236 static int writenote(struct memelfnote *men, struct file *file)
1237 {
1238         struct elf_note en;
1239
1240         en.n_namesz = strlen(men->name) + 1;
1241         en.n_descsz = men->datasz;
1242         en.n_type = men->type;
1243
1244         DUMP_WRITE(&en, sizeof(en));
1245         DUMP_WRITE(men->name, en.n_namesz);
1246         /* XXX - cast from long long to long to avoid need for libgcc.a */
1247         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1248         DUMP_WRITE(men->data, men->datasz);
1249         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1250
1251         return 1;
1252 }
1253 #undef DUMP_WRITE
1254 #undef DUMP_SEEK
1255
1256 #define DUMP_WRITE(addr, nr)    \
1257         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1258                 goto end_coredump;
1259 #define DUMP_SEEK(off)  \
1260         if (!dump_seek(file, (off))) \
1261                 goto end_coredump;
1262
1263 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1264 {
1265         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1266         elf->e_ident[EI_CLASS] = ELF_CLASS;
1267         elf->e_ident[EI_DATA] = ELF_DATA;
1268         elf->e_ident[EI_VERSION] = EV_CURRENT;
1269         elf->e_ident[EI_OSABI] = ELF_OSABI;
1270         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1271
1272         elf->e_type = ET_CORE;
1273         elf->e_machine = ELF_ARCH;
1274         elf->e_version = EV_CURRENT;
1275         elf->e_entry = 0;
1276         elf->e_phoff = sizeof(struct elfhdr);
1277         elf->e_shoff = 0;
1278         elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1279         elf->e_ehsize = sizeof(struct elfhdr);
1280         elf->e_phentsize = sizeof(struct elf_phdr);
1281         elf->e_phnum = segs;
1282         elf->e_shentsize = 0;
1283         elf->e_shnum = 0;
1284         elf->e_shstrndx = 0;
1285         return;
1286 }
1287
1288 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1289 {
1290         phdr->p_type = PT_NOTE;
1291         phdr->p_offset = offset;
1292         phdr->p_vaddr = 0;
1293         phdr->p_paddr = 0;
1294         phdr->p_filesz = sz;
1295         phdr->p_memsz = 0;
1296         phdr->p_flags = 0;
1297         phdr->p_align = 0;
1298         return;
1299 }
1300
1301 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1302                 unsigned int sz, void *data)
1303 {
1304         note->name = name;
1305         note->type = type;
1306         note->datasz = sz;
1307         note->data = data;
1308         return;
1309 }
1310
1311 /*
1312  * fill up all the fields in prstatus from the given task struct, except
1313  * registers which need to be filled up seperately.
1314  */
1315 static void fill_prstatus(struct elf_prstatus *prstatus,
1316                           struct task_struct *p, long signr)
1317 {
1318         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1319         prstatus->pr_sigpend = p->pending.signal.sig[0];
1320         prstatus->pr_sighold = p->blocked.sig[0];
1321         prstatus->pr_pid = p->pid;
1322         prstatus->pr_ppid = p->parent->pid;
1323         prstatus->pr_pgrp = process_group(p);
1324         prstatus->pr_sid = process_session(p);
1325         if (thread_group_leader(p)) {
1326                 /*
1327                  * This is the record for the group leader.  Add in the
1328                  * cumulative times of previous dead threads.  This total
1329                  * won't include the time of each live thread whose state
1330                  * is included in the core dump.  The final total reported
1331                  * to our parent process when it calls wait4 will include
1332                  * those sums as well as the little bit more time it takes
1333                  * this and each other thread to finish dying after the
1334                  * core dump synchronization phase.
