~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/binfmt_elf.c

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * linux/fs/binfmt_elf.c
  3  *
  4  * These are the functions used to load ELF format executables as used
  5  * on SVr4 machines.  Information on the format may be found in the book
  6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
  7  * Tools".
  8  *
  9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
 10  */
 11 
 12 #include <linux/module.h>
 13 #include <linux/kernel.h>
 14 #include <linux/fs.h>
 15 #include <linux/mm.h>
 16 #include <linux/mman.h>
 17 #include <linux/errno.h>
 18 #include <linux/signal.h>
 19 #include <linux/binfmts.h>
 20 #include <linux/string.h>
 21 #include <linux/file.h>
 22 #include <linux/slab.h>
 23 #include <linux/personality.h>
 24 #include <linux/elfcore.h>
 25 #include <linux/init.h>
 26 #include <linux/highuid.h>
 27 #include <linux/compiler.h>
 28 #include <linux/highmem.h>
 29 #include <linux/pagemap.h>
 30 #include <linux/vmalloc.h>
 31 #include <linux/security.h>
 32 #include <linux/random.h>
 33 #include <linux/elf.h>
 34 #include <linux/elf-randomize.h>
 35 #include <linux/utsname.h>
 36 #include <linux/coredump.h>
 37 #include <linux/sched.h>
 38 #include <linux/sched/coredump.h>
 39 #include <linux/sched/task_stack.h>
 40 #include <linux/sched/cputime.h>
 41 #include <linux/cred.h>
 42 #include <linux/dax.h>
 43 #include <linux/uaccess.h>
 44 #include <asm/param.h>
 45 #include <asm/page.h>
 46 
 47 #ifndef user_long_t
 48 #define user_long_t long
 49 #endif
 50 #ifndef user_siginfo_t
 51 #define user_siginfo_t siginfo_t
 52 #endif
 53 
 54 /* That's for binfmt_elf_fdpic to deal with */
 55 #ifndef elf_check_fdpic
 56 #define elf_check_fdpic(ex) false
 57 #endif
 58 
 59 static int load_elf_binary(struct linux_binprm *bprm);
 60 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
 61                                 int, int, unsigned long);
 62 
 63 #ifdef CONFIG_USELIB
 64 static int load_elf_library(struct file *);
 65 #else
 66 #define load_elf_library NULL
 67 #endif
 68 
 69 /*
 70  * If we don't support core dumping, then supply a NULL so we
 71  * don't even try.
 72  */
 73 #ifdef CONFIG_ELF_CORE
 74 static int elf_core_dump(struct coredump_params *cprm);
 75 #else
 76 #define elf_core_dump   NULL
 77 #endif
 78 
 79 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
 80 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
 81 #else
 82 #define ELF_MIN_ALIGN   PAGE_SIZE
 83 #endif
 84 
 85 #ifndef ELF_CORE_EFLAGS
 86 #define ELF_CORE_EFLAGS 0
 87 #endif
 88 
 89 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
 90 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
 91 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
 92 
 93 static struct linux_binfmt elf_format = {
 94         .module         = THIS_MODULE,
 95         .load_binary    = load_elf_binary,
 96         .load_shlib     = load_elf_library,
 97         .core_dump      = elf_core_dump,
 98         .min_coredump   = ELF_EXEC_PAGESIZE,
 99 };
100 
101 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
102 
103 static int set_brk(unsigned long start, unsigned long end, int prot)
104 {
105         start = ELF_PAGEALIGN(start);
106         end = ELF_PAGEALIGN(end);
107         if (end > start) {
108                 /*
109                  * Map the last of the bss segment.
110                  * If the header is requesting these pages to be
111                  * executable, honour that (ppc32 needs this).
112                  */
113                 int error = vm_brk_flags(start, end - start,
114                                 prot & PROT_EXEC ? VM_EXEC : 0);
115                 if (error)
116                         return error;
117         }
118         current->mm->start_brk = current->mm->brk = end;
119         return 0;
120 }
121 
122 /* We need to explicitly zero any fractional pages
123    after the data section (i.e. bss).  This would
124    contain the junk from the file that should not
125    be in memory
126  */
127 static int padzero(unsigned long elf_bss)
128 {
129         unsigned long nbyte;
130 
131         nbyte = ELF_PAGEOFFSET(elf_bss);
132         if (nbyte) {
133                 nbyte = ELF_MIN_ALIGN - nbyte;
134                 if (clear_user((void __user *) elf_bss, nbyte))
135                         return -EFAULT;
136         }
137         return 0;
138 }
139 
140 /* Let's use some macros to make this stack manipulation a little clearer */
141 #ifdef CONFIG_STACK_GROWSUP
142 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
143 #define STACK_ROUND(sp, items) \
144         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
145 #define STACK_ALLOC(sp, len) ({ \
146         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
147         old_sp; })
148 #else
149 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
150 #define STACK_ROUND(sp, items) \
151         (((unsigned long) (sp - items)) &~ 15UL)
152 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
153 #endif
154 
155 #ifndef ELF_BASE_PLATFORM
156 /*
157  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
158  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
159  * will be copied to the user stack in the same manner as AT_PLATFORM.
160  */
161 #define ELF_BASE_PLATFORM NULL
162 #endif
163 
164 static int
165 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
166                 unsigned long load_addr, unsigned long interp_load_addr)
167 {
168         unsigned long p = bprm->p;
169         int argc = bprm->argc;
170         int envc = bprm->envc;
171         elf_addr_t __user *sp;
172         elf_addr_t __user *u_platform;
173         elf_addr_t __user *u_base_platform;
174         elf_addr_t __user *u_rand_bytes;
175         const char *k_platform = ELF_PLATFORM;
176         const char *k_base_platform = ELF_BASE_PLATFORM;
177         unsigned char k_rand_bytes[16];
178         int items;
179         elf_addr_t *elf_info;
180         int ei_index = 0;
181         const struct cred *cred = current_cred();
182         struct vm_area_struct *vma;
183 
184         /*
185          * In some cases (e.g. Hyper-Threading), we want to avoid L1
186          * evictions by the processes running on the same package. One
187          * thing we can do is to shuffle the initial stack for them.
188          */
189 
190         p = arch_align_stack(p);
191 
192         /*
193          * If this architecture has a platform capability string, copy it
194          * to userspace.  In some cases (Sparc), this info is impossible
195          * for userspace to get any other way, in others (i386) it is
196          * merely difficult.
197          */
198         u_platform = NULL;
199         if (k_platform) {
200                 size_t len = strlen(k_platform) + 1;
201 
202                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
203                 if (__copy_to_user(u_platform, k_platform, len))
204                         return -EFAULT;
205         }
206 
207         /*
208          * If this architecture has a "base" platform capability
209          * string, copy it to userspace.
210          */
211         u_base_platform = NULL;
212         if (k_base_platform) {
213                 size_t len = strlen(k_base_platform) + 1;
214 
215                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
216                 if (__copy_to_user(u_base_platform, k_base_platform, len))
217                         return -EFAULT;
218         }
219 
220         /*
221          * Generate 16 random bytes for userspace PRNG seeding.
222          */
223         get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
224         u_rand_bytes = (elf_addr_t __user *)
225                        STACK_ALLOC(p, sizeof(k_rand_bytes));
226         if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
227                 return -EFAULT;
228 
229         /* Create the ELF interpreter info */
230         elf_info = (elf_addr_t *)current->mm->saved_auxv;
231         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
232 #define NEW_AUX_ENT(id, val) \
233         do { \
234                 elf_info[ei_index++] = id; \
235                 elf_info[ei_index++] = val; \
236         } while (0)
237 
238 #ifdef ARCH_DLINFO
239         /* 
240          * ARCH_DLINFO must come first so PPC can do its special alignment of
241          * AUXV.
242          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
243          * ARCH_DLINFO changes
244          */
245         ARCH_DLINFO;
246 #endif
247         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
248         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
249         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
250         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
251         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
252         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
253         NEW_AUX_ENT(AT_BASE, interp_load_addr);
254         NEW_AUX_ENT(AT_FLAGS, 0);
255         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
256         NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
257         NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
258         NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
259         NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
260         NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
261         NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
262 #ifdef ELF_HWCAP2
263         NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
264 #endif
265         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
266         if (k_platform) {
267                 NEW_AUX_ENT(AT_PLATFORM,
268                             (elf_addr_t)(unsigned long)u_platform);
269         }
270         if (k_base_platform) {
271                 NEW_AUX_ENT(AT_BASE_PLATFORM,
272                             (elf_addr_t)(unsigned long)u_base_platform);
273         }
274         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
275                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
276         }
277 #undef NEW_AUX_ENT
278         /* AT_NULL is zero; clear the rest too */
279         memset(&elf_info[ei_index], 0,
280                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
281 
282         /* And advance past the AT_NULL entry.  */
283         ei_index += 2;
284 
285         sp = STACK_ADD(p, ei_index);
286 
287         items = (argc + 1) + (envc + 1) + 1;
288         bprm->p = STACK_ROUND(sp, items);
289 
290         /* Point sp at the lowest address on the stack */
291 #ifdef CONFIG_STACK_GROWSUP
292         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
293         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
294 #else
295         sp = (elf_addr_t __user *)bprm->p;
296 #endif
297 
298 
299         /*
300          * Grow the stack manually; some architectures have a limit on how
301          * far ahead a user-space access may be in order to grow the stack.
