CRED: Use RCU to access another task's creds and to release a task's own creds
[linux-2.6.git] / 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/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.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 *, unsigned long limit);
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 (unlikely(retval != size))
141                 return retval < 0 ? retval : -ENOEXEC;
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         kdebug("____ LOAD %d ____", current->pid);
183
184         memset(&exec_params, 0, sizeof(exec_params));
185         memset(&interp_params, 0, sizeof(interp_params));
186
187         exec_params.hdr = *(struct elfhdr *) bprm->buf;
188         exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
189
190         /* check that this is a binary we know how to deal with */
191         retval = -ENOEXEC;
192         if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
193                 goto error;
194
195         /* read the program header table */
196         retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
197         if (retval < 0)
198                 goto error;
199
200         /* scan for a program header that specifies an interpreter */
201         phdr = exec_params.phdrs;
202
203         for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
204                 switch (phdr->p_type) {
205                 case PT_INTERP:
206                         retval = -ENOMEM;
207                         if (phdr->p_filesz > PATH_MAX)
208                                 goto error;
209                         retval = -ENOENT;
210                         if (phdr->p_filesz < 2)
211                                 goto error;
212
213                         /* read the name of the interpreter into memory */
214                         interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
215                         if (!interpreter_name)
216                                 goto error;
217
218                         retval = kernel_read(bprm->file,
219                                              phdr->p_offset,
220                                              interpreter_name,
221                                              phdr->p_filesz);
222                         if (unlikely(retval != phdr->p_filesz)) {
223                                 if (retval >= 0)
224                                         retval = -ENOEXEC;
225                                 goto error;
226                         }
227
228                         retval = -ENOENT;
229                         if (interpreter_name[phdr->p_filesz - 1] != '\0')
230                                 goto error;
231
232                         kdebug("Using ELF interpreter %s", interpreter_name);
233
234                         /* replace the program with the interpreter */
235                         interpreter = open_exec(interpreter_name);
236                         retval = PTR_ERR(interpreter);
237                         if (IS_ERR(interpreter)) {
238                                 interpreter = NULL;
239                                 goto error;
240                         }
241
242                         /*
243                          * If the binary is not readable then enforce
244                          * mm->dumpable = 0 regardless of the interpreter's
245                          * permissions.
246                          */
247                         if (file_permission(interpreter, MAY_READ) < 0)
248                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
249
250                         retval = kernel_read(interpreter, 0, bprm->buf,
251                                              BINPRM_BUF_SIZE);
252                         if (unlikely(retval != BINPRM_BUF_SIZE)) {
253                                 if (retval >= 0)
254                                         retval = -ENOEXEC;
255                                 goto error;
256                         }
257
258                         interp_params.hdr = *((struct elfhdr *) bprm->buf);
259                         break;
260
261                 case PT_LOAD:
262 #ifdef CONFIG_MMU
263                         if (exec_params.load_addr == 0)
264                                 exec_params.load_addr = phdr->p_vaddr;
265 #endif
266                         break;
267                 }
268
269         }
270
271         if (elf_check_const_displacement(&exec_params.hdr))
272                 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
273
274         /* perform insanity checks on the interpreter */
275         if (interpreter_name) {
276                 retval = -ELIBBAD;
277                 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
278                         goto error;
279
280                 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
281
282                 /* read the interpreter's program header table */
283                 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
284                 if (retval < 0)
285                         goto error;
286         }
287
288         stack_size = exec_params.stack_size;
289         if (stack_size < interp_params.stack_size)
290                 stack_size = interp_params.stack_size;
291
292         if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
293                 executable_stack = EXSTACK_ENABLE_X;
294         else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
295                 executable_stack = EXSTACK_DISABLE_X;
296         else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
297                 executable_stack = EXSTACK_ENABLE_X;
298         else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
299                 executable_stack = EXSTACK_DISABLE_X;
300         else
301                 executable_stack = EXSTACK_DEFAULT;
302
303         retval = -ENOEXEC;
304         if (stack_size == 0)
305                 goto error;
306
307         if (elf_check_const_displacement(&interp_params.hdr))
308                 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
309
310         /* flush all traces of the currently running executable */
311         retval = flush_old_exec(bprm);
312         if (retval)
313                 goto error;
314
315         /* there's now no turning back... the old userspace image is dead,
316          * defunct, deceased, etc. after this point we have to exit via
317          * error_kill */
318         set_personality(PER_LINUX_FDPIC);
319         set_binfmt(&elf_fdpic_format);
320
321         current->mm->start_code = 0;
322         current->mm->end_code = 0;
323         current->mm->start_stack = 0;
324         current->mm->start_data = 0;
325         current->mm->end_data = 0;
326         current->mm->context.exec_fdpic_loadmap = 0;
327         current->mm->context.interp_fdpic_loadmap = 0;
328
329         current->flags &= ~PF_FORKNOEXEC;
330
331 #ifdef CONFIG_MMU
332         elf_fdpic_arch_lay_out_mm(&exec_params,
333                                   &interp_params,
334                                   &current->mm->start_stack,
335                                   &current->mm->start_brk);
336
337         retval = setup_arg_pages(bprm, current->mm->start_stack,
338                                  executable_stack);
339         if (retval < 0) {
340                 send_sig(SIGKILL, current, 0);
341                 goto error_kill;
342         }
343 #endif
344
345         /* load the executable and interpreter into memory */
346         retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
347                                     "executable");
348         if (retval < 0)
349                 goto error_kill;
350
351         if (interpreter_name) {
352                 retval = elf_fdpic_map_file(&interp_params, interpreter,
353                                             current->mm, "interpreter");
354                 if (retval < 0) {
355                         printk(KERN_ERR "Unable to load interpreter\n");
356                         goto error_kill;
357                 }
358
359                 allow_write_access(interpreter);
360                 fput(interpreter);
361                 interpreter = NULL;
362         }
363
364 #ifdef CONFIG_MMU
365         if (!current->mm->start_brk)
366                 current->mm->start_brk = current->mm->end_data;
367
368         current->mm->brk = current->mm->start_brk =
369                 PAGE_ALIGN(current->mm->start_brk);
370
371 #else
372         /* create a stack and brk area big enough for everyone
373          * - the brk heap starts at the bottom and works up
374          * - the stack starts at the top and works down
375          */
376         stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
377         if (stack_size < PAGE_SIZE * 2)
378                 stack_size = PAGE_SIZE * 2;
379
380         down_write(&current->mm->mmap_sem);
381         current->mm->start_brk = do_mmap(NULL, 0, stack_size,
382                                          PROT_READ | PROT_WRITE | PROT_EXEC,
383                                          MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
384                                          0);
385
386         if (IS_ERR_VALUE(current->mm->start_brk)) {
387                 up_write(&current->mm->mmap_sem);
388                 retval = current->mm->start_brk;
389                 current->mm->start_brk = 0;
390                 goto error_kill;
391         }
392
393         /* expand the stack mapping to use up the entire allocation granule */
394         fullsize = kobjsize((char *) current->mm->start_brk);
395         if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
396                                     fullsize, 0, 0)))
397                 stack_size = fullsize;
398         up_write(&current->mm->mmap_sem);
399
400         current->mm->brk = current->mm->start_brk;
401         current->mm->context.end_brk = current->mm->start_brk;
402         current->mm->context.end_brk +=
403                 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
404         current->mm->start_stack = current->mm->start_brk + stack_size;
405 #endif
406
407         compute_creds(bprm);
408         current->flags &= ~PF_FORKNOEXEC;
409         if (create_elf_fdpic_tables(bprm, current->mm,
410                                     &exec_params, &interp_params) < 0)
411                 goto error_kill;
412
413         kdebug("- start_code  %lx", current->mm->start_code);
414         kdebug("- end_code    %lx", current->mm->end_code);
415         kdebug("- start_data  %lx", current->mm->start_data);
416         kdebug("- end_data    %lx", current->mm->end_data);
417         kdebug("- start_brk   %lx", current->mm->start_brk);
418         kdebug("- brk         %lx", current->mm->brk);
419         kdebug("- start_stack %lx", current->mm->start_stack);
420
421 #ifdef ELF_FDPIC_PLAT_INIT
422         /*
423          * The ABI may specify that certain registers be set up in special
424          * ways (on i386 %edx is the address of a DT_FINI function, for
425          * example.  This macro performs whatever initialization to
426          * the regs structure is required.