1335                  */
1336                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1337                                    &prstatus->pr_utime);
1338                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1339                                    &prstatus->pr_stime);
1340         } else {
1341                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1342                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1343         }
1344         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1345         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1346
1347         prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1348         prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1349 }
1350
1351 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1352                        struct mm_struct *mm)
1353 {
1354         unsigned int i, len;
1355
1356         /* first copy the parameters from user space */
1357         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1358
1359         len = mm->arg_end - mm->arg_start;
1360         if (len >= ELF_PRARGSZ)
1361                 len = ELF_PRARGSZ - 1;
1362         if (copy_from_user(&psinfo->pr_psargs,
1363                            (const char __user *) mm->arg_start, len))
1364                 return -EFAULT;
1365         for (i = 0; i < len; i++)
1366                 if (psinfo->pr_psargs[i] == 0)
1367                         psinfo->pr_psargs[i] = ' ';
1368         psinfo->pr_psargs[len] = 0;
1369
1370         psinfo->pr_pid = p->pid;
1371         psinfo->pr_ppid = p->parent->pid;
1372         psinfo->pr_pgrp = process_group(p);
1373         psinfo->pr_sid = process_session(p);
1374
1375         i = p->state ? ffz(~p->state) + 1 : 0;
1376         psinfo->pr_state = i;
1377         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1378         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1379         psinfo->pr_nice = task_nice(p);
1380         psinfo->pr_flag = p->flags;
1381         SET_UID(psinfo->pr_uid, p->uid);
1382         SET_GID(psinfo->pr_gid, p->gid);
1383         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1384
1385         return 0;
1386 }
1387
1388 /* Here is the structure in which status of each thread is captured. */
1389 struct elf_thread_status
1390 {
1391         struct list_head list;
1392         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1393         elf_fpregset_t fpu;             /* NT_PRFPREG */
1394         struct task_struct *thread;
1395 #ifdef ELF_CORE_COPY_XFPREGS
1396         elf_fpxregset_t xfpu;           /* NT_PRXFPREG */
1397 #endif
1398         struct memelfnote notes[3];
1399         int num_notes;
1400 };
1401
1402 /*
1403  * In order to add the specific thread information for the elf file format,
1404  * we need to keep a linked list of every thread's pr_status and then create
1405  * a single section for them in the final core file.
1406  */
1407 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1408 {
1409         struct task_struct *p = t->thread;
1410         int sz = 0;
1411
1412         t->num_notes = 0;
1413
1414         fill_prstatus(&t->prstatus, p, signr);
1415         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1416
1417         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1418                   &t->prstatus);
1419         t->num_notes++;
1420         sz += notesize(&t->notes[0]);
1421
1422         t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1423         if (t->prstatus.pr_fpvalid) {
1424                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1425                           &t->fpu);
1426                 t->num_notes++;
1427                 sz += notesize(&t->notes[1]);
1428         }
1429
1430 #ifdef ELF_CORE_COPY_XFPREGS
1431         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1432                 fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu),
1433                           &t->xfpu);
1434                 t->num_notes++;
1435                 sz += notesize(&t->notes[2]);
1436         }
1437 #endif
1438         return sz;
1439 }
1440
1441 /*
1442  * dump the segments for an MMU process
1443  */
1444 #ifdef CONFIG_MMU
1445 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm,
1446                                    size_t *size, unsigned long *limit)
1447 {
1448         struct vm_area_struct *vma;
1449
1450         for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1451                 unsigned long addr;
1452
1453                 if (!maydump(vma))
1454                         continue;
1455
1456                 for (addr = vma->vm_start;
1457                      addr < vma->vm_end;
1458                      addr += PAGE_SIZE
1459                      ) {
1460                         struct vm_area_struct *vma;
1461                         struct page *page;
1462
1463                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1464                                            &page, &vma) <= 0) {
1465                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1466                         }
1467                         else if (page == ZERO_PAGE(addr)) {
1468                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1469                                 page_cache_release(page);
1470                         }
1471                         else {
1472                                 void *kaddr;
1473
1474                                 flush_cache_page(vma, addr, page_to_pfn(page));
1475                                 kaddr = kmap(page);
1476                                 if ((*size += PAGE_SIZE) > *limit ||
1477                                     !dump_write(file, kaddr, PAGE_SIZE)
1478                                     ) {
1479                                         kunmap(page);
1480                                         page_cache_release(page);
1481                                         return -EIO;
1482                                 }
1483                                 kunmap(page);
1484                                 page_cache_release(page);
1485                         }
1486                 }
1487         }
1488
1489         return 0;
1490
1491 end_coredump:
1492         return -EFBIG;
1493 }
1494 #endif
1495
1496 /*
1497  * dump the segments for a NOMMU process
1498  */
1499 #ifndef CONFIG_MMU
1500 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm,
1501                                    size_t *size, unsigned long *limit)
1502 {
1503         struct vm_list_struct *vml;
1504
1505         for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1506         struct vm_area_struct *vma = vml->vma;
1507
1508                 if (!maydump(vma))
1509                         continue;
1510
1511                 if ((*size += PAGE_SIZE) > *limit)
1512                         return -EFBIG;
1513
1514                 if (!dump_write(file, (void *) vma->vm_start,
1515                                 vma->vm_end - vma->vm_start))
1516                         return -EIO;
1517         }
1518
1519         return 0;
1520 }
1521 #endif
1522
1523 /*
1524  * Actual dumper
1525  *
1526  * This is a two-pass process; first we find the offsets of the bits,
1527  * and then they are actually written out.  If we run out of core limit
1528  * we just truncate.