302          */
303         vma = find_extend_vma(current->mm, bprm->p);
304         if (!vma)
305                 return -EFAULT;
306 
307         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
308         if (__put_user(argc, sp++))
309                 return -EFAULT;
310 
311         /* Populate list of argv pointers back to argv strings. */
312         p = current->mm->arg_end = current->mm->arg_start;
313         while (argc-- > 0) {
314                 size_t len;
315                 if (__put_user((elf_addr_t)p, sp++))
316                         return -EFAULT;
317                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
318                 if (!len || len > MAX_ARG_STRLEN)
319                         return -EINVAL;
320                 p += len;
321         }
322         if (__put_user(0, sp++))
323                 return -EFAULT;
324         current->mm->arg_end = p;
325 
326         /* Populate list of envp pointers back to envp strings. */
327         current->mm->env_end = current->mm->env_start = p;
328         while (envc-- > 0) {
329                 size_t len;
330                 if (__put_user((elf_addr_t)p, sp++))
331                         return -EFAULT;
332                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
333                 if (!len || len > MAX_ARG_STRLEN)
334                         return -EINVAL;
335                 p += len;
336         }
337         if (__put_user(0, sp++))
338                 return -EFAULT;
339         current->mm->env_end = p;
340 
341         /* Put the elf_info on the stack in the right place.  */
342         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
343                 return -EFAULT;
344         return 0;
345 }
346 
347 #ifndef elf_map
348 
349 static unsigned long elf_map(struct file *filep, unsigned long addr,
350                 struct elf_phdr *eppnt, int prot, int type,
351                 unsigned long total_size)
352 {
353         unsigned long map_addr;
354         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
355         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
356         addr = ELF_PAGESTART(addr);
357         size = ELF_PAGEALIGN(size);
358 
359         /* mmap() will return -EINVAL if given a zero size, but a
360          * segment with zero filesize is perfectly valid */
361         if (!size)
362                 return addr;
363 
364         /*
365         * total_size is the size of the ELF (interpreter) image.
366         * The _first_ mmap needs to know the full size, otherwise
367         * randomization might put this image into an overlapping
368         * position with the ELF binary image. (since size < total_size)
369         * So we first map the 'big' image - and unmap the remainder at
370         * the end. (which unmap is needed for ELF images with holes.)
371         */
372         if (total_size) {
373                 total_size = ELF_PAGEALIGN(total_size);
374                 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
375                 if (!BAD_ADDR(map_addr))
376                         vm_munmap(map_addr+size, total_size-size);
377         } else
378                 map_addr = vm_mmap(filep, addr, size, prot, type, off);
379 
380         return(map_addr);
381 }
382 
383 #endif /* !elf_map */
384 
385 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
386 {
387         int i, first_idx = -1, last_idx = -1;
388 
389         for (i = 0; i < nr; i++) {
390                 if (cmds[i].p_type == PT_LOAD) {
391                         last_idx = i;
392                         if (first_idx == -1)
393                                 first_idx = i;
394                 }
395         }
396         if (first_idx == -1)
397                 return 0;
398 
399         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
400                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
401 }
402 
403 /**
404  * load_elf_phdrs() - load ELF program headers
405  * @elf_ex:   ELF header of the binary whose program headers should be loaded
406  * @elf_file: the opened ELF binary file
407  *
408  * Loads ELF program headers from the binary file elf_file, which has the ELF
409  * header pointed to by elf_ex, into a newly allocated array. The caller is
410  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
411  */
412 static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
413                                        struct file *elf_file)
414 {
415         struct elf_phdr *elf_phdata = NULL;
416         int retval, size, err = -1;
417         loff_t pos = elf_ex->e_phoff;
418 
419         /*
420          * If the size of this structure has changed, then punt, since
421          * we will be doing the wrong thing.
422          */
423         if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
424                 goto out;
425 
426         /* Sanity check the number of program headers... */
427         if (elf_ex->e_phnum < 1 ||
428                 elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
429                 goto out;
430 
431         /* ...and their total size. */
432         size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
433         if (size > ELF_MIN_ALIGN)
434                 goto out;
435 
436         elf_phdata = kmalloc(size, GFP_KERNEL);
437         if (!elf_phdata)
438                 goto out;
439 
440         /* Read in the program headers */
441         retval = kernel_read(elf_file, elf_phdata, size, &pos);
442         if (retval != size) {
443                 err = (retval < 0) ? retval : -EIO;
444                 goto out;
445         }
446 
447         /* Success! */
448         err = 0;
449 out:
450         if (err) {
451                 kfree(elf_phdata);
452                 elf_phdata = NULL;
453         }
454         return elf_phdata;
455 }
456 
457 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
458 
459 /**
460  * struct arch_elf_state - arch-specific ELF loading state
461  *
462  * This structure is used to preserve architecture specific data during
463  * the loading of an ELF file, throughout the checking of architecture
464  * specific ELF headers & through to the point where the ELF load is
465  * known to be proceeding (ie. SET_PERSONALITY).
466  *
467  * This implementation is a dummy for architectures which require no
468  * specific state.
469  */
470 struct arch_elf_state {
471 };
472 
473 #define INIT_ARCH_ELF_STATE {}
474 
475 /**
476  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
477  * @ehdr:       The main ELF header
478  * @phdr:       The program header to check
479  * @elf:        The open ELF file
480  * @is_interp:  True if the phdr is from the interpreter of the ELF being
481  *              loaded, else false.
482  * @state:      Architecture-specific state preserved throughout the process
483  *              of loading the ELF.
484  *
485  * Inspects the program header phdr to validate its correctness and/or
486  * suitability for the system. Called once per ELF program header in the
487  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
488  * interpreter.
489  *
490  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
491  *         with that return code.
492  */
493 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
494                                    struct elf_phdr *phdr,
495                                    struct file *elf, bool is_interp,
496                                    struct arch_elf_state *state)
497 {
498         /* Dummy implementation, always proceed */
499         return 0;
500 }
501 
502 /**
503  * arch_check_elf() - check an ELF executable
504  * @ehdr:       The main ELF header
505  * @has_interp: True if the ELF has an interpreter, else false.
506  * @interp_ehdr: The interpreter's ELF header
507  * @state:      Architecture-specific state preserved throughout the process
508  *              of loading the ELF.
509  *
510  * Provides a final opportunity for architecture code to reject the loading
511  * of the ELF & cause an exec syscall to return an error. This is called after
512  * all program headers to be checked by arch_elf_pt_proc have been.
513  *
514  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
515  *         with that return code.
516  */
517 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
518                                  struct elfhdr *interp_ehdr,
519                                  struct arch_elf_state *state)
520 {
521         /* Dummy implementation, always proceed */
522         return 0;
523 }
524 
525 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
526 
527 /* This is much more generalized than the library routine read function,
528    so we keep this separate.  Technically the library read function
529    is only provided so that we can read a.out libraries that have
530    an ELF header */
531 
532 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
533                 struct file *interpreter, unsigned long *interp_map_addr,
534                 unsigned long no_base, struct elf_phdr *interp_elf_phdata)
535 {
536         struct elf_phdr *eppnt;
537         unsigned long load_addr = 0;
538         int load_addr_set = 0;
539         unsigned long last_bss = 0, elf_bss = 0;
540         int bss_prot = 0;
541         unsigned long error = ~0UL;
542         unsigned long total_size;
543         int i;
544 
545         /* First of all, some simple consistency checks */
546         if (interp_elf_ex->e_type != ET_EXEC &&
547             interp_elf_ex->e_type != ET_DYN)
548                 goto out;
549         if (!elf_check_arch(interp_elf_ex) ||
550             elf_check_fdpic(interp_elf_ex))
551                 goto out;
552         if (!interpreter->f_op->mmap)
553                 goto out;
554 
555         total_size = total_mapping_size(interp_elf_phdata,
556                                         interp_elf_ex->e_phnum);
557         if (!total_size) {
558                 error = -EINVAL;
559                 goto out;
560         }
561 
562         eppnt = interp_elf_phdata;
563         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
564                 if (eppnt->p_type == PT_LOAD) {
565                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
566                         int elf_prot = 0;
567                         unsigned long vaddr = 0;
568                         unsigned long k, map_addr;
569 
570                         if (eppnt->p_flags & PF_R)
571                                 elf_prot = PROT_READ;
572                         if (eppnt->p_flags & PF_W)
573                                 elf_prot |= PROT_WRITE;
574                         if (eppnt->p_flags & PF_X)
575                                 elf_prot |= PROT_EXEC;
576                         vaddr = eppnt->p_vaddr;
577                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
578                                 elf_type |= MAP_FIXED;
579                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
580                                 load_addr = -vaddr;
581 
582                         map_addr = elf_map(interpreter, load_addr + vaddr,
583                                         eppnt, elf_prot, elf_type, total_size);
584                         total_size = 0;
585                         if (!*interp_map_addr)
586                                 *interp_map_addr = map_addr;
587                         error = map_addr;
588                         if (BAD_ADDR(map_addr))
589                                 goto out;
590 
591                         if (!load_addr_set &&
592                             interp_elf_ex->e_type == ET_DYN) {
593                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
594                                 load_addr_set = 1;
595                         }
596 
597                         /*
598                          * Check to see if the section's size will overflow the
599                          * allowed task size. Note that p_filesz must always be
600                          * <= p_memsize so it's only necessary to check p_memsz.
601                          */
602                         k = load_addr + eppnt->p_vaddr;
603                         if (BAD_ADDR(k) ||
604                             eppnt->p_filesz > eppnt->p_memsz ||
605                             eppnt->p_memsz > TASK_SIZE ||
606                             TASK_SIZE - eppnt->p_memsz < k) {
607                                 error = -ENOMEM;
608                                 goto out;
609                         }
610 
611                         /*
612                          * Find the end of the file mapping for this phdr, and
613                          * keep track of the largest address we see for this.
614                          */
615                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
616                         if (k > elf_bss)
617                                 elf_bss = k;
618 
619                         /*
620                          * Do the same thing for the memory mapping - between
621                          * elf_bss and last_bss is the bss section.
