tcp: do_tcp_sendpages() must try to push data out on oom conditions
[linux-flexiantxendom0-3.2.10.git] / fs / binfmt_elf.c
index 7c1e65d..16f7354 100644 (file)
 #include <linux/random.h>
 #include <linux/elf.h>
 #include <linux/utsname.h>
+#include <linux/coredump.h>
 #include <asm/uaccess.h>
 #include <asm/param.h>
 #include <asm/page.h>
+#include <asm/exec.h>
 
 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
 static int load_elf_library(struct file *);
@@ -44,8 +46,8 @@ static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  * If we don't support core dumping, then supply a NULL so we
  * don't even try.
  */
-#if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
-static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit);
+#ifdef CONFIG_ELF_CORE
+static int elf_core_dump(struct coredump_params *cprm);
 #else
 #define elf_core_dump  NULL
 #endif
@@ -65,12 +67,11 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
 
 static struct linux_binfmt elf_format = {
-               .module         = THIS_MODULE,
-               .load_binary    = load_elf_binary,
-               .load_shlib     = load_elf_library,
-               .core_dump      = elf_core_dump,
-               .min_coredump   = ELF_EXEC_PAGESIZE,
-               .hasvdso        = 1
+       .module         = THIS_MODULE,
+       .load_binary    = load_elf_binary,
+       .load_shlib     = load_elf_library,
+       .core_dump      = elf_core_dump,
+       .min_coredump   = ELF_EXEC_PAGESIZE,
 };
 
 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
@@ -81,9 +82,7 @@ static int set_brk(unsigned long start, unsigned long end)
        end = ELF_PAGEALIGN(end);
        if (end > start) {
                unsigned long addr;
-               down_write(&current->mm->mmap_sem);
-               addr = do_brk(start, end - start);
-               up_write(&current->mm->mmap_sem);
+               addr = vm_brk(start, end - start);
                if (BAD_ADDR(addr))
                        return addr;
        }
@@ -315,8 +314,6 @@ create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
        return 0;
 }
 
-#ifndef elf_map
-
 static unsigned long elf_map(struct file *filep, unsigned long addr,
                struct elf_phdr *eppnt, int prot, int type,
                unsigned long total_size)
@@ -353,8 +350,6 @@ static unsigned long elf_map(struct file *filep, unsigned long addr,
        return(map_addr);
 }
 
-#endif /* !elf_map */
-
 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 {
        int i, first_idx = -1, last_idx = -1;
@@ -420,7 +415,7 @@ static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
                goto out;
 
        retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
-                            (char *)elf_phdata,size);
+                            (char *)elf_phdata, size);
        error = -EIO;
        if (retval != size) {
                if (retval < 0)
@@ -517,9 +512,7 @@ static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
                elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 
                /* Map the last of the bss segment */
-               down_write(&current->mm->mmap_sem);
-               error = do_brk(elf_bss, last_bss - elf_bss);
-               up_write(&current->mm->mmap_sem);
+               error = vm_brk(elf_bss, last_bss - elf_bss);
                if (BAD_ADDR(error))
                        goto out_close;
        }
@@ -574,7 +567,7 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
        unsigned long elf_entry;
        unsigned long interp_load_addr = 0;
        unsigned long start_code, end_code, start_data, end_data;
-       unsigned long reloc_func_desc = 0;
+       unsigned long reloc_func_desc __maybe_unused = 0;
        int executable_stack = EXSTACK_DEFAULT;
        unsigned long def_flags = 0;
        struct {
@@ -600,7 +593,7 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                goto out;
        if (!elf_check_arch(&loc->elf_ex))
                goto out;
-       if (!bprm->file->f_op||!bprm->file->f_op->mmap)
+       if (!bprm->file->f_op || !bprm->file->f_op->mmap)
                goto out;
 
        /* Now read in all of the header information */
@@ -662,27 +655,6 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                        if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
                                goto out_free_interp;
 
-                       /*
-                        * The early SET_PERSONALITY here is so that the lookup
-                        * for the interpreter happens in the namespace of the 
-                        * to-be-execed image.  SET_PERSONALITY can select an
-                        * alternate root.
-                        *
-                        * However, SET_PERSONALITY is NOT allowed to switch
-                        * this task into the new images's memory mapping
-                        * policy - that is, TASK_SIZE must still evaluate to
-                        * that which is appropriate to the execing application.
-                        * This is because exit_mmap() needs to have TASK_SIZE
-                        * evaluate to the size of the old image.
-                        *
-                        * So if (say) a 64-bit application is execing a 32-bit
-                        * application it is the architecture's responsibility
-                        * to defer changing the value of TASK_SIZE until the
-                        * switch really is going to happen - do this in
-                        * flush_thread().      - akpm
-                        */
-                       SET_PERSONALITY(loc->elf_ex);
-
                        interpreter = open_exec(elf_interpreter);
                        retval = PTR_ERR(interpreter);
                        if (IS_ERR(interpreter))
@@ -693,8 +665,7 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                         * mm->dumpable = 0 regardless of the interpreter's
                         * permissions.
                         */
-                       if (file_permission(interpreter, MAY_READ) < 0)
-                               bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
+                       would_dump(bprm, interpreter);
 
                        retval = kernel_read(interpreter, 0, bprm->buf,
                                             BINPRM_BUF_SIZE);
@@ -730,9 +701,6 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                /* Verify the interpreter has a valid arch */
                if (!elf_check_arch(&loc->interp_elf_ex))
                        goto out_free_dentry;
-       } else {
-               /* Executables without an interpreter also need a personality  */
-               SET_PERSONALITY(loc->elf_ex);
        }
 
        /* Flush all traces of the currently running executable */
@@ -741,7 +709,6 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                goto out_free_dentry;
 
        /* OK, This is the point of no return */
-       current->flags &= ~PF_FORKNOEXEC;
        current->mm->def_flags = def_flags;
 
        /* Do this immediately, since STACK_TOP as used in setup_arg_pages
@@ -752,7 +719,8 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 
        if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
                current->flags |= PF_RANDOMIZE;
-       arch_pick_mmap_layout(current->mm);
+
+       setup_new_exec(bprm);
 
        /* Do this so that we can load the interpreter, if need be.  We will
           change some of these later */
@@ -767,7 +735,7 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
        
        current->mm->start_stack = bprm->p;
 
-       /* Now we do a little grungy work by mmaping the ELF image into
+       /* Now we do a little grungy work by mmapping the ELF image into
           the correct location in memory. */
        for(i = 0, elf_ppnt = elf_phdata;
            i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
@@ -783,8 +751,8 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                        /* There was a PT_LOAD segment with p_memsz > p_filesz
                           before this one. Map anonymous pages, if needed,
                           and clear the area.  */
-                       retval = set_brk (elf_bss + load_bias,
-                                         elf_brk + load_bias);
+                       retval = set_brk(elf_bss + load_bias,
+                                        elf_brk + load_bias);
                        if (retval) {
                                send_sig(SIGKILL, current, 0);
                                goto out_free_dentry;
@@ -822,8 +790,17 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                         * default mmap base, as well as whatever program they
                         * might try to exec.  This is because the brk will
                         * follow the loader, and is not movable.  */
-#ifdef CONFIG_X86
-                       load_bias = 0;
+#ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
+                       /* Memory randomization might have been switched off
+                        * in runtime via sysctl.
+                        * If that is the case, retain the original non-zero
+                        * load_bias value in order to establish proper
+                        * non-randomized mappings.
+                        */
+                       if (current->flags & PF_RANDOMIZE)
+                               load_bias = 0;
+                       else
+                               load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 #else
                        load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 #endif
@@ -953,7 +930,6 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 
        install_exec_creds(bprm);
-       current->flags &= ~PF_FORKNOEXEC;
        retval = create_elf_tables(bprm, &loc->elf_ex,
                          load_addr, interp_load_addr);
        if (retval < 0) {
@@ -968,9 +944,13 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
        current->mm->start_stack = bprm->p;
 
 #ifdef arch_randomize_brk
-       if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
+       if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
                current->mm->brk = current->mm->start_brk =
                        arch_randomize_brk(current->mm);
+#ifdef CONFIG_COMPAT_BRK
+               current->brk_randomized = 1;
+#endif
+       }
 #endif
 
        if (current->personality & MMAP_PAGE_ZERO) {
@@ -978,10 +958,8 @@ static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
                   and some applications "depend" upon this behavior.
                   Since we do not have the power to recompile these, we
                   emulate the SVr4 behavior. Sigh. */
-               down_write(&current->mm->mmap_sem);
-               error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
+               error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
                                MAP_FIXED | MAP_PRIVATE, 0);
-               up_write(&current->mm->mmap_sem);
        }
 
 #ifdef ELF_PLAT_INIT
@@ -1066,8 +1044,7 @@ static int load_elf_library(struct file *file)
                eppnt++;
 
        /* Now use mmap to map the library into memory. */
-       down_write(&current->mm->mmap_sem);
-       error = do_mmap(file,
+       error = vm_mmap(file,
                        ELF_PAGESTART(eppnt->p_vaddr),
                        (eppnt->p_filesz +
                         ELF_PAGEOFFSET(eppnt->p_vaddr)),
@@ -1075,7 +1052,6 @@ static int load_elf_library(struct file *file)
                        MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
                        (eppnt->p_offset -
                         ELF_PAGEOFFSET(eppnt->p_vaddr)));
-       up_write(&current->mm->mmap_sem);
        if (error != ELF_PAGESTART(eppnt->p_vaddr))
                goto out_free_ph;
 
@@ -1088,11 +1064,8 @@ static int load_elf_library(struct file *file)
        len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
                            ELF_MIN_ALIGN - 1);
        bss = eppnt->p_memsz + eppnt->p_vaddr;
-       if (bss > len) {
-               down_write(&current->mm->mmap_sem);
-               do_brk(len, bss - len);
-               up_write(&current->mm->mmap_sem);
-       }
+       if (bss > len)
+               vm_brk(len, bss - len);
        error = 0;
 
 out_free_ph:
@@ -1101,47 +1074,35 @@ out:
        return error;
 }
 
-/*
- * Note that some platforms still use traditional core dumps and not
- * the ELF core dump.  Each platform can select it as appropriate.
- */
-#if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
-
+#ifdef CONFIG_ELF_CORE
 /*
  * ELF core dumper
  *
  * Modelled on fs/exec.c:aout_core_dump()
  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
  */
+
 /*
- * These are the only things you should do on a core-file: use only these
- * functions to write out all the necessary info.
+ * The purpose of always_dump_vma() is to make sure that special kernel mappings
+ * that are useful for post-mortem analysis are included in every core dump.
+ * In that way we ensure that the core dump is fully interpretable later
+ * without matching up the same kernel and hardware config to see what PC values
+ * meant. These special mappings include - vDSO, vsyscall, and other
+ * architecture specific mappings
  */
-static int dump_write(struct file *file, const void *addr, int nr)
+static bool always_dump_vma(struct vm_area_struct *vma)
 {
-       return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
-}
+       /* Any vsyscall mappings? */
+       if (vma == get_gate_vma(vma->vm_mm))
+               return true;
+       /*
+        * arch_vma_name() returns non-NULL for special architecture mappings,
+        * such as vDSO sections.
+        */
+       if (arch_vma_name(vma))
+               return true;
 
-static int dump_seek(struct file *file, loff_t off)
-{
-       if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
-               if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
-                       return 0;
-       } else {
-               char *buf = (char *)get_zeroed_page(GFP_KERNEL);
-               if (!buf)
-                       return 0;
-               while (off > 0) {
-                       unsigned long n = off;
-                       if (n > PAGE_SIZE)
-                               n = PAGE_SIZE;
-                       if (!dump_write(file, buf, n))
-                               return 0;
-                       off -= n;
-               }
-               free_page((unsigned long)buf);
-       }
-       return 1;
+       return false;
 }
 
 /*
@@ -1152,10 +1113,13 @@ static unsigned long vma_dump_size(struct vm_area_struct *vma,
 {
 #define FILTER(type)   (mm_flags & (1UL << MMF_DUMP_##type))
 
-       /* The vma can be set up to tell us the answer directly.  */
-       if (vma->vm_flags & VM_ALWAYSDUMP)
+       /* always dump the vdso and vsyscall sections */
+       if (always_dump_vma(vma))
                goto whole;
 
+       if (vma->vm_flags & VM_NODUMP)
+               return 0;
+
        /* Hugetlb memory check */
        if (vma->vm_flags & VM_HUGETLB) {
                if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
@@ -1277,13 +1241,6 @@ static int writenote(struct memelfnote *men, struct file *file,
 }
 #undef DUMP_WRITE
 
-#define DUMP_WRITE(addr, nr)   \
-       if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
-               goto end_coredump;
-#define DUMP_SEEK(off) \
-       if (!dump_seek(file, (off))) \
-               goto end_coredump;
-
 static void fill_elf_header(struct elfhdr *elf, int segs,
                            u16 machine, u32 flags, u8 osabi)
 {
@@ -1447,6 +1404,22 @@ static void do_thread_regset_writeback(struct task_struct *task,
                regset->writeback(task, regset, 1);
 }
 
+#ifndef PR_REG_SIZE
+#define PR_REG_SIZE(S) sizeof(S)
+#endif
+
+#ifndef PRSTATUS_SIZE
+#define PRSTATUS_SIZE(S) sizeof(S)
+#endif
+
+#ifndef PR_REG_PTR
+#define PR_REG_PTR(S) (&((S)->pr_reg))
+#endif
+
+#ifndef SET_PR_FPVALID
+#define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
+#endif
+
 static int fill_thread_core_info(struct elf_thread_core_info *t,
                                 const struct user_regset_view *view,
                                 long signr, size_t *total)
@@ -1461,11 +1434,11 @@ static int fill_thread_core_info(struct elf_thread_core_info *t,
         */
        fill_prstatus(&t->prstatus, t->task, signr);
        (void) view->regsets[0].get(t->task, &view->regsets[0],
-                                   0, sizeof(t->prstatus.pr_reg),
-                                   &t->prstatus.pr_reg, NULL);
+                                   0, PR_REG_SIZE(t->prstatus.pr_reg),
+                                   PR_REG_PTR(&t->prstatus), NULL);
 
        fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
-                 sizeof(t->prstatus), &t->prstatus);
+                 PRSTATUS_SIZE(t->prstatus), &t->prstatus);
        *total += notesize(&t->notes[0]);
 
        do_thread_regset_writeback(t->task, &view->regsets[0]);
@@ -1478,7 +1451,7 @@ static int fill_thread_core_info(struct elf_thread_core_info *t,
        for (i = 1; i < view->n; ++i) {
                const struct user_regset *regset = &view->regsets[i];
                do_thread_regset_writeback(t->task, regset);
-               if (regset->core_note_type &&
+               if (regset->core_note_type && regset->get &&
                    (!regset->active || regset->active(t->task, regset))) {
                        int ret;
                        size_t size = regset->n * regset->size;
@@ -1495,7 +1468,7 @@ static int fill_thread_core_info(struct elf_thread_core_info *t,
                                                  regset->core_note_type,
                                                  size, data);
                                else {
-                                       t->prstatus.pr_fpvalid = 1;
+                                       SET_PR_FPVALID(&t->prstatus, 1);
                                        fill_note(&t->notes[i], "CORE",
                                                  NT_PRFPREG, size, data);
                                }
@@ -1714,42 +1687,52 @@ struct elf_note_info {
        int numnote;
 };
 
-static int fill_note_info(struct elfhdr *elf, int phdrs,
-                         struct elf_note_info *info,
-                         long signr, struct pt_regs *regs)
+static int elf_note_info_init(struct elf_note_info *info)
 {
-#define        NUM_NOTES       6
-       struct list_head *t;
-
-       info->notes = NULL;
-       info->prstatus = NULL;
-       info->psinfo = NULL;
-       info->fpu = NULL;
-#ifdef ELF_CORE_COPY_XFPREGS
-       info->xfpu = NULL;
-#endif
+       memset(info, 0, sizeof(*info));
        INIT_LIST_HEAD(&info->thread_list);
 
-       info->notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote),
-                             GFP_KERNEL);
+       /* Allocate space for six ELF notes */
+       info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
        if (!info->notes)
                return 0;
        info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
        if (!info->psinfo)
-               return 0;
+               goto notes_free;
        info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
        if (!info->prstatus)
-               return 0;
+               goto psinfo_free;
        info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
        if (!info->fpu)
-               return 0;
+               goto prstatus_free;
 #ifdef ELF_CORE_COPY_XFPREGS
        info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
        if (!info->xfpu)
-               return 0;
+               goto fpu_free;
 #endif
+       return 1;
+#ifdef ELF_CORE_COPY_XFPREGS
+ fpu_free:
+       kfree(info->fpu);
+#endif
+ prstatus_free:
+       kfree(info->prstatus);
+ psinfo_free:
+       kfree(info->psinfo);
+ notes_free:
+       kfree(info->notes);
+       return 0;
+}
+
+static int fill_note_info(struct elfhdr *elf, int phdrs,
+                         struct elf_note_info *info,
+                         long signr, struct pt_regs *regs)
+{
+       struct list_head *t;
+
+       if (!elf_note_info_init(info))
+               return 0;
 
-       info->thread_status_size = 0;
        if (signr) {
                struct core_thread *ct;
                struct elf_thread_status *ets;
@@ -1809,8 +1792,6 @@ static int fill_note_info(struct elfhdr *elf, int phdrs,
 #endif
 
        return 1;
-
-#undef NUM_NOTES
 }
 
 static size_t get_note_info_size(struct elf_note_info *info)
@@ -1894,6 +1875,34 @@ static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
        return gate_vma;
 }
 
+static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
+                            elf_addr_t e_shoff, int segs)
+{
+       elf->e_shoff = e_shoff;
+       elf->e_shentsize = sizeof(*shdr4extnum);
+       elf->e_shnum = 1;
+       elf->e_shstrndx = SHN_UNDEF;
+
+       memset(shdr4extnum, 0, sizeof(*shdr4extnum));
+
+       shdr4extnum->sh_type = SHT_NULL;
+       shdr4extnum->sh_size = elf->e_shnum;
+       shdr4extnum->sh_link = elf->e_shstrndx;
+       shdr4extnum->sh_info = segs;
+}
+
+static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
+                                    unsigned long mm_flags)
+{
+       struct vm_area_struct *vma;
+       size_t size = 0;
+
+       for (vma = first_vma(current, gate_vma); vma != NULL;
+            vma = next_vma(vma, gate_vma))
+               size += vma_dump_size(vma, mm_flags);
+       return size;
+}
+
 /*
  * Actual dumper
  *
@@ -1901,7 +1910,7 @@ static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
  * and then they are actually written out.  If we run out of core limit
  * we just truncate.
  */
-static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit)
+static int elf_core_dump(struct coredump_params *cprm)
 {
        int has_dumped = 0;
        mm_segment_t fs;
@@ -1910,8 +1919,11 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
        struct vm_area_struct *vma, *gate_vma;
        struct elfhdr *elf = NULL;
        loff_t offset = 0, dataoff, foffset;
-       unsigned long mm_flags;
        struct elf_note_info info;
+       struct elf_phdr *phdr4note = NULL;
+       struct elf_shdr *shdr4extnum = NULL;
+       Elf_Half e_phnum;
+       elf_addr_t e_shoff;
 
        /*
         * We no longer stop all VM operations.
@@ -1934,20 +1946,25 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
         * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
         */
        segs = current->mm->map_count;
-#ifdef ELF_CORE_EXTRA_PHDRS
-       segs += ELF_CORE_EXTRA_PHDRS;
-#endif
+       segs += elf_core_extra_phdrs();
 
-       gate_vma = get_gate_vma(current);
+       gate_vma = get_gate_vma(current->mm);
        if (gate_vma != NULL)
                segs++;
 
+       /* for notes section */
+       segs++;
+
+       /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
+        * this, kernel supports extended numbering. Have a look at
+        * include/linux/elf.h for further information. */
+       e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
+
        /*
         * Collect all the non-memory information about the process for the
         * notes.  This also sets up the file header.
         */
-       if (!fill_note_info(elf, segs + 1, /* including notes section */
-                           &info, signr, regs))
+       if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
                goto cleanup;
 
        has_dumped = 1;
@@ -1956,31 +1973,47 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
        fs = get_fs();
        set_fs(KERNEL_DS);
 
-       DUMP_WRITE(elf, sizeof(*elf));
        offset += sizeof(*elf);                         /* Elf header */
-       offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
+       offset += segs * sizeof(struct elf_phdr);       /* Program headers */
        foffset = offset;
 
        /* Write notes phdr entry */
        {
-               struct elf_phdr phdr;
                size_t sz = get_note_info_size(&info);
 
                sz += elf_coredump_extra_notes_size();
 
-               fill_elf_note_phdr(&phdr, sz, offset);
+               phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
+               if (!phdr4note)
+                       goto end_coredump;
+
+               fill_elf_note_phdr(phdr4note, sz, offset);
                offset += sz;
-               DUMP_WRITE(&phdr, sizeof(phdr));
        }
 
        dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
 
-       /*
-        * We must use the same mm->flags while dumping core to avoid
-        * inconsistency between the program headers and bodies, otherwise an
-        * unusable core file can be generated.
-        */
-       mm_flags = current->mm->flags;
+       offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
+       offset += elf_core_extra_data_size();
+       e_shoff = offset;
+
+       if (e_phnum == PN_XNUM) {
+               shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
+               if (!shdr4extnum)
+                       goto end_coredump;
+               fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
+       }
+
+       offset = dataoff;
+
+       size += sizeof(*elf);
+       if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
+               goto end_coredump;
+
+       size += sizeof(*phdr4note);
+       if (size > cprm->limit
+           || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
+               goto end_coredump;
 
        /* Write program headers for segments dump */
        for (vma = first_vma(current, gate_vma); vma != NULL;
@@ -1991,7 +2024,7 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
                phdr.p_offset = offset;
                phdr.p_vaddr = vma->vm_start;
                phdr.p_paddr = 0;
-               phdr.p_filesz = vma_dump_size(vma, mm_flags);
+               phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
                phdr.p_memsz = vma->vm_end - vma->vm_start;
                offset += phdr.p_filesz;
                phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
@@ -2001,81 +2034,81 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
                        phdr.p_flags |= PF_X;
                phdr.p_align = ELF_EXEC_PAGESIZE;
 
-               DUMP_WRITE(&phdr, sizeof(phdr));
+               size += sizeof(phdr);
+               if (size > cprm->limit
+                   || !dump_write(cprm->file, &phdr, sizeof(phdr)))
+                       goto end_coredump;
        }
 
-#ifdef ELF_CORE_WRITE_EXTRA_PHDRS
-       ELF_CORE_WRITE_EXTRA_PHDRS;
-#endif
+       if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
+               goto end_coredump;
 
        /* write out the notes section */
-       if (!write_note_info(&info, file, &foffset))
+       if (!write_note_info(&info, cprm->file, &foffset))
                goto end_coredump;
 
-       if (elf_coredump_extra_notes_write(file, &foffset))
+       if (elf_coredump_extra_notes_write(cprm->file, &foffset))
                goto end_coredump;
 
        /* Align to page */
-       DUMP_SEEK(dataoff - foffset);
+       if (!dump_seek(cprm->file, dataoff - foffset))
+               goto end_coredump;
 
        for (vma = first_vma(current, gate_vma); vma != NULL;
                        vma = next_vma(vma, gate_vma)) {
                unsigned long addr;
                unsigned long end;
 
-               end = vma->vm_start + vma_dump_size(vma, mm_flags);
+               end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
 
                for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
                        struct page *page;
-                       struct vm_area_struct *tmp_vma;
-
-                       if (get_user_pages(current, current->mm, addr, 1, 0, 1,
-                                               &page, &tmp_vma) <= 0) {
-                               DUMP_SEEK(PAGE_SIZE);
-                       } else {
-                               if (page == ZERO_PAGE(0)) {
-                                       if (!dump_seek(file, PAGE_SIZE)) {
-                                               page_cache_release(page);
-                                               goto end_coredump;
-                                       }
-                               } else {
-                                       void *kaddr;
-                                       flush_cache_page(tmp_vma, addr,
-                                                        page_to_pfn(page));
-                                       kaddr = kmap(page);
-                                       if ((size += PAGE_SIZE) > limit ||
-                                           !dump_write(file, kaddr,
-                                           PAGE_SIZE)) {
-                                               kunmap(page);
-                                               page_cache_release(page);
-                                               goto end_coredump;
-                                       }
-                                       kunmap(page);
-                               }
+                       int stop;
+
+                       page = get_dump_page(addr);
+                       if (page) {
+                               void *kaddr = kmap(page);
+                               stop = ((size += PAGE_SIZE) > cprm->limit) ||
+                                       !dump_write(cprm->file, kaddr,
+                                                   PAGE_SIZE);
+                               kunmap(page);
                                page_cache_release(page);
-                       }
+                       } else
+                               stop = !dump_seek(cprm->file, PAGE_SIZE);
+                       if (stop)
+                               goto end_coredump;
                }
        }
 
-#ifdef ELF_CORE_WRITE_EXTRA_DATA
-       ELF_CORE_WRITE_EXTRA_DATA;
-#endif
+       if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
+               goto end_coredump;
+
+       if (e_phnum == PN_XNUM) {
+               size += sizeof(*shdr4extnum);
+               if (size > cprm->limit
+                   || !dump_write(cprm->file, shdr4extnum,
+                                  sizeof(*shdr4extnum)))
+                       goto end_coredump;
+       }
 
 end_coredump:
        set_fs(fs);
 
 cleanup:
        free_note_info(&info);
+       kfree(shdr4extnum);
+       kfree(phdr4note);
        kfree(elf);
 out:
        return has_dumped;
 }
 
-#endif         /* USE_ELF_CORE_DUMP */
+#endif         /* CONFIG_ELF_CORE */
 
 static int __init init_elf_binfmt(void)
 {
-       return register_binfmt(&elf_format);
+       register_binfmt(&elf_format);
+       return 0;
 }
 
 static void __exit exit_elf_binfmt(void)