binfmt_elf_fdpic: convert initial stack alignment to arch_align_stack()
[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/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/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/pgalloc.h>
40
41 typedef char *elf_caddr_t;
42
43 #if 0
44 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
45 #else
46 #define kdebug(fmt, ...) do {} while(0)
47 #endif
48
49 #if 0
50 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
51 #else
52 #define kdcore(fmt, ...) do {} while(0)
53 #endif
54
55 MODULE_LICENSE("GPL");
56
57 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
58 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
59 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
60                               struct mm_struct *, const char *);
61
62 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
63                                    struct elf_fdpic_params *,
64                                    struct elf_fdpic_params *);
65
66 #ifndef CONFIG_MMU
67 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
68                                             unsigned long *);
69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
70                                                    struct file *,
71                                                    struct mm_struct *);
72 #endif
73
74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
75                                              struct file *, struct mm_struct *);
76
77 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
78 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
79 #endif
80
81 static struct linux_binfmt elf_fdpic_format = {
82         .module         = THIS_MODULE,
83         .load_binary    = load_elf_fdpic_binary,
84 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
85         .core_dump      = elf_fdpic_core_dump,
86 #endif
87         .min_coredump   = ELF_EXEC_PAGESIZE,
88 };
89
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92         return register_binfmt(&elf_fdpic_format);
93 }
94
95 static void __exit exit_elf_fdpic_binfmt(void)
96 {
97         unregister_binfmt(&elf_fdpic_format);
98 }
99
100 core_initcall(init_elf_fdpic_binfmt);
101 module_exit(exit_elf_fdpic_binfmt);
102
103 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
104 {
105         if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
106                 return 0;
107         if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
108                 return 0;
109         if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
110                 return 0;
111         if (!file->f_op || !file->f_op->mmap)
112                 return 0;
113         return 1;
114 }
115
116 /*****************************************************************************/
117 /*
118  * read the program headers table into memory
119  */
120 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
121                                  struct file *file)
122 {
123         struct elf32_phdr *phdr;
124         unsigned long size;
125         int retval, loop;
126
127         if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
128                 return -ENOMEM;
129         if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
130                 return -ENOMEM;
131
132         size = params->hdr.e_phnum * sizeof(struct elf_phdr);
133         params->phdrs = kmalloc(size, GFP_KERNEL);
134         if (!params->phdrs)
135                 return -ENOMEM;
136
137         retval = kernel_read(file, params->hdr.e_phoff,
138                              (char *) params->phdrs, size);
139         if (unlikely(retval != size))
140                 return retval < 0 ? retval : -ENOEXEC;
141
142         /* determine stack size for this binary */
143         phdr = params->phdrs;
144         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
145                 if (phdr->p_type != PT_GNU_STACK)
146                         continue;
147
148                 if (phdr->p_flags & PF_X)
149                         params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
150                 else
151                         params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
152
153                 params->stack_size = phdr->p_memsz;
154                 break;
155         }
156
157         return 0;
158 }
159
160 /*****************************************************************************/
161 /*
162  * load an fdpic binary into various bits of memory
163  */
164 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
165                                  struct pt_regs *regs)
166 {
167         struct elf_fdpic_params exec_params, interp_params;
168         struct elf_phdr *phdr;
169         unsigned long stack_size, entryaddr;
170 #ifndef CONFIG_MMU
171         unsigned long fullsize;
172 #endif
173 #ifdef ELF_FDPIC_PLAT_INIT
174         unsigned long dynaddr;
175 #endif
176         struct file *interpreter = NULL; /* to shut gcc up */
177         char *interpreter_name = NULL;
178         int executable_stack;
179         int retval, i;
180
181         kdebug("____ LOAD %d ____", current->pid);
182
183         memset(&exec_params, 0, sizeof(exec_params));
184         memset(&interp_params, 0, sizeof(interp_params));
185
186         exec_params.hdr = *(struct elfhdr *) bprm->buf;
187         exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
188
189         /* check that this is a binary we know how to deal with */
190         retval = -ENOEXEC;
191         if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
192                 goto error;
193
194         /* read the program header table */
195         retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
196         if (retval < 0)
197                 goto error;
198
199         /* scan for a program header that specifies an interpreter */
200         phdr = exec_params.phdrs;
201
202         for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
203                 switch (phdr->p_type) {
204                 case PT_INTERP:
205                         retval = -ENOMEM;
206                         if (phdr->p_filesz > PATH_MAX)
207                                 goto error;
208                         retval = -ENOENT;
209                         if (phdr->p_filesz < 2)
210                                 goto error;
211
212                         /* read the name of the interpreter into memory */
213                         interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
214                         if (!interpreter_name)
215                                 goto error;
216
217                         retval = kernel_read(bprm->file,
218                                              phdr->p_offset,
219                                              interpreter_name,
220                                              phdr->p_filesz);
221                         if (unlikely(retval != phdr->p_filesz)) {
222                                 if (retval >= 0)
223                                         retval = -ENOEXEC;
224                                 goto error;
225                         }
226
227                         retval = -ENOENT;
228                         if (interpreter_name[phdr->p_filesz - 1] != '\0')
229                                 goto error;
230
231                         kdebug("Using ELF interpreter %s", interpreter_name);
232
233                         /* replace the program with the interpreter */
234                         interpreter = open_exec(interpreter_name);
235                         retval = PTR_ERR(interpreter);
236                         if (IS_ERR(interpreter)) {
237                                 interpreter = NULL;
238                                 goto error;
239                         }
240
241                         /*
242                          * If the binary is not readable then enforce
243                          * mm->dumpable = 0 regardless of the interpreter's
244                          * permissions.