622                          */
623                         k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
624                         if (k > last_bss) {
625                                 last_bss = k;
626                                 bss_prot = elf_prot;
627                         }
628                 }
629         }
630 
631         /*
632          * Now fill out the bss section: first pad the last page from
633          * the file up to the page boundary, and zero it from elf_bss
634          * up to the end of the page.
635          */
636         if (padzero(elf_bss)) {
637                 error = -EFAULT;
638                 goto out;
639         }
640         /*
641          * Next, align both the file and mem bss up to the page size,
642          * since this is where elf_bss was just zeroed up to, and where
643          * last_bss will end after the vm_brk_flags() below.
644          */
645         elf_bss = ELF_PAGEALIGN(elf_bss);
646         last_bss = ELF_PAGEALIGN(last_bss);
647         /* Finally, if there is still more bss to allocate, do it. */
648         if (last_bss > elf_bss) {
649                 error = vm_brk_flags(elf_bss, last_bss - elf_bss,
650                                 bss_prot & PROT_EXEC ? VM_EXEC : 0);
651                 if (error)
652                         goto out;
653         }
654 
655         error = load_addr;
656 out:
657         return error;
658 }
659 
660 /*
661  * These are the functions used to load ELF style executables and shared
662  * libraries.  There is no binary dependent code anywhere else.
663  */
664 
665 #ifndef STACK_RND_MASK
666 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
667 #endif
668 
669 static unsigned long randomize_stack_top(unsigned long stack_top)
670 {
671         unsigned long random_variable = 0;
672 
673         if (current->flags & PF_RANDOMIZE) {
674                 random_variable = get_random_long();
675                 random_variable &= STACK_RND_MASK;
676                 random_variable <<= PAGE_SHIFT;
677         }
678 #ifdef CONFIG_STACK_GROWSUP
679         return PAGE_ALIGN(stack_top) + random_variable;
680 #else
681         return PAGE_ALIGN(stack_top) - random_variable;
682 #endif
683 }
684 
685 static int load_elf_binary(struct linux_binprm *bprm)
686 {
687         struct file *interpreter = NULL; /* to shut gcc up */
688         unsigned long load_addr = 0, load_bias = 0;
689         int load_addr_set = 0;
690         char * elf_interpreter = NULL;
691         unsigned long error;
692         struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
693         unsigned long elf_bss, elf_brk;
694         int bss_prot = 0;
695         int retval, i;
696         unsigned long elf_entry;
697         unsigned long interp_load_addr = 0;
698         unsigned long start_code, end_code, start_data, end_data;
699         unsigned long reloc_func_desc __maybe_unused = 0;
700         int executable_stack = EXSTACK_DEFAULT;
701         struct pt_regs *regs = current_pt_regs();
702         struct {
703                 struct elfhdr elf_ex;
704                 struct elfhdr interp_elf_ex;
705         } *loc;
706         struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
707         loff_t pos;
708 
709         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
710         if (!loc) {
711                 retval = -ENOMEM;
712                 goto out_ret;
713         }
714         
715         /* Get the exec-header */
716         loc->elf_ex = *((struct elfhdr *)bprm->buf);
717 
718         retval = -ENOEXEC;
719         /* First of all, some simple consistency checks */
720         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
721                 goto out;
722 
723         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
724                 goto out;
725         if (!elf_check_arch(&loc->elf_ex))
726                 goto out;
727         if (elf_check_fdpic(&loc->elf_ex))
728                 goto out;
729         if (!bprm->file->f_op->mmap)
730                 goto out;
731 
732         elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
733         if (!elf_phdata)
734                 goto out;
735 
736         elf_ppnt = elf_phdata;
737         elf_bss = 0;
738         elf_brk = 0;
739 
740         start_code = ~0UL;
741         end_code = 0;
742         start_data = 0;
743         end_data = 0;
744 
745         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
746                 if (elf_ppnt->p_type == PT_INTERP) {
747                         /* This is the program interpreter used for
748                          * shared libraries - for now assume that this
749                          * is an a.out format binary
750                          */
751                         retval = -ENOEXEC;
752                         if (elf_ppnt->p_filesz > PATH_MAX || 
753                             elf_ppnt->p_filesz < 2)
754                                 goto out_free_ph;
755 
756                         retval = -ENOMEM;
757                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
758                                                   GFP_KERNEL);
759                         if (!elf_interpreter)
760                                 goto out_free_ph;
761 
762                         pos = elf_ppnt->p_offset;
763                         retval = kernel_read(bprm->file, elf_interpreter,
764                                              elf_ppnt->p_filesz, &pos);
765                         if (retval != elf_ppnt->p_filesz) {
766                                 if (retval >= 0)
767                                         retval = -EIO;
768                                 goto out_free_interp;
769                         }
770                         /* make sure path is NULL terminated */
771                         retval = -ENOEXEC;
772                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
773                                 goto out_free_interp;
774 
775                         interpreter = open_exec(elf_interpreter);
776                         retval = PTR_ERR(interpreter);
777                         if (IS_ERR(interpreter))
778                                 goto out_free_interp;
779 
780                         /*
781                          * If the binary is not readable then enforce
782                          * mm->dumpable = 0 regardless of the interpreter's
783                          * permissions.
784                          */
785                         would_dump(bprm, interpreter);
786 
787                         /* Get the exec headers */
788                         pos = 0;
789                         retval = kernel_read(interpreter, &loc->interp_elf_ex,
790                                              sizeof(loc->interp_elf_ex), &pos);
791                         if (retval != sizeof(loc->interp_elf_ex)) {
792                                 if (retval >= 0)
793                                         retval = -EIO;
794                                 goto out_free_dentry;
795                         }
796 
797                         break;
798                 }
799                 elf_ppnt++;
800         }
801 
802         elf_ppnt = elf_phdata;
803         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
804                 switch (elf_ppnt->p_type) {
805                 case PT_GNU_STACK:
806                         if (elf_ppnt->p_flags & PF_X)
807                                 executable_stack = EXSTACK_ENABLE_X;
808                         else
809                                 executable_stack = EXSTACK_DISABLE_X;
810                         break;
811 
812                 case PT_LOPROC ... PT_HIPROC:
813                         retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
814                                                   bprm->file, false,
815                                                   &arch_state);
816                         if (retval)
817                                 goto out_free_dentry;
818                         break;
819                 }
820 
821         /* Some simple consistency checks for the interpreter */
822         if (elf_interpreter) {
823                 retval = -ELIBBAD;
824                 /* Not an ELF interpreter */
825                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
826                         goto out_free_dentry;
827                 /* Verify the interpreter has a valid arch */
828                 if (!elf_check_arch(&loc->interp_elf_ex) ||
829                     elf_check_fdpic(&loc->interp_elf_ex))
830                         goto out_free_dentry;
831 
832                 /* Load the interpreter program headers */
833                 interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
834                                                    interpreter);
835                 if (!interp_elf_phdata)
836                         goto out_free_dentry;
837 
838                 /* Pass PT_LOPROC..PT_HIPROC headers to arch code */
839                 elf_ppnt = interp_elf_phdata;
840                 for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
841                         switch (elf_ppnt->p_type) {
842                         case PT_LOPROC ... PT_HIPROC:
843                                 retval = arch_elf_pt_proc(&loc->interp_elf_ex,
844                                                           elf_ppnt, interpreter,
845                                                           true, &arch_state);
846                                 if (retval)
847                                         goto out_free_dentry;
848                                 break;
849                         }
850         }
851 
852         /*
853          * Allow arch code to reject the ELF at this point, whilst it's
854          * still possible to return an error to the code that invoked
855          * the exec syscall.
856          */
857         retval = arch_check_elf(&loc->elf_ex,
858                                 !!interpreter, &loc->interp_elf_ex,
859                                 &arch_state);
860         if (retval)
861                 goto out_free_dentry;
862 
863         /* Flush all traces of the currently running executable */
864         retval = flush_old_exec(bprm);
865         if (retval)
866                 goto out_free_dentry;
867 
868         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
869            may depend on the personality.  */
870         SET_PERSONALITY2(loc->elf_ex, &arch_state);
871         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
872                 current->personality |= READ_IMPLIES_EXEC;
873 
874         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
875                 current->flags |= PF_RANDOMIZE;
876 
877         setup_new_exec(bprm);
878         install_exec_creds(bprm);
879 
880         /* Do this so that we can load the interpreter, if need be.  We will
881            change some of these later */
882         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
883                                  executable_stack);
884         if (retval < 0)
885                 goto out_free_dentry;
886         
887         current->mm->start_stack = bprm->p;
888 
889         /* Now we do a little grungy work by mmapping the ELF image into
890            the correct location in memory. */
891         for(i = 0, elf_ppnt = elf_phdata;
892             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
893                 int elf_prot = 0, elf_flags;
894                 unsigned long k, vaddr;
895                 unsigned long total_size = 0;
896 
897                 if (elf_ppnt->p_type != PT_LOAD)
898                         continue;
899 
900                 if (unlikely (elf_brk > elf_bss)) {
901                         unsigned long nbyte;
902                     
903                         /* There was a PT_LOAD segment with p_memsz > p_filesz
904                            before this one. Map anonymous pages, if needed,
905                            and clear the area.  */
906                         retval = set_brk(elf_bss + load_bias,
907                                          elf_brk + load_bias,
908                                          bss_prot);
909                         if (retval)
910                                 goto out_free_dentry;
911                         nbyte = ELF_PAGEOFFSET(elf_bss);
912                         if (nbyte) {
913                                 nbyte = ELF_MIN_ALIGN - nbyte;
914                                 if (nbyte > elf_brk - elf_bss)
915                                         nbyte = elf_brk - elf_bss;
916                                 if (clear_user((void __user *)elf_bss +
917                                                         load_bias, nbyte)) {
918                                         /*
919                                          * This bss-zeroing can fail if the ELF
920                                          * file specifies odd protections. So
921                                          * we don't check the return value
922                                          */
923                                 }
924                         }
925                 }
926 
927                 if (elf_ppnt->p_flags & PF_R)
928                         elf_prot |= PROT_READ;
929                 if (elf_ppnt->p_flags & PF_W)
930                         elf_prot |= PROT_WRITE;
931                 if (elf_ppnt->p_flags & PF_X)
932                         elf_prot |= PROT_EXEC;
933 
934                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
935 
936                 vaddr = elf_ppnt->p_vaddr;
937                 /*
938                  * If we are loading ET_EXEC or we have already performed
939                  * the ET_DYN load_addr calculations, proceed normally.
