Commit | Line | Data |
---|---|---|
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> |
af65d648 | 27 | #include <asm/proto.h> |
1dbf527c JF |
28 | |
29 | enum { | |
30 | VDSO_DISABLED = 0, | |
31 | VDSO_ENABLED = 1, | |
32 | VDSO_COMPAT = 2, | |
33 | }; | |
34 | ||
35 | #ifdef CONFIG_COMPAT_VDSO | |
36 | #define VDSO_DEFAULT VDSO_COMPAT | |
37 | #else | |
38 | #define VDSO_DEFAULT VDSO_ENABLED | |
39 | #endif | |
1da177e4 | 40 | |
af65d648 RM |
41 | #ifdef CONFIG_X86_64 |
42 | #define vdso_enabled sysctl_vsyscall32 | |
43 | #define arch_setup_additional_pages syscall32_setup_pages | |
44 | #endif | |
45 | ||
46 | /* | |
47 | * This is the difference between the prelinked addresses in the vDSO images | |
48 | * and the VDSO_HIGH_BASE address where CONFIG_COMPAT_VDSO places the vDSO | |
49 | * in the user address space. | |
50 | */ | |
51 | #define VDSO_ADDR_ADJUST (VDSO_HIGH_BASE - (unsigned long)VDSO32_PRELINK) | |
52 | ||
e6e5494c IM |
53 | /* |
54 | * Should the kernel map a VDSO page into processes and pass its | |
55 | * address down to glibc upon exec()? | |
56 | */ | |
1dbf527c | 57 | unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT; |
e6e5494c | 58 | |
e6e5494c IM |
59 | static int __init vdso_setup(char *s) |
60 | { | |
61 | vdso_enabled = simple_strtoul(s, NULL, 0); | |
62 | ||
63 | return 1; | |
64 | } | |
65 | ||
af65d648 RM |
66 | /* |
67 | * For consistency, the argument vdso32=[012] affects the 32-bit vDSO | |
68 | * behavior on both 64-bit and 32-bit kernels. | |
69 | * On 32-bit kernels, vdso=[012] means the same thing. | |
70 | */ | |
71 | __setup("vdso32=", vdso_setup); | |
e6e5494c | 72 | |
af65d648 RM |
73 | #ifdef CONFIG_X86_32 |
74 | __setup_param("vdso=", vdso32_setup, vdso_setup, 0); | |
75 | ||
76 | EXPORT_SYMBOL_GPL(vdso_enabled); | |
77 | #endif | |
1da177e4 | 78 | |
d4f7a2c1 JF |
79 | static __init void reloc_symtab(Elf32_Ehdr *ehdr, |
80 | unsigned offset, unsigned size) | |
81 | { | |
82 | Elf32_Sym *sym = (void *)ehdr + offset; | |
83 | unsigned nsym = size / sizeof(*sym); | |
84 | unsigned i; | |
85 | ||
86 | for(i = 0; i < nsym; i++, sym++) { | |
87 | if (sym->st_shndx == SHN_UNDEF || | |
88 | sym->st_shndx == SHN_ABS) | |
89 | continue; /* skip */ | |
90 | ||
91 | if (sym->st_shndx > SHN_LORESERVE) { | |
92 | printk(KERN_INFO "VDSO: unexpected st_shndx %x\n", | |
93 | sym->st_shndx); | |
94 | continue; | |
95 | } | |
96 | ||
97 | switch(ELF_ST_TYPE(sym->st_info)) { | |
98 | case STT_OBJECT: | |