Back to home page

LXR

 
 

    


0001 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
0002  *
0003  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
0004  * Written by David Howells (dhowells@redhat.com)
0005  * Derived from binfmt_elf.c
0006  *
0007  * This program is free software; you can redistribute it and/or
0008  * modify it under the terms of the GNU General Public License
0009  * as published by the Free Software Foundation; either version
0010  * 2 of the License, or (at your option) any later version.
0011  */
0012 
0013 #include <linux/module.h>
0014 
0015 #include <linux/fs.h>
0016 #include <linux/stat.h>
0017 #include <linux/sched.h>
0018 #include <linux/mm.h>
0019 #include <linux/mman.h>
0020 #include <linux/errno.h>
0021 #include <linux/signal.h>
0022 #include <linux/binfmts.h>
0023 #include <linux/string.h>
0024 #include <linux/file.h>
0025 #include <linux/fcntl.h>
0026 #include <linux/slab.h>
0027 #include <linux/pagemap.h>
0028 #include <linux/security.h>
0029 #include <linux/highmem.h>
0030 #include <linux/highuid.h>
0031 #include <linux/personality.h>
0032 #include <linux/ptrace.h>
0033 #include <linux/init.h>
0034 #include <linux/elf.h>
0035 #include <linux/elf-fdpic.h>
0036 #include <linux/elfcore.h>
0037 #include <linux/coredump.h>
0038 #include <linux/dax.h>
0039 
0040 #include <linux/uaccess.h>
0041 #include <asm/param.h>
0042 #include <asm/pgalloc.h>
0043 
0044 typedef char *elf_caddr_t;
0045 
0046 #if 0
0047 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
0048 #else
0049 #define kdebug(fmt, ...) do {} while(0)
0050 #endif
0051 
0052 #if 0
0053 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
0054 #else
0055 #define kdcore(fmt, ...) do {} while(0)
0056 #endif
0057 
0058 MODULE_LICENSE("GPL");
0059 
0060 static int load_elf_fdpic_binary(struct linux_binprm *);
0061 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
0062 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
0063                   struct mm_struct *, const char *);
0064 
0065 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
0066                    struct elf_fdpic_params *,
0067                    struct elf_fdpic_params *);
0068 
0069 #ifndef CONFIG_MMU
0070 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
0071                            struct file *,
0072                            struct mm_struct *);
0073 #endif
0074 
0075 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
0076                          struct file *, struct mm_struct *);
0077 
0078 #ifdef CONFIG_ELF_CORE
0079 static int elf_fdpic_core_dump(struct coredump_params *cprm);
0080 #endif
0081 
0082 static struct linux_binfmt elf_fdpic_format = {
0083     .module     = THIS_MODULE,
0084     .load_binary    = load_elf_fdpic_binary,
0085 #ifdef CONFIG_ELF_CORE
0086     .core_dump  = elf_fdpic_core_dump,
0087 #endif
0088     .min_coredump   = ELF_EXEC_PAGESIZE,
0089 };
0090 
0091 static int __init init_elf_fdpic_binfmt(void)
0092 {
0093     register_binfmt(&elf_fdpic_format);
0094     return 0;
0095 }
0096 
0097 static void __exit exit_elf_fdpic_binfmt(void)
0098 {
0099     unregister_binfmt(&elf_fdpic_format);
0100 }
0101 
0102 core_initcall(init_elf_fdpic_binfmt);
0103 module_exit(exit_elf_fdpic_binfmt);
0104 
0105 static int is_elf(struct elfhdr *hdr, struct file *file)
0106 {
0107     if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
0108         return 0;
0109     if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
0110         return 0;
0111     if (!elf_check_arch(hdr))
0112         return 0;
0113     if (!file->f_op->mmap)
0114         return 0;
0115     return 1;
0116 }
0117 
0118 #ifndef elf_check_fdpic
0119 #define elf_check_fdpic(x) 0
0120 #endif
0121 
0122 #ifndef elf_check_const_displacement
0123 #define elf_check_const_displacement(x) 0
0124 #endif
0125 
0126 static int is_constdisp(struct elfhdr *hdr)
0127 {
0128     if (!elf_check_fdpic(hdr))
0129         return 1;
0130     if (elf_check_const_displacement(hdr))
0131         return 1;
0132     return 0;
0133 }
0134 
0135 /*****************************************************************************/
0136 /*
0137  * read the program headers table into memory
0138  */
0139 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
0140                  struct file *file)
0141 {
0142     struct elf32_phdr *phdr;
0143     unsigned long size;
0144     int retval, loop;
0145 
0146     if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
0147         return -ENOMEM;
0148     if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
0149         return -ENOMEM;
0150 
0151     size = params->hdr.e_phnum * sizeof(struct elf_phdr);
0152     params->phdrs = kmalloc(size, GFP_KERNEL);
0153     if (!params->phdrs)
0154         return -ENOMEM;
0155 
0156     retval = kernel_read(file, params->hdr.e_phoff,
0157                  (char *) params->phdrs, size);
0158     if (unlikely(retval != size))
0159         return retval < 0 ? retval : -ENOEXEC;
0160 
0161     /* determine stack size for this binary */
0162     phdr = params->phdrs;
0163     for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
0164         if (phdr->p_type != PT_GNU_STACK)
0165             continue;
0166 
0167         if (phdr->p_flags & PF_X)
0168             params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
0169         else
0170             params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
0171 
0172         params->stack_size = phdr->p_memsz;
0173         break;
0174     }
0175 
0176     return 0;
0177 }
0178 
0179 /*****************************************************************************/
0180 /*
0181  * load an fdpic binary into various bits of memory
0182  */
0183 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
0184 {
0185     struct elf_fdpic_params exec_params, interp_params;
0186     struct pt_regs *regs = current_pt_regs();
0187     struct elf_phdr *phdr;
0188     unsigned long stack_size, entryaddr;
0189 #ifdef ELF_FDPIC_PLAT_INIT
0190     unsigned long dynaddr;
0191 #endif
0192 #ifndef CONFIG_MMU
0193     unsigned long stack_prot;
0194 #endif
0195     struct file *interpreter = NULL; /* to shut gcc up */
0196     char *interpreter_name = NULL;
0197     int executable_stack;
0198     int retval, i;
0199 
0200     kdebug("____ LOAD %d ____", current->pid);
0201 
0202     memset(&exec_params, 0, sizeof(exec_params));
0203     memset(&interp_params, 0, sizeof(interp_params));
0204 
0205     exec_params.hdr = *(struct elfhdr *) bprm->buf;
0206     exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
0207 
0208     /* check that this is a binary we know how to deal with */
0209     retval = -ENOEXEC;
0210     if (!is_elf(&exec_params.hdr, bprm->file))
0211         goto error;
0212     if (!elf_check_fdpic(&exec_params.hdr)) {
0213 #ifdef CONFIG_MMU
0214         /* binfmt_elf handles non-fdpic elf except on nommu */
0215         goto error;
0216 #else
0217         /* nommu can only load ET_DYN (PIE) ELF */
0218         if (exec_params.hdr.e_type != ET_DYN)
0219             goto error;
0220 #endif
0221     }
0222 
0223     /* read the program header table */
0224     retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
0225     if (retval < 0)
0226         goto error;
0227 
0228     /* scan for a program header that specifies an interpreter */
0229     phdr = exec_params.phdrs;
0230 
0231     for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
0232         switch (phdr->p_type) {
0233         case PT_INTERP:
0234             retval = -ENOMEM;
0235             if (phdr->p_filesz > PATH_MAX)
0236                 goto error;
0237             retval = -ENOENT;
0238             if (phdr->p_filesz < 2)
0239                 goto error;
0240 
0241             /* read the name of the interpreter into memory */
0242             interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
0243             if (!interpreter_name)
0244                 goto error;
0245 
0246             retval = kernel_read(bprm->file,
0247                          phdr->p_offset,
0248                          interpreter_name,
0249                          phdr->p_filesz);
0250             if (unlikely(retval != phdr->p_filesz)) {
0251                 if (retval >= 0)
0252                     retval = -ENOEXEC;
0253                 goto error;
0254             }
0255 
0256             retval = -ENOENT;
0257             if (interpreter_name[phdr->p_filesz - 1] != '\0')
0258                 goto error;
0259 
0260             kdebug("Using ELF interpreter %s", interpreter_name);
0261 
0262             /* replace the program with the interpreter */
0263             interpreter = open_exec(interpreter_name);
0264             retval = PTR_ERR(interpreter);
0265             if (IS_ERR(interpreter)) {
0266                 interpreter = NULL;
0267                 goto error;
0268             }
0269 
0270             /*
0271              * If the binary is not readable then enforce
0272              * mm->dumpable = 0 regardless of the interpreter's
0273              * permissions.