1529  */
1530 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1531                                struct file *file)
1532 {
1533 #define NUM_NOTES       6
1534         int has_dumped = 0;
1535         mm_segment_t fs;
1536         int segs;
1537         size_t size = 0;
1538         int i;
1539         struct vm_area_struct *vma;
1540         struct elfhdr *elf = NULL;
1541         loff_t offset = 0, dataoff;
1542         unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1543         int numnote;
1544         struct memelfnote *notes = NULL;
1545         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1546         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1547         struct task_struct *g, *p;
1548         LIST_HEAD(thread_list);
1549         struct list_head *t;
1550         elf_fpregset_t *fpu = NULL;
1551 #ifdef ELF_CORE_COPY_XFPREGS
1552         elf_fpxregset_t *xfpu = NULL;
1553 #endif
1554         int thread_status_size = 0;
1555 #ifndef CONFIG_MMU
1556         struct vm_list_struct *vml;
1557 #endif
1558         elf_addr_t *auxv;
1559
1560         /*
1561          * We no longer stop all VM operations.
1562          *
1563          * This is because those proceses that could possibly change map_count
1564          * or the mmap / vma pages are now blocked in do_exit on current
1565          * finishing this core dump.
1566          *
1567          * Only ptrace can touch these memory addresses, but it doesn't change
1568          * the map_count or the pages allocated. So no possibility of crashing
1569          * exists while dumping the mm->vm_next areas to the core file.
1570          */
1571
1572         /* alloc memory for large data structures: too large to be on stack */
1573         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1574         if (!elf)
1575                 goto cleanup;
1576         prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1577         if (!prstatus)
1578                 goto cleanup;
1579         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1580         if (!psinfo)
1581                 goto cleanup;
1582         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1583         if (!notes)
1584                 goto cleanup;
1585         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1586         if (!fpu)
1587                 goto cleanup;
1588 #ifdef ELF_CORE_COPY_XFPREGS
1589         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1590         if (!xfpu)
1591                 goto cleanup;
1592 #endif
1593
1594         if (signr) {
1595                 struct elf_thread_status *tmp;
1596                 rcu_read_lock();
1597                 do_each_thread(g,p)
1598                         if (current->mm == p->mm && current != p) {
1599                                 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1600                                 if (!tmp) {
1601                                         rcu_read_unlock();
1602                                         goto cleanup;
1603                                 }
1604                                 tmp->thread = p;
1605                                 list_add(&tmp->list, &thread_list);
1606                         }
1607                 while_each_thread(g,p);
1608                 rcu_read_unlock();
1609                 list_for_each(t, &thread_list) {
1610                         struct elf_thread_status *tmp;
1611                         int sz;
1612
1613                         tmp = list_entry(t, struct elf_thread_status, list);
1614                         sz = elf_dump_thread_status(signr, tmp);
1615                         thread_status_size += sz;
1616                 }
1617         }
1618
1619         /* now collect the dump for the current */
1620         fill_prstatus(prstatus, current, signr);
1621         elf_core_copy_regs(&prstatus->pr_reg, regs);
1622
1623 #ifdef CONFIG_MMU
1624         segs = current->mm->map_count;
1625 #else
1626         segs = 0;
1627         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1628             segs++;
1629 #endif
1630 #ifdef ELF_CORE_EXTRA_PHDRS
1631         segs += ELF_CORE_EXTRA_PHDRS;
1632 #endif
1633
1634         /* Set up header */
1635         fill_elf_fdpic_header(elf, segs + 1);   /* including notes section */
1636
1637         has_dumped = 1;
1638         current->flags |= PF_DUMPCORE;
1639
1640         /*
1641          * Set up the notes in similar form to SVR4 core dumps made
1642          * with info from their /proc.