427          */
428         dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
429         ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
430                             dynaddr);
431 #endif
432
433         /* everything is now ready... get the userspace context ready to roll */
434         entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
435         start_thread(regs, entryaddr, current->mm->start_stack);
436
437         retval = 0;
438
439 error:
440         if (interpreter) {
441                 allow_write_access(interpreter);
442                 fput(interpreter);
443         }
444         kfree(interpreter_name);
445         kfree(exec_params.phdrs);
446         kfree(exec_params.loadmap);
447         kfree(interp_params.phdrs);
448         kfree(interp_params.loadmap);
449         return retval;
450
451         /* unrecoverable error - kill the process */
452 error_kill:
453         send_sig(SIGSEGV, current, 0);
454         goto error;
455
456 }
457
458 /*****************************************************************************/
459
460 #ifndef ELF_BASE_PLATFORM
461 /*
462  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
463  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
464  * will be copied to the user stack in the same manner as AT_PLATFORM.
465  */
466 #define ELF_BASE_PLATFORM NULL
467 #endif
468
469 /*
470  * present useful information to the program by shovelling it onto the new
471  * process's stack
472  */
473 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
474                                    struct mm_struct *mm,
475                                    struct elf_fdpic_params *exec_params,
476                                    struct elf_fdpic_params *interp_params)
477 {
478         const struct cred *cred = current_cred();
479         unsigned long sp, csp, nitems;
480         elf_caddr_t __user *argv, *envp;
481         size_t platform_len = 0, len;
482         char *k_platform, *k_base_platform;
483         char __user *u_platform, *u_base_platform, *p;
484         long hwcap;
485         int loop;
486         int nr; /* reset for each csp adjustment */
487
488 #ifdef CONFIG_MMU
489         /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
490          * by the processes running on the same package. One thing we can do is
491          * to shuffle the initial stack for them, so we give the architecture
492          * an opportunity to do so here.
493          */
494         sp = arch_align_stack(bprm->p);
495 #else
496         sp = mm->start_stack;
497
498         /* stack the program arguments and environment */
499         if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
500                 return -EFAULT;
501 #endif
502
503         hwcap = ELF_HWCAP;
504
505         /*
506          * If this architecture has a platform capability string, copy it
507          * to userspace.  In some cases (Sparc), this info is impossible
508          * for userspace to get any other way, in others (i386) it is
509          * merely difficult.
510          */
511         k_platform = ELF_PLATFORM;
512         u_platform = NULL;
513
514         if (k_platform) {
515                 platform_len = strlen(k_platform) + 1;
516                 sp -= platform_len;
517                 u_platform = (char __user *) sp;
518                 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
519                         return -EFAULT;
520         }
521
522         /*
523          * If this architecture has a "base" platform capability
524          * string, copy it to userspace.
525          */
526         k_base_platform = ELF_BASE_PLATFORM;
527         u_base_platform = NULL;
528
529         if (k_base_platform) {
530                 platform_len = strlen(k_base_platform) + 1;
531                 sp -= platform_len;
532                 u_base_platform = (char __user *) sp;
533                 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
534                         return -EFAULT;
535         }
536
537         sp &= ~7UL;
538
539         /* stack the load map(s) */
540         len = sizeof(struct elf32_fdpic_loadmap);
541         len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
542         sp = (sp - len) & ~7UL;
543         exec_params->map_addr = sp;
544
545         if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
546                 return -EFAULT;
547
548         current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
549
550         if (interp_params->loadmap) {
551                 len = sizeof(struct elf32_fdpic_loadmap);
552                 len += sizeof(struct elf32_fdpic_loadseg) *
553                         interp_params->loadmap->nsegs;
554                 sp = (sp - len) & ~7UL;
555                 interp_params->map_addr = sp;
556
557                 if (copy_to_user((void __user *) sp, interp_params->loadmap,
558                                  len) != 0)
559                         return -EFAULT;
560
561                 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
562         }
563
564         /* force 16 byte _final_ alignment here for generality */
565 #define DLINFO_ITEMS 15
566
567         nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
568                 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
569
570         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
571                 nitems++;
572
573         csp = sp;
574         sp -= nitems * 2 * sizeof(unsigned long);
575         sp -= (bprm->envc + 1) * sizeof(char *);        /* envv[] */
576         sp -= (bprm->argc + 1) * sizeof(char *);        /* argv[] */
577         sp -= 1 * sizeof(unsigned long);                /* argc */
578
579         csp -= sp & 15UL;
580         sp -= sp & 15UL;
581
582         /* put the ELF interpreter info on the stack */
583 #define NEW_AUX_ENT(id, val)                                            \
584         do {                                                            \
585                 struct { unsigned long _id, _val; } __user *ent;        \
586                                                                         \
587                 ent = (void __user *) csp;                              \
588                 __put_user((id), &ent[nr]._id);                         \
589                 __put_user((val), &ent[nr]._val);                       \
590                 nr++;                                                   \
591         } while (0)
592
593         nr = 0;
594         csp -= 2 * sizeof(unsigned long);
595         NEW_AUX_ENT(AT_NULL, 0);
596         if (k_platform) {
597                 nr = 0;
598                 csp -= 2 * sizeof(unsigned long);
599                 NEW_AUX_ENT(AT_PLATFORM,
600                             (elf_addr_t) (unsigned long) u_platform);
601         }
602
603         if (k_base_platform) {
604                 nr = 0;
605                 csp -= 2 * sizeof(unsigned long);
606                 NEW_AUX_ENT(AT_BASE_PLATFORM,
607                             (elf_addr_t) (unsigned long) u_base_platform);
608         }
609
610         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
611                 nr = 0;
612                 csp -= 2 * sizeof(unsigned long);
613                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
614         }
615
616         nr = 0;
617         csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
618         NEW_AUX_ENT(AT_HWCAP,   hwcap);
619         NEW_AUX_ENT(AT_PAGESZ,  PAGE_SIZE);
620         NEW_AUX_ENT(AT_CLKTCK,  CLOCKS_PER_SEC);
621         NEW_AUX_ENT(AT_PHDR,    exec_params->ph_addr);
622         NEW_AUX_ENT(AT_PHENT,   sizeof(struct elf_phdr));
623         NEW_AUX_ENT(AT_PHNUM,   exec_params->hdr.e_phnum);
624         NEW_AUX_ENT(AT_BASE,    interp_params->elfhdr_addr);
625         NEW_AUX_ENT(AT_FLAGS,   0);
626         NEW_AUX_ENT(AT_ENTRY,   exec_params->entry_addr);
627         NEW_AUX_ENT(AT_UID,     (elf_addr_t) cred->uid);