0274              */
0275             would_dump(bprm, interpreter);
0276 
0277             retval = kernel_read(interpreter, 0, bprm->buf,
0278                          BINPRM_BUF_SIZE);
0279             if (unlikely(retval != BINPRM_BUF_SIZE)) {
0280                 if (retval >= 0)
0281                     retval = -ENOEXEC;
0282                 goto error;
0283             }
0284 
0285             interp_params.hdr = *((struct elfhdr *) bprm->buf);
0286             break;
0287 
0288         case PT_LOAD:
0289 #ifdef CONFIG_MMU
0290             if (exec_params.load_addr == 0)
0291                 exec_params.load_addr = phdr->p_vaddr;
0292 #endif
0293             break;
0294         }
0295 
0296     }
0297 
0298     if (is_constdisp(&exec_params.hdr))
0299         exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
0300 
0301     /* perform insanity checks on the interpreter */
0302     if (interpreter_name) {
0303         retval = -ELIBBAD;
0304         if (!is_elf(&interp_params.hdr, interpreter))
0305             goto error;
0306 
0307         interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
0308 
0309         /* read the interpreter's program header table */
0310         retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
0311         if (retval < 0)
0312             goto error;
0313     }
0314 
0315     stack_size = exec_params.stack_size;
0316     if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
0317         executable_stack = EXSTACK_ENABLE_X;
0318     else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
0319         executable_stack = EXSTACK_DISABLE_X;
0320     else
0321         executable_stack = EXSTACK_DEFAULT;
0322 
0323     if (stack_size == 0) {
0324         stack_size = interp_params.stack_size;
0325         if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
0326             executable_stack = EXSTACK_ENABLE_X;
0327         else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
0328             executable_stack = EXSTACK_DISABLE_X;
0329         else
0330             executable_stack = EXSTACK_DEFAULT;
0331     }
0332 
0333     retval = -ENOEXEC;
0334     if (stack_size == 0)
0335         stack_size = 131072UL; /* same as exec.c's default commit */
0336 
0337     if (is_constdisp(&interp_params.hdr))
0338         interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
0339 
0340     /* flush all traces of the currently running executable */
0341     retval = flush_old_exec(bprm);
0342     if (retval)
0343         goto error;
0344 
0345     /* there's now no turning back... the old userspace image is dead,
0346      * defunct, deceased, etc.
0347      */
0348     if (elf_check_fdpic(&exec_params.hdr))
0349         set_personality(PER_LINUX_FDPIC);
0350     else
0351         set_personality(PER_LINUX);
0352     if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
0353         current->personality |= READ_IMPLIES_EXEC;
0354 
0355     setup_new_exec(bprm);
0356 
0357     set_binfmt(&elf_fdpic_format);
0358 
0359     current->mm->start_code = 0;
0360     current->mm->end_code = 0;
0361     current->mm->start_stack = 0;
0362     current->mm->start_data = 0;
0363     current->mm->end_data = 0;
0364     current->mm->context.exec_fdpic_loadmap = 0;
0365     current->mm->context.interp_fdpic_loadmap = 0;
0366 
0367 #ifdef CONFIG_MMU
0368     elf_fdpic_arch_lay_out_mm(&exec_params,
0369                   &interp_params,
0370                   &current->mm->start_stack,
0371                   &current->mm->start_brk);
0372 
0373     retval = setup_arg_pages(bprm, current->mm->start_stack,
0374                  executable_stack);
0375     if (retval < 0)
0376         goto error;
0377 #endif
0378 
0379     /* load the executable and interpreter into memory */
0380     retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
0381                     "executable");
0382     if (retval < 0)
0383         goto error;
0384 
0385     if (interpreter_name) {
0386         retval = elf_fdpic_map_file(&interp_params, interpreter,
0387                         current->mm, "interpreter");
0388         if (retval < 0) {
0389             printk(KERN_ERR "Unable to load interpreter\n");
0390             goto error;
0391         }
0392 
0393         allow_write_access(interpreter);
0394         fput(interpreter);
0395         interpreter = NULL;
0396     }
0397 
0398 #ifdef CONFIG_MMU
0399     if (!current->mm->start_brk)
0400         current->mm->start_brk = current->mm->end_data;
0401 
0402     current->mm->brk = current->mm->start_brk =
0403         PAGE_ALIGN(current->mm->start_brk);
0404 
0405 #else
0406     /* create a stack area and zero-size brk area */
0407     stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
0408     if (stack_size < PAGE_SIZE * 2)
0409         stack_size = PAGE_SIZE * 2;
0410 
0411     stack_prot = PROT_READ | PROT_WRITE;
0412     if (executable_stack == EXSTACK_ENABLE_X ||
0413         (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
0414         stack_prot |= PROT_EXEC;
0415 
0416     current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
0417                      MAP_PRIVATE | MAP_ANONYMOUS |
0418                      MAP_UNINITIALIZED | MAP_GROWSDOWN,
0419                      0);
0420 
0421     if (IS_ERR_VALUE(current->mm->start_brk)) {
0422         retval = current->mm->start_brk;
0423         current->mm->start_brk = 0;
0424         goto error;
0425     }
0426 
0427     current->mm->brk = current->mm->start_brk;
0428     current->mm->context.end_brk = current->mm->start_brk;
0429     current->mm->start_stack = current->mm->start_brk + stack_size;
0430 #endif
0431 
0432     install_exec_creds(bprm);
0433     if (create_elf_fdpic_tables(bprm, current->mm,
0434                     &exec_params, &interp_params) < 0)
0435         goto error;
0436 
0437     kdebug("- start_code  %lx", current->mm->start_code);
0438     kdebug("- end_code    %lx", current->mm->end_code);
0439     kdebug("- start_data  %lx", current->mm->start_data);
0440     kdebug("- end_data    %lx", current->mm->end_data);
0441     kdebug("- start_brk   %lx", current->mm->start_brk);
0442     kdebug("- brk         %lx", current->mm->brk);
0443     kdebug("- start_stack %lx", current->mm->start_stack);
0444 
0445 #ifdef ELF_FDPIC_PLAT_INIT
0446     /*
0447      * The ABI may specify that certain registers be set up in special
0448      * ways (on i386 %edx is the address of a DT_FINI function, for
0449      * example.  This macro performs whatever initialization to
0450      * the regs structure is required.