1643          */
1644
1645         fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1646         fill_psinfo(psinfo, current->group_leader, current->mm);
1647         fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1648
1649         numnote = 2;
1650
1651         auxv = (elf_addr_t *) current->mm->saved_auxv;
1652
1653         i = 0;
1654         do
1655                 i += 2;
1656         while (auxv[i - 2] != AT_NULL);
1657         fill_note(&notes[numnote++], "CORE", NT_AUXV,
1658                   i * sizeof(elf_addr_t), auxv);
1659
1660         /* Try to dump the FPU. */
1661         if ((prstatus->pr_fpvalid =
1662              elf_core_copy_task_fpregs(current, regs, fpu)))
1663                 fill_note(notes + numnote++,
1664                           "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1665 #ifdef ELF_CORE_COPY_XFPREGS
1666         if (elf_core_copy_task_xfpregs(current, xfpu))
1667                 fill_note(notes + numnote++,
1668                           "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu);
1669 #endif
1670
1671         fs = get_fs();
1672         set_fs(KERNEL_DS);
1673
1674         DUMP_WRITE(elf, sizeof(*elf));
1675         offset += sizeof(*elf);                         /* Elf header */
1676         offset += (segs+1) * sizeof(struct elf_phdr);   /* Program headers */
1677
1678         /* Write notes phdr entry */
1679         {
1680                 struct elf_phdr phdr;
1681                 int sz = 0;
1682
1683                 for (i = 0; i < numnote; i++)
1684                         sz += notesize(notes + i);
1685
1686                 sz += thread_status_size;
1687
1688                 fill_elf_note_phdr(&phdr, sz, offset);
1689                 offset += sz;
1690                 DUMP_WRITE(&phdr, sizeof(phdr));
1691         }
1692
1693         /* Page-align dumped data */
1694         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1695
1696         /* write program headers for segments dump */
1697         for (
1698 #ifdef CONFIG_MMU
1699                 vma = current->mm->mmap; vma; vma = vma->vm_next
1700 #else
1701                         vml = current->mm->context.vmlist; vml; vml = vml->next
1702 #endif
1703              ) {
1704                 struct elf_phdr phdr;
1705                 size_t sz;
1706
1707 #ifndef CONFIG_MMU
1708                 vma = vml->vma;
1709 #endif
1710
1711                 sz = vma->vm_end - vma->vm_start;
1712
1713                 phdr.p_type = PT_LOAD;
1714                 phdr.p_offset = offset;
1715                 phdr.p_vaddr = vma->vm_start;
1716                 phdr.p_paddr = 0;
1717                 phdr.p_filesz = maydump(vma) ? sz : 0;
1718                 phdr.p_memsz = sz;
1719                 offset += phdr.p_filesz;
1720                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1721                 if (vma->vm_flags & VM_WRITE)
1722                         phdr.p_flags |= PF_W;
1723                 if (vma->vm_flags & VM_EXEC)
1724                         phdr.p_flags |= PF_X;
1725                 phdr.p_align = ELF_EXEC_PAGESIZE;
1726
1727                 DUMP_WRITE(&phdr, sizeof(phdr));
1728         }
1729
1730 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1731         ELF_CORE_WRITE_EXTRA_PHDRS;
1732 #endif
1733
1734         /* write out the notes section */
1735         for (i = 0; i < numnote; i++)
1736                 if (!writenote(notes + i, file))
1737                         goto end_coredump;
1738
1739         /* write out the thread status notes section */
1740         list_for_each(t, &thread_list) {
1741                 struct elf_thread_status *tmp =
1742                                 list_entry(t, struct elf_thread_status, list);
1743
1744                 for (i = 0; i < tmp->num_notes; i++)
1745                         if (!writenote(&tmp->notes[i], file))
1746                                 goto end_coredump;
1747         }
1748
1749         DUMP_SEEK(dataoff);
1750
1751         if (elf_fdpic_dump_segments(file, current->mm, &size, &limit) < 0)
1752                 goto end_coredump;
1753
1754 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1755         ELF_CORE_WRITE_EXTRA_DATA;
1756 #endif
1757
1758         if (file->f_pos != offset) {
1759                 /* Sanity check */
1760                 printk(KERN_WARNING
1761                        "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1762                        file->f_pos, offset);
1763         }
1764
1765 end_coredump:
1766         set_fs(fs);
1767
1768 cleanup:
1769         while (!list_empty(&thread_list)) {
1770                 struct list_head *tmp = thread_list.next;
1771                 list_del(tmp);
1772                 kfree(list_entry(tmp, struct elf_thread_status, list));
1773         }
1774
1775         kfree(elf);
1776         kfree(prstatus);
1777         kfree(psinfo);
1778         kfree(notes);
1779         kfree(fpu);
1780 #ifdef ELF_CORE_COPY_XFPREGS
1781         kfree(xfpu);
1782 #endif
1783         return has_dumped;
1784 #undef NUM_NOTES
1785 }
1786
1787 #endif          /* USE_ELF_CORE_DUMP */