245                          */
246                         if (file_permission(interpreter, MAY_READ) < 0)
247                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
248
249                         retval = kernel_read(interpreter, 0, bprm->buf,
250                                              BINPRM_BUF_SIZE);
251                         if (unlikely(retval != BINPRM_BUF_SIZE)) {
252                                 if (retval >= 0)
253                                         retval = -ENOEXEC;
254                                 goto error;
255                         }
256
257                         interp_params.hdr = *((struct elfhdr *) bprm->buf);
258                         break;
259
260                 case PT_LOAD:
261 #ifdef CONFIG_MMU
262                         if (exec_params.load_addr == 0)
263                                 exec_params.load_addr = phdr->p_vaddr;
264 #endif
265                         break;
266                 }
267
268         }
269
270         if (elf_check_const_displacement(&exec_params.hdr))
271                 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
272
273         /* perform insanity checks on the interpreter */
274         if (interpreter_name) {
275                 retval = -ELIBBAD;
276                 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
277                         goto error;
278
279                 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
280
281                 /* read the interpreter's program header table */
282                 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
283                 if (retval < 0)
284                         goto error;
285         }
286
287         stack_size = exec_params.stack_size;
288         if (stack_size < interp_params.stack_size)
289                 stack_size = interp_params.stack_size;
290
291         if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
292                 executable_stack = EXSTACK_ENABLE_X;
293         else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
294                 executable_stack = EXSTACK_DISABLE_X;
295         else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
296                 executable_stack = EXSTACK_ENABLE_X;
297         else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
298                 executable_stack = EXSTACK_DISABLE_X;
299         else
300                 executable_stack = EXSTACK_DEFAULT;
301
302         retval = -ENOEXEC;
303         if (stack_size == 0)
304                 goto error;
305
306         if (elf_check_const_displacement(&interp_params.hdr))
307                 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
308
309         /* flush all traces of the currently running executable */
310         retval = flush_old_exec(bprm);
311         if (retval)
312                 goto error;
313
314         /* there's now no turning back... the old userspace image is dead,
315          * defunct, deceased, etc. after this point we have to exit via
316          * error_kill */
317         set_personality(PER_LINUX_FDPIC);
318         set_binfmt(&elf_fdpic_format);
319
320         current->mm->start_code = 0;
321         current->mm->end_code = 0;
322         current->mm->start_stack = 0;
323         current->mm->start_data = 0;
324         current->mm->end_data = 0;
325         current->mm->context.exec_fdpic_loadmap = 0;
326         current->mm->context.interp_fdpic_loadmap = 0;
327
328         current->flags &= ~PF_FORKNOEXEC;
329
330 #ifdef CONFIG_MMU
331         elf_fdpic_arch_lay_out_mm(&exec_params,
332                                   &interp_params,
333                                   &current->mm->start_stack,
334                                   &current->mm->start_brk);
335
336         retval = setup_arg_pages(bprm, current->mm->start_stack,
337                                  executable_stack);
338         if (retval < 0) {
339                 send_sig(SIGKILL, current, 0);
340                 goto error_kill;
341         }
342 #endif
343
344         /* load the executable and interpreter into memory */
345         retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
346                                     "executable");
347         if (retval < 0)
348                 goto error_kill;
349
350         if (interpreter_name) {
351                 retval = elf_fdpic_map_file(&interp_params, interpreter,
352                                             current->mm, "interpreter");
353                 if (retval < 0) {
354                         printk(KERN_ERR "Unable to load interpreter\n");
355                         goto error_kill;
356                 }
357
358                 allow_write_access(interpreter);
359                 fput(interpreter);
360                 interpreter = NULL;
361         }
362
363 #ifdef CONFIG_MMU
364         if (!current->mm->start_brk)
365                 current->mm->start_brk = current->mm->end_data;
366
367         current->mm->brk = current->mm->start_brk =
368                 PAGE_ALIGN(current->mm->start_brk);
369
370 #else
371         /* create a stack and brk area big enough for everyone
372          * - the brk heap starts at the bottom and works up
373          * - the stack starts at the top and works down
374          */
375         stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
376         if (stack_size < PAGE_SIZE * 2)
377                 stack_size = PAGE_SIZE * 2;
378
379         down_write(&current->mm->mmap_sem);
380         current->mm->start_brk = do_mmap(NULL, 0, stack_size,
381                                          PROT_READ | PROT_WRITE | PROT_EXEC,
382                                          MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
383                                          0);
384
385         if (IS_ERR_VALUE(current->mm->start_brk)) {
386                 up_write(&current->mm->mmap_sem);
387                 retval = current->mm->start_brk;
388                 current->mm->start_brk = 0;
389                 goto error_kill;
390         }
391
392         /* expand the stack mapping to use up the entire allocation granule */
393         fullsize = kobjsize((char *) current->mm->start_brk);
394         if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
395                                     fullsize, 0, 0)))
396                 stack_size = fullsize;
397         up_write(&current->mm->mmap_sem);
398
399         current->mm->brk = current->mm->start_brk;
400         current->mm->context.end_brk = current->mm->start_brk;
401         current->mm->context.end_brk +=
402                 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
403         current->mm->start_stack = current->mm->start_brk + stack_size;
404 #endif
405
406         compute_creds(bprm);
407         current->flags &= ~PF_FORKNOEXEC;
408         if (create_elf_fdpic_tables(bprm, current->mm,
409                                     &exec_params, &interp_params) < 0)
410                 goto error_kill;
411
412         kdebug("- start_code  %lx", current->mm->start_code);
413         kdebug("- end_code    %lx", current->mm->end_code);
414         kdebug("- start_data  %lx", current->mm->start_data);
415         kdebug("- end_data    %lx", current->mm->end_data);
416         kdebug("- start_brk   %lx", current->mm->start_brk);
417         kdebug("- brk         %lx", current->mm->brk);
418         kdebug("- start_stack %lx", current->mm->start_stack);
419
420 #ifdef ELF_FDPIC_PLAT_INIT
421         /*
422          * The ABI may specify that certain registers be set up in special
423          * ways (on i386 %edx is the address of a DT_FINI function, for
424          * example.  This macro performs whatever initialization to
425          * the regs structure is required.
426          */
427         dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
428         ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
429                             dynaddr);
430 #endif
431
432         /* everything is now ready... get the userspace context ready to roll */
433         entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
434         start_thread(regs, entryaddr, current->mm->start_stack);
435
436         retval = 0;
437
438 error:
439         if (interpreter) {
440                 allow_write_access(interpreter);
441                 fput(interpreter);
442         }
443         kfree(interpreter_name);
444         kfree(exec_params.phdrs);
445         kfree(exec_params.loadmap);
446         kfree(interp_params.phdrs);
447         kfree(interp_params.loadmap);
448         return retval;
449
450         /* unrecoverable error - kill the process */
451 error_kill:
452         send_sig(SIGSEGV, current, 0);
453         goto error;
454
455 }
456
457 /*****************************************************************************/
458
459 #ifndef ELF_BASE_PLATFORM
460 /*
461  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
462  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
463  * will be copied to the user stack in the same manner as AT_PLATFORM.
464  */
465 #define ELF_BASE_PLATFORM NULL
466 #endif
467
468 /*
469  * present useful information to the program by shovelling it onto the new
470  * process's stack
471  */
472 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
473                                    struct mm_struct *mm,
474                                    struct elf_fdpic_params *exec_params,
475                                    struct elf_fdpic_params *interp_params)
476 {
477         unsigned long sp, csp, nitems;
478         elf_caddr_t __user *argv, *envp;
479         size_t platform_len = 0, len;
480         char *k_platform, *k_base_platform;
481         char __user *u_platform, *u_base_platform, *p;
482         long hwcap;
483         int loop;
484         int nr; /* reset for each csp adjustment */
485
486 #ifdef CONFIG_MMU
487         /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
488          * by the processes running on the same package. One thing we can do is
489          * to shuffle the initial stack for them, so we give the architecture
490          * an opportunity to do so here.
491          */
492         sp = arch_align_stack(bprm->p);
493 #else
494         sp = mm->start_stack;
495
496         /* stack the program arguments and environment */
497         if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
498                 return -EFAULT;
499 #endif
500
501         hwcap = ELF_HWCAP;
502
503         /*
504          * If this architecture has a platform capability string, copy it
505          * to userspace.  In some cases (Sparc), this info is impossible
506          * for userspace to get any other way, in others (i386) it is
507          * merely difficult.
508          */
509         k_platform = ELF_PLATFORM;
510         u_platform = NULL;
511
512         if (k_platform) {
513                 platform_len = strlen(k_platform) + 1;
514                 sp -= platform_len;
515                 u_platform = (char __user *) sp;
516                 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
517                         return -EFAULT;
518         }
519
520         /*
521          * If this architecture has a "base" platform capability
522          * string, copy it to userspace.