0451      */
0452     dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
0453     ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
0454                 dynaddr);
0455 #endif
0456 
0457     /* everything is now ready... get the userspace context ready to roll */
0458     entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
0459     start_thread(regs, entryaddr, current->mm->start_stack);
0460 
0461     retval = 0;
0462 
0463 error:
0464     if (interpreter) {
0465         allow_write_access(interpreter);
0466         fput(interpreter);
0467     }
0468     kfree(interpreter_name);
0469     kfree(exec_params.phdrs);
0470     kfree(exec_params.loadmap);
0471     kfree(interp_params.phdrs);
0472     kfree(interp_params.loadmap);
0473     return retval;
0474 }
0475 
0476 /*****************************************************************************/
0477 
0478 #ifndef ELF_BASE_PLATFORM
0479 /*
0480  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
0481  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
0482  * will be copied to the user stack in the same manner as AT_PLATFORM.
0483  */
0484 #define ELF_BASE_PLATFORM NULL
0485 #endif
0486 
0487 /*
0488  * present useful information to the program by shovelling it onto the new
0489  * process's stack
0490  */
0491 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
0492                    struct mm_struct *mm,
0493                    struct elf_fdpic_params *exec_params,
0494                    struct elf_fdpic_params *interp_params)
0495 {
0496     const struct cred *cred = current_cred();
0497     unsigned long sp, csp, nitems;
0498     elf_caddr_t __user *argv, *envp;
0499     size_t platform_len = 0, len;
0500     char *k_platform, *k_base_platform;
0501     char __user *u_platform, *u_base_platform, *p;
0502     int loop;
0503     int nr; /* reset for each csp adjustment */
0504 
0505 #ifdef CONFIG_MMU
0506     /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
0507      * by the processes running on the same package. One thing we can do is
0508      * to shuffle the initial stack for them, so we give the architecture
0509      * an opportunity to do so here.
0510      */
0511     sp = arch_align_stack(bprm->p);
0512 #else
0513     sp = mm->start_stack;
0514 
0515     /* stack the program arguments and environment */
0516     if (transfer_args_to_stack(bprm, &sp) < 0)
0517         return -EFAULT;
0518     sp &= ~15;
0519 #endif
0520 
0521     /*
0522      * If this architecture has a platform capability string, copy it
0523      * to userspace.  In some cases (Sparc), this info is impossible
0524      * for userspace to get any other way, in others (i386) it is
0525      * merely difficult.
0526      */
0527     k_platform = ELF_PLATFORM;
0528     u_platform = NULL;
0529 
0530     if (k_platform) {
0531         platform_len = strlen(k_platform) + 1;
0532         sp -= platform_len;
0533         u_platform = (char __user *) sp;
0534         if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
0535             return -EFAULT;
0536     }
0537 
0538     /*
0539      * If this architecture has a "base" platform capability
0540      * string, copy it to userspace.
0541      */
0542     k_base_platform = ELF_BASE_PLATFORM;
0543     u_base_platform = NULL;
0544 
0545     if (k_base_platform) {
0546         platform_len = strlen(k_base_platform) + 1;
0547         sp -= platform_len;
0548         u_base_platform = (char __user *) sp;
0549         if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
0550             return -EFAULT;
0551     }
0552 
0553     sp &= ~7UL;
0554 
0555     /* stack the load map(s) */
0556     len = sizeof(struct elf32_fdpic_loadmap);
0557     len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
0558     sp = (sp - len) & ~7UL;
0559     exec_params->map_addr = sp;
0560 
0561     if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
0562         return -EFAULT;
0563 
0564     current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
0565 
0566     if (interp_params->loadmap) {
0567         len = sizeof(struct elf32_fdpic_loadmap);
0568         len += sizeof(struct elf32_fdpic_loadseg) *
0569             interp_params->loadmap->nsegs;
0570         sp = (sp - len) & ~7UL;
0571         interp_params->map_addr = sp;
0572 
0573         if (copy_to_user((void __user *) sp, interp_params->loadmap,
0574                  len) != 0)
0575             return -EFAULT;
0576 
0577         current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
0578     }
0579 
0580     /* force 16 byte _final_ alignment here for generality */
0581 #define DLINFO_ITEMS 15
0582 
0583     nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
0584         (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
0585 
0586     if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
0587         nitems++;
0588 
0589     csp = sp;
0590     sp -= nitems * 2 * sizeof(unsigned long);
0591     sp -= (bprm->envc + 1) * sizeof(char *);    /* envv[] */
0592     sp -= (bprm->argc + 1) * sizeof(char *);    /* argv[] */
0593     sp -= 1 * sizeof(unsigned long);        /* argc */
0594 
0595     csp -= sp & 15UL;
0596     sp -= sp & 15UL;
0597 
0598     /* put the ELF interpreter info on the stack */
0599 #define NEW_AUX_ENT(id, val)                        \
0600     do {                                \
0601         struct { unsigned long _id, _val; } __user *ent;    \
0602                                     \
0603         ent = (void __user *) csp;              \
0604         __put_user((id), &ent[nr]._id);             \
0605         __put_user((val), &ent[nr]._val);           \
0606         nr++;                           \
0607     } while (0)
0608 
0609     nr = 0;
0610     csp -= 2 * sizeof(unsigned long);
0611     NEW_AUX_ENT(AT_NULL, 0);
0612     if (k_platform) {
0613         nr = 0;
0614         csp -= 2 * sizeof(unsigned long);
0615         NEW_AUX_ENT(AT_PLATFORM,
0616                 (elf_addr_t) (unsigned long) u_platform);
0617     }
0618 
0619     if (k_base_platform) {
0620         nr = 0;
0621         csp -= 2 * sizeof(unsigned long);
0622         NEW_AUX_ENT(AT_BASE_PLATFORM,
0623                 (elf_addr_t) (unsigned long) u_base_platform);
0624     }
0625 
0626     if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
0627         nr = 0;
0628         csp -= 2 * sizeof(unsigned long);
0629         NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
0630     }
0631 
0632     nr = 0;
0633     csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
0634     NEW_AUX_ENT(AT_HWCAP,   ELF_HWCAP);
0635 #ifdef ELF_HWCAP2
0636     NEW_AUX_ENT(AT_HWCAP2,  ELF_HWCAP2);
0637 #endif
0638     NEW_AUX_ENT(AT_PAGESZ,  PAGE_SIZE);
0639     NEW_AUX_ENT(AT_CLKTCK,  CLOCKS_PER_SEC);
0640     NEW_AUX_ENT(AT_PHDR,    exec_params->ph_addr);
0641     NEW_AUX_ENT(AT_PHENT,   sizeof(struct elf_phdr));
0642     NEW_AUX_ENT(AT_PHNUM,   exec_params->hdr.e_phnum);
0643     NEW_AUX_ENT(AT_BASE,    interp_params->elfhdr_addr);
0644     NEW_AUX_ENT(AT_FLAGS,   0);
0645     NEW_AUX_ENT(AT_ENTRY,   exec_params->entry_addr);
0646     NEW_AUX_ENT(AT_UID, (elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
0647     NEW_AUX_ENT(AT_EUID,    (elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
0648     NEW_AUX_ENT(AT_GID, (elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
0649     NEW_AUX_ENT(AT_EGID,    (elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
0650     NEW_AUX_ENT(AT_SECURE,  security_bprm_secureexec(bprm));
0651     NEW_AUX_ENT(AT_EXECFN,  bprm->exec);
0652 
0653 #ifdef ARCH_DLINFO
0654     nr = 0;
0655     csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
0656 
0657     /* ARCH_DLINFO must come last so platform specific code can enforce
0658      * special alignment requirements on the AUXV if necessary (eg. PPC).
