parisc: add flexible mmap memory layout support

Add support for the flexible mmap memory layout (as described in
http://lwn.net/Articles/91829). This is especially very interesting on
parisc since we currently only support 32bit userspace (even with a
64bit Linux kernel).

Signed-off-by: Helge Deller <deller@gmx.de>

Changed files
+243 -53
arch
+4
arch/parisc/include/asm/elf.h
··· 348 348 349 349 #define ELF_HWCAP 0 350 350 351 + struct mm_struct; 352 + extern unsigned long arch_randomize_brk(struct mm_struct *); 353 + #define arch_randomize_brk arch_randomize_brk 354 + 351 355 #endif
+1
arch/parisc/include/asm/pgtable.h
··· 511 511 /* We provide our own get_unmapped_area to provide cache coherency */ 512 512 513 513 #define HAVE_ARCH_UNMAPPED_AREA 514 + #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 514 515 515 516 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG 516 517 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
+2
arch/parisc/include/asm/processor.h
··· 30 30 #endif 31 31 #define current_text_addr() ({ void *pc; current_ia(pc); pc; }) 32 32 33 + #define HAVE_ARCH_PICK_MMAP_LAYOUT 34 + 33 35 #define TASK_SIZE_OF(tsk) ((tsk)->thread.task_size) 34 36 #define TASK_SIZE TASK_SIZE_OF(current) 35 37 #define TASK_UNMAPPED_BASE (current->thread.map_base)
+10
arch/parisc/include/asm/thread_info.h
··· 76 76 #define _TIF_SYSCALL_TRACE_MASK (_TIF_SYSCALL_TRACE | _TIF_SINGLESTEP | \ 77 77 _TIF_BLOCKSTEP | _TIF_SYSCALL_AUDIT) 78 78 79 + #ifdef CONFIG_64BIT 80 + # ifdef CONFIG_COMPAT 81 + # define is_32bit_task() (test_thread_flag(TIF_32BIT)) 82 + # else 83 + # define is_32bit_task() (0) 84 + # endif 85 + #else 86 + # define is_32bit_task() (1) 87 + #endif 88 + 79 89 #endif /* __KERNEL__ */ 80 90 81 91 #endif /* _ASM_PARISC_THREAD_INFO_H */
+20 -1
arch/parisc/kernel/process.c
··· 13 13 * Copyright (C) 2000 Grant Grundler <grundler with parisc-linux.org> 14 14 * Copyright (C) 2001 Alan Modra <amodra at parisc-linux.org> 15 15 * Copyright (C) 2001-2002 Ryan Bradetich <rbrad at parisc-linux.org> 16 - * Copyright (C) 2001-2007 Helge Deller <deller at parisc-linux.org> 16 + * Copyright (C) 2001-2014 Helge Deller <deller@gmx.de> 17 17 * Copyright (C) 2002 Randolph Chung <tausq with parisc-linux.org> 18 18 * 19 19 * ··· 49 49 #include <linux/kallsyms.h> 50 50 #include <linux/uaccess.h> 51 51 #include <linux/rcupdate.h> 52 + #include <linux/random.h> 52 53 53 54 #include <asm/io.h> 54 55 #include <asm/asm-offsets.h> ··· 287 286 return ptr; 288 287 } 289 288 #endif 289 + 290 + static inline unsigned long brk_rnd(void) 291 + { 292 + /* 8MB for 32bit, 1GB for 64bit */ 293 + if (is_32bit_task()) 294 + return (get_random_int() & 0x7ffUL) << PAGE_SHIFT; 295 + else 296 + return (get_random_int() & 0x3ffffUL) << PAGE_SHIFT; 297 + } 298 + 299 + unsigned long arch_randomize_brk(struct mm_struct *mm) 300 + { 301 + unsigned long ret = PAGE_ALIGN(mm->brk + brk_rnd()); 302 + 303 + if (ret < mm->brk) 304 + return mm->brk; 305 + return ret; 306 + }
+206 -52
arch/parisc/kernel/sys_parisc.c
··· 5 5 * Copyright (C) 1999-2003 Matthew Wilcox <willy at parisc-linux.org> 6 6 * Copyright (C) 2000-2003 Paul Bame <bame at parisc-linux.org> 7 7 * Copyright (C) 2001 Thomas Bogendoerfer <tsbogend at parisc-linux.org> 8 + * Copyright (C) 1999-2014 Helge Deller <deller@gmx.de> 8 9 * 9 10 * 10 11 * This program is free software; you can redistribute it and/or modify ··· 24 23 */ 25 24 26 25 #include <asm/uaccess.h> 26 + #include <asm/elf.h> 27 27 #include <linux/file.h> 28 28 #include <linux/fs.h> 29 29 #include <linux/linkage.h> ··· 34 32 #include <linux/syscalls.h> 35 33 #include <linux/utsname.h> 36 34 #include <linux/personality.h> 35 + #include <linux/random.h> 37 36 38 - static unsigned long get_unshared_area(unsigned long addr, unsigned long len) 37 + /* we construct an artificial offset for the mapping based on the physical 38 + * address of the kernel mapping variable */ 39 + #define GET_LAST_MMAP(filp) \ 40 + (filp ? ((unsigned long) filp->f_mapping) >> 8 : 0UL) 41 + #define SET_LAST_MMAP(filp, val) \ 42 + { /* nothing */ } 43 + 44 + static int get_offset(unsigned int last_mmap) 39 45 { 40 - struct vm_unmapped_area_info info; 46 + return (last_mmap & (SHMLBA-1)) >> PAGE_SHIFT; 47 + } 41 48 42 - info.flags = 0; 43 - info.length = len; 44 - info.low_limit = PAGE_ALIGN(addr); 45 - info.high_limit = TASK_SIZE; 46 - info.align_mask = 0; 47 - info.align_offset = 0; 48 - return vm_unmapped_area(&info); 49 + static unsigned long shared_align_offset(unsigned int last_mmap, 50 + unsigned long pgoff) 51 + { 52 + return (get_offset(last_mmap) + pgoff) << PAGE_SHIFT; 53 + } 54 + 55 + static inline unsigned long COLOR_ALIGN(unsigned long addr, 56 + unsigned int last_mmap, unsigned long pgoff) 57 + { 58 + unsigned long base = (addr+SHMLBA-1) & ~(SHMLBA-1); 59 + unsigned long off = (SHMLBA-1) & 60 + (shared_align_offset(last_mmap, pgoff) << PAGE_SHIFT); 61 + 62 + return base + off; 49 63 } 50 64 51 65 /* 52 - * We need to know the offset to use. Old scheme was to look for 53 - * existing mapping and use the same offset. New scheme is to use the 54 - * address of the kernel data structure as the seed for the offset. 55 - * We'll see how that works... 56 - * 57 - * The mapping is cacheline aligned, so there's no information in the bottom 58 - * few bits of the address. We're looking for 10 bits (4MB / 4k), so let's 59 - * drop the bottom 8 bits and use bits 8-17. 66 + * Top of mmap area (just below the process stack). 60 67 */ 61 - static int get_offset(struct address_space *mapping) 68 + 69 + static unsigned long mmap_upper_limit(void) 62 70 { 63 - return (unsigned long) mapping >> 8; 71 + unsigned long stack_base; 72 + 73 + /* Limit stack size to 1GB - see setup_arg_pages() in fs/exec.c */ 74 + stack_base = rlimit_max(RLIMIT_STACK); 75 + if (stack_base > (1 << 30)) 76 + stack_base = 1 << 30; 77 + 78 + return PAGE_ALIGN(STACK_TOP - stack_base); 64 79 } 65 80 66 - static unsigned long shared_align_offset(struct file *filp, unsigned long pgoff) 67 - { 68 - struct address_space *mapping = filp ? filp->f_mapping : NULL; 69 - 70 - return (get_offset(mapping) + pgoff) << PAGE_SHIFT; 71 - } 72 - 73 - static unsigned long get_shared_area(struct file *filp, unsigned long addr, 74 - unsigned long len, unsigned long pgoff) 75 - { 76 - struct vm_unmapped_area_info info; 77 - 78 - info.flags = 0; 79 - info.length = len; 80 - info.low_limit = PAGE_ALIGN(addr); 81 - info.high_limit = TASK_SIZE; 82 - info.align_mask = PAGE_MASK & (SHMLBA - 1); 83 - info.align_offset = shared_align_offset(filp, pgoff); 84 - return vm_unmapped_area(&info); 85 - } 86 81 87 82 unsigned long arch_get_unmapped_area(struct file *filp, unsigned long addr, 88 83 unsigned long len, unsigned long pgoff, unsigned long flags) 89 84 { 90 - if (len > TASK_SIZE) 91 - return -ENOMEM; 92 - if (flags & MAP_FIXED) { 93 - if ((flags & MAP_SHARED) && 94 - (addr - shared_align_offset(filp, pgoff)) & (SHMLBA - 1)) 95 - return -EINVAL; 96 - return addr; 97 - } 98 - if (!addr) 99 - addr = TASK_UNMAPPED_BASE; 85 + struct mm_struct *mm = current->mm; 86 + struct vm_area_struct *vma; 87 + unsigned long task_size = TASK_SIZE; 88 + int do_color_align, last_mmap; 89 + struct vm_unmapped_area_info info; 100 90 91 + if (len > task_size) 92 + return -ENOMEM; 93 + 94 + do_color_align = 0; 101 95 if (filp || (flags & MAP_SHARED)) 102 - addr = get_shared_area(filp, addr, len, pgoff); 103 - else 104 - addr = get_unshared_area(addr, len); 96 + do_color_align = 1; 97 + last_mmap = GET_LAST_MMAP(filp); 98 + 99 + if (flags & MAP_FIXED) { 100 + if ((flags & MAP_SHARED) && last_mmap && 101 + (addr - shared_align_offset(last_mmap, pgoff)) 102 + & (SHMLBA - 1)) 103 + return -EINVAL; 104 + goto found_addr; 105 + } 106 + 107 + if (addr) { 108 + if (do_color_align && last_mmap) 109 + addr = COLOR_ALIGN(addr, last_mmap, pgoff); 110 + else 111 + addr = PAGE_ALIGN(addr); 112 + 113 + vma = find_vma(mm, addr); 114 + if (task_size - len >= addr && 115 + (!vma || addr + len <= vma->vm_start)) 116 + goto found_addr; 117 + } 118 + 119 + info.flags = 0; 120 + info.length = len; 121 + info.low_limit = mm->mmap_legacy_base; 122 + info.high_limit = mmap_upper_limit(); 123 + info.align_mask = last_mmap ? (PAGE_MASK & (SHMLBA - 1)) : 0; 124 + info.align_offset = shared_align_offset(last_mmap, pgoff); 125 + addr = vm_unmapped_area(&info); 126 + 127 + found_addr: 128 + if (do_color_align && !last_mmap && !(addr & ~PAGE_MASK)) 129 + SET_LAST_MMAP(filp, addr - (pgoff << PAGE_SHIFT)); 105 130 106 131 return addr; 107 132 } 133 + 134 + unsigned long 135 + arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 136 + const unsigned long len, const unsigned long pgoff, 137 + const unsigned long flags) 138 + { 139 + struct vm_area_struct *vma; 140 + struct mm_struct *mm = current->mm; 141 + unsigned long addr = addr0; 142 + int do_color_align, last_mmap; 143 + struct vm_unmapped_area_info info; 144 + 145 + #ifdef CONFIG_64BIT 146 + /* This should only ever run for 32-bit processes. */ 147 + BUG_ON(!test_thread_flag(TIF_32BIT)); 148 + #endif 149 + 150 + /* requested length too big for entire address space */ 151 + if (len > TASK_SIZE) 152 + return -ENOMEM; 153 + 154 + do_color_align = 0; 155 + if (filp || (flags & MAP_SHARED)) 156 + do_color_align = 1; 157 + last_mmap = GET_LAST_MMAP(filp); 158 + 159 + if (flags & MAP_FIXED) { 160 + if ((flags & MAP_SHARED) && last_mmap && 161 + (addr - shared_align_offset(last_mmap, pgoff)) 162 + & (SHMLBA - 1)) 163 + return -EINVAL; 164 + goto found_addr; 165 + } 166 + 167 + /* requesting a specific address */ 168 + if (addr) { 169 + if (do_color_align && last_mmap) 170 + addr = COLOR_ALIGN(addr, last_mmap, pgoff); 171 + else 172 + addr = PAGE_ALIGN(addr); 173 + vma = find_vma(mm, addr); 174 + if (TASK_SIZE - len >= addr && 175 + (!vma || addr + len <= vma->vm_start)) 176 + goto found_addr; 177 + } 178 + 179 + info.flags = VM_UNMAPPED_AREA_TOPDOWN; 180 + info.length = len; 181 + info.low_limit = PAGE_SIZE; 182 + info.high_limit = mm->mmap_base; 183 + info.align_mask = last_mmap ? (PAGE_MASK & (SHMLBA - 1)) : 0; 184 + info.align_offset = shared_align_offset(last_mmap, pgoff); 185 + addr = vm_unmapped_area(&info); 186 + if (!(addr & ~PAGE_MASK)) 187 + goto found_addr; 188 + VM_BUG_ON(addr != -ENOMEM); 189 + 190 + /* 191 + * A failed mmap() very likely causes application failure, 192 + * so fall back to the bottom-up function here. This scenario 193 + * can happen with large stack limits and large mmap() 194 + * allocations. 195 + */ 196 + return arch_get_unmapped_area(filp, addr0, len, pgoff, flags); 197 + 198 + found_addr: 199 + if (do_color_align && !last_mmap && !(addr & ~PAGE_MASK)) 200 + SET_LAST_MMAP(filp, addr - (pgoff << PAGE_SHIFT)); 201 + 202 + return addr; 203 + } 204 + 205 + static int mmap_is_legacy(void) 206 + { 207 + if (current->personality & ADDR_COMPAT_LAYOUT) 208 + return 1; 209 + 210 + /* parisc stack always grows up - so a unlimited stack should 211 + * not be an indicator to use the legacy memory layout. 212 + * if (rlimit(RLIMIT_STACK) == RLIM_INFINITY) 213 + * return 1; 214 + */ 215 + 216 + return sysctl_legacy_va_layout; 217 + } 218 + 219 + static unsigned long mmap_rnd(void) 220 + { 221 + unsigned long rnd = 0; 222 + 223 + /* 224 + * 8 bits of randomness in 32bit mmaps, 20 address space bits 225 + * 28 bits of randomness in 64bit mmaps, 40 address space bits 226 + */ 227 + if (current->flags & PF_RANDOMIZE) { 228 + if (is_32bit_task()) 229 + rnd = get_random_int() % (1<<8); 230 + else 231 + rnd = get_random_int() % (1<<28); 232 + } 233 + return rnd << PAGE_SHIFT; 234 + } 235 + 236 + static unsigned long mmap_legacy_base(void) 237 + { 238 + return TASK_UNMAPPED_BASE + mmap_rnd(); 239 + } 240 + 241 + /* 242 + * This function, called very early during the creation of a new 243 + * process VM image, sets up which VM layout function to use: 244 + */ 245 + void arch_pick_mmap_layout(struct mm_struct *mm) 246 + { 247 + mm->mmap_legacy_base = mmap_legacy_base(); 248 + mm->mmap_base = mmap_upper_limit(); 249 + 250 + if (mmap_is_legacy()) { 251 + mm->mmap_base = mm->mmap_legacy_base; 252 + mm->get_unmapped_area = arch_get_unmapped_area; 253 + } else { 254 + mm->get_unmapped_area = arch_get_unmapped_area_topdown; 255 + } 256 + } 257 + 108 258 109 259 asmlinkage unsigned long sys_mmap2(unsigned long addr, unsigned long len, 110 260 unsigned long prot, unsigned long flags, unsigned long fd,