523          */
524         k_base_platform = ELF_BASE_PLATFORM;
525         u_base_platform = NULL;
526
527         if (k_base_platform) {
528                 platform_len = strlen(k_base_platform) + 1;
529                 sp -= platform_len;
530                 u_base_platform = (char __user *) sp;
531                 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
532                         return -EFAULT;
533         }
534
535         sp &= ~7UL;
536
537         /* stack the load map(s) */
538         len = sizeof(struct elf32_fdpic_loadmap);
539         len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
540         sp = (sp - len) & ~7UL;
541         exec_params->map_addr = sp;
542
543         if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
544                 return -EFAULT;
545
546         current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
547
548         if (interp_params->loadmap) {
549                 len = sizeof(struct elf32_fdpic_loadmap);
550                 len += sizeof(struct elf32_fdpic_loadseg) *
551                         interp_params->loadmap->nsegs;
552                 sp = (sp - len) & ~7UL;
553                 interp_params->map_addr = sp;
554
555                 if (copy_to_user((void __user *) sp, interp_params->loadmap,
556                                  len) != 0)
557                         return -EFAULT;
558
559                 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
560         }
561
562         /* force 16 byte _final_ alignment here for generality */
563 #define DLINFO_ITEMS 13
564
565         nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
566                 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
567
568         csp = sp;
569         sp -= nitems * 2 * sizeof(unsigned long);
570         sp -= (bprm->envc + 1) * sizeof(char *);        /* envv[] */
571         sp -= (bprm->argc + 1) * sizeof(char *);        /* argv[] */
572         sp -= 1 * sizeof(unsigned long);                /* argc */
573
574         csp -= sp & 15UL;
575         sp -= sp & 15UL;
576
577         /* put the ELF interpreter info on the stack */
578 #define NEW_AUX_ENT(id, val)                                            \
579         do {                                                            \
580                 struct { unsigned long _id, _val; } __user *ent;        \
581                                                                         \
582                 ent = (void __user *) csp;                              \
583                 __put_user((id), &ent[nr]._id);                         \
584                 __put_user((val), &ent[nr]._val);                       \
585                 nr++;                                                   \
586         } while (0)
587
588         nr = 0;
589         csp -= 2 * sizeof(unsigned long);
590         NEW_AUX_ENT(AT_NULL, 0);
591         if (k_platform) {
592                 nr = 0;
593                 csp -= 2 * sizeof(unsigned long);
594                 NEW_AUX_ENT(AT_PLATFORM,
595                             (elf_addr_t) (unsigned long) u_platform);
596         }
597
598         if (k_base_platform) {
599                 nr = 0;
600                 csp -= 2 * sizeof(unsigned long);
601                 NEW_AUX_ENT(AT_BASE_PLATFORM,
602                             (elf_addr_t) (unsigned long) u_base_platform);
603         }
604
605         nr = 0;
606         csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
607         NEW_AUX_ENT(AT_HWCAP,   hwcap);
608         NEW_AUX_ENT(AT_PAGESZ,  PAGE_SIZE);
609         NEW_AUX_ENT(AT_CLKTCK,  CLOCKS_PER_SEC);
610         NEW_AUX_ENT(AT_PHDR,    exec_params->ph_addr);
611         NEW_AUX_ENT(AT_PHENT,   sizeof(struct elf_phdr));
612         NEW_AUX_ENT(AT_PHNUM,   exec_params->hdr.e_phnum);
613         NEW_AUX_ENT(AT_BASE,    interp_params->elfhdr_addr);
614         NEW_AUX_ENT(AT_FLAGS,   0);
615         NEW_AUX_ENT(AT_ENTRY,   exec_params->entry_addr);
616         NEW_AUX_ENT(AT_UID,     (elf_addr_t) current->uid);
617         NEW_AUX_ENT(AT_EUID,    (elf_addr_t) current->euid);
618         NEW_AUX_ENT(AT_GID,     (elf_addr_t) current->gid);
619         NEW_AUX_ENT(AT_EGID,    (elf_addr_t) current->egid);
620
621 #ifdef ARCH_DLINFO
622         nr = 0;
623         csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
624
625         /* ARCH_DLINFO must come last so platform specific code can enforce
626          * special alignment requirements on the AUXV if necessary (eg. PPC).
627          */
628         ARCH_DLINFO;
629 #endif
630 #undef NEW_AUX_ENT
631
632         /* allocate room for argv[] and envv[] */
633         csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
634         envp = (elf_caddr_t __user *) csp;
635         csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
636         argv = (elf_caddr_t __user *) csp;
637
638         /* stack argc */
639         csp -= sizeof(unsigned long);
640         __put_user(bprm->argc, (unsigned long __user *) csp);
641
642         BUG_ON(csp != sp);
643
644         /* fill in the argv[] array */
645 #ifdef CONFIG_MMU
646         current->mm->arg_start = bprm->p;
647 #else
648         current->mm->arg_start = current->mm->start_stack -
649                 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
650 #endif
651
652         p = (char __user *) current->mm->arg_start;
653         for (loop = bprm->argc; loop > 0; loop--) {
654                 __put_user((elf_caddr_t) p, argv++);
655                 len = strnlen_user(p, MAX_ARG_STRLEN);
656                 if (!len || len > MAX_ARG_STRLEN)
657                         return -EINVAL;
658                 p += len;
659         }
660         __put_user(NULL, argv);
661         current->mm->arg_end = (unsigned long) p;
662
663         /* fill in the envv[] array */
664         current->mm->env_start = (unsigned long) p;
665         for (loop = bprm->envc; loop > 0; loop--) {
666                 __put_user((elf_caddr_t)(unsigned long) p, envp++);
667                 len = strnlen_user(p, MAX_ARG_STRLEN);
668                 if (!len || len > MAX_ARG_STRLEN)
669                         return -EINVAL;
670                 p += len;
671         }
672         __put_user(NULL, envp);
673         current->mm->env_end = (unsigned long) p;
674
675         mm->start_stack = (unsigned long) sp;
676         return 0;
677 }
678
679 /*****************************************************************************/
680 /*
681  * transfer the program arguments and environment from the holding pages onto
682  * the stack
683  */
684 #ifndef CONFIG_MMU
685 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
686                                             unsigned long *_sp)
687 {
688         unsigned long index, stop, sp;
689         char *src;
690         int ret = 0;
691
692         stop = bprm->p >> PAGE_SHIFT;
693         sp = *_sp;
694
695         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
696                 src = kmap(bprm->page[index]);
697                 sp -= PAGE_SIZE;
698                 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
699                         ret = -EFAULT;
700                 kunmap(bprm->page[index]);
701                 if (ret < 0)
702                         goto out;
703         }
704
705         *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
706
707 out:
708         return ret;
709 }
710 #endif
711
712 /*****************************************************************************/
713 /*
714  * load the appropriate binary image (executable or interpreter) into memory
715  * - we assume no MMU is available
716  * - if no other PIC bits are set in params->hdr->e_flags
717  *   - we assume that the LOADable segments in the binary are independently relocatable
718  *   - we assume R/O executable segments are shareable
719  * - else
720  *   - we assume the loadable parts of the image to require fixed displacement
721  *   - the image is not shareable
722  */
723 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
724                               struct file *file,
725                               struct mm_struct *mm,
726                               const char *what)
727 {
728         struct elf32_fdpic_loadmap *loadmap;
729 #ifdef CONFIG_MMU
730         struct elf32_fdpic_loadseg *mseg;
731 #endif
732         struct elf32_fdpic_loadseg *seg;
733         struct elf32_phdr *phdr;
734         unsigned long load_addr, stop;
735         unsigned nloads, tmp;
736         size_t size;
737         int loop, ret;
738
739         /* allocate a load map table */
740         nloads = 0;
741         for (loop = 0; loop < params->hdr.e_phnum; loop++)
742                 if (params->phdrs[loop].p_type == PT_LOAD)
743                         nloads++;
744
745         if (nloads == 0)
746                 return -ELIBBAD;
747
748         size = sizeof(*loadmap) + nloads * sizeof(*seg);
749         loadmap = kzalloc(size, GFP_KERNEL);
750         if (!loadmap)
751                 return -ENOMEM;
752
753         params->loadmap = loadmap;
754
755         loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
756         loadmap->nsegs = nloads;
757
758         load_addr = params->load_addr;