0659      */
0660     ARCH_DLINFO;
0661 #endif
0662 #undef NEW_AUX_ENT
0663 
0664     /* allocate room for argv[] and envv[] */
0665     csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
0666     envp = (elf_caddr_t __user *) csp;
0667     csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
0668     argv = (elf_caddr_t __user *) csp;
0669 
0670     /* stack argc */
0671     csp -= sizeof(unsigned long);
0672     __put_user(bprm->argc, (unsigned long __user *) csp);
0673 
0674     BUG_ON(csp != sp);
0675 
0676     /* fill in the argv[] array */
0677 #ifdef CONFIG_MMU
0678     current->mm->arg_start = bprm->p;
0679 #else
0680     current->mm->arg_start = current->mm->start_stack -
0681         (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
0682 #endif
0683 
0684     p = (char __user *) current->mm->arg_start;
0685     for (loop = bprm->argc; loop > 0; loop--) {
0686         __put_user((elf_caddr_t) p, argv++);
0687         len = strnlen_user(p, MAX_ARG_STRLEN);
0688         if (!len || len > MAX_ARG_STRLEN)
0689             return -EINVAL;
0690         p += len;
0691     }
0692     __put_user(NULL, argv);
0693     current->mm->arg_end = (unsigned long) p;
0694 
0695     /* fill in the envv[] array */
0696     current->mm->env_start = (unsigned long) p;
0697     for (loop = bprm->envc; loop > 0; loop--) {
0698         __put_user((elf_caddr_t)(unsigned long) p, envp++);
0699         len = strnlen_user(p, MAX_ARG_STRLEN);
0700         if (!len || len > MAX_ARG_STRLEN)
0701             return -EINVAL;
0702         p += len;
0703     }
0704     __put_user(NULL, envp);
0705     current->mm->env_end = (unsigned long) p;
0706 
0707     mm->start_stack = (unsigned long) sp;
0708     return 0;
0709 }
0710 
0711 /*****************************************************************************/
0712 /*
0713  * load the appropriate binary image (executable or interpreter) into memory
0714  * - we assume no MMU is available
0715  * - if no other PIC bits are set in params->hdr->e_flags
0716  *   - we assume that the LOADable segments in the binary are independently relocatable
0717  *   - we assume R/O executable segments are shareable
0718  * - else
0719  *   - we assume the loadable parts of the image to require fixed displacement
0720  *   - the image is not shareable
0721  */
0722 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
0723                   struct file *file,
0724                   struct mm_struct *mm,
0725                   const char *what)
0726 {
0727     struct elf32_fdpic_loadmap *loadmap;
0728 #ifdef CONFIG_MMU
0729     struct elf32_fdpic_loadseg *mseg;
0730 #endif
0731     struct elf32_fdpic_loadseg *seg;
0732     struct elf32_phdr *phdr;
0733     unsigned long load_addr, stop;
0734     unsigned nloads, tmp;
0735     size_t size;
0736     int loop, ret;
0737 
0738     /* allocate a load map table */
0739     nloads = 0;
0740     for (loop = 0; loop < params->hdr.e_phnum; loop++)
0741         if (params->phdrs[loop].p_type == PT_LOAD)
0742             nloads++;
0743 
0744     if (nloads == 0)
0745         return -ELIBBAD;
0746 
0747     size = sizeof(*loadmap) + nloads * sizeof(*seg);
0748     loadmap = kzalloc(size, GFP_KERNEL);
0749     if (!loadmap)
0750         return -ENOMEM;
0751 
0752     params->loadmap = loadmap;
0753 
0754     loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
0755     loadmap->nsegs = nloads;
0756 
0757     load_addr = params->load_addr;
0758     seg = loadmap->segs;
0759 
0760     /* map the requested LOADs into the memory space */
0761     switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
0762     case ELF_FDPIC_FLAG_CONSTDISP:
0763     case ELF_FDPIC_FLAG_CONTIGUOUS:
0764 #ifndef CONFIG_MMU
0765         ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
0766         if (ret < 0)
0767             return ret;
0768         break;
0769 #endif
0770     default:
0771         ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
0772         if (ret < 0)
0773             return ret;
0774         break;
0775     }
0776 
0777     /* map the entry point */
0778     if (params->hdr.e_entry) {
0779         seg = loadmap->segs;
0780         for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
0781             if (params->hdr.e_entry >= seg->p_vaddr &&
0782                 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
0783                 params->entry_addr =
0784                     (params->hdr.e_entry - seg->p_vaddr) +
0785                     seg->addr;
0786                 break;
0787             }
0788         }
0789     }
0790 
0791     /* determine where the program header table has wound up if mapped */
0792     stop = params->hdr.e_phoff;
0793     stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
0794     phdr = params->phdrs;
0795 
0796     for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
0797         if (phdr->p_type != PT_LOAD)
0798             continue;
0799 
0800         if (phdr->p_offset > params->hdr.e_phoff ||
0801             phdr->p_offset + phdr->p_filesz < stop)
0802             continue;
0803 
0804         seg = loadmap->segs;
0805         for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
0806             if (phdr->p_vaddr >= seg->p_vaddr &&
0807                 phdr->p_vaddr + phdr->p_filesz <=
0808                 seg->p_vaddr + seg->p_memsz) {
0809                 params->ph_addr =
0810                     (phdr->p_vaddr - seg->p_vaddr) +
0811                     seg->addr +
0812                     params->hdr.e_phoff - phdr->p_offset;
0813                 break;
0814             }
0815         }
0816         break;
0817     }
0818 
0819     /* determine where the dynamic section has wound up if there is one */
0820     phdr = params->phdrs;
0821     for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
0822         if (phdr->p_type != PT_DYNAMIC)
0823             continue;
0824 
0825         seg = loadmap->segs;
0826         for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
0827             if (phdr->p_vaddr >= seg->p_vaddr &&
0828                 phdr->p_vaddr + phdr->p_memsz <=
0829                 seg->p_vaddr + seg->p_memsz) {
0830                 params->dynamic_addr =
0831                     (phdr->p_vaddr - seg->p_vaddr) +
0832                     seg->addr;
0833 
0834                 /* check the dynamic section contains at least
0835                  * one item, and that the last item is a NULL
0836                  * entry */
0837                 if (phdr->p_memsz == 0 ||
0838                     phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
0839                     goto dynamic_error;
0840 
0841                 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
0842                 if (((Elf32_Dyn *)
0843                      params->dynamic_addr)[tmp - 1].d_tag != 0)
0844                     goto dynamic_error;
0845                 break;
0846             }
0847         }
0848         break;
0849     }
0850 
0851     /* now elide adjacent segments in the load map on MMU linux
0852      * - on uClinux the holes between may actually be filled with system
0853      *   stuff or stuff from other processes
0854      */
0855 #ifdef CONFIG_MMU
0856     nloads = loadmap->nsegs;
0857     mseg = loadmap->segs;
0858     seg = mseg + 1;
0859     for (loop = 1; loop < nloads; loop++) {
0860         /* see if we have a candidate for merging */
0861         if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
0862             load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
0863             if (load_addr == (seg->addr & PAGE_MASK)) {
0864                 mseg->p_memsz +=
0865                     load_addr -
0866                     (mseg->addr + mseg->p_memsz);
0867                 mseg->p_memsz += seg->addr & ~PAGE_MASK;
0868                 mseg->p_memsz += seg->p_memsz;
0869                 loadmap->nsegs--;
0870                 continue;
0871             }
0872         }
0873 
0874         mseg++;