940                  */
941                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
942                         elf_flags |= MAP_FIXED;
943                 } else if (loc->elf_ex.e_type == ET_DYN) {
944                         /*
945                          * This logic is run once for the first LOAD Program
946                          * Header for ET_DYN binaries to calculate the
947                          * randomization (load_bias) for all the LOAD
948                          * Program Headers, and to calculate the entire
949                          * size of the ELF mapping (total_size). (Note that
950                          * load_addr_set is set to true later once the
951                          * initial mapping is performed.)
952                          *
953                          * There are effectively two types of ET_DYN
954                          * binaries: programs (i.e. PIE: ET_DYN with INTERP)
955                          * and loaders (ET_DYN without INTERP, since they
956                          * _are_ the ELF interpreter). The loaders must
957                          * be loaded away from programs since the program
958                          * may otherwise collide with the loader (especially
959                          * for ET_EXEC which does not have a randomized
960                          * position). For example to handle invocations of
961                          * "./ld.so someprog" to test out a new version of
962                          * the loader, the subsequent program that the
963                          * loader loads must avoid the loader itself, so
964                          * they cannot share the same load range. Sufficient
965                          * room for the brk must be allocated with the
966                          * loader as well, since brk must be available with
967                          * the loader.
968                          *
969                          * Therefore, programs are loaded offset from
970                          * ELF_ET_DYN_BASE and loaders are loaded into the
971                          * independently randomized mmap region (0 load_bias
972                          * without MAP_FIXED).
973                          */
974                         if (elf_interpreter) {
975                                 load_bias = ELF_ET_DYN_BASE;
976                                 if (current->flags & PF_RANDOMIZE)
977                                         load_bias += arch_mmap_rnd();
978                                 elf_flags |= MAP_FIXED;
979                         } else
980                                 load_bias = 0;
981 
982                         /*
983                          * Since load_bias is used for all subsequent loading
984                          * calculations, we must lower it by the first vaddr
985                          * so that the remaining calculations based on the
986                          * ELF vaddrs will be correctly offset. The result
987                          * is then page aligned.
988                          */
989                         load_bias = ELF_PAGESTART(load_bias - vaddr);
990 
991                         total_size = total_mapping_size(elf_phdata,
992                                                         loc->elf_ex.e_phnum);
993                         if (!total_size) {
994                                 retval = -EINVAL;
995                                 goto out_free_dentry;
996                         }
997                 }
998 
999                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1000                                 elf_prot, elf_flags, total_size);
1001                 if (BAD_ADDR(error)) {
1002                         retval = IS_ERR((void *)error) ?
1003                                 PTR_ERR((void*)error) : -EINVAL;
1004                         goto out_free_dentry;
1005                 }
1006 
1007                 if (!load_addr_set) {
1008                         load_addr_set = 1;
1009                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1010                         if (loc->elf_ex.e_type == ET_DYN) {
1011                                 load_bias += error -
1012                                              ELF_PAGESTART(load_bias + vaddr);
1013                                 load_addr += load_bias;
1014                                 reloc_func_desc = load_bias;
1015                         }
1016                 }
1017                 k = elf_ppnt->p_vaddr;
1018                 if (k < start_code)
1019                         start_code = k;
1020                 if (start_data < k)
1021                         start_data = k;
1022 
1023                 /*
1024                  * Check to see if the section's size will overflow the
1025                  * allowed task size. Note that p_filesz must always be
1026                  * <= p_memsz so it is only necessary to check p_memsz.
1027                  */
1028                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1029                     elf_ppnt->p_memsz > TASK_SIZE ||
1030                     TASK_SIZE - elf_ppnt->p_memsz < k) {
1031                         /* set_brk can never work. Avoid overflows. */
1032                         retval = -EINVAL;
1033                         goto out_free_dentry;
1034                 }
1035 
1036                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1037 
1038                 if (k > elf_bss)
1039                         elf_bss = k;
1040                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1041                         end_code = k;
1042                 if (end_data < k)
1043                         end_data = k;
1044                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1045                 if (k > elf_brk) {
1046                         bss_prot = elf_prot;
1047                         elf_brk = k;
1048                 }
1049         }
1050 
1051         loc->elf_ex.e_entry += load_bias;
1052         elf_bss += load_bias;
1053         elf_brk += load_bias;
1054         start_code += load_bias;
1055         end_code += load_bias;
1056         start_data += load_bias;
1057         end_data += load_bias;
1058 
1059         /* Calling set_brk effectively mmaps the pages that we need
1060          * for the bss and break sections.  We must do this before
1061          * mapping in the interpreter, to make sure it doesn't wind
1062          * up getting placed where the bss needs to go.
1063          */
1064         retval = set_brk(elf_bss, elf_brk, bss_prot);
1065         if (retval)
1066                 goto out_free_dentry;
1067         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1068                 retval = -EFAULT; /* Nobody gets to see this, but.. */
1069                 goto out_free_dentry;
1070         }
1071 
1072         if (elf_interpreter) {
1073                 unsigned long interp_map_addr = 0;
1074 
1075                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
1076                                             interpreter,
1077                                             &interp_map_addr,
1078                                             load_bias, interp_elf_phdata);
1079                 if (!IS_ERR((void *)elf_entry)) {
1080                         /*
1081                          * load_elf_interp() returns relocation
1082                          * adjustment
1083                          */
1084                         interp_load_addr = elf_entry;
1085                         elf_entry += loc->interp_elf_ex.e_entry;
1086                 }
1087                 if (BAD_ADDR(elf_entry)) {
1088                         retval = IS_ERR((void *)elf_entry) ?
1089                                         (int)elf_entry : -EINVAL;
1090                         goto out_free_dentry;
1091                 }
1092                 reloc_func_desc = interp_load_addr;
1093 
1094                 allow_write_access(interpreter);
1095                 fput(interpreter);
1096                 kfree(elf_interpreter);
1097         } else {
1098                 elf_entry = loc->elf_ex.e_entry;
1099                 if (BAD_ADDR(elf_entry)) {
1100                         retval = -EINVAL;
1101                         goto out_free_dentry;
1102                 }
1103         }
1104 
1105         kfree(interp_elf_phdata);
1106         kfree(elf_phdata);
1107 
1108         set_binfmt(&elf_format);
1109 
1110 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1111         retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1112         if (retval < 0)
1113                 goto out;
1114 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1115 
1116         retval = create_elf_tables(bprm, &loc->elf_ex,
1117                           load_addr, interp_load_addr);
1118         if (retval < 0)
1119                 goto out;
1120         /* N.B. passed_fileno might not be initialized? */
1121         current->mm->end_code = end_code;
1122         current->mm->start_code = start_code;
1123         current->mm->start_data = start_data;
1124         current->mm->end_data = end_data;
1125         current->mm->start_stack = bprm->p;
1126 
1127         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1128                 current->mm->brk = current->mm->start_brk =
1129                         arch_randomize_brk(current->mm);
1130 #ifdef compat_brk_randomized
1131                 current->brk_randomized = 1;
1132 #endif
1133         }
1134 
1135         if (current->personality & MMAP_PAGE_ZERO) {
1136                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1137                    and some applications "depend" upon this behavior.
1138                    Since we do not have the power to recompile these, we
1139                    emulate the SVr4 behavior. Sigh. */
1140                 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1141                                 MAP_FIXED | MAP_PRIVATE, 0);
1142         }
1143 
1144 #ifdef ELF_PLAT_INIT
1145         /*
1146          * The ABI may specify that certain registers be set up in special
1147          * ways (on i386 %edx is the address of a DT_FINI function, for
1148          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1149          * that the e_entry field is the address of the function descriptor
1150          * for the startup routine, rather than the address of the startup
1151          * routine itself.  This macro performs whatever initialization to
1152          * the regs structure is required as well as any relocations to the
1153          * function descriptor entries when executing dynamically links apps.