628         NEW_AUX_ENT(AT_EUID,    (elf_addr_t) cred->euid);
629         NEW_AUX_ENT(AT_GID,     (elf_addr_t) cred->gid);
630         NEW_AUX_ENT(AT_EGID,    (elf_addr_t) cred->egid);
631         NEW_AUX_ENT(AT_SECURE,  security_bprm_secureexec(bprm));
632         NEW_AUX_ENT(AT_EXECFN,  bprm->exec);
633
634 #ifdef ARCH_DLINFO
635         nr = 0;
636         csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
637
638         /* ARCH_DLINFO must come last so platform specific code can enforce
639          * special alignment requirements on the AUXV if necessary (eg. PPC).
640          */
641         ARCH_DLINFO;
642 #endif
643 #undef NEW_AUX_ENT
644
645         /* allocate room for argv[] and envv[] */
646         csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
647         envp = (elf_caddr_t __user *) csp;
648         csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
649         argv = (elf_caddr_t __user *) csp;
650
651         /* stack argc */
652         csp -= sizeof(unsigned long);
653         __put_user(bprm->argc, (unsigned long __user *) csp);
654
655         BUG_ON(csp != sp);
656
657         /* fill in the argv[] array */
658 #ifdef CONFIG_MMU
659         current->mm->arg_start = bprm->p;
660 #else
661         current->mm->arg_start = current->mm->start_stack -
662                 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
663 #endif
664
665         p = (char __user *) current->mm->arg_start;
666         for (loop = bprm->argc; loop > 0; loop--) {
667                 __put_user((elf_caddr_t) p, argv++);
668                 len = strnlen_user(p, MAX_ARG_STRLEN);
669                 if (!len || len > MAX_ARG_STRLEN)
670                         return -EINVAL;
671                 p += len;
672         }
673         __put_user(NULL, argv);
674         current->mm->arg_end = (unsigned long) p;
675
676         /* fill in the envv[] array */
677         current->mm->env_start = (unsigned long) p;
678         for (loop = bprm->envc; loop > 0; loop--) {
679                 __put_user((elf_caddr_t)(unsigned long) p, envp++);
680                 len = strnlen_user(p, MAX_ARG_STRLEN);
681                 if (!len || len > MAX_ARG_STRLEN)
682                         return -EINVAL;
683                 p += len;
684         }
685         __put_user(NULL, envp);
686         current->mm->env_end = (unsigned long) p;
687
688         mm->start_stack = (unsigned long) sp;
689         return 0;
690 }
691
692 /*****************************************************************************/
693 /*
694  * transfer the program arguments and environment from the holding pages onto
695  * the stack
696  */
697 #ifndef CONFIG_MMU
698 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
699                                             unsigned long *_sp)
700 {
701         unsigned long index, stop, sp;
702         char *src;
703         int ret = 0;
704
705         stop = bprm->p >> PAGE_SHIFT;
706         sp = *_sp;
707
708         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
709                 src = kmap(bprm->page[index]);
710                 sp -= PAGE_SIZE;
711                 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
712                         ret = -EFAULT;
713                 kunmap(bprm->page[index]);
714                 if (ret < 0)
715                         goto out;
716         }
717
718         *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
719
720 out:
721         return ret;
722 }
723 #endif
724
725 /*****************************************************************************/
726 /*
727  * load the appropriate binary image (executable or interpreter) into memory
728  * - we assume no MMU is available
729  * - if no other PIC bits are set in params->hdr->e_flags
730  *   - we assume that the LOADable segments in the binary are independently relocatable
731  *   - we assume R/O executable segments are shareable
732  * - else
733  *   - we assume the loadable parts of the image to require fixed displacement
734  *   - the image is not shareable
735  */
736 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
737                               struct file *file,
738                               struct mm_struct *mm,
739                               const char *what)
740 {
741         struct elf32_fdpic_loadmap *loadmap;
742 #ifdef CONFIG_MMU
743         struct elf32_fdpic_loadseg *mseg;
744 #endif
745         struct elf32_fdpic_loadseg *seg;
746         struct elf32_phdr *phdr;
747         unsigned long load_addr, stop;
748         unsigned nloads, tmp;
749         size_t size;
750         int loop, ret;
751
752         /* allocate a load map table */
753         nloads = 0;
754         for (loop = 0; loop < params->hdr.e_phnum; loop++)
755                 if (params->phdrs[loop].p_type == PT_LOAD)
756                         nloads++;
757
758         if (nloads == 0)
759                 return -ELIBBAD;
760
761         size = sizeof(*loadmap) + nloads * sizeof(*seg);
762         loadmap = kzalloc(size, GFP_KERNEL);
763         if (!loadmap)
764                 return -ENOMEM;
765
766         params->loadmap = loadmap;
767
768         loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
769         loadmap->nsegs = nloads;
770
771         load_addr = params->load_addr;
772         seg = loadmap->segs;
773
774         /* map the requested LOADs into the memory space */
775         switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
776         case ELF_FDPIC_FLAG_CONSTDISP:
777         case ELF_FDPIC_FLAG_CONTIGUOUS:
778 #ifndef CONFIG_MMU
779                 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
780                 if (ret < 0)
781                         return ret;
782                 break;
783 #endif
784         default:
785                 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
786                 if (ret < 0)
787                         return ret;
788                 break;
789         }
790
791         /* map the entry point */
792         if (params->hdr.e_entry) {
793                 seg = loadmap->segs;
794                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
795                         if (params->hdr.e_entry >= seg->p_vaddr &&
796                             params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
797                                 params->entry_addr =
798                                         (params->hdr.e_entry - seg->p_vaddr) +
799                                         seg->addr;
800                                 break;
801                         }
802                 }
803         }
804
805         /* determine where the program header table has wound up if mapped */
806         stop = params->hdr.e_phoff;
807         stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
808         phdr = params->phdrs;
809
810         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
811                 if (phdr->p_type != PT_LOAD)
812                         continue;
813
814                 if (phdr->p_offset > params->hdr.e_phoff ||
815                     phdr->p_offset + phdr->p_filesz < stop)
816                         continue;
817
818                 seg = loadmap->segs;
819                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
820                         if (phdr->p_vaddr >= seg->p_vaddr &&
821                             phdr->p_vaddr + phdr->p_filesz <=
822                             seg->p_vaddr + seg->p_memsz) {
823                                 params->ph_addr =
824                                         (phdr->p_vaddr - seg->p_vaddr) +
825                                         seg->addr +
826                                         params->hdr.e_phoff - phdr->p_offset;
827                                 break;
828                         }
829                 }
830                 break;
831         }
832