0875         if (mseg != seg)
0876             *mseg = *seg;
0877     }
0878 #endif
0879 
0880     kdebug("Mapped Object [%s]:", what);
0881     kdebug("- elfhdr   : %lx", params->elfhdr_addr);
0882     kdebug("- entry    : %lx", params->entry_addr);
0883     kdebug("- PHDR[]   : %lx", params->ph_addr);
0884     kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
0885     seg = loadmap->segs;
0886     for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
0887         kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
0888                loop,
0889                seg->addr, seg->addr + seg->p_memsz - 1,
0890                seg->p_vaddr, seg->p_memsz);
0891 
0892     return 0;
0893 
0894 dynamic_error:
0895     printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
0896            what, file_inode(file)->i_ino);
0897     return -ELIBBAD;
0898 }
0899 
0900 /*****************************************************************************/
0901 /*
0902  * map a file with constant displacement under uClinux
0903  */
0904 #ifndef CONFIG_MMU
0905 static int elf_fdpic_map_file_constdisp_on_uclinux(
0906     struct elf_fdpic_params *params,
0907     struct file *file,
0908     struct mm_struct *mm)
0909 {
0910     struct elf32_fdpic_loadseg *seg;
0911     struct elf32_phdr *phdr;
0912     unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
0913     int loop, ret;
0914 
0915     load_addr = params->load_addr;
0916     seg = params->loadmap->segs;
0917 
0918     /* determine the bounds of the contiguous overall allocation we must
0919      * make */
0920     phdr = params->phdrs;
0921     for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
0922         if (params->phdrs[loop].p_type != PT_LOAD)
0923             continue;
0924 
0925         if (base > phdr->p_vaddr)
0926             base = phdr->p_vaddr;
0927         if (top < phdr->p_vaddr + phdr->p_memsz)
0928             top = phdr->p_vaddr + phdr->p_memsz;
0929     }
0930 
0931     /* allocate one big anon block for everything */
0932     mflags = MAP_PRIVATE;
0933     if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
0934         mflags |= MAP_EXECUTABLE;
0935 
0936     maddr = vm_mmap(NULL, load_addr, top - base,
0937             PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
0938     if (IS_ERR_VALUE(maddr))
0939         return (int) maddr;
0940 
0941     if (load_addr != 0)
0942         load_addr += PAGE_ALIGN(top - base);
0943 
0944     /* and then load the file segments into it */
0945     phdr = params->phdrs;
0946     for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
0947         if (params->phdrs[loop].p_type != PT_LOAD)
0948             continue;
0949 
0950         seg->addr = maddr + (phdr->p_vaddr - base);
0951         seg->p_vaddr = phdr->p_vaddr;
0952         seg->p_memsz = phdr->p_memsz;
0953 
0954         ret = read_code(file, seg->addr, phdr->p_offset,
0955                        phdr->p_filesz);
0956         if (ret < 0)
0957             return ret;
0958 
0959         /* map the ELF header address if in this segment */
0960         if (phdr->p_offset == 0)
0961             params->elfhdr_addr = seg->addr;
0962 
0963         /* clear any space allocated but not loaded */
0964         if (phdr->p_filesz < phdr->p_memsz) {
0965             if (clear_user((void *) (seg->addr + phdr->p_filesz),
0966                        phdr->p_memsz - phdr->p_filesz))
0967                 return -EFAULT;
0968         }
0969 
0970         if (mm) {
0971             if (phdr->p_flags & PF_X) {
0972                 if (!mm->start_code) {
0973                     mm->start_code = seg->addr;
0974                     mm->end_code = seg->addr +
0975                         phdr->p_memsz;
0976                 }
0977             } else if (!mm->start_data) {
0978                 mm->start_data = seg->addr;
0979                 mm->end_data = seg->addr + phdr->p_memsz;
0980             }
0981         }
0982 
0983         seg++;
0984     }
0985 
0986     return 0;
0987 }
0988 #endif
0989 
0990 /*****************************************************************************/
0991 /*
0992  * map a binary by direct mmap() of the individual PT_LOAD segments
0993  */
0994 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
0995                          struct file *file,
0996                          struct mm_struct *mm)
0997 {
0998     struct elf32_fdpic_loadseg *seg;
0999     struct elf32_phdr *phdr;
1000     unsigned long load_addr, delta_vaddr;
1001     int loop, dvset;
1002 
1003     load_addr = params->load_addr;
1004     delta_vaddr = 0;
1005     dvset = 0;
1006 
1007     seg = params->loadmap->segs;
1008 
1009     /* deal with each load segment separately */
1010     phdr = params->phdrs;
1011     for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1012         unsigned long maddr, disp, excess, excess1;
1013         int prot = 0, flags;
1014 
1015         if (phdr->p_type != PT_LOAD)
1016             continue;
1017 
1018         kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1019                (unsigned long) phdr->p_vaddr,
1020                (unsigned long) phdr->p_offset,
1021                (unsigned long) phdr->p_filesz,
1022                (unsigned long) phdr->p_memsz);
1023 
1024         /* determine the mapping parameters */
1025         if (phdr->p_flags & PF_R) prot |= PROT_READ;
1026         if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1027         if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1028 
1029         flags = MAP_PRIVATE | MAP_DENYWRITE;
1030         if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1031             flags |= MAP_EXECUTABLE;
1032 
1033         maddr = 0;
1034 
1035         switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1036         case ELF_FDPIC_FLAG_INDEPENDENT:
1037             /* PT_LOADs are independently locatable */
1038             break;
1039 
1040         case ELF_FDPIC_FLAG_HONOURVADDR:
1041             /* the specified virtual address must be honoured */
1042             maddr = phdr->p_vaddr;
1043             flags |= MAP_FIXED;
1044             break;
1045 
1046         case ELF_FDPIC_FLAG_CONSTDISP:
1047             /* constant displacement
1048              * - can be mapped anywhere, but must be mapped as a
1049              *   unit
1050              */
1051             if (!dvset) {
1052                 maddr = load_addr;
1053                 delta_vaddr = phdr->p_vaddr;
1054                 dvset = 1;
1055             } else {
1056                 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1057                 flags |= MAP_FIXED;
1058             }
1059             break;
1060 
1061         case ELF_FDPIC_FLAG_CONTIGUOUS:
1062             /* contiguity handled later */
1063             break;
1064 
1065         default:
1066             BUG();
1067         }
1068 
1069         maddr &= PAGE_MASK;
1070 
1071         /* create the mapping */
1072         disp = phdr->p_vaddr & ~PAGE_MASK;
1073         maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1074                 phdr->p_offset - disp);
1075 
1076         kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1077                loop, phdr->p_memsz + disp, prot, flags,
1078                phdr->p_offset - disp, maddr);
1079 
1080         if (IS_ERR_VALUE(maddr))
1081             return (int) maddr;
1082 
1083         if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1084             ELF_FDPIC_FLAG_CONTIGUOUS)
1085             load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1086 
1087         seg->addr = maddr + disp;
1088         seg->p_vaddr = phdr->p_vaddr;
1089         seg->p_memsz = phdr->p_memsz;
1090 
1091         /* map the ELF header address if in this segment */
1092         if (phdr->p_offset == 0)
1093             params->elfhdr_addr = seg->addr;
1094 
1095         /* clear the bit between beginning of mapping and beginning of
1096          * PT_LOAD */
1097         if (prot & PROT_WRITE && disp > 0) {
1098             kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1099             if (clear_user((void __user *) maddr, disp))
1100                 return -EFAULT;
1101             maddr += disp;
1102         }
1103 
1104         /* clear any space allocated but not loaded
1105          * - on uClinux we can just clear the lot
1106          * - on MMU linux we'll get a SIGBUS beyond the last page
1107          *   extant in the file
1108          */
1109         excess = phdr->p_memsz - phdr->p_filesz;
1110         excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1111 
1112 #ifdef CONFIG_MMU
1113         if (excess > excess1) {
1114             unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1115             unsigned long xmaddr;
1116 
1117             flags |= MAP_FIXED | MAP_ANONYMOUS;
1118             xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1119                      prot, flags, 0);
1120 
1121             kdebug("mmap[%d] <anon>"
1122                    " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1123                    loop, xaddr, excess - excess1, prot, flags,
1124                    xmaddr);
1125 
1126             if (xmaddr != xaddr)
1127                 return -ENOMEM;
1128         }
1129 
1130         if (prot & PROT_WRITE && excess1 > 0) {
1131             kdebug("clear[%d] ad=%lx sz=%lx",
1132                    loop, maddr + phdr->p_filesz, excess1);
1133             if (clear_user((void __user *) maddr + phdr->p_filesz,
1134                        excess1))
1135                 return -EFAULT;
1136         }
1137 
1138 #else
1139         if (excess > 0) {
1140             kdebug("clear[%d] ad=%lx sz=%lx",
1141                    loop, maddr + phdr->p_filesz, excess);
1142             if (clear_user((void *) maddr + phdr->p_filesz, excess))
1143                 return -EFAULT;
1144         }
1145 #endif
1146 
1147         if (mm) {
1148             if (phdr->p_flags & PF_X) {
1149                 if (!mm->start_code) {
1150                     mm->start_code = maddr;
1151                     mm->end_code = maddr + phdr->p_memsz;
1152                 }
1153             } else if (!mm->start_data) {
1154                 mm->start_data = maddr;
1155                 mm->end_data = maddr + phdr->p_memsz;
1156             }
1157         }
1158 
1159         seg++;
1160     }
1161 
1162     return 0;
1163 }
1164 
1165 /*****************************************************************************/
1166 /*
1167  * ELF-FDPIC core dumper
1168  *
1169  * Modelled on fs/exec.c:aout_core_dump()
1170  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1171  *
1172  * Modelled on fs/binfmt_elf.c core dumper
1173  */
1174 #ifdef CONFIG_ELF_CORE
1175 
1176 /*
1177  * Decide whether a segment is worth dumping; default is yes to be
1178  * sure (missing info is worse than too much; etc).
1179  * Personally I'd include everything, and use the coredump limit...
1180  *
1181  * I think we should skip something. But I am not sure how. H.J.
1182  */
1183 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1184 {
1185     int dump_ok;
1186 
1187     /* Do not dump I/O mapped devices or special mappings */
1188     if (vma->vm_flags & VM_IO) {
1189         kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1190         return 0;
1191     }
1192 
1193     /* If we may not read the contents, don't allow us to dump
1194      * them either. "dump_write()" can't handle it anyway.
1195      */
1196     if (!(vma->vm_flags & VM_READ)) {
1197         kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1198         return 0;
1199     }
1200 
1201     /* support for DAX */
1202     if (vma_is_dax(vma)) {
1203         if (vma->vm_flags & VM_SHARED) {
1204             dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1205             kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1206                    vma->vm_flags, dump_ok ? "yes" : "no");
1207         } else {
1208             dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1209             kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1210                    vma->vm_flags, dump_ok ? "yes" : "no");
1211         }
1212         return dump_ok;
1213     }
1214 
1215     /* By default, dump shared memory if mapped from an anonymous file. */
1216     if (vma->vm_flags & VM_SHARED) {
1217         if (file_inode(vma->vm_file)->i_nlink == 0) {
1218             dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1219             kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1220                    vma->vm_flags, dump_ok ? "yes" : "no");
1221             return dump_ok;
1222         }
1223 
1224         dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1225         kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1226                vma->vm_flags, dump_ok ? "yes" : "no");
1227         return dump_ok;
1228     }
1229 
1230 #ifdef CONFIG_MMU
1231     /* By default, if it hasn't been written to, don't write it out */
1232     if (!vma->anon_vma) {
1233         dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1234         kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1235                vma->vm_flags, dump_ok ? "yes" : "no");
1236         return dump_ok;
1237     }
1238 #endif
1239 
1240     dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1241     kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1242            dump_ok ? "yes" : "no");
1243     return dump_ok;
1244 }
1245 
1246 /* An ELF note in memory */
1247 struct memelfnote
1248 {
1249     const char *name;
1250     int type;
1251     unsigned int datasz;
1252     void *data;
1253 };
1254 
1255 static int notesize(struct memelfnote *en)
1256 {
1257     int sz;
1258 
1259     sz = sizeof(struct elf_note);
1260     sz += roundup(strlen(en->name) + 1, 4);
1261     sz += roundup(en->datasz, 4);
1262 
1263     return sz;
1264 }
1265 
1266 /* #define DEBUG */
1267 
1268 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1269 {
1270     struct elf_note en;
1271     en.n_namesz = strlen(men->name) + 1;
1272     en.n_descsz = men->datasz;
1273     en.n_type = men->type;
1274 
1275     return dump_emit(cprm, &en, sizeof(en)) &&
1276         dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1277         dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1278 }
1279 
1280 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1281 {
1282     memcpy(elf->e_ident, ELFMAG, SELFMAG);
1283     elf->e_ident[EI_CLASS] = ELF_CLASS;
1284     elf->e_ident[EI_DATA] = ELF_DATA;
1285     elf->e_ident[EI_VERSION] = EV_CURRENT;
1286     elf->e_ident[EI_OSABI] = ELF_OSABI;
1287     memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1288 
1289     elf->e_type = ET_CORE;
1290     elf->e_machine = ELF_ARCH;
1291     elf->e_version = EV_CURRENT;
1292     elf->e_entry = 0;
1293     elf->e_phoff = sizeof(struct elfhdr);
1294     elf->e_shoff = 0;
1295     elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1296     elf->e_ehsize = sizeof(struct elfhdr);
1297     elf->e_phentsize = sizeof(struct elf_phdr);
1298     elf->e_phnum = segs;
1299     elf->e_shentsize = 0;
1300     elf->e_shnum = 0;
1301     elf->e_shstrndx = 0;
1302     return;
1303 }
1304 
1305 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1306 {
1307     phdr->p_type = PT_NOTE;
1308     phdr->p_offset = offset;
1309     phdr->p_vaddr = 0;
1310     phdr->p_paddr = 0;
1311     phdr->p_filesz = sz;
1312     phdr->p_memsz = 0;
1313     phdr->p_flags = 0;
1314     phdr->p_align = 0;
1315     return;
1316 }
1317 
1318 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1319         unsigned int sz, void *data)
1320 {
1321     note->name = name;
1322     note->type = type;
1323     note->datasz = sz;
1324     note->data = data;
1325     return;
1326 }
1327 
1328 /*
1329  * fill up all the fields in prstatus from the given task struct, except
1330  * registers which need to be filled up separately.
1331  */
1332 static void fill_prstatus(struct elf_prstatus *prstatus,
1333               struct task_struct *p, long signr)
1334 {
1335     prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1336     prstatus->pr_sigpend = p->pending.signal.sig[0];
1337     prstatus->pr_sighold = p->blocked.sig[0];
1338     rcu_read_lock();
1339     prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1340     rcu_read_unlock();
1341     prstatus->pr_pid = task_pid_vnr(p);
1342     prstatus->pr_pgrp = task_pgrp_vnr(p);
1343     prstatus->pr_sid = task_session_vnr(p);
1344     if (thread_group_leader(p)) {
1345         struct task_cputime cputime;
1346 
1347         /*
1348          * This is the record for the group leader.  It shows the
1349          * group-wide total, not its individual thread total.
1350          */
1351         thread_group_cputime(p, &cputime);
1352         cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1353         cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1354     } else {
1355         cputime_t utime, stime;
1356 
1357         task_cputime(p, &utime, &stime);
1358         cputime_to_timeval(utime, &prstatus->pr_utime);
1359         cputime_to_timeval(stime, &prstatus->pr_stime);
1360     }
1361     cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1362     cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1363 
1364     prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1365     prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1366 }
1367 
1368 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1369                struct mm_struct *mm)
1370 {
1371     const struct cred *cred;
1372     unsigned int i, len;
1373 
1374     /* first copy the parameters from user space */
1375     memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1376 
1377     len = mm->arg_end - mm->arg_start;
1378     if (len >= ELF_PRARGSZ)
1379         len = ELF_PRARGSZ - 1;
1380     if (copy_from_user(&psinfo->pr_psargs,
1381                    (const char __user *) mm->arg_start, len))
1382         return -EFAULT;
1383     for (i = 0; i < len; i++)
1384         if (psinfo->pr_psargs[i] == 0)
1385             psinfo->pr_psargs[i] = ' ';
1386     psinfo->pr_psargs[len] = 0;
1387 
1388     rcu_read_lock();
1389     psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1390     rcu_read_unlock();
1391     psinfo->pr_pid = task_pid_vnr(p);
1392     psinfo->pr_pgrp = task_pgrp_vnr(p);
1393     psinfo->pr_sid = task_session_vnr(p);
1394 
1395     i = p->state ? ffz(~p->state) + 1 : 0;
1396     psinfo->pr_state = i;
1397     psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1398     psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1399     psinfo->pr_nice = task_nice(p);
1400     psinfo->pr_flag = p->flags;
1401     rcu_read_lock();
1402     cred = __task_cred(p);
1403     SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1404     SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1405     rcu_read_unlock();
1406     strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1407 
1408     return 0;
1409 }
1410 
1411 /* Here is the structure in which status of each thread is captured. */
1412 struct elf_thread_status
1413 {
1414     struct list_head list;
1415     struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1416     elf_fpregset_t fpu;     /* NT_PRFPREG */
1417     struct task_struct *thread;
1418 #ifdef ELF_CORE_COPY_XFPREGS
1419     elf_fpxregset_t xfpu;       /* ELF_CORE_XFPREG_TYPE */
1420 #endif
1421     struct memelfnote notes[3];
1422     int num_notes;
1423 };
1424 
1425 /*
1426  * In order to add the specific thread information for the elf file format,
1427  * we need to keep a linked list of every thread's pr_status and then create
1428  * a single section for them in the final core file.
1429  */
1430 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1431 {
1432     struct task_struct *p = t->thread;
1433     int sz = 0;
1434 
1435     t->num_notes = 0;
1436 
1437     fill_prstatus(&t->prstatus, p, signr);
1438     elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1439 
1440     fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1441           &t->prstatus);
1442     t->num_notes++;
1443     sz += notesize(&t->notes[0]);
1444 
1445     t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1446     if (t->prstatus.pr_fpvalid) {
1447         fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1448               &t->fpu);
1449         t->num_notes++;
1450         sz += notesize(&t->notes[1]);
1451     }
1452 
1453 #ifdef ELF_CORE_COPY_XFPREGS
1454     if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1455         fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1456               sizeof(t->xfpu), &t->xfpu);
1457         t->num_notes++;
1458         sz += notesize(&t->notes[2]);
1459     }
1460 #endif
1461     return sz;
1462 }
1463 
1464 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1465                  elf_addr_t e_shoff, int segs)
1466 {
1467     elf->e_shoff = e_shoff;
1468     elf->e_shentsize = sizeof(*shdr4extnum);
1469     elf->e_shnum = 1;
1470     elf->e_shstrndx = SHN_UNDEF;
1471 
1472     memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1473 
1474     shdr4extnum->sh_type = SHT_NULL;
1475     shdr4extnum->sh_size = elf->e_shnum;
1476     shdr4extnum->sh_link = elf->e_shstrndx;
1477     shdr4extnum->sh_info = segs;
1478 }
1479 
1480 /*
1481  * dump the segments for an MMU process
1482  */
1483 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1484 {
1485     struct vm_area_struct *vma;
1486 
1487     for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1488         unsigned long addr;
1489 
1490         if (!maydump(vma, cprm->mm_flags))
1491             continue;
1492 
1493 #ifdef CONFIG_MMU
1494         for (addr = vma->vm_start; addr < vma->vm_end;
1495                             addr += PAGE_SIZE) {
1496             bool res;
1497             struct page *page = get_dump_page(addr);
1498             if (page) {
1499                 void *kaddr = kmap(page);
1500                 res = dump_emit(cprm, kaddr, PAGE_SIZE);
1501                 kunmap(page);
1502                 put_page(page);
1503             } else {
1504                 res = dump_skip(cprm, PAGE_SIZE);
1505             }
1506             if (!res)
1507                 return false;
1508         }
1509 #else
1510         if (!dump_emit(cprm, (void *) vma->vm_start,
1511                 vma->vm_end - vma->vm_start))
1512             return false;
1513 #endif
1514     }
1515     return true;
1516 }
1517 
1518 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1519 {
1520     struct vm_area_struct *vma;
1521     size_t size = 0;
1522 
1523     for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1524         if (maydump(vma, mm_flags))
1525             size += vma->vm_end - vma->vm_start;
1526     return size;
1527 }
1528 
1529 /*
1530  * Actual dumper
1531  *
1532  * This is a two-pass process; first we find the offsets of the bits,
1533  * and then they are actually written out.  If we run out of core limit
1534  * we just truncate.
1535  */
1536 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1537 {
1538 #define NUM_NOTES   6
1539     int has_dumped = 0;
1540     mm_segment_t fs;
1541     int segs;
1542     int i;
1543     struct vm_area_struct *vma;
1544     struct elfhdr *elf = NULL;
1545     loff_t offset = 0, dataoff;
1546     int numnote;
1547     struct memelfnote *notes = NULL;
1548     struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1549     struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1550     LIST_HEAD(thread_list);
1551     struct list_head *t;
1552     elf_fpregset_t *fpu = NULL;
1553 #ifdef ELF_CORE_COPY_XFPREGS
1554     elf_fpxregset_t *xfpu = NULL;
1555 #endif
1556     int thread_status_size = 0;
1557     elf_addr_t *auxv;
1558     struct elf_phdr *phdr4note = NULL;
1559     struct elf_shdr *shdr4extnum = NULL;
1560     Elf_Half e_phnum;
1561     elf_addr_t e_shoff;
1562     struct core_thread *ct;
1563     struct elf_thread_status *tmp;
1564 
1565     /*
1566      * We no longer stop all VM operations.
1567      *
1568      * This is because those proceses that could possibly change map_count
1569      * or the mmap / vma pages are now blocked in do_exit on current
1570      * finishing this core dump.
1571      *
1572      * Only ptrace can touch these memory addresses, but it doesn't change
1573      * the map_count or the pages allocated. So no possibility of crashing
1574      * exists while dumping the mm->vm_next areas to the core file.
1575      */
1576 
1577     /* alloc memory for large data structures: too large to be on stack */
1578     elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1579     if (!elf)
1580         goto cleanup;
1581     prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1582     if (!prstatus)
1583         goto cleanup;
1584     psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1585     if (!psinfo)
1586         goto cleanup;
1587     notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1588     if (!notes)
1589         goto cleanup;
1590     fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1591     if (!fpu)
1592         goto cleanup;
1593 #ifdef ELF_CORE_COPY_XFPREGS
1594     xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1595     if (!xfpu)
1596         goto cleanup;
1597 #endif
1598 
1599     for (ct = current->mm->core_state->dumper.next;
1600                     ct; ct = ct->next) {
1601         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1602         if (!tmp)
1603             goto cleanup;
1604 
1605         tmp->thread = ct->task;
1606         list_add(&tmp->list, &thread_list);
1607     }
1608 
1609     list_for_each(t, &thread_list) {
1610         struct elf_thread_status *tmp;
1611         int sz;
1612 
1613         tmp = list_entry(t, struct elf_thread_status, list);
1614         sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1615         thread_status_size += sz;
1616     }
1617 
1618     /* now collect the dump for the current */
1619     fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1620     elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1621 
1622     segs = current->mm->map_count;
1623     segs += elf_core_extra_phdrs();
1624 
1625     /* for notes section */
1626     segs++;
1627 
1628     /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1629      * this, kernel supports extended numbering. Have a look at
1630      * include/linux/elf.h for further information. */
1631     e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1632 
1633     /* Set up header */
1634     fill_elf_fdpic_header(elf, e_phnum);
1635 
1636     has_dumped = 1;
1637     /*
1638      * Set up the notes in similar form to SVR4 core dumps made
1639      * with info from their /proc.
1640      */
1641 
1642     fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1643     fill_psinfo(psinfo, current->group_leader, current->mm);
1644     fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1645 
1646     numnote = 2;
1647 
1648     auxv = (elf_addr_t *) current->mm->saved_auxv;
1649 
1650     i = 0;
1651     do
1652         i += 2;
1653     while (auxv[i - 2] != AT_NULL);
1654     fill_note(&notes[numnote++], "CORE", NT_AUXV,
1655           i * sizeof(elf_addr_t), auxv);
1656 
1657     /* Try to dump the FPU. */
1658     if ((prstatus->pr_fpvalid =
1659          elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1660         fill_note(notes + numnote++,
1661               "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1662 #ifdef ELF_CORE_COPY_XFPREGS
1663     if (elf_core_copy_task_xfpregs(current, xfpu))
1664         fill_note(notes + numnote++,
1665               "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1666 #endif
1667 
1668     fs = get_fs();
1669     set_fs(KERNEL_DS);
1670 
1671     offset += sizeof(*elf);             /* Elf header */
1672     offset += segs * sizeof(struct elf_phdr);   /* Program headers */
1673 
1674     /* Write notes phdr entry */
1675     {
1676         int sz = 0;
1677 
1678         for (i = 0; i < numnote; i++)
1679             sz += notesize(notes + i);
1680 
1681         sz += thread_status_size;
1682 
1683         phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1684         if (!phdr4note)
1685             goto end_coredump;
1686 
1687         fill_elf_note_phdr(phdr4note, sz, offset);
1688         offset += sz;
1689     }
1690 
1691     /* Page-align dumped data */
1692     dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1693 
1694     offset += elf_core_vma_data_size(cprm->mm_flags);
1695     offset += elf_core_extra_data_size();
1696     e_shoff = offset;
1697 
1698     if (e_phnum == PN_XNUM) {
1699         shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1700         if (!shdr4extnum)
1701             goto end_coredump;
1702         fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1703     }
1704 
1705     offset = dataoff;
1706 
1707     if (!dump_emit(cprm, elf, sizeof(*elf)))
1708         goto end_coredump;
1709 
1710     if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1711         goto end_coredump;
1712 
1713     /* write program headers for segments dump */
1714     for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1715         struct elf_phdr phdr;
1716         size_t sz;
1717 
1718         sz = vma->vm_end - vma->vm_start;
1719 
1720         phdr.p_type = PT_LOAD;
1721         phdr.p_offset = offset;
1722         phdr.p_vaddr = vma->vm_start;
1723         phdr.p_paddr = 0;
1724         phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1725         phdr.p_memsz = sz;
1726         offset += phdr.p_filesz;
1727         phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1728         if (vma->vm_flags & VM_WRITE)
1729             phdr.p_flags |= PF_W;
1730         if (vma->vm_flags & VM_EXEC)
1731             phdr.p_flags |= PF_X;
1732         phdr.p_align = ELF_EXEC_PAGESIZE;
1733 
1734         if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1735             goto end_coredump;
1736     }
1737 
1738     if (!elf_core_write_extra_phdrs(cprm, offset))
1739         goto end_coredump;
1740 
1741     /* write out the notes section */
1742     for (i = 0; i < numnote; i++)
1743         if (!writenote(notes + i, cprm))
1744             goto end_coredump;
1745 
1746     /* write out the thread status notes section */
1747     list_for_each(t, &thread_list) {
1748         struct elf_thread_status *tmp =
1749                 list_entry(t, struct elf_thread_status, list);
1750 
1751         for (i = 0; i < tmp->num_notes; i++)
1752             if (!writenote(&tmp->notes[i], cprm))
1753                 goto end_coredump;
1754     }
1755 
1756     if (!dump_skip(cprm, dataoff - cprm->pos))
1757         goto end_coredump;
1758 
1759     if (!elf_fdpic_dump_segments(cprm))
1760         goto end_coredump;
1761 
1762     if (!elf_core_write_extra_data(cprm))
1763         goto end_coredump;
1764 
1765     if (e_phnum == PN_XNUM) {
1766         if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1767             goto end_coredump;
1768     }
1769 
1770     if (cprm->file->f_pos != offset) {
1771         /* Sanity check */
1772         printk(KERN_WARNING
1773                "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1774                cprm->file->f_pos, offset);
1775     }
1776 
1777 end_coredump:
1778     set_fs(fs);
1779 
1780 cleanup:
1781     while (!list_empty(&thread_list)) {
1782         struct list_head *tmp = thread_list.next;
1783         list_del(tmp);
1784         kfree(list_entry(tmp, struct elf_thread_status, list));
1785     }
1786     kfree(phdr4note);
1787     kfree(elf);
1788     kfree(prstatus);
1789     kfree(psinfo);
1790     kfree(notes);
1791     kfree(fpu);
1792     kfree(shdr4extnum);
1793 #ifdef ELF_CORE_COPY_XFPREGS
1794     kfree(xfpu);
1795 #endif
1796     return has_dumped;
1797 #undef NUM_NOTES
1798 }
1799 
1800 #endif      /* CONFIG_ELF_CORE */