1154          */
1155         ELF_PLAT_INIT(regs, reloc_func_desc);
1156 #endif
1157 
1158         start_thread(regs, elf_entry, bprm->p);
1159         retval = 0;
1160 out:
1161         kfree(loc);
1162 out_ret:
1163         return retval;
1164 
1165         /* error cleanup */
1166 out_free_dentry:
1167         kfree(interp_elf_phdata);
1168         allow_write_access(interpreter);
1169         if (interpreter)
1170                 fput(interpreter);
1171 out_free_interp:
1172         kfree(elf_interpreter);
1173 out_free_ph:
1174         kfree(elf_phdata);
1175         goto out;
1176 }
1177 
1178 #ifdef CONFIG_USELIB
1179 /* This is really simpleminded and specialized - we are loading an
1180    a.out library that is given an ELF header. */
1181 static int load_elf_library(struct file *file)
1182 {
1183         struct elf_phdr *elf_phdata;
1184         struct elf_phdr *eppnt;
1185         unsigned long elf_bss, bss, len;
1186         int retval, error, i, j;
1187         struct elfhdr elf_ex;
1188         loff_t pos = 0;
1189 
1190         error = -ENOEXEC;
1191         retval = kernel_read(file, &elf_ex, sizeof(elf_ex), &pos);
1192         if (retval != sizeof(elf_ex))
1193                 goto out;
1194 
1195         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1196                 goto out;
1197 
1198         /* First of all, some simple consistency checks */
1199         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1200             !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1201                 goto out;
1202         if (elf_check_fdpic(&elf_ex))
1203                 goto out;
1204 
1205         /* Now read in all of the header information */
1206 
1207         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1208         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1209 
1210         error = -ENOMEM;
1211         elf_phdata = kmalloc(j, GFP_KERNEL);
1212         if (!elf_phdata)
1213                 goto out;
1214 
1215         eppnt = elf_phdata;
1216         error = -ENOEXEC;
1217         pos =  elf_ex.e_phoff;
1218         retval = kernel_read(file, eppnt, j, &pos);
1219         if (retval != j)
1220                 goto out_free_ph;
1221 
1222         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1223                 if ((eppnt + i)->p_type == PT_LOAD)
1224                         j++;
1225         if (j != 1)
1226                 goto out_free_ph;
1227 
1228         while (eppnt->p_type != PT_LOAD)
1229                 eppnt++;
1230 
1231         /* Now use mmap to map the library into memory. */
1232         error = vm_mmap(file,
1233                         ELF_PAGESTART(eppnt->p_vaddr),
1234                         (eppnt->p_filesz +
1235                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1236                         PROT_READ | PROT_WRITE | PROT_EXEC,
1237                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1238                         (eppnt->p_offset -
1239                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1240         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1241                 goto out_free_ph;
1242 
1243         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1244         if (padzero(elf_bss)) {
1245                 error = -EFAULT;
1246                 goto out_free_ph;
1247         }
1248 
1249         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1250                             ELF_MIN_ALIGN - 1);
1251         bss = eppnt->p_memsz + eppnt->p_vaddr;
1252         if (bss > len) {
1253                 error = vm_brk(len, bss - len);
1254                 if (error)
1255                         goto out_free_ph;
1256         }
1257         error = 0;
1258 
1259 out_free_ph:
1260         kfree(elf_phdata);
1261 out:
1262         return error;
1263 }
1264 #endif /* #ifdef CONFIG_USELIB */
1265 
1266 #ifdef CONFIG_ELF_CORE
1267 /*
1268  * ELF core dumper
1269  *
1270  * Modelled on fs/exec.c:aout_core_dump()
1271  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1272  */
1273 
1274 /*
1275  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1276  * that are useful for post-mortem analysis are included in every core dump.
1277  * In that way we ensure that the core dump is fully interpretable later
1278  * without matching up the same kernel and hardware config to see what PC values
1279  * meant. These special mappings include - vDSO, vsyscall, and other
1280  * architecture specific mappings
1281  */
1282 static bool always_dump_vma(struct vm_area_struct *vma)
1283 {
1284         /* Any vsyscall mappings? */
1285         if (vma == get_gate_vma(vma->vm_mm))
1286                 return true;
1287 
1288         /*
1289          * Assume that all vmas with a .name op should always be dumped.
1290          * If this changes, a new vm_ops field can easily be added.
1291          */
1292         if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1293                 return true;
1294 
1295         /*
1296          * arch_vma_name() returns non-NULL for special architecture mappings,
1297          * such as vDSO sections.
1298          */
1299         if (arch_vma_name(vma))
1300                 return true;
1301 
1302         return false;
1303 }
1304 
1305 /*
1306  * Decide what to dump of a segment, part, all or none.
1307  */
1308 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1309                                    unsigned long mm_flags)
1310 {
1311 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1312 
1313         /* always dump the vdso and vsyscall sections */
1314         if (always_dump_vma(vma))
1315                 goto whole;
1316 
1317         if (vma->vm_flags & VM_DONTDUMP)
1318                 return 0;
1319 
1320         /* support for DAX */
1321         if (vma_is_dax(vma)) {
1322                 if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1323                         goto whole;
1324                 if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1325                         goto whole;
1326                 return 0;
1327         }
1328 
1329         /* Hugetlb memory check */
1330         if (vma->vm_flags & VM_HUGETLB) {
1331                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1332                         goto whole;
1333                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1334                         goto whole;
1335                 return 0;
1336         }
1337 
1338         /* Do not dump I/O mapped devices or special mappings */
1339         if (vma->vm_flags & VM_IO)
1340                 return 0;
1341 
1342         /* By default, dump shared memory if mapped from an anonymous file. */
1343         if (vma->vm_flags & VM_SHARED) {
1344                 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1345                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1346                         goto whole;
1347                 return 0;
1348         }
1349 
1350         /* Dump segments that have been written to.  */
1351         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1352                 goto whole;
1353         if (vma->vm_file == NULL)
1354                 return 0;
1355 
1356         if (FILTER(MAPPED_PRIVATE))
1357                 goto whole;
1358 
1359         /*
1360          * If this looks like the beginning of a DSO or executable mapping,
1361          * check for an ELF header.  If we find one, dump the first page to
1362          * aid in determining what was mapped here.
1363          */
1364         if (FILTER(ELF_HEADERS) &&
1365             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1366                 u32 __user *header = (u32 __user *) vma->vm_start;
1367                 u32 word;
1368                 mm_segment_t fs = get_fs();
1369                 /*
1370                  * Doing it this way gets the constant folded by GCC.
1371                  */
1372                 union {
1373                         u32 cmp;
1374                         char elfmag[SELFMAG];
1375                 } magic;
1376                 BUILD_BUG_ON(SELFMAG != sizeof word);
1377                 magic.elfmag[EI_MAG0] = ELFMAG0;
1378                 magic.elfmag[EI_MAG1] = ELFMAG1;
1379                 magic.elfmag[EI_MAG2] = ELFMAG2;
1380                 magic.elfmag[EI_MAG3] = ELFMAG3;
1381                 /*
1382                  * Switch to the user "segment" for get_user(),
1383                  * then put back what elf_core_dump() had in place.
1384                  */
1385                 set_fs(USER_DS);
1386                 if (unlikely(get_user(word, header)))
1387                         word = 0;
1388                 set_fs(fs);
1389                 if (word == magic.cmp)
1390                         return PAGE_SIZE;
1391         }
1392 
1393 #undef  FILTER
1394 
1395         return 0;
1396 
1397 whole:
1398         return vma->vm_end - vma->vm_start;
1399 }
1400 
1401 /* An ELF note in memory */
1402 struct memelfnote
1403 {
1404         const char *name;
1405         int type;
1406         unsigned int datasz;
1407         void *data;
1408 };
1409 
1410 static int notesize(struct memelfnote *en)
1411 {
1412         int sz;
1413 
1414         sz = sizeof(struct elf_note);
1415         sz += roundup(strlen(en->name) + 1, 4);
1416         sz += roundup(en->datasz, 4);
1417 
1418         return sz;
1419 }
1420 
1421 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1422 {
1423         struct elf_note en;
1424         en.n_namesz = strlen(men->name) + 1;
1425         en.n_descsz = men->datasz;
1426         en.n_type = men->type;
1427 
1428         return dump_emit(cprm, &en, sizeof(en)) &&
1429             dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1430             dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1431 }
1432 
1433 static void fill_elf_header(struct elfhdr *elf, int segs,
1434                             u16 machine, u32 flags)
1435 {
1436         memset(elf, 0, sizeof(*elf));
1437 
1438         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1439         elf->e_ident[EI_CLASS] = ELF_CLASS;
1440         elf->e_ident[EI_DATA] = ELF_DATA;
1441         elf->e_ident[EI_VERSION] = EV_CURRENT;
1442         elf->e_ident[EI_OSABI] = ELF_OSABI;
1443 
1444         elf->e_type = ET_CORE;
1445         elf->e_machine = machine;
1446         elf->e_version = EV_CURRENT;
1447         elf->e_phoff = sizeof(struct elfhdr);
1448         elf->e_flags = flags;
1449         elf->e_ehsize = sizeof(struct elfhdr);
1450         elf->e_phentsize = sizeof(struct elf_phdr);
1451         elf->e_phnum = segs;
1452 
1453         return;
1454 }
1455 
1456 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1457 {
1458         phdr->p_type = PT_NOTE;
1459         phdr->p_offset = offset;
1460         phdr->p_vaddr = 0;
1461         phdr->p_paddr = 0;
1462         phdr->p_filesz = sz;
1463         phdr->p_memsz = 0;
1464         phdr->p_flags = 0;
1465         phdr->p_align = 0;
1466         return;
1467 }
1468 
1469 static void fill_note(struct memelfnote *note, const char *name, int type, 
1470                 unsigned int sz, void *data)
1471 {
1472         note->name = name;
1473         note->type = type;
1474         note->datasz = sz;
1475         note->data = data;
1476         return;
1477 }
1478 
1479 /*
1480  * fill up all the fields in prstatus from the given task struct, except
1481  * registers which need to be filled up separately.
1482  */
1483 static void fill_prstatus(struct elf_prstatus *prstatus,
1484                 struct task_struct *p, long signr)
1485 {
1486         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1487         prstatus->pr_sigpend = p->pending.signal.sig[0];
1488         prstatus->pr_sighold = p->blocked.sig[0];
1489         rcu_read_lock();
1490         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1491         rcu_read_unlock();
1492         prstatus->pr_pid = task_pid_vnr(p);
1493         prstatus->pr_pgrp = task_pgrp_vnr(p);
1494         prstatus->pr_sid = task_session_vnr(p);
1495         if (thread_group_leader(p)) {
1496                 struct task_cputime cputime;
1497 
1498                 /*
1499                  * This is the record for the group leader.  It shows the
1500                  * group-wide total, not its individual thread total.