833         /* determine where the dynamic section has wound up if there is one */
834         phdr = params->phdrs;
835         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
836                 if (phdr->p_type != PT_DYNAMIC)
837                         continue;
838
839                 seg = loadmap->segs;
840                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
841                         if (phdr->p_vaddr >= seg->p_vaddr &&
842                             phdr->p_vaddr + phdr->p_memsz <=
843                             seg->p_vaddr + seg->p_memsz) {
844                                 params->dynamic_addr =
845                                         (phdr->p_vaddr - seg->p_vaddr) +
846                                         seg->addr;
847
848                                 /* check the dynamic section contains at least
849                                  * one item, and that the last item is a NULL
850                                  * entry */
851                                 if (phdr->p_memsz == 0 ||
852                                     phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
853                                         goto dynamic_error;
854
855                                 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
856                                 if (((Elf32_Dyn *)
857                                      params->dynamic_addr)[tmp - 1].d_tag != 0)
858                                         goto dynamic_error;
859                                 break;
860                         }
861                 }
862                 break;
863         }
864
865         /* now elide adjacent segments in the load map on MMU linux
866          * - on uClinux the holes between may actually be filled with system
867          *   stuff or stuff from other processes
868          */
869 #ifdef CONFIG_MMU
870         nloads = loadmap->nsegs;
871         mseg = loadmap->segs;
872         seg = mseg + 1;
873         for (loop = 1; loop < nloads; loop++) {
874                 /* see if we have a candidate for merging */
875                 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
876                         load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
877                         if (load_addr == (seg->addr & PAGE_MASK)) {
878                                 mseg->p_memsz +=
879                                         load_addr -
880                                         (mseg->addr + mseg->p_memsz);
881                                 mseg->p_memsz += seg->addr & ~PAGE_MASK;
882                                 mseg->p_memsz += seg->p_memsz;
883                                 loadmap->nsegs--;
884                                 continue;
885                         }
886                 }
887
888                 mseg++;
889                 if (mseg != seg)
890                         *mseg = *seg;
891         }
892 #endif
893
894         kdebug("Mapped Object [%s]:", what);
895         kdebug("- elfhdr   : %lx", params->elfhdr_addr);
896         kdebug("- entry    : %lx", params->entry_addr);
897         kdebug("- PHDR[]   : %lx", params->ph_addr);
898         kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
899         seg = loadmap->segs;
900         for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
901                 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
902                        loop,
903                        seg->addr, seg->addr + seg->p_memsz - 1,
904                        seg->p_vaddr, seg->p_memsz);
905
906         return 0;
907
908 dynamic_error:
909         printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
910                what, file->f_path.dentry->d_inode->i_ino);
911         return -ELIBBAD;
912 }
913
914 /*****************************************************************************/
915 /*
916  * map a file with constant displacement under uClinux
917  */
918 #ifndef CONFIG_MMU
919 static int elf_fdpic_map_file_constdisp_on_uclinux(
920         struct elf_fdpic_params *params,
921         struct file *file,
922         struct mm_struct *mm)
923 {
924         struct elf32_fdpic_loadseg *seg;
925         struct elf32_phdr *phdr;
926         unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
927         loff_t fpos;
928         int loop, ret;
929
930         load_addr = params->load_addr;
931         seg = params->loadmap->segs;
932
933         /* determine the bounds of the contiguous overall allocation we must
934          * make */
935         phdr = params->phdrs;
936         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
937                 if (params->phdrs[loop].p_type != PT_LOAD)
938                         continue;
939
940                 if (base > phdr->p_vaddr)
941                         base = phdr->p_vaddr;
942                 if (top < phdr->p_vaddr + phdr->p_memsz)
943                         top = phdr->p_vaddr + phdr->p_memsz;
944         }
945
946         /* allocate one big anon block for everything */
947         mflags = MAP_PRIVATE;
948         if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
949                 mflags |= MAP_EXECUTABLE;
950
951         down_write(&mm->mmap_sem);
952         maddr = do_mmap(NULL, load_addr, top - base,
953                         PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
954         up_write(&mm->mmap_sem);
955         if (IS_ERR_VALUE(maddr))
956                 return (int) maddr;
957
958         if (load_addr != 0)
959                 load_addr += PAGE_ALIGN(top - base);
960
961         /* and then load the file segments into it */
962         phdr = params->phdrs;
963         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
964                 if (params->phdrs[loop].p_type != PT_LOAD)
965                         continue;
966
967                 fpos = phdr->p_offset;
968
969                 seg->addr = maddr + (phdr->p_vaddr - base);
970                 seg->p_vaddr = phdr->p_vaddr;
971                 seg->p_memsz = phdr->p_memsz;
972
973                 ret = file->f_op->read(file, (void *) seg->addr,
974                                        phdr->p_filesz, &fpos);
975                 if (ret < 0)
976                         return ret;
977
978                 /* map the ELF header address if in this segment */
979                 if (phdr->p_offset == 0)
980                         params->elfhdr_addr = seg->addr;
981
982                 /* clear any space allocated but not loaded */
983                 if (phdr->p_filesz < phdr->p_memsz)
984                         clear_user((void *) (seg->addr + phdr->p_filesz),
985                                    phdr->p_memsz - phdr->p_filesz);
986
987                 if (mm) {
988                         if (phdr->p_flags & PF_X) {
989                                 if (!mm->start_code) {
990                                         mm->start_code = seg->addr;
991                                         mm->end_code = seg->addr +
992                                                 phdr->p_memsz;
993                                 }
994                         } else if (!mm->start_data) {
995                                 mm->start_data = seg->addr;
996 #ifndef CONFIG_MMU
997                                 mm->end_data = seg->addr + phdr->p_memsz;
998 #endif
999                         }
1000
1001 #ifdef CONFIG_MMU
1002                         if (seg->addr + phdr->p_memsz > mm->end_data)
1003                                 mm->end_data = seg->addr + phdr->p_memsz;
1004 #endif
1005                 }
1006
1007                 seg++;
1008         }
1009
1010         return 0;
1011 }
1012 #endif
1013
1014 /*****************************************************************************/
1015 /*
1016  * map a binary by direct mmap() of the individual PT_LOAD segments
1017  */
1018 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1019                                              struct file *file,