759         seg = loadmap->segs;
760
761         /* map the requested LOADs into the memory space */
762         switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
763         case ELF_FDPIC_FLAG_CONSTDISP:
764         case ELF_FDPIC_FLAG_CONTIGUOUS:
765 #ifndef CONFIG_MMU
766                 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
767                 if (ret < 0)
768                         return ret;
769                 break;
770 #endif
771         default:
772                 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
773                 if (ret < 0)
774                         return ret;
775                 break;
776         }
777
778         /* map the entry point */
779         if (params->hdr.e_entry) {
780                 seg = loadmap->segs;
781                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
782                         if (params->hdr.e_entry >= seg->p_vaddr &&
783                             params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
784                                 params->entry_addr =
785                                         (params->hdr.e_entry - seg->p_vaddr) +
786                                         seg->addr;
787                                 break;
788                         }
789                 }
790         }
791
792         /* determine where the program header table has wound up if mapped */
793         stop = params->hdr.e_phoff;
794         stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
795         phdr = params->phdrs;
796
797         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
798                 if (phdr->p_type != PT_LOAD)
799                         continue;
800
801                 if (phdr->p_offset > params->hdr.e_phoff ||
802                     phdr->p_offset + phdr->p_filesz < stop)
803                         continue;
804
805                 seg = loadmap->segs;
806                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
807                         if (phdr->p_vaddr >= seg->p_vaddr &&
808                             phdr->p_vaddr + phdr->p_filesz <=
809                             seg->p_vaddr + seg->p_memsz) {
810                                 params->ph_addr =
811                                         (phdr->p_vaddr - seg->p_vaddr) +
812                                         seg->addr +
813                                         params->hdr.e_phoff - phdr->p_offset;
814                                 break;
815                         }
816                 }
817                 break;
818         }
819
820         /* determine where the dynamic section has wound up if there is one */
821         phdr = params->phdrs;
822         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
823                 if (phdr->p_type != PT_DYNAMIC)
824                         continue;
825
826                 seg = loadmap->segs;
827                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
828                         if (phdr->p_vaddr >= seg->p_vaddr &&
829                             phdr->p_vaddr + phdr->p_memsz <=
830                             seg->p_vaddr + seg->p_memsz) {
831                                 params->dynamic_addr =
832                                         (phdr->p_vaddr - seg->p_vaddr) +
833                                         seg->addr;
834
835                                 /* check the dynamic section contains at least
836                                  * one item, and that the last item is a NULL
837                                  * entry */
838                                 if (phdr->p_memsz == 0 ||
839                                     phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
840                                         goto dynamic_error;
841
842                                 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
843                                 if (((Elf32_Dyn *)
844                                      params->dynamic_addr)[tmp - 1].d_tag != 0)
845                                         goto dynamic_error;
846                                 break;
847                         }
848                 }
849                 break;
850         }
851
852         /* now elide adjacent segments in the load map on MMU linux
853          * - on uClinux the holes between may actually be filled with system
854          *   stuff or stuff from other processes
855          */
856 #ifdef CONFIG_MMU
857         nloads = loadmap->nsegs;
858         mseg = loadmap->segs;
859         seg = mseg + 1;
860         for (loop = 1; loop < nloads; loop++) {
861                 /* see if we have a candidate for merging */
862                 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
863                         load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
864                         if (load_addr == (seg->addr & PAGE_MASK)) {
865                                 mseg->p_memsz +=
866                                         load_addr -
867                                         (mseg->addr + mseg->p_memsz);
868                                 mseg->p_memsz += seg->addr & ~PAGE_MASK;
869                                 mseg->p_memsz += seg->p_memsz;
870                                 loadmap->nsegs--;
871                                 continue;
872                         }
873                 }
874
875                 mseg++;
876                 if (mseg != seg)
877                         *mseg = *seg;
878         }
879 #endif
880
881         kdebug("Mapped Object [%s]:", what);
882         kdebug("- elfhdr   : %lx", params->elfhdr_addr);
883         kdebug("- entry    : %lx", params->entry_addr);
884         kdebug("- PHDR[]   : %lx", params->ph_addr);
885         kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
886         seg = loadmap->segs;
887         for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
888                 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
889                        loop,
890                        seg->addr, seg->addr + seg->p_memsz - 1,
891                        seg->p_vaddr, seg->p_memsz);
892
893         return 0;
894
895 dynamic_error:
896         printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
897                what, file->f_path.dentry->d_inode->i_ino);
898         return -ELIBBAD;
899 }
900
901 /*****************************************************************************/
902 /*
903  * map a file with constant displacement under uClinux
904  */
905 #ifndef CONFIG_MMU
906 static int elf_fdpic_map_file_constdisp_on_uclinux(
907         struct elf_fdpic_params *params,
908         struct file *file,
909         struct mm_struct *mm)
910 {
911         struct elf32_fdpic_loadseg *seg;
912         struct elf32_phdr *phdr;
913         unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
914         loff_t fpos;
915         int loop, ret;
916
917         load_addr = params->load_addr;
918         seg = params->loadmap->segs;
919
920         /* determine the bounds of the contiguous overall allocation we must
921          * make */
922         phdr = params->phdrs;
923         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
924                 if (params->phdrs[loop].p_type != PT_LOAD)
925                         continue;
926
927                 if (base > phdr->p_vaddr)
928                         base = phdr->p_vaddr;
929                 if (top < phdr->p_vaddr + phdr->p_memsz)
930                         top = phdr->p_vaddr + phdr->p_memsz;
931         }
932
933         /* allocate one big anon block for everything */
934         mflags = MAP_PRIVATE;
935         if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
936                 mflags |= MAP_EXECUTABLE;
937
938         down_write(&mm->mmap_sem);
939         maddr = do_mmap(NULL, load_addr, top - base,
940                         PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
941         up_write(&mm->mmap_sem);
942         if (IS_ERR_VALUE(maddr))
943                 return (int) maddr;
944
945         if (load_addr != 0)
946                 load_addr += PAGE_ALIGN(top - base);
947
948         /* and then load the file segments into it */
949         phdr = params->phdrs;
950         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
951                 if (params->phdrs[loop].p_type != PT_LOAD)
952                         continue;
953
954                 fpos = phdr->p_offset;
955
956                 seg->addr = maddr + (phdr->p_vaddr - base);
957                 seg->p_vaddr = phdr->p_vaddr;
958                 seg->p_memsz = phdr->p_memsz;
959
960                 ret = file->f_op->read(file, (void *) seg->addr,
961                                        phdr->p_filesz, &fpos);
962                 if (ret < 0)
963                         return ret;
964
965                 /* map the ELF header address if in this segment */
966                 if (phdr->p_offset == 0)
967                         params->elfhdr_addr = seg->addr;
968
969                 /* clear any space allocated but not loaded */
970                 if (phdr->p_filesz < phdr->p_memsz)
971                         clear_user((void *) (seg->addr + phdr->p_filesz),
972                                    phdr->p_memsz - phdr->p_filesz);
973
974                 if (mm) {
975                         if (phdr->p_flags & PF_X) {
976                                 if (!mm->start_code) {
977                                         mm->start_code = seg->addr;
978                                         mm->end_code = seg->addr +
979                                                 phdr->p_memsz;
980                                 }
981                         } else if (!mm->start_data) {
982                                 mm->start_data = seg->addr;
983 #ifndef CONFIG_MMU
984                                 mm->end_data = seg->addr + phdr->p_memsz;
985 #endif
986                         }
987
988 #ifdef CONFIG_MMU
989                         if (seg->addr + phdr->p_memsz > mm->end_data)
990                                 mm->end_data = seg->addr + phdr->p_memsz;
991 #endif
992                 }
993
994                 seg++;
995         }
996
997         return 0;
998 }
999 #endif
1000
1001 /*****************************************************************************/
1002 /*
1003  * map a binary by direct mmap() of the individual PT_LOAD segments
1004  */
1005 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1006                                              struct file *file,
1007                                              struct mm_struct *mm)
1008 {
1009         struct elf32_fdpic_loadseg *seg;
1010         struct elf32_phdr *phdr;
1011         unsigned long load_addr, delta_vaddr;
1012         int loop, dvset;
1013
1014         load_addr = params->load_addr;
1015         delta_vaddr = 0;
1016         dvset = 0;
1017
1018         seg = params->loadmap->segs;
1019
1020         /* deal with each load segment separately */
1021         phdr = params->phdrs;
1022         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1023                 unsigned long maddr, disp, excess, excess1;
1024                 int prot = 0, flags;
1025
1026                 if (phdr->p_type != PT_LOAD)
1027                         continue;
1028
1029                 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1030                        (unsigned long) phdr->p_vaddr,
1031                        (unsigned long) phdr->p_offset,
1032                        (unsigned long) phdr->p_filesz,
1033                        (unsigned long) phdr->p_memsz);
1034
1035                 /* determine the mapping parameters */
1036                 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1037                 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1038                 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1039
1040                 flags = MAP_PRIVATE | MAP_DENYWRITE;
1041                 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1042                         flags |= MAP_EXECUTABLE;
1043
1044                 maddr = 0;
1045
1046                 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1047                 case ELF_FDPIC_FLAG_INDEPENDENT:
1048                         /* PT_LOADs are independently locatable */
1049                         break;
1050
1051                 case ELF_FDPIC_FLAG_HONOURVADDR:
1052                         /* the specified virtual address must be honoured */
1053                         maddr = phdr->p_vaddr;
1054                         flags |= MAP_FIXED;
1055                         break;
1056
1057                 case ELF_FDPIC_FLAG_CONSTDISP:
1058                         /* constant displacement
1059                          * - can be mapped anywhere, but must be mapped as a
1060                          *   unit
1061                          */
1062                         if (!dvset) {
1063                                 maddr = load_addr;
1064                                 delta_vaddr = phdr->p_vaddr;
1065                                 dvset = 1;
1066                         } else {
1067                                 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1068                                 flags |= MAP_FIXED;
1069                         }
1070                         break;
1071
1072                 case ELF_FDPIC_FLAG_CONTIGUOUS:
1073                         /* contiguity handled later */
1074                         break;
1075
1076                 default:
1077                         BUG();
1078                 }
1079
1080                 maddr &= PAGE_MASK;
1081
1082                 /* create the mapping */
1083                 disp = phdr->p_vaddr & ~PAGE_MASK;
1084                 down_write(&mm->mmap_sem);
1085                 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1086                                 phdr->p_offset - disp);
1087                 up_write(&mm->mmap_sem);
1088
1089                 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1090                        loop, phdr->p_memsz + disp, prot, flags,
1091                        phdr->p_offset - disp, maddr);
1092
1093                 if (IS_ERR_VALUE(maddr))
1094                         return (int) maddr;
1095
1096                 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1097                     ELF_FDPIC_FLAG_CONTIGUOUS)
1098                         load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1099
1100                 seg->addr = maddr + disp;
1101                 seg->p_vaddr = phdr->p_vaddr;
1102                 seg->p_memsz = phdr->p_memsz;
1103
1104                 /* map the ELF header address if in this segment */
1105                 if (phdr->p_offset == 0)
1106                         params->elfhdr_addr = seg->addr;
1107
1108                 /* clear the bit between beginning of mapping and beginning of
1109                  * PT_LOAD */
1110                 if (prot & PROT_WRITE && disp > 0) {
1111                         kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1112                         clear_user((void __user *) maddr, disp);
1113                         maddr += disp;
1114                 }
1115
1116                 /* clear any space allocated but not loaded
1117                  * - on uClinux we can just clear the lot
1118                  * - on MMU linux we'll get a SIGBUS beyond the last page
1119                  *   extant in the file
1120                  */
1121                 excess = phdr->p_memsz - phdr->p_filesz;
1122                 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1123
1124 #ifdef CONFIG_MMU
1125                 if (excess > excess1) {
1126                         unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1127                         unsigned long xmaddr;
1128
1129                         flags |= MAP_FIXED | MAP_ANONYMOUS;
1130                         down_write(&mm->mmap_sem);
1131                         xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1132                                          prot, flags, 0);
1133                         up_write(&mm->mmap_sem);
1134
1135                         kdebug("mmap[%d] <anon>"
1136                                " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1137                                loop, xaddr, excess - excess1, prot, flags,
1138                                xmaddr);
1139
1140                         if (xmaddr != xaddr)
1141                                 return -ENOMEM;
1142                 }
1143
1144                 if (prot & PROT_WRITE && excess1 > 0) {
1145                         kdebug("clear[%d] ad=%lx sz=%lx",
1146                                loop, maddr + phdr->p_filesz, excess1);
1147                         clear_user((void __user *) maddr + phdr->p_filesz,
1148                                    excess1);
1149                 }
1150
1151 #else
1152                 if (excess > 0) {
1153                         kdebug("clear[%d] ad=%lx sz=%lx",
1154                                loop, maddr + phdr->p_filesz, excess);
1155                         clear_user((void *) maddr + phdr->p_filesz, excess);
1156                 }
1157 #endif
1158
1159                 if (mm) {
1160                         if (phdr->p_flags & PF_X) {
1161                                 if (!mm->start_code) {
1162                                         mm->start_code = maddr;
1163                                         mm->end_code = maddr + phdr->p_memsz;
1164                                 }
1165                         } else if (!mm->start_data) {
1166                                 mm->start_data = maddr;
1167                                 mm->end_data = maddr + phdr->p_memsz;
1168                         }
1169                 }
1170
1171                 seg++;
1172         }
1173
1174         return 0;
1175 }
1176
1177 /*****************************************************************************/
1178 /*
1179  * ELF-FDPIC core dumper
1180  *
1181  * Modelled on fs/exec.c:aout_core_dump()
1182  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1183  *
1184  * Modelled on fs/binfmt_elf.c core dumper
1185  */
1186 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1187
1188 /*
1189  * These are the only things you should do on a core-file: use only these
1190  * functions to write out all the necessary info.
1191  */
1192 static int dump_write(struct file *file, const void *addr, int nr)
1193 {
1194         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1195 }
1196
1197 static int dump_seek(struct file *file, loff_t off)
1198 {
1199         if (file->f_op->llseek) {
1200                 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1201                         return 0;
1202         } else {
1203                 file->f_pos = off;
1204         }
1205         return 1;
1206 }
1207
1208 /*
1209  * Decide whether a segment is worth dumping; default is yes to be
1210  * sure (missing info is worse than too much; etc).
1211  * Personally I'd include everything, and use the coredump limit...
1212  *
1213  * I think we should skip something. But I am not sure how. H.J.
1214  */
1215 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1216 {
1217         int dump_ok;
1218
1219         /* Do not dump I/O mapped devices or special mappings */
1220         if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1221                 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1222                 return 0;
1223         }
1224
1225         /* If we may not read the contents, don't allow us to dump
1226          * them either. "dump_write()" can't handle it anyway.
1227          */
1228         if (!(vma->vm_flags & VM_READ)) {
1229                 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1230                 return 0;
1231         }
1232
1233         /* By default, dump shared memory if mapped from an anonymous file. */
1234         if (vma->vm_flags & VM_SHARED) {
1235                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1236                         dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1237                         kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1238                                vma->vm_flags, dump_ok ? "yes" : "no");
1239                         return dump_ok;
1240                 }
1241
1242                 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1243                 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1244                        vma->vm_flags, dump_ok ? "yes" : "no");
1245                 return dump_ok;
1246         }
1247
1248 #ifdef CONFIG_MMU
1249         /* By default, if it hasn't been written to, don't write it out */
1250         if (!vma->anon_vma) {
1251                 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1252                 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1253                        vma->vm_flags, dump_ok ? "yes" : "no");
1254                 return dump_ok;
1255         }
1256 #endif
1257
1258         dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1259         kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1260                dump_ok ? "yes" : "no");
1261         return dump_ok;
1262 }
1263
1264 /* An ELF note in memory */
1265 struct memelfnote
1266 {
1267         const char *name;
1268         int type;
1269         unsigned int datasz;
1270         void *data;
1271 };
1272
1273 static int notesize(struct memelfnote *en)
1274 {
1275         int sz;
1276
1277         sz = sizeof(struct elf_note);
1278         sz += roundup(strlen(en->name) + 1, 4);
1279         sz += roundup(en->datasz, 4);
1280
1281         return sz;
1282 }
1283
1284 /* #define DEBUG */
1285
1286 #define DUMP_WRITE(addr, nr)    \
1287         do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1288 #define DUMP_SEEK(off)  \
1289         do { if (!dump_seek(file, (off))) return 0; } while(0)
1290
1291 static int writenote(struct memelfnote *men, struct file *file)
1292 {
1293         struct elf_note en;
1294
1295         en.n_namesz = strlen(men->name) + 1;
1296         en.n_descsz = men->datasz;
1297         en.n_type = men->type;
1298
1299         DUMP_WRITE(&en, sizeof(en));
1300         DUMP_WRITE(men->name, en.n_namesz);
1301         /* XXX - cast from long long to long to avoid need for libgcc.a */
1302         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1303         DUMP_WRITE(men->data, men->datasz);
1304         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1305
1306         return 1;
1307 }
1308 #undef DUMP_WRITE
1309 #undef DUMP_SEEK
1310
1311 #define DUMP_WRITE(addr, nr)    \
1312         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1313                 goto end_coredump;
1314 #define DUMP_SEEK(off)  \
1315         if (!dump_seek(file, (off))) \
1316                 goto end_coredump;
1317
1318 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1319 {
1320         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1321         elf->e_ident[EI_CLASS] = ELF_CLASS;
1322         elf->e_ident[EI_DATA] = ELF_DATA;
1323         elf->e_ident[EI_VERSION] = EV_CURRENT;
1324         elf->e_ident[EI_OSABI] = ELF_OSABI;
1325         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1326
1327         elf->e_type = ET_CORE;
1328         elf->e_machine = ELF_ARCH;
1329         elf->e_version = EV_CURRENT;
1330         elf->e_entry = 0;
1331         elf->e_phoff = sizeof(struct elfhdr);
1332         elf->e_shoff = 0;
1333         elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1334         elf->e_ehsize = sizeof(struct elfhdr);
1335         elf->e_phentsize = sizeof(struct elf_phdr);
1336         elf->e_phnum = segs;
1337         elf->e_shentsize = 0;
1338         elf->e_shnum = 0;
1339         elf->e_shstrndx = 0;
1340         return;
1341 }
1342
1343 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1344 {
1345         phdr->p_type = PT_NOTE;
1346         phdr->p_offset = offset;
1347         phdr->p_vaddr = 0;
1348         phdr->p_paddr = 0;
1349         phdr->p_filesz = sz;
1350         phdr->p_memsz = 0;
1351         phdr->p_flags = 0;
1352         phdr->p_align = 0;
1353         return;
1354 }
1355
1356 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1357                 unsigned int sz, void *data)
1358 {
1359         note->name = name;
1360         note->type = type;
1361         note->datasz = sz;
1362         note->data = data;
1363         return;
1364 }
1365
1366 /*
1367  * fill up all the fields in prstatus from the given task struct, except
1368  * registers which need to be filled up seperately.
1369  */
1370 static void fill_prstatus(struct elf_prstatus *prstatus,
1371                           struct task_struct *p, long signr)
1372 {
1373         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1374         prstatus->pr_sigpend = p->pending.signal.sig[0];
1375         prstatus->pr_sighold = p->blocked.sig[0];
1376         prstatus->pr_pid = task_pid_vnr(p);
1377         prstatus->pr_ppid = task_pid_vnr(p->parent);
1378         prstatus->pr_pgrp = task_pgrp_vnr(p);
1379         prstatus->pr_sid = task_session_vnr(p);
1380         if (thread_group_leader(p)) {
1381                 /*
1382                  * This is the record for the group leader.  Add in the
1383                  * cumulative times of previous dead threads.  This total
1384                  * won't include the time of each live thread whose state
1385                  * is included in the core dump.  The final total reported
1386                  * to our parent process when it calls wait4 will include
1387                  * those sums as well as the little bit more time it takes
1388                  * this and each other thread to finish dying after the
1389                  * core dump synchronization phase.
1390                  */
1391                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1392                                    &prstatus->pr_utime);
1393                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1394                                    &prstatus->pr_stime);
1395         } else {
1396                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1397                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1398         }
1399         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1400         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1401
1402         prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1403         prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1404 }
1405
1406 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1407                        struct mm_struct *mm)
1408 {
1409         unsigned int i, len;
1410
1411         /* first copy the parameters from user space */
1412         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1413
1414         len = mm->arg_end - mm->arg_start;
1415         if (len >= ELF_PRARGSZ)
1416                 len = ELF_PRARGSZ - 1;
1417         if (copy_from_user(&psinfo->pr_psargs,
1418                            (const char __user *) mm->arg_start, len))
1419                 return -EFAULT;
1420         for (i = 0; i < len; i++)
1421                 if (psinfo->pr_psargs[i] == 0)
1422                         psinfo->pr_psargs[i] = ' ';
1423         psinfo->pr_psargs[len] = 0;
1424
1425         psinfo->pr_pid = task_pid_vnr(p);
1426         psinfo->pr_ppid = task_pid_vnr(p->parent);
1427         psinfo->pr_pgrp = task_pgrp_vnr(p);
1428         psinfo->pr_sid = task_session_vnr(p);
1429
1430         i = p->state ? ffz(~p->state) + 1 : 0;
1431         psinfo->pr_state = i;
1432         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1433         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1434         psinfo->pr_nice = task_nice(p);
1435         psinfo->pr_flag = p->flags;
1436         SET_UID(psinfo->pr_uid, p->uid);
1437         SET_GID(psinfo->pr_gid, p->gid);
1438         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1439
1440         return 0;
1441 }
1442
1443 /* Here is the structure in which status of each thread is captured. */
1444 struct elf_thread_status
1445 {
1446         struct list_head list;
1447         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1448         elf_fpregset_t fpu;             /* NT_PRFPREG */
1449         struct task_struct *thread;
1450 #ifdef ELF_CORE_COPY_XFPREGS
1451         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1452 #endif
1453         struct memelfnote notes[3];
1454         int num_notes;
1455 };
1456
1457 /*
1458  * In order to add the specific thread information for the elf file format,
1459  * we need to keep a linked list of every thread's pr_status and then create
1460  * a single section for them in the final core file.
1461  */
1462 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1463 {
1464         struct task_struct *p = t->thread;
1465         int sz = 0;
1466
1467         t->num_notes = 0;
1468
1469         fill_prstatus(&t->prstatus, p, signr);
1470         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1471
1472         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1473                   &t->prstatus);
1474         t->num_notes++;
1475         sz += notesize(&t->notes[0]);
1476
1477         t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1478         if (t->prstatus.pr_fpvalid) {
1479                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1480                           &t->fpu);
1481                 t->num_notes++;
1482                 sz += notesize(&t->notes[1]);
1483         }
1484
1485 #ifdef ELF_CORE_COPY_XFPREGS
1486         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1487                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1488                           sizeof(t->xfpu), &t->xfpu);
1489                 t->num_notes++;
1490                 sz += notesize(&t->notes[2]);
1491         }
1492 #endif
1493         return sz;
1494 }
1495
1496 /*
1497  * dump the segments for an MMU process
1498  */
1499 #ifdef CONFIG_MMU
1500 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1501                            unsigned long *limit, unsigned long mm_flags)
1502 {
1503         struct vm_area_struct *vma;
1504
1505         for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1506                 unsigned long addr;
1507
1508                 if (!maydump(vma, mm_flags))
1509                         continue;
1510
1511                 for (addr = vma->vm_start;
1512                      addr < vma->vm_end;
1513                      addr += PAGE_SIZE
1514                      ) {
1515                         struct vm_area_struct *vma;
1516                         struct page *page;
1517
1518                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1519                                            &page, &vma) <= 0) {
1520                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1521                         }
1522                         else if (page == ZERO_PAGE(0)) {
1523                                 page_cache_release(page);
1524                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1525                         }
1526                         else {
1527                                 void *kaddr;
1528
1529                                 flush_cache_page(vma, addr, page_to_pfn(page));
1530                                 kaddr = kmap(page);
1531                                 if ((*size += PAGE_SIZE) > *limit ||
1532                                     !dump_write(file, kaddr, PAGE_SIZE)
1533                                     ) {
1534                                         kunmap(page);
1535                                         page_cache_release(page);
1536                                         return -EIO;
1537                                 }
1538                                 kunmap(page);
1539                                 page_cache_release(page);
1540                         }
1541                 }
1542         }
1543
1544         return 0;
1545
1546 end_coredump:
1547         return -EFBIG;
1548 }
1549 #endif
1550
1551 /*
1552  * dump the segments for a NOMMU process
1553  */
1554 #ifndef CONFIG_MMU
1555 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1556                            unsigned long *limit, unsigned long mm_flags)
1557 {
1558         struct vm_list_struct *vml;
1559
1560         for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1561         struct vm_area_struct *vma = vml->vma;
1562
1563                 if (!maydump(vma, mm_flags))
1564                         continue;
1565
1566                 if ((*size += PAGE_SIZE) > *limit)
1567                         return -EFBIG;
1568
1569                 if (!dump_write(file, (void *) vma->vm_start,
1570                                 vma->vm_end - vma->vm_start))
1571                         return -EIO;
1572         }
1573
1574         return 0;
1575 }
1576 #endif
1577
1578 /*
1579  * Actual dumper
1580  *
1581  * This is a two-pass process; first we find the offsets of the bits,
1582  * and then they are actually written out.  If we run out of core limit
1583  * we just truncate.
1584  */
1585 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1586                                struct file *file, unsigned long limit)
1587 {
1588 #define NUM_NOTES       6
1589         int has_dumped = 0;
1590         mm_segment_t fs;
1591         int segs;
1592         size_t size = 0;
1593         int i;
1594         struct vm_area_struct *vma;
1595         struct elfhdr *elf = NULL;
1596         loff_t offset = 0, dataoff;
1597         int numnote;
1598         struct memelfnote *notes = NULL;
1599         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1600         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1601         LIST_HEAD(thread_list);
1602         struct list_head *t;
1603         elf_fpregset_t *fpu = NULL;
1604 #ifdef ELF_CORE_COPY_XFPREGS
1605         elf_fpxregset_t *xfpu = NULL;
1606 #endif
1607         int thread_status_size = 0;
1608 #ifndef CONFIG_MMU
1609         struct vm_list_struct *vml;
1610 #endif
1611         elf_addr_t *auxv;
1612         unsigned long mm_flags;
1613
1614         /*
1615          * We no longer stop all VM operations.
1616          *
1617          * This is because those proceses that could possibly change map_count
1618          * or the mmap / vma pages are now blocked in do_exit on current
1619          * finishing this core dump.
1620          *
1621          * Only ptrace can touch these memory addresses, but it doesn't change
1622          * the map_count or the pages allocated. So no possibility of crashing
1623          * exists while dumping the mm->vm_next areas to the core file.
1624          */
1625
1626         /* alloc memory for large data structures: too large to be on stack */
1627         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1628         if (!elf)
1629                 goto cleanup;
1630         prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1631         if (!prstatus)
1632                 goto cleanup;
1633         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1634         if (!psinfo)
1635                 goto cleanup;
1636         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1637         if (!notes)
1638                 goto cleanup;
1639         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1640         if (!fpu)
1641                 goto cleanup;
1642 #ifdef ELF_CORE_COPY_XFPREGS
1643         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1644         if (!xfpu)
1645                 goto cleanup;
1646 #endif
1647
1648         if (signr) {
1649                 struct core_thread *ct;
1650                 struct elf_thread_status *tmp;
1651
1652                 for (ct = current->mm->core_state->dumper.next;
1653                                                 ct; ct = ct->next) {
1654                         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1655                         if (!tmp)
1656                                 goto cleanup;
1657
1658                         tmp->thread = ct->task;
1659                         list_add(&tmp->list, &thread_list);
1660                 }
1661
1662                 list_for_each(t, &thread_list) {
1663                         struct elf_thread_status *tmp;
1664                         int sz;
1665
1666                         tmp = list_entry(t, struct elf_thread_status, list);
1667                         sz = elf_dump_thread_status(signr, tmp);
1668                         thread_status_size += sz;
1669                 }
1670         }
1671
1672         /* now collect the dump for the current */
1673         fill_prstatus(prstatus, current, signr);
1674         elf_core_copy_regs(&prstatus->pr_reg, regs);
1675
1676 #ifdef CONFIG_MMU
1677         segs = current->mm->map_count;
1678 #else
1679         segs = 0;
1680         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1681             segs++;
1682 #endif
1683 #ifdef ELF_CORE_EXTRA_PHDRS
1684         segs += ELF_CORE_EXTRA_PHDRS;
1685 #endif
1686
1687         /* Set up header */
1688         fill_elf_fdpic_header(elf, segs + 1);   /* including notes section */
1689
1690         has_dumped = 1;
1691         current->flags |= PF_DUMPCORE;
1692
1693         /*
1694          * Set up the notes in similar form to SVR4 core dumps made
1695          * with info from their /proc.
1696          */
1697
1698         fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1699         fill_psinfo(psinfo, current->group_leader, current->mm);
1700         fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1701
1702         numnote = 2;
1703
1704         auxv = (elf_addr_t *) current->mm->saved_auxv;
1705
1706         i = 0;
1707         do
1708                 i += 2;
1709         while (auxv[i - 2] != AT_NULL);
1710         fill_note(&notes[numnote++], "CORE", NT_AUXV,
1711                   i * sizeof(elf_addr_t), auxv);
1712
1713         /* Try to dump the FPU. */
1714         if ((prstatus->pr_fpvalid =
1715              elf_core_copy_task_fpregs(current, regs, fpu)))
1716                 fill_note(notes + numnote++,
1717                           "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1718 #ifdef ELF_CORE_COPY_XFPREGS
1719         if (elf_core_copy_task_xfpregs(current, xfpu))
1720                 fill_note(notes + numnote++,
1721                           "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1722 #endif
1723
1724         fs = get_fs();
1725         set_fs(KERNEL_DS);
1726
1727         DUMP_WRITE(elf, sizeof(*elf));
1728         offset += sizeof(*elf);                         /* Elf header */
1729         offset += (segs+1) * sizeof(struct elf_phdr);   /* Program headers */
1730
1731         /* Write notes phdr entry */
1732         {
1733                 struct elf_phdr phdr;
1734                 int sz = 0;
1735
1736                 for (i = 0; i < numnote; i++)
1737                         sz += notesize(notes + i);
1738
1739                 sz += thread_status_size;
1740
1741                 fill_elf_note_phdr(&phdr, sz, offset);
1742                 offset += sz;
1743                 DUMP_WRITE(&phdr, sizeof(phdr));
1744         }
1745
1746         /* Page-align dumped data */
1747         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1748
1749         /*
1750          * We must use the same mm->flags while dumping core to avoid
1751          * inconsistency between the program headers and bodies, otherwise an
1752          * unusable core file can be generated.
1753          */
1754         mm_flags = current->mm->flags;
1755
1756         /* write program headers for segments dump */
1757         for (
1758 #ifdef CONFIG_MMU
1759                 vma = current->mm->mmap; vma; vma = vma->vm_next
1760 #else
1761                         vml = current->mm->context.vmlist; vml; vml = vml->next
1762 #endif
1763              ) {
1764                 struct elf_phdr phdr;
1765                 size_t sz;
1766
1767 #ifndef CONFIG_MMU
1768                 vma = vml->vma;
1769 #endif
1770
1771                 sz = vma->vm_end - vma->vm_start;
1772
1773                 phdr.p_type = PT_LOAD;
1774                 phdr.p_offset = offset;
1775                 phdr.p_vaddr = vma->vm_start;
1776                 phdr.p_paddr = 0;
1777                 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1778                 phdr.p_memsz = sz;
1779                 offset += phdr.p_filesz;
1780                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1781                 if (vma->vm_flags & VM_WRITE)
1782                         phdr.p_flags |= PF_W;
1783                 if (vma->vm_flags & VM_EXEC)
1784                         phdr.p_flags |= PF_X;
1785                 phdr.p_align = ELF_EXEC_PAGESIZE;
1786
1787                 DUMP_WRITE(&phdr, sizeof(phdr));
1788         }
1789
1790 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1791         ELF_CORE_WRITE_EXTRA_PHDRS;
1792 #endif
1793
1794         /* write out the notes section */
1795         for (i = 0; i < numnote; i++)
1796                 if (!writenote(notes + i, file))
1797                         goto end_coredump;
1798
1799         /* write out the thread status notes section */
1800         list_for_each(t, &thread_list) {
1801                 struct elf_thread_status *tmp =
1802                                 list_entry(t, struct elf_thread_status, list);
1803
1804                 for (i = 0; i < tmp->num_notes; i++)
1805                         if (!writenote(&tmp->notes[i], file))
1806                                 goto end_coredump;
1807         }
1808
1809         DUMP_SEEK(dataoff);
1810
1811         if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1812                 goto end_coredump;
1813
1814 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1815         ELF_CORE_WRITE_EXTRA_DATA;
1816 #endif
1817
1818         if (file->f_pos != offset) {
1819                 /* Sanity check */
1820                 printk(KERN_WARNING
1821                        "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1822                        file->f_pos, offset);
1823         }
1824
1825 end_coredump:
1826         set_fs(fs);
1827
1828 cleanup:
1829         while (!list_empty(&thread_list)) {
1830                 struct list_head *tmp = thread_list.next;
1831                 list_del(tmp);
1832                 kfree(list_entry(tmp, struct elf_thread_status, list));
1833         }
1834
1835         kfree(elf);
1836         kfree(prstatus);
1837         kfree(psinfo);
1838         kfree(notes);
1839         kfree(fpu);
1840 #ifdef ELF_CORE_COPY_XFPREGS
1841         kfree(xfpu);
1842 #endif
1843         return has_dumped;
1844 #undef NUM_NOTES
1845 }
1846
1847 #endif          /* USE_ELF_CORE_DUMP */