1501                  */
1502                 thread_group_cputime(p, &cputime);
1503                 prstatus->pr_utime = ns_to_timeval(cputime.utime);
1504                 prstatus->pr_stime = ns_to_timeval(cputime.stime);
1505         } else {
1506                 u64 utime, stime;
1507 
1508                 task_cputime(p, &utime, &stime);
1509                 prstatus->pr_utime = ns_to_timeval(utime);
1510                 prstatus->pr_stime = ns_to_timeval(stime);
1511         }
1512 
1513         prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1514         prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1515 }
1516 
1517 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1518                        struct mm_struct *mm)
1519 {
1520         const struct cred *cred;
1521         unsigned int i, len;
1522         
1523         /* first copy the parameters from user space */
1524         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1525 
1526         len = mm->arg_end - mm->arg_start;
1527         if (len >= ELF_PRARGSZ)
1528                 len = ELF_PRARGSZ-1;
1529         if (copy_from_user(&psinfo->pr_psargs,
1530                            (const char __user *)mm->arg_start, len))
1531                 return -EFAULT;
1532         for(i = 0; i < len; i++)
1533                 if (psinfo->pr_psargs[i] == 0)
1534                         psinfo->pr_psargs[i] = ' ';
1535         psinfo->pr_psargs[len] = 0;
1536 
1537         rcu_read_lock();
1538         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1539         rcu_read_unlock();
1540         psinfo->pr_pid = task_pid_vnr(p);
1541         psinfo->pr_pgrp = task_pgrp_vnr(p);
1542         psinfo->pr_sid = task_session_vnr(p);
1543 
1544         i = p->state ? ffz(~p->state) + 1 : 0;
1545         psinfo->pr_state = i;
1546         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1547         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1548         psinfo->pr_nice = task_nice(p);
1549         psinfo->pr_flag = p->flags;
1550         rcu_read_lock();
1551         cred = __task_cred(p);
1552         SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1553         SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1554         rcu_read_unlock();
1555         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1556         
1557         return 0;
1558 }
1559 
1560 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1561 {
1562         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1563         int i = 0;
1564         do
1565                 i += 2;
1566         while (auxv[i - 2] != AT_NULL);
1567         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1568 }
1569 
1570 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1571                 const siginfo_t *siginfo)
1572 {
1573         mm_segment_t old_fs = get_fs();
1574         set_fs(KERNEL_DS);
1575         copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1576         set_fs(old_fs);
1577         fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1578 }
1579 
1580 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1581 /*
1582  * Format of NT_FILE note:
1583  *
1584  * long count     -- how many files are mapped
1585  * long page_size -- units for file_ofs
1586  * array of [COUNT] elements of
1587  *   long start
1588  *   long end
1589  *   long file_ofs
1590  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1591  */
1592 static int fill_files_note(struct memelfnote *note)
1593 {
1594         struct vm_area_struct *vma;
1595         unsigned count, size, names_ofs, remaining, n;
1596         user_long_t *data;
1597         user_long_t *start_end_ofs;
1598         char *name_base, *name_curpos;
1599 
1600         /* *Estimated* file count and total data size needed */
1601         count = current->mm->map_count;
1602         size = count * 64;
1603 
1604         names_ofs = (2 + 3 * count) * sizeof(data[0]);
1605  alloc:
1606         if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1607                 return -EINVAL;
1608         size = round_up(size, PAGE_SIZE);
1609         data = vmalloc(size);
1610         if (!data)
1611                 return -ENOMEM;
1612 
1613         start_end_ofs = data + 2;
1614         name_base = name_curpos = ((char *)data) + names_ofs;
1615         remaining = size - names_ofs;
1616         count = 0;
1617         for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1618                 struct file *file;
1619                 const char *filename;
1620 
1621                 file = vma->vm_file;
1622                 if (!file)
1623                         continue;
1624                 filename = file_path(file, name_curpos, remaining);
1625                 if (IS_ERR(filename)) {
1626                         if (PTR_ERR(filename) == -ENAMETOOLONG) {
1627                                 vfree(data);
1628                                 size = size * 5 / 4;
1629                                 goto alloc;
1630                         }
1631                         continue;
1632                 }
1633 
1634                 /* file_path() fills at the end, move name down */
1635                 /* n = strlen(filename) + 1: */
1636                 n = (name_curpos + remaining) - filename;
1637                 remaining = filename - name_curpos;
1638                 memmove(name_curpos, filename, n);
1639                 name_curpos += n;
1640 
1641                 *start_end_ofs++ = vma->vm_start;
1642                 *start_end_ofs++ = vma->vm_end;
1643                 *start_end_ofs++ = vma->vm_pgoff;
1644                 count++;
1645         }
1646 
1647         /* Now we know exact count of files, can store it */
1648         data[0] = count;
1649         data[1] = PAGE_SIZE;
1650         /*
1651          * Count usually is less than current->mm->map_count,
1652          * we need to move filenames down.
1653          */
1654         n = current->mm->map_count - count;
1655         if (n != 0) {
1656                 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1657                 memmove(name_base - shift_bytes, name_base,
1658                         name_curpos - name_base);
1659                 name_curpos -= shift_bytes;
1660         }
1661 
1662         size = name_curpos - (char *)data;
1663         fill_note(note, "CORE", NT_FILE, size, data);
1664         return 0;
1665 }
1666 
1667 #ifdef CORE_DUMP_USE_REGSET
1668 #include <linux/regset.h>
1669 
1670 struct elf_thread_core_info {
1671         struct elf_thread_core_info *next;
1672         struct task_struct *task;
1673         struct elf_prstatus prstatus;
1674         struct memelfnote notes[0];
1675 };
1676 
1677 struct elf_note_info {
1678         struct elf_thread_core_info *thread;
1679         struct memelfnote psinfo;
1680         struct memelfnote signote;
1681         struct memelfnote auxv;
1682         struct memelfnote files;
1683         user_siginfo_t csigdata;
1684         size_t size;
1685         int thread_notes;
1686 };
1687 
1688 /*
1689  * When a regset has a writeback hook, we call it on each thread before
1690  * dumping user memory.  On register window machines, this makes sure the
1691  * user memory backing the register data is up to date before we read it.
1692  */
1693 static void do_thread_regset_writeback(struct task_struct *task,
1694                                        const struct user_regset *regset)
1695 {
1696         if (regset->writeback)
1697                 regset->writeback(task, regset, 1);
1698 }
1699 
1700 #ifndef PRSTATUS_SIZE
1701 #define PRSTATUS_SIZE(S, R) sizeof(S)
1702 #endif
1703 
1704 #ifndef SET_PR_FPVALID
1705 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1706 #endif
1707 
1708 static int fill_thread_core_info(struct elf_thread_core_info *t,
1709                                  const struct user_regset_view *view,
1710                                  long signr, size_t *total)
1711 {
1712         unsigned int i;
1713         unsigned int regset0_size = regset_size(t->task, &view->regsets[0]);
1714 
1715         /*
1716          * NT_PRSTATUS is the one special case, because the regset data
1717          * goes into the pr_reg field inside the note contents, rather
1718          * than being the whole note contents.  We fill the reset in here.
1719          * We assume that regset 0 is NT_PRSTATUS.
1720          */
1721         fill_prstatus(&t->prstatus, t->task, signr);
1722         (void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset0_size,
1723                                     &t->prstatus.pr_reg, NULL);
1724 
1725         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1726                   PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1727         *total += notesize(&t->notes[0]);
1728 
1729         do_thread_regset_writeback(t->task, &view->regsets[0]);
1730 
1731         /*
1732          * Each other regset might generate a note too.  For each regset
1733          * that has no core_note_type or is inactive, we leave t->notes[i]
1734          * all zero and we'll know to skip writing it later.
1735          */
1736         for (i = 1; i < view->n; ++i) {
1737                 const struct user_regset *regset = &view->regsets[i];
1738                 do_thread_regset_writeback(t->task, regset);
1739                 if (regset->core_note_type && regset->get &&
1740                     (!regset->active || regset->active(t->task, regset))) {
1741                         int ret;
1742                         size_t size = regset_size(t->task, regset);
1743                         void *data = kmalloc(size, GFP_KERNEL);
1744                         if (unlikely(!data))
1745                                 return 0;
1746                         ret = regset->get(t->task, regset,
1747                                           0, size, data, NULL);
1748                         if (unlikely(ret))
1749                                 kfree(data);
1750                         else {
1751                                 if (regset->core_note_type != NT_PRFPREG)
1752                                         fill_note(&t->notes[i], "LINUX",
1753                                                   regset->core_note_type,
1754                                                   size, data);
1755                                 else {
1756                                         SET_PR_FPVALID(&t->prstatus,
1757                                                         1, regset0_size);
1758                                         fill_note(&t->notes[i], "CORE",
1759                                                   NT_PRFPREG, size, data);
1760                                 }
1761                                 *total += notesize(&t->notes[i]);
1762                         }
1763                 }
1764         }
1765 
1766         return 1;
1767 }
1768 
1769 static int fill_note_info(struct elfhdr *elf, int phdrs,
1770                           struct elf_note_info *info,
1771                           const siginfo_t *siginfo, struct pt_regs *regs)
1772 {
1773         struct task_struct *dump_task = current;
1774         const struct user_regset_view *view = task_user_regset_view(dump_task);
1775         struct elf_thread_core_info *t;
1776         struct elf_prpsinfo *psinfo;
1777         struct core_thread *ct;
1778         unsigned int i;
1779 
1780         info->size = 0;
1781         info->thread = NULL;
1782 
1783         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1784         if (psinfo == NULL) {
1785                 info->psinfo.data = NULL; /* So we don't free this wrongly */
1786                 return 0;
1787         }
1788 
1789         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1790 
1791         /*
1792          * Figure out how many notes we're going to need for each thread.
1793          */
1794         info->thread_notes = 0;
1795         for (i = 0; i < view->n; ++i)
1796                 if (view->regsets[i].core_note_type != 0)
1797                         ++info->thread_notes;
1798 
1799         /*
1800          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1801          * since it is our one special case.
1802          */
1803         if (unlikely(info->thread_notes == 0) ||
1804             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1805                 WARN_ON(1);
1806                 return 0;
1807         }
1808 
1809         /*
1810          * Initialize the ELF file header.
1811          */
1812         fill_elf_header(elf, phdrs,
1813                         view->e_machine, view->e_flags);
1814 
1815         /*
1816          * Allocate a structure for each thread.
1817          */
1818         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1819                 t = kzalloc(offsetof(struct elf_thread_core_info,
1820                                      notes[info->thread_notes]),
1821                             GFP_KERNEL);
1822                 if (unlikely(!t))
1823                         return 0;
1824 
1825                 t->task = ct->task;
1826                 if (ct->task == dump_task || !info->thread) {
1827                         t->next = info->thread;
1828                         info->thread = t;
1829                 } else {
1830                         /*
1831                          * Make sure to keep the original task at
1832                          * the head of the list.
1833                          */
1834                         t->next = info->thread->next;
1835                         info->thread->next = t;
1836                 }
1837         }
1838 
1839         /*
1840          * Now fill in each thread's information.
1841          */
1842         for (t = info->thread; t != NULL; t = t->next)
1843                 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1844                         return 0;
1845 
1846         /*
1847          * Fill in the two process-wide notes.
1848          */
1849         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1850         info->size += notesize(&info->psinfo);
1851 
1852         fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1853         info->size += notesize(&info->signote);
1854 
1855         fill_auxv_note(&info->auxv, current->mm);
1856         info->size += notesize(&info->auxv);
1857 
1858         if (fill_files_note(&info->files) == 0)
1859                 info->size += notesize(&info->files);
1860 
1861         return 1;
1862 }
1863 
1864 static size_t get_note_info_size(struct elf_note_info *info)
1865 {
1866         return info->size;
1867 }
1868 
1869 /*
1870  * Write all the notes for each thread.  When writing the first thread, the
1871  * process-wide notes are interleaved after the first thread-specific note.
1872  */
1873 static int write_note_info(struct elf_note_info *info,
1874                            struct coredump_params *cprm)
1875 {
1876         bool first = true;
1877         struct elf_thread_core_info *t = info->thread;
1878 
1879         do {
1880                 int i;
1881 
1882                 if (!writenote(&t->notes[0], cprm))
1883                         return 0;
1884 
1885                 if (first && !writenote(&info->psinfo, cprm))
1886                         return 0;
1887                 if (first && !writenote(&info->signote, cprm))
1888                         return 0;
1889                 if (first && !writenote(&info->auxv, cprm))
1890                         return 0;
1891                 if (first && info->files.data &&
1892                                 !writenote(&info->files, cprm))
1893                         return 0;
1894 
1895                 for (i = 1; i < info->thread_notes; ++i)
1896                         if (t->notes[i].data &&
1897                             !writenote(&t->notes[i], cprm))
1898                                 return 0;
1899 
1900                 first = false;
1901                 t = t->next;
1902         } while (t);
1903 
1904         return 1;
1905 }
1906 
1907 static void free_note_info(struct elf_note_info *info)
1908 {
1909         struct elf_thread_core_info *threads = info->thread;
1910         while (threads) {
1911                 unsigned int i;
1912                 struct elf_thread_core_info *t = threads;
1913                 threads = t->next;
1914                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1915                 for (i = 1; i < info->thread_notes; ++i)
1916                         kfree(t->notes[i].data);
1917                 kfree(t);
1918         }
1919         kfree(info->psinfo.data);
1920         vfree(info->files.data);
1921 }
1922 
1923 #else
1924 
1925 /* Here is the structure in which status of each thread is captured. */
1926 struct elf_thread_status
1927 {
1928         struct list_head list;
1929         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1930         elf_fpregset_t fpu;             /* NT_PRFPREG */
1931         struct task_struct *thread;
1932 #ifdef ELF_CORE_COPY_XFPREGS
1933         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1934 #endif
1935         struct memelfnote notes[3];
1936         int num_notes;
1937 };
1938 
1939 /*
1940  * In order to add the specific thread information for the elf file format,
1941  * we need to keep a linked list of every threads pr_status and then create
1942  * a single section for them in the final core file.
1943  */
1944 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1945 {
1946         int sz = 0;
1947         struct task_struct *p = t->thread;
1948         t->num_notes = 0;
1949 
1950         fill_prstatus(&t->prstatus, p, signr);
1951         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);        
1952         
1953         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1954                   &(t->prstatus));
1955         t->num_notes++;
1956         sz += notesize(&t->notes[0]);
1957 
1958         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1959                                                                 &t->fpu))) {
1960                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1961                           &(t->fpu));
1962                 t->num_notes++;
1963                 sz += notesize(&t->notes[1]);
1964         }
1965 
1966 #ifdef ELF_CORE_COPY_XFPREGS
1967         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1968                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1969                           sizeof(t->xfpu), &t->xfpu);
1970                 t->num_notes++;
1971                 sz += notesize(&t->notes[2]);
1972         }
1973 #endif  
1974         return sz;
1975 }
1976 
1977 struct elf_note_info {
1978         struct memelfnote *notes;
1979         struct memelfnote *notes_files;
1980         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1981         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1982         struct list_head thread_list;
1983         elf_fpregset_t *fpu;
1984 #ifdef ELF_CORE_COPY_XFPREGS
1985         elf_fpxregset_t *xfpu;
1986 #endif
1987         user_siginfo_t csigdata;
1988         int thread_status_size;
1989         int numnote;
1990 };
1991 
1992 static int elf_note_info_init(struct elf_note_info *info)
1993 {
1994         memset(info, 0, sizeof(*info));
1995         INIT_LIST_HEAD(&info->thread_list);
1996 
1997         /* Allocate space for ELF notes */
1998         info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1999         if (!info->notes)
2000                 return 0;
2001         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2002         if (!info->psinfo)
2003                 return 0;
2004         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2005         if (!info->prstatus)
2006                 return 0;
2007         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2008         if (!info->fpu)
2009                 return 0;
2010 #ifdef ELF_CORE_COPY_XFPREGS
2011         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
2012         if (!info->xfpu)
2013                 return 0;
2014 #endif
2015         return 1;
2016 }
2017 
2018 static int fill_note_info(struct elfhdr *elf, int phdrs,
2019                           struct elf_note_info *info,
2020                           const siginfo_t *siginfo, struct pt_regs *regs)
2021 {
2022         struct list_head *t;
2023         struct core_thread *ct;
2024         struct elf_thread_status *ets;
2025 
2026         if (!elf_note_info_init(info))
2027                 return 0;
2028 
2029         for (ct = current->mm->core_state->dumper.next;
2030                                         ct; ct = ct->next) {
2031                 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2032                 if (!ets)
2033                         return 0;
2034 
2035                 ets->thread = ct->task;
2036                 list_add(&ets->list, &info->thread_list);
2037         }
2038 
2039         list_for_each(t, &info->thread_list) {
2040                 int sz;
2041 
2042                 ets = list_entry(t, struct elf_thread_status, list);
2043                 sz = elf_dump_thread_status(siginfo->si_signo, ets);
2044                 info->thread_status_size += sz;
2045         }
2046         /* now collect the dump for the current */
2047         memset(info->prstatus, 0, sizeof(*info->prstatus));
2048         fill_prstatus(info->prstatus, current, siginfo->si_signo);
2049         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2050 
2051         /* Set up header */
2052         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2053 
2054         /*
2055          * Set up the notes in similar form to SVR4 core dumps made
2056          * with info from their /proc.
2057          */
2058 
2059         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2060                   sizeof(*info->prstatus), info->prstatus);
2061         fill_psinfo(info->psinfo, current->group_leader, current->mm);
2062         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2063                   sizeof(*info->psinfo), info->psinfo);
2064 
2065         fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2066         fill_auxv_note(info->notes + 3, current->mm);
2067         info->numnote = 4;
2068 
2069         if (fill_files_note(info->notes + info->numnote) == 0) {
2070                 info->notes_files = info->notes + info->numnote;
2071                 info->numnote++;
2072         }
2073 
2074         /* Try to dump the FPU. */
2075         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2076                                                                info->fpu);
2077         if (info->prstatus->pr_fpvalid)
2078                 fill_note(info->notes + info->numnote++,
2079                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2080 #ifdef ELF_CORE_COPY_XFPREGS
2081         if (elf_core_copy_task_xfpregs(current, info->xfpu))
2082                 fill_note(info->notes + info->numnote++,
2083                           "LINUX", ELF_CORE_XFPREG_TYPE,
2084                           sizeof(*info->xfpu), info->xfpu);
2085 #endif
2086 
2087         return 1;
2088 }
2089 
2090 static size_t get_note_info_size(struct elf_note_info *info)
2091 {
2092         int sz = 0;
2093         int i;
2094 
2095         for (i = 0; i < info->numnote; i++)
2096                 sz += notesize(info->notes + i);
2097 
2098         sz += info->thread_status_size;
2099 
2100         return sz;
2101 }
2102 
2103 static int write_note_info(struct elf_note_info *info,
2104                            struct coredump_params *cprm)
2105 {
2106         int i;
2107         struct list_head *t;
2108 
2109         for (i = 0; i < info->numnote; i++)
2110                 if (!writenote(info->notes + i, cprm))
2111                         return 0;
2112 
2113         /* write out the thread status notes section */
2114         list_for_each(t, &info->thread_list) {
2115                 struct elf_thread_status *tmp =
2116                                 list_entry(t, struct elf_thread_status, list);
2117 
2118                 for (i = 0; i < tmp->num_notes; i++)
2119                         if (!writenote(&tmp->notes[i], cprm))
2120                                 return 0;
2121         }
2122 
2123         return 1;
2124 }
2125 
2126 static void free_note_info(struct elf_note_info *info)
2127 {
2128         while (!list_empty(&info->thread_list)) {
2129                 struct list_head *tmp = info->thread_list.next;
2130                 list_del(tmp);
2131                 kfree(list_entry(tmp, struct elf_thread_status, list));
2132         }
2133 
2134         /* Free data possibly allocated by fill_files_note(): */
2135         if (info->notes_files)
2136                 vfree(info->notes_files->data);
2137 
2138         kfree(info->prstatus);
2139         kfree(info->psinfo);
2140         kfree(info->notes);
2141         kfree(info->fpu);
2142 #ifdef ELF_CORE_COPY_XFPREGS
2143         kfree(info->xfpu);
2144 #endif
2145 }
2146 
2147 #endif
2148 
2149 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2150                                         struct vm_area_struct *gate_vma)
2151 {
2152         struct vm_area_struct *ret = tsk->mm->mmap;
2153 
2154         if (ret)
2155                 return ret;
2156         return gate_vma;
2157 }
2158 /*
2159  * Helper function for iterating across a vma list.  It ensures that the caller
2160  * will visit `gate_vma' prior to terminating the search.
2161  */
2162 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2163                                         struct vm_area_struct *gate_vma)
2164 {
2165         struct vm_area_struct *ret;
2166 
2167         ret = this_vma->vm_next;
2168         if (ret)
2169                 return ret;
2170         if (this_vma == gate_vma)
2171                 return NULL;
2172         return gate_vma;
2173 }
2174 
2175 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2176                              elf_addr_t e_shoff, int segs)
2177 {
2178         elf->e_shoff = e_shoff;
2179         elf->e_shentsize = sizeof(*shdr4extnum);
2180         elf->e_shnum = 1;
2181         elf->e_shstrndx = SHN_UNDEF;
2182 
2183         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2184 
2185         shdr4extnum->sh_type = SHT_NULL;
2186         shdr4extnum->sh_size = elf->e_shnum;
2187         shdr4extnum->sh_link = elf->e_shstrndx;
2188         shdr4extnum->sh_info = segs;
2189 }
2190 
2191 /*
2192  * Actual dumper
2193  *
2194  * This is a two-pass process; first we find the offsets of the bits,
2195  * and then they are actually written out.  If we run out of core limit
2196  * we just truncate.
2197  */
2198 static int elf_core_dump(struct coredump_params *cprm)
2199 {
2200         int has_dumped = 0;
2201         mm_segment_t fs;
2202         int segs, i;
2203         size_t vma_data_size = 0;
2204         struct vm_area_struct *vma, *gate_vma;
2205         struct elfhdr *elf = NULL;
2206         loff_t offset = 0, dataoff;
2207         struct elf_note_info info = { };
2208         struct elf_phdr *phdr4note = NULL;
2209         struct elf_shdr *shdr4extnum = NULL;
2210         Elf_Half e_phnum;
2211         elf_addr_t e_shoff;
2212         elf_addr_t *vma_filesz = NULL;
2213 
2214         /*
2215          * We no longer stop all VM operations.
2216          * 
2217          * This is because those proceses that could possibly change map_count
2218          * or the mmap / vma pages are now blocked in do_exit on current
2219          * finishing this core dump.
2220          *
2221          * Only ptrace can touch these memory addresses, but it doesn't change
2222          * the map_count or the pages allocated. So no possibility of crashing
2223          * exists while dumping the mm->vm_next areas to the core file.
2224          */
2225   
2226         /* alloc memory for large data structures: too large to be on stack */
2227         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2228         if (!elf)
2229                 goto out;
2230         /*
2231          * The number of segs are recored into ELF header as 16bit value.
2232          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2233          */
2234         segs = current->mm->map_count;
2235         segs += elf_core_extra_phdrs();
2236 
2237         gate_vma = get_gate_vma(current->mm);
2238         if (gate_vma != NULL)
2239                 segs++;
2240 
2241         /* for notes section */
2242         segs++;
2243 
2244         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2245          * this, kernel supports extended numbering. Have a look at
2246          * include/linux/elf.h for further information. */
2247         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2248 
2249         /*
2250          * Collect all the non-memory information about the process for the
2251          * notes.  This also sets up the file header.
2252          */
2253         if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2254                 goto cleanup;
2255 
2256         has_dumped = 1;
2257 
2258         fs = get_fs();
2259         set_fs(KERNEL_DS);
2260 
2261         offset += sizeof(*elf);                         /* Elf header */
2262         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
2263 
2264         /* Write notes phdr entry */
2265         {
2266                 size_t sz = get_note_info_size(&info);
2267 
2268                 sz += elf_coredump_extra_notes_size();
2269 
2270                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2271                 if (!phdr4note)
2272                         goto end_coredump;
2273 
2274                 fill_elf_note_phdr(phdr4note, sz, offset);
2275                 offset += sz;
2276         }
2277 
2278         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2279 
2280         if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2281                 goto end_coredump;
2282         vma_filesz = vmalloc((segs - 1) * sizeof(*vma_filesz));
2283         if (!vma_filesz)
2284                 goto end_coredump;
2285 
2286         for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2287                         vma = next_vma(vma, gate_vma)) {
2288                 unsigned long dump_size;
2289 
2290                 dump_size = vma_dump_size(vma, cprm->mm_flags);
2291                 vma_filesz[i++] = dump_size;
2292                 vma_data_size += dump_size;
2293         }
2294 
2295         offset += vma_data_size;
2296         offset += elf_core_extra_data_size();
2297         e_shoff = offset;
2298 
2299         if (e_phnum == PN_XNUM) {
2300                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2301                 if (!shdr4extnum)
2302                         goto end_coredump;
2303                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2304         }
2305 
2306         offset = dataoff;
2307 
2308         if (!dump_emit(cprm, elf, sizeof(*elf)))
2309                 goto end_coredump;
2310 
2311         if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2312                 goto end_coredump;
2313 
2314         /* Write program headers for segments dump */
2315         for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2316                         vma = next_vma(vma, gate_vma)) {
2317                 struct elf_phdr phdr;
2318 
2319                 phdr.p_type = PT_LOAD;
2320                 phdr.p_offset = offset;
2321                 phdr.p_vaddr = vma->vm_start;
2322                 phdr.p_paddr = 0;
2323                 phdr.p_filesz = vma_filesz[i++];
2324                 phdr.p_memsz = vma->vm_end - vma->vm_start;
2325                 offset += phdr.p_filesz;
2326                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2327                 if (vma->vm_flags & VM_WRITE)
2328                         phdr.p_flags |= PF_W;
2329                 if (vma->vm_flags & VM_EXEC)
2330                         phdr.p_flags |= PF_X;
2331                 phdr.p_align = ELF_EXEC_PAGESIZE;
2332 
2333                 if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2334                         goto end_coredump;
2335         }
2336 
2337         if (!elf_core_write_extra_phdrs(cprm, offset))
2338                 goto end_coredump;
2339 
2340         /* write out the notes section */
2341         if (!write_note_info(&info, cprm))
2342                 goto end_coredump;
2343 
2344         if (elf_coredump_extra_notes_write(cprm))
2345                 goto end_coredump;
2346 
2347         /* Align to page */
2348         if (!dump_skip(cprm, dataoff - cprm->pos))
2349                 goto end_coredump;
2350 
2351         for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2352                         vma = next_vma(vma, gate_vma)) {
2353                 unsigned long addr;
2354                 unsigned long end;
2355 
2356                 end = vma->vm_start + vma_filesz[i++];
2357 
2358                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2359                         struct page *page;
2360                         int stop;
2361 
2362                         page = get_dump_page(addr);
2363                         if (page) {
2364                                 void *kaddr = kmap(page);
2365                                 stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2366                                 kunmap(page);
2367                                 put_page(page);
2368                         } else
2369                                 stop = !dump_skip(cprm, PAGE_SIZE);
2370                         if (stop)
2371                                 goto end_coredump;
2372                 }
2373         }
2374         dump_truncate(cprm);
2375 
2376         if (!elf_core_write_extra_data(cprm))
2377                 goto end_coredump;
2378 
2379         if (e_phnum == PN_XNUM) {
2380                 if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2381                         goto end_coredump;
2382         }
2383 
2384 end_coredump:
2385         set_fs(fs);
2386 
2387 cleanup:
2388         free_note_info(&info);
2389         kfree(shdr4extnum);
2390         vfree(vma_filesz);
2391         kfree(phdr4note);
2392         kfree(elf);
2393 out:
2394         return has_dumped;
2395 }
2396 
2397 #endif          /* CONFIG_ELF_CORE */
2398 
2399 static int __init init_elf_binfmt(void)
2400 {
2401         register_binfmt(&elf_format);
2402         return 0;
2403 }
2404 
2405 static void __exit exit_elf_binfmt(void)
2406 {
2407         /* Remove the COFF and ELF loaders. */
2408         unregister_binfmt(&elf_format);
2409 }
2410 
2411 core_initcall(init_elf_binfmt);
2412 module_exit(exit_elf_binfmt);
2413 MODULE_LICENSE("GPL");
2414 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp