x86 vDSO: ia32 vdso32-syscall build
[deliverable/linux.git] / arch / x86 / vdso / vdso32-setup.c
CommitLineData
1da177e4 1/*
1da177e4 2 * (C) Copyright 2002 Linus Torvalds
e6e5494c
IM
3 * Portions based on the vdso-randomization code from exec-shield:
4 * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar
1da177e4
LT
5 *
6 * This file contains the needed initializations to support sysenter.
7 */
8
9#include <linux/init.h>
10#include <linux/smp.h>
11#include <linux/thread_info.h>
12#include <linux/sched.h>
13#include <linux/gfp.h>
14#include <linux/string.h>
15#include <linux/elf.h>
e6e5494c 16#include <linux/mm.h>
4e950f6f 17#include <linux/err.h>
e6e5494c 18#include <linux/module.h>
1da177e4
LT
19
20#include <asm/cpufeature.h>
21#include <asm/msr.h>
22#include <asm/pgtable.h>
23#include <asm/unistd.h>
d4f7a2c1 24#include <asm/elf.h>
1dbf527c 25#include <asm/tlbflush.h>
6c3652ef 26#include <asm/vdso.h>
1dbf527c
JF
27
28enum {
29 VDSO_DISABLED = 0,
30 VDSO_ENABLED = 1,
31 VDSO_COMPAT = 2,
32};
33
34#ifdef CONFIG_COMPAT_VDSO
35#define VDSO_DEFAULT VDSO_COMPAT
36#else
37#define VDSO_DEFAULT VDSO_ENABLED
38#endif
1da177e4 39
e6e5494c
IM
40/*
41 * Should the kernel map a VDSO page into processes and pass its
42 * address down to glibc upon exec()?
43 */
1dbf527c 44unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
e6e5494c
IM
45
46EXPORT_SYMBOL_GPL(vdso_enabled);
47
48static int __init vdso_setup(char *s)
49{
50 vdso_enabled = simple_strtoul(s, NULL, 0);
51
52 return 1;
53}
54
55__setup("vdso=", vdso_setup);
56
0aa97fb2 57extern asmlinkage void ia32_sysenter_target(void);
1da177e4 58
d4f7a2c1
JF
59static __init void reloc_symtab(Elf32_Ehdr *ehdr,
60 unsigned offset, unsigned size)
61{
62 Elf32_Sym *sym = (void *)ehdr + offset;
63 unsigned nsym = size / sizeof(*sym);
64 unsigned i;
65
66 for(i = 0; i < nsym; i++, sym++) {
67 if (sym->st_shndx == SHN_UNDEF ||
68 sym->st_shndx == SHN_ABS)
69 continue; /* skip */
70
71 if (sym->st_shndx > SHN_LORESERVE) {
72 printk(KERN_INFO "VDSO: unexpected st_shndx %x\n",
73 sym->st_shndx);
74 continue;
75 }
76
77 switch(ELF_ST_TYPE(sym->st_info)) {
78 case STT_OBJECT:
79 case STT_FUNC:
80 case STT_SECTION:
81 case STT_FILE:
82 sym->st_value += VDSO_HIGH_BASE;
83 }
84 }
85}
86
87static __init void reloc_dyn(Elf32_Ehdr *ehdr, unsigned offset)
88{
89 Elf32_Dyn *dyn = (void *)ehdr + offset;
90
91 for(; dyn->d_tag != DT_NULL; dyn++)
92 switch(dyn->d_tag) {
93 case DT_PLTGOT:
94 case DT_HASH:
95 case DT_STRTAB:
96 case DT_SYMTAB:
97 case DT_RELA:
98 case DT_INIT:
99 case DT_FINI:
100 case DT_REL:
101 case DT_DEBUG:
102 case DT_JMPREL:
103 case DT_VERSYM:
104 case DT_VERDEF:
105 case DT_VERNEED:
106 case DT_ADDRRNGLO ... DT_ADDRRNGHI:
107 /* definitely pointers needing relocation */
108 dyn->d_un.d_ptr += VDSO_HIGH_BASE;
109 break;
110
111 case DT_ENCODING ... OLD_DT_LOOS-1:
112 case DT_LOOS ... DT_HIOS-1:
113 /* Tags above DT_ENCODING are pointers if
114 they're even */
115 if (dyn->d_tag >= DT_ENCODING &&
116 (dyn->d_tag & 1) == 0)
117 dyn->d_un.d_ptr += VDSO_HIGH_BASE;
118 break;
119
120 case DT_VERDEFNUM:
121 case DT_VERNEEDNUM:
122 case DT_FLAGS_1:
123 case DT_RELACOUNT:
124 case DT_RELCOUNT:
125 case DT_VALRNGLO ... DT_VALRNGHI:
126 /* definitely not pointers */
127 break;
128
129 case OLD_DT_LOOS ... DT_LOOS-1:
130 case DT_HIOS ... DT_VALRNGLO-1:
131 default:
132 if (dyn->d_tag > DT_ENCODING)
133 printk(KERN_INFO "VDSO: unexpected DT_tag %x\n",
134 dyn->d_tag);
135 break;
136 }
137}
138
139static __init void relocate_vdso(Elf32_Ehdr *ehdr)
140{
141 Elf32_Phdr *phdr;
142 Elf32_Shdr *shdr;
143 int i;
144
145 BUG_ON(memcmp(ehdr->e_ident, ELFMAG, 4) != 0 ||
146 !elf_check_arch(ehdr) ||
147 ehdr->e_type != ET_DYN);
148
149 ehdr->e_entry += VDSO_HIGH_BASE;
150
151 /* rebase phdrs */
152 phdr = (void *)ehdr + ehdr->e_phoff;
153 for (i = 0; i < ehdr->e_phnum; i++) {
154 phdr[i].p_vaddr += VDSO_HIGH_BASE;
155
156 /* relocate dynamic stuff */
157 if (phdr[i].p_type == PT_DYNAMIC)
158 reloc_dyn(ehdr, phdr[i].p_offset);
159 }
160
161 /* rebase sections */
162 shdr = (void *)ehdr + ehdr->e_shoff;
163 for(i = 0; i < ehdr->e_shnum; i++) {
164 if (!(shdr[i].sh_flags & SHF_ALLOC))
165 continue;
166
167 shdr[i].sh_addr += VDSO_HIGH_BASE;
168
169 if (shdr[i].sh_type == SHT_SYMTAB ||
170 shdr[i].sh_type == SHT_DYNSYM)
171 reloc_symtab(ehdr, shdr[i].sh_offset,
172 shdr[i].sh_size);
173 }
174}
d4f7a2c1 175
6fe940d6 176void enable_sep_cpu(void)
1da177e4
LT
177{
178 int cpu = get_cpu();
179 struct tss_struct *tss = &per_cpu(init_tss, cpu);
180
6fe940d6
LS
181 if (!boot_cpu_has(X86_FEATURE_SEP)) {
182 put_cpu();
183 return;
184 }
185
a75c54f9
RR
186 tss->x86_tss.ss1 = __KERNEL_CS;
187 tss->x86_tss.esp1 = sizeof(struct tss_struct) + (unsigned long) tss;
1da177e4 188 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
a75c54f9 189 wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.esp1, 0);
0aa97fb2 190 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0);
1da177e4
LT
191 put_cpu();
192}
193
1dbf527c
JF
194static struct vm_area_struct gate_vma;
195
196static int __init gate_vma_init(void)
197{
198 gate_vma.vm_mm = NULL;
199 gate_vma.vm_start = FIXADDR_USER_START;
200 gate_vma.vm_end = FIXADDR_USER_END;
201 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
202 gate_vma.vm_page_prot = __P101;
203 /*
204 * Make sure the vDSO gets into every core dump.
205 * Dumping its contents makes post-mortem fully interpretable later
206 * without matching up the same kernel and hardware config to see
207 * what PC values meant.
208 */
209 gate_vma.vm_flags |= VM_ALWAYSDUMP;
210 return 0;
211}
212
1da177e4
LT
213/*
214 * These symbols are defined by vsyscall.o to mark the bounds
215 * of the ELF DSO images included therein.
216 */
217extern const char vsyscall_int80_start, vsyscall_int80_end;
218extern const char vsyscall_sysenter_start, vsyscall_sysenter_end;
7d91d531 219static struct page *syscall_pages[1];
1da177e4 220
1dbf527c
JF
221static void map_compat_vdso(int map)
222{
223 static int vdso_mapped;
224
225 if (map == vdso_mapped)
226 return;
227
228 vdso_mapped = map;
229
230 __set_fixmap(FIX_VDSO, page_to_pfn(syscall_pages[0]) << PAGE_SHIFT,
231 map ? PAGE_READONLY_EXEC : PAGE_NONE);
232
233 /* flush stray tlbs */
234 flush_tlb_all();
235}
236
a6c4e076 237int __init sysenter_setup(void)
1da177e4 238{
7d91d531 239 void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
d4f7a2c1
JF
240 const void *vsyscall;
241 size_t vsyscall_len;
242
7d91d531 243 syscall_pages[0] = virt_to_page(syscall_page);
1da177e4 244
1dbf527c
JF
245 gate_vma_init();
246
e6e5494c 247 printk("Compat vDSO mapped to %08lx.\n", __fix_to_virt(FIX_VDSO));
1da177e4
LT
248
249 if (!boot_cpu_has(X86_FEATURE_SEP)) {
d4f7a2c1
JF
250 vsyscall = &vsyscall_int80_start;
251 vsyscall_len = &vsyscall_int80_end - &vsyscall_int80_start;
252 } else {
253 vsyscall = &vsyscall_sysenter_start;
254 vsyscall_len = &vsyscall_sysenter_end - &vsyscall_sysenter_start;
1da177e4
LT
255 }
256
d4f7a2c1
JF
257 memcpy(syscall_page, vsyscall, vsyscall_len);
258 relocate_vdso(syscall_page);
1da177e4 259
1da177e4
LT
260 return 0;
261}
e6e5494c 262
e6e5494c
IM
263/* Setup a VMA at program startup for the vsyscall page */
264int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
265{
e6e5494c
IM
266 struct mm_struct *mm = current->mm;
267 unsigned long addr;
752783c0 268 int ret = 0;
1dbf527c 269 bool compat;
e6e5494c
IM
270
271 down_write(&mm->mmap_sem);
e6e5494c 272
1dbf527c
JF
273 /* Test compat mode once here, in case someone
274 changes it via sysctl */
275 compat = (vdso_enabled == VDSO_COMPAT);
276
277 map_compat_vdso(compat);
278
279 if (compat)
280 addr = VDSO_HIGH_BASE;
281 else {
282 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
283 if (IS_ERR_VALUE(addr)) {
284 ret = addr;
285 goto up_fail;
286 }
287
288 /*
289 * MAYWRITE to allow gdb to COW and set breakpoints
290 *
291 * Make sure the vDSO gets into every core dump.
292 * Dumping its contents makes post-mortem fully
293 * interpretable later without matching up the same
294 * kernel and hardware config to see what PC values
295 * meant.
296 */
297 ret = install_special_mapping(mm, addr, PAGE_SIZE,
298 VM_READ|VM_EXEC|
299 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
300 VM_ALWAYSDUMP,
301 syscall_pages);
302
303 if (ret)
304 goto up_fail;
305 }
e6e5494c
IM
306
307 current->mm->context.vdso = (void *)addr;
308 current_thread_info()->sysenter_return =
6c3652ef 309 VDSO32_SYMBOL(addr, SYSENTER_RETURN);
1dbf527c
JF
310
311 up_fail:
e6e5494c 312 up_write(&mm->mmap_sem);
1dbf527c 313
e6e5494c 314 return ret;
e6e5494c
IM
315}
316
317const char *arch_vma_name(struct vm_area_struct *vma)
318{
319 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
320 return "[vdso]";
321 return NULL;
322}
323
324struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
325{
1dbf527c
JF
326 struct mm_struct *mm = tsk->mm;
327
328 /* Check to see if this task was created in compat vdso mode */
329 if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
330 return &gate_vma;
e6e5494c
IM
331 return NULL;
332}
333
334int in_gate_area(struct task_struct *task, unsigned long addr)
335{
2f30c007
JB
336 const struct vm_area_struct *vma = get_gate_vma(task);
337
338 return vma && addr >= vma->vm_start && addr < vma->vm_end;
e6e5494c
IM
339}
340
341int in_gate_area_no_task(unsigned long addr)
342{
343 return 0;
344}
This page took 0.31219 seconds and 5 git commands to generate.