1020                                              struct mm_struct *mm)
1021 {
1022         struct elf32_fdpic_loadseg *seg;
1023         struct elf32_phdr *phdr;
1024         unsigned long load_addr, delta_vaddr;
1025         int loop, dvset;
1026
1027         load_addr = params->load_addr;
1028         delta_vaddr = 0;
1029         dvset = 0;
1030
1031         seg = params->loadmap->segs;
1032
1033         /* deal with each load segment separately */
1034         phdr = params->phdrs;
1035         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1036                 unsigned long maddr, disp, excess, excess1;
1037                 int prot = 0, flags;
1038
1039                 if (phdr->p_type != PT_LOAD)
1040                         continue;
1041
1042                 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1043                        (unsigned long) phdr->p_vaddr,
1044                        (unsigned long) phdr->p_offset,
1045                        (unsigned long) phdr->p_filesz,
1046                        (unsigned long) phdr->p_memsz);
1047
1048                 /* determine the mapping parameters */
1049                 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1050                 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1051                 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1052
1053                 flags = MAP_PRIVATE | MAP_DENYWRITE;
1054                 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1055                         flags |= MAP_EXECUTABLE;
1056
1057                 maddr = 0;
1058
1059                 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1060                 case ELF_FDPIC_FLAG_INDEPENDENT:
1061                         /* PT_LOADs are independently locatable */
1062                         break;
1063
1064                 case ELF_FDPIC_FLAG_HONOURVADDR:
1065                         /* the specified virtual address must be honoured */
1066                         maddr = phdr->p_vaddr;
1067                         flags |= MAP_FIXED;
1068                         break;
1069
1070                 case ELF_FDPIC_FLAG_CONSTDISP:
1071                         /* constant displacement
1072                          * - can be mapped anywhere, but must be mapped as a
1073                          *   unit
1074                          */
1075                         if (!dvset) {
1076                                 maddr = load_addr;
1077                                 delta_vaddr = phdr->p_vaddr;
1078                                 dvset = 1;
1079                         } else {
1080                                 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1081                                 flags |= MAP_FIXED;
1082                         }
1083                         break;
1084
1085                 case ELF_FDPIC_FLAG_CONTIGUOUS:
1086                         /* contiguity handled later */
1087                         break;
1088
1089                 default:
1090                         BUG();
1091                 }
1092
1093                 maddr &= PAGE_MASK;
1094
1095                 /* create the mapping */
1096                 disp = phdr->p_vaddr & ~PAGE_MASK;
1097                 down_write(&mm->mmap_sem);
1098                 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1099                                 phdr->p_offset - disp);
1100                 up_write(&mm->mmap_sem);
1101
1102                 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1103                        loop, phdr->p_memsz + disp, prot, flags,
1104                        phdr->p_offset - disp, maddr);
1105
1106                 if (IS_ERR_VALUE(maddr))
1107                         return (int) maddr;
1108
1109                 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1110                     ELF_FDPIC_FLAG_CONTIGUOUS)
1111                         load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1112
1113                 seg->addr = maddr + disp;
1114                 seg->p_vaddr = phdr->p_vaddr;
1115                 seg->p_memsz = phdr->p_memsz;
1116
1117                 /* map the ELF header address if in this segment */
1118                 if (phdr->p_offset == 0)
1119                         params->elfhdr_addr = seg->addr;
1120
1121                 /* clear the bit between beginning of mapping and beginning of
1122                  * PT_LOAD */
1123                 if (prot & PROT_WRITE && disp > 0) {
1124                         kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1125                         clear_user((void __user *) maddr, disp);
1126                         maddr += disp;
1127                 }
1128
1129                 /* clear any space allocated but not loaded
1130                  * - on uClinux we can just clear the lot
1131                  * - on MMU linux we'll get a SIGBUS beyond the last page
1132                  *   extant in the file
1133                  */
1134                 excess = phdr->p_memsz - phdr->p_filesz;
1135                 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1136
1137 #ifdef CONFIG_MMU
1138                 if (excess > excess1) {
1139                         unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1140                         unsigned long xmaddr;
1141
1142                         flags |= MAP_FIXED | MAP_ANONYMOUS;
1143                         down_write(&mm->mmap_sem);
1144                         xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1145                                          prot, flags, 0);
1146                         up_write(&mm->mmap_sem);
1147
1148                         kdebug("mmap[%d] <anon>"
1149                                " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1150                                loop, xaddr, excess - excess1, prot, flags,
1151                                xmaddr);
1152
1153                         if (xmaddr != xaddr)
1154                                 return -ENOMEM;
1155                 }
1156
1157                 if (prot & PROT_WRITE && excess1 > 0) {
1158                         kdebug("clear[%d] ad=%lx sz=%lx",
1159                                loop, maddr + phdr->p_filesz, excess1);
1160                         clear_user((void __user *) maddr + phdr->p_filesz,
1161                                    excess1);
1162                 }
1163
1164 #else
1165                 if (excess > 0) {
1166                         kdebug("clear[%d] ad=%lx sz=%lx",
1167                                loop, maddr + phdr->p_filesz, excess);
1168                         clear_user((void *) maddr + phdr->p_filesz, excess);
1169                 }
1170 #endif
1171
1172                 if (mm) {
1173                         if (phdr->p_flags & PF_X) {
1174                                 if (!mm->start_code) {
1175                                         mm->start_code = maddr;
1176                                         mm->end_code = maddr + phdr->p_memsz;
1177                                 }
1178                         } else if (!mm->start_data) {
1179                                 mm->start_data = maddr;
1180                                 mm->end_data = maddr + phdr->p_memsz;
1181                         }
1182                 }
1183
1184                 seg++;
1185         }
1186
1187         return 0;
1188 }
1189
1190 /*****************************************************************************/
1191 /*
1192  * ELF-FDPIC core dumper
1193  *
1194  * Modelled on fs/exec.c:aout_core_dump()
1195  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1196  *
1197  * Modelled on fs/binfmt_elf.c core dumper
1198  */
1199 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1200
1201 /*
1202  * These are the only things you should do on a core-file: use only these
1203  * functions to write out all the necessary info.
1204  */
1205 static int dump_write(struct file *file, const void *addr, int nr)
1206 {
1207         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1208 }
1209
1210 static int dump_seek(struct file *file, loff_t off)
1211 {
1212         if (file->f_op->llseek) {
1213                 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1214                         return 0;
1215         } else {
1216                 file->f_pos = off;
1217         }
1218         return 1;
1219 }
1220
1221 /*
1222  * Decide whether a segment is worth dumping; default is yes to be
1223  * sure (missing info is worse than too much; etc).
1224  * Personally I'd include everything, and use the coredump limit...
1225  *
1226  * I think we should skip something. But I am not sure how. H.J.
1227  */
1228 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1229 {
1230         int dump_ok;
1231
1232         /* Do not dump I/O mapped devices or special mappings */
1233         if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1234                 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1235                 return 0;
1236         }
1237
1238         /* If we may not read the contents, don't allow us to dump
1239          * them either. "dump_write()" can't handle it anyway.
1240          */
1241         if (!(vma->vm_flags & VM_READ)) {
1242                 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1243                 return 0;
1244         }
1245
1246         /* By default, dump shared memory if mapped from an anonymous file. */
1247         if (vma->vm_flags & VM_SHARED) {
1248                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1249                         dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1250                         kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1251                                vma->vm_flags, dump_ok ? "yes" : "no");
1252                         return dump_ok;
1253                 }
1254
1255                 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1256                 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1257                        vma->vm_flags, dump_ok ? "yes" : "no");
1258                 return dump_ok;
1259         }
1260
1261 #ifdef CONFIG_MMU
1262         /* By default, if it hasn't been written to, don't write it out */
1263         if (!vma->anon_vma) {
1264                 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1265                 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1266                        vma->vm_flags, dump_ok ? "yes" : "no");
1267                 return dump_ok;
1268         }
1269 #endif
1270
1271         dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1272         kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1273                dump_ok ? "yes" : "no");
1274         return dump_ok;
1275 }
1276
1277 /* An ELF note in memory */
1278 struct memelfnote
1279 {
1280         const char *name;
1281         int type;
1282         unsigned int datasz;
1283         void *data;
1284 };
1285
1286 static int notesize(struct memelfnote *en)
1287 {
1288         int sz;
1289
1290         sz = sizeof(struct elf_note);
1291         sz += roundup(strlen(en->name) + 1, 4);
1292         sz += roundup(en->datasz, 4);
1293
1294         return sz;
1295 }
1296
1297 /* #define DEBUG */
1298
1299 #define DUMP_WRITE(addr, nr)    \
1300         do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1301 #define DUMP_SEEK(off)  \
1302         do { if (!dump_seek(file, (off))) return 0; } while(0)
1303
1304 static int writenote(struct memelfnote *men, struct file *file)
1305 {
1306         struct elf_note en;
1307
1308         en.n_namesz = strlen(men->name) + 1;
1309         en.n_descsz = men->datasz;
1310         en.n_type = men->type;
1311
1312         DUMP_WRITE(&en, sizeof(en));
1313         DUMP_WRITE(men->name, en.n_namesz);
1314         /* XXX - cast from long long to long to avoid need for libgcc.a */
1315         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1316         DUMP_WRITE(men->data, men->datasz);
1317         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1318
1319         return 1;
1320 }
1321 #undef DUMP_WRITE
1322 #undef DUMP_SEEK
1323
1324 #define DUMP_WRITE(addr, nr)    \
1325         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1326                 goto end_coredump;
1327 #define DUMP_SEEK(off)  \
1328         if (!dump_seek(file, (off))) \
1329                 goto end_coredump;
1330
1331 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1332 {
1333         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1334         elf->e_ident[EI_CLASS] = ELF_CLASS;
1335         elf->e_ident[EI_DATA] = ELF_DATA;
1336         elf->e_ident[EI_VERSION] = EV_CURRENT;
1337         elf->e_ident[EI_OSABI] = ELF_OSABI;
1338         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1339
1340         elf->e_type = ET_CORE;
1341         elf->e_machine = ELF_ARCH;
1342         elf->e_version = EV_CURRENT;
1343         elf->e_entry = 0;
1344         elf->e_phoff = sizeof(struct elfhdr);
1345         elf->e_shoff = 0;
1346         elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1347         elf->e_ehsize = sizeof(struct elfhdr);
1348         elf->e_phentsize = sizeof(struct elf_phdr);
1349         elf->e_phnum = segs;
1350         elf->e_shentsize = 0;
1351         elf->e_shnum = 0;
1352         elf->e_shstrndx = 0;
1353         return;
1354 }
1355
1356 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1357 {
1358         phdr->p_type = PT_NOTE;
1359         phdr->p_offset = offset;
1360         phdr->p_vaddr = 0;
1361         phdr->p_paddr = 0;
1362         phdr->p_filesz = sz;
1363         phdr->p_memsz = 0;
1364         phdr->p_flags = 0;
1365         phdr->p_align = 0;
1366         return;
1367 }
1368
1369 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1370                 unsigned int sz, void *data)
1371 {
1372         note->name = name;
1373         note->type = type;
1374         note->datasz = sz;
1375         note->data = data;
1376         return;
1377 }
1378
1379 /*
1380  * fill up all the fields in prstatus from the given task struct, except
1381  * registers which need to be filled up seperately.
1382  */
1383 static void fill_prstatus(struct elf_prstatus *prstatus,
1384                           struct task_struct *p, long signr)
1385 {
1386         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1387         prstatus->pr_sigpend = p->pending.signal.sig[0];
1388         prstatus->pr_sighold = p->blocked.sig[0];
1389         prstatus->pr_pid = task_pid_vnr(p);
1390         prstatus->pr_ppid = task_pid_vnr(p->parent);
1391         prstatus->pr_pgrp = task_pgrp_vnr(p);
1392         prstatus->pr_sid = task_session_vnr(p);
1393         if (thread_group_leader(p)) {
1394                 struct task_cputime cputime;
1395
1396                 /*
1397                  * This is the record for the group leader.  It shows the
1398                  * group-wide total, not its individual thread total.
1399                  */
1400                 thread_group_cputime(p, &cputime);
1401                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1402                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1403         } else {
1404                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1405                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1406         }
1407         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1408         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1409
1410         prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1411         prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1412 }
1413
1414 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1415                        struct mm_struct *mm)
1416 {
1417         const struct cred *cred;
1418         unsigned int i, len;
1419
1420         /* first copy the parameters from user space */
1421         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1422
1423         len = mm->arg_end - mm->arg_start;
1424         if (len >= ELF_PRARGSZ)
1425                 len = ELF_PRARGSZ - 1;
1426         if (copy_from_user(&psinfo->pr_psargs,
1427                            (const char __user *) mm->arg_start, len))
1428                 return -EFAULT;
1429         for (i = 0; i < len; i++)
1430                 if (psinfo->pr_psargs[i] == 0)
1431                         psinfo->pr_psargs[i] = ' ';
1432         psinfo->pr_psargs[len] = 0;
1433
1434         psinfo->pr_pid = task_pid_vnr(p);
1435         psinfo->pr_ppid = task_pid_vnr(p->parent);
1436         psinfo->pr_pgrp = task_pgrp_vnr(p);
1437         psinfo->pr_sid = task_session_vnr(p);
1438
1439         i = p->state ? ffz(~p->state) + 1 : 0;
1440         psinfo->pr_state = i;
1441         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1442         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1443         psinfo->pr_nice = task_nice(p);
1444         psinfo->pr_flag = p->flags;
1445         rcu_read_lock();
1446         cred = __task_cred(p);
1447         SET_UID(psinfo->pr_uid, cred->uid);
1448         SET_GID(psinfo->pr_gid, cred->gid);
1449         rcu_read_unlock();
1450         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1451
1452         return 0;
1453 }
1454
1455 /* Here is the structure in which status of each thread is captured. */
1456 struct elf_thread_status
1457 {
1458         struct list_head list;
1459         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1460         elf_fpregset_t fpu;             /* NT_PRFPREG */
1461         struct task_struct *thread;
1462 #ifdef ELF_CORE_COPY_XFPREGS
1463         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1464 #endif
1465         struct memelfnote notes[3];
1466         int num_notes;
1467 };
1468
1469 /*
1470  * In order to add the specific thread information for the elf file format,
1471  * we need to keep a linked list of every thread's pr_status and then create
1472  * a single section for them in the final core file.
1473  */
1474 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1475 {
1476         struct task_struct *p = t->thread;
1477         int sz = 0;
1478
1479         t->num_notes = 0;
1480
1481         fill_prstatus(&t->prstatus, p, signr);
1482         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1483
1484         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1485                   &t->prstatus);
1486         t->num_notes++;
1487         sz += notesize(&t->notes[0]);
1488
1489         t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1490         if (t->prstatus.pr_fpvalid) {
1491                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1492                           &t->fpu);
1493                 t->num_notes++;
1494                 sz += notesize(&t->notes[1]);
1495         }
1496
1497 #ifdef ELF_CORE_COPY_XFPREGS
1498         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1499                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1500                           sizeof(t->xfpu), &t->xfpu);
1501                 t->num_notes++;
1502                 sz += notesize(&t->notes[2]);
1503         }
1504 #endif
1505         return sz;
1506 }
1507
1508 /*
1509  * dump the segments for an MMU process
1510  */
1511 #ifdef CONFIG_MMU
1512 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1513                            unsigned long *limit, unsigned long mm_flags)
1514 {
1515         struct vm_area_struct *vma;
1516
1517         for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1518                 unsigned long addr;
1519
1520                 if (!maydump(vma, mm_flags))
1521                         continue;
1522
1523                 for (addr = vma->vm_start;
1524                      addr < vma->vm_end;
1525                      addr += PAGE_SIZE
1526                      ) {
1527                         struct vm_area_struct *vma;
1528                         struct page *page;
1529
1530                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1531                                            &page, &vma) <= 0) {
1532                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1533                         }
1534                         else if (page == ZERO_PAGE(0)) {
1535                                 page_cache_release(page);
1536                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1537                         }
1538                         else {
1539                                 void *kaddr;
1540
1541                                 flush_cache_page(vma, addr, page_to_pfn(page));
1542                                 kaddr = kmap(page);
1543                                 if ((*size += PAGE_SIZE) > *limit ||
1544                                     !dump_write(file, kaddr, PAGE_SIZE)
1545                                     ) {
1546                                         kunmap(page);
1547                                         page_cache_release(page);
1548                                         return -EIO;
1549                                 }
1550                                 kunmap(page);
1551                                 page_cache_release(page);
1552                         }
1553                 }
1554         }
1555
1556         return 0;
1557
1558 end_coredump:
1559         return -EFBIG;
1560 }
1561 #endif
1562
1563 /*
1564  * dump the segments for a NOMMU process
1565  */
1566 #ifndef CONFIG_MMU
1567 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1568                            unsigned long *limit, unsigned long mm_flags)
1569 {
1570         struct vm_list_struct *vml;
1571
1572         for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1573         struct vm_area_struct *vma = vml->vma;
1574
1575                 if (!maydump(vma, mm_flags))
1576                         continue;
1577
1578                 if ((*size += PAGE_SIZE) > *limit)
1579                         return -EFBIG;
1580
1581                 if (!dump_write(file, (void *) vma->vm_start,
1582                                 vma->vm_end - vma->vm_start))
1583                         return -EIO;
1584         }
1585
1586         return 0;
1587 }
1588 #endif
1589
1590 /*
1591  * Actual dumper
1592  *
1593  * This is a two-pass process; first we find the offsets of the bits,
1594  * and then they are actually written out.  If we run out of core limit
1595  * we just truncate.
1596  */
1597 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1598                                struct file *file, unsigned long limit)
1599 {
1600 #define NUM_NOTES       6
1601         int has_dumped = 0;
1602         mm_segment_t fs;
1603         int segs;
1604         size_t size = 0;
1605         int i;
1606         struct vm_area_struct *vma;
1607         struct elfhdr *elf = NULL;
1608         loff_t offset = 0, dataoff;
1609         int numnote;
1610         struct memelfnote *notes = NULL;
1611         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1612         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1613         LIST_HEAD(thread_list);
1614         struct list_head *t;
1615         elf_fpregset_t *fpu = NULL;
1616 #ifdef ELF_CORE_COPY_XFPREGS
1617         elf_fpxregset_t *xfpu = NULL;
1618 #endif
1619         int thread_status_size = 0;
1620 #ifndef CONFIG_MMU
1621         struct vm_list_struct *vml;
1622 #endif
1623         elf_addr_t *auxv;
1624         unsigned long mm_flags;
1625
1626         /*
1627          * We no longer stop all VM operations.
1628          *
1629          * This is because those proceses that could possibly change map_count
1630          * or the mmap / vma pages are now blocked in do_exit on current
1631          * finishing this core dump.
1632          *
1633          * Only ptrace can touch these memory addresses, but it doesn't change
1634          * the map_count or the pages allocated. So no possibility of crashing
1635          * exists while dumping the mm->vm_next areas to the core file.
1636          */
1637
1638         /* alloc memory for large data structures: too large to be on stack */
1639         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1640         if (!elf)
1641                 goto cleanup;
1642         prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1643         if (!prstatus)
1644                 goto cleanup;
1645         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1646         if (!psinfo)
1647                 goto cleanup;
1648         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1649         if (!notes)
1650                 goto cleanup;
1651         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1652         if (!fpu)
1653                 goto cleanup;
1654 #ifdef ELF_CORE_COPY_XFPREGS
1655         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1656         if (!xfpu)
1657                 goto cleanup;
1658 #endif
1659
1660         if (signr) {
1661                 struct core_thread *ct;
1662                 struct elf_thread_status *tmp;
1663
1664                 for (ct = current->mm->core_state->dumper.next;
1665                                                 ct; ct = ct->next) {
1666                         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1667                         if (!tmp)
1668                                 goto cleanup;
1669
1670                         tmp->thread = ct->task;
1671                         list_add(&tmp->list, &thread_list);
1672                 }
1673
1674                 list_for_each(t, &thread_list) {
1675                         struct elf_thread_status *tmp;
1676                         int sz;
1677
1678                         tmp = list_entry(t, struct elf_thread_status, list);
1679                         sz = elf_dump_thread_status(signr, tmp);
1680                         thread_status_size += sz;
1681                 }
1682         }
1683
1684         /* now collect the dump for the current */
1685         fill_prstatus(prstatus, current, signr);
1686         elf_core_copy_regs(&prstatus->pr_reg, regs);
1687
1688 #ifdef CONFIG_MMU
1689         segs = current->mm->map_count;
1690 #else
1691         segs = 0;
1692         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1693             segs++;
1694 #endif
1695 #ifdef ELF_CORE_EXTRA_PHDRS
1696         segs += ELF_CORE_EXTRA_PHDRS;
1697 #endif
1698
1699         /* Set up header */
1700         fill_elf_fdpic_header(elf, segs + 1);   /* including notes section */
1701
1702         has_dumped = 1;
1703         current->flags |= PF_DUMPCORE;
1704
1705         /*
1706          * Set up the notes in similar form to SVR4 core dumps made
1707          * with info from their /proc.
1708          */
1709
1710         fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1711         fill_psinfo(psinfo, current->group_leader, current->mm);
1712         fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1713
1714         numnote = 2;
1715
1716         auxv = (elf_addr_t *) current->mm->saved_auxv;
1717
1718         i = 0;
1719         do
1720                 i += 2;
1721         while (auxv[i - 2] != AT_NULL);
1722         fill_note(&notes[numnote++], "CORE", NT_AUXV,
1723                   i * sizeof(elf_addr_t), auxv);
1724
1725         /* Try to dump the FPU. */
1726         if ((prstatus->pr_fpvalid =
1727              elf_core_copy_task_fpregs(current, regs, fpu)))
1728                 fill_note(notes + numnote++,
1729                           "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1730 #ifdef ELF_CORE_COPY_XFPREGS
1731         if (elf_core_copy_task_xfpregs(current, xfpu))
1732                 fill_note(notes + numnote++,
1733                           "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1734 #endif
1735
1736         fs = get_fs();
1737         set_fs(KERNEL_DS);
1738
1739         DUMP_WRITE(elf, sizeof(*elf));
1740         offset += sizeof(*elf);                         /* Elf header */
1741         offset += (segs+1) * sizeof(struct elf_phdr);   /* Program headers */
1742
1743         /* Write notes phdr entry */
1744         {
1745                 struct elf_phdr phdr;
1746                 int sz = 0;
1747
1748                 for (i = 0; i < numnote; i++)
1749                         sz += notesize(notes + i);
1750
1751                 sz += thread_status_size;
1752
1753                 fill_elf_note_phdr(&phdr, sz, offset);
1754                 offset += sz;
1755                 DUMP_WRITE(&phdr, sizeof(phdr));
1756         }
1757
1758         /* Page-align dumped data */
1759         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1760
1761         /*
1762          * We must use the same mm->flags while dumping core to avoid
1763          * inconsistency between the program headers and bodies, otherwise an
1764          * unusable core file can be generated.
1765          */
1766         mm_flags = current->mm->flags;
1767
1768         /* write program headers for segments dump */
1769         for (
1770 #ifdef CONFIG_MMU
1771                 vma = current->mm->mmap; vma; vma = vma->vm_next
1772 #else
1773                         vml = current->mm->context.vmlist; vml; vml = vml->next
1774 #endif
1775              ) {
1776                 struct elf_phdr phdr;
1777                 size_t sz;
1778
1779 #ifndef CONFIG_MMU
1780                 vma = vml->vma;
1781 #endif
1782
1783                 sz = vma->vm_end - vma->vm_start;
1784
1785                 phdr.p_type = PT_LOAD;
1786                 phdr.p_offset = offset;
1787                 phdr.p_vaddr = vma->vm_start;
1788                 phdr.p_paddr = 0;
1789                 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1790                 phdr.p_memsz = sz;
1791                 offset += phdr.p_filesz;
1792                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1793                 if (vma->vm_flags & VM_WRITE)
1794                         phdr.p_flags |= PF_W;
1795                 if (vma->vm_flags & VM_EXEC)
1796                         phdr.p_flags |= PF_X;
1797                 phdr.p_align = ELF_EXEC_PAGESIZE;
1798
1799                 DUMP_WRITE(&phdr, sizeof(phdr));
1800         }
1801
1802 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1803         ELF_CORE_WRITE_EXTRA_PHDRS;
1804 #endif
1805
1806         /* write out the notes section */
1807         for (i = 0; i < numnote; i++)
1808                 if (!writenote(notes + i, file))
1809                         goto end_coredump;
1810
1811         /* write out the thread status notes section */
1812         list_for_each(t, &thread_list) {
1813                 struct elf_thread_status *tmp =
1814                                 list_entry(t, struct elf_thread_status, list);
1815
1816                 for (i = 0; i < tmp->num_notes; i++)
1817                         if (!writenote(&tmp->notes[i], file))
1818                                 goto end_coredump;
1819         }
1820
1821         DUMP_SEEK(dataoff);
1822
1823         if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1824                 goto end_coredump;
1825
1826 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1827         ELF_CORE_WRITE_EXTRA_DATA;
1828 #endif
1829
1830         if (file->f_pos != offset) {
1831                 /* Sanity check */
1832                 printk(KERN_WARNING
1833                        "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1834                        file->f_pos, offset);
1835         }
1836
1837 end_coredump:
1838         set_fs(fs);
1839
1840 cleanup:
1841         while (!list_empty(&thread_list)) {
1842                 struct list_head *tmp = thread_list.next;
1843                 list_del(tmp);
1844                 kfree(list_entry(tmp, struct elf_thread_status, list));
1845         }
1846
1847         kfree(elf);
1848         kfree(prstatus);
1849         kfree(psinfo);
1850         kfree(notes);
1851         kfree(fpu);
1852 #ifdef ELF_CORE_COPY_XFPREGS
1853         kfree(xfpu);
1854 #endif
1855         return has_dumped;
1856 #undef NUM_NOTES
1857 }
1858
1859 #endif          /* USE_ELF_CORE_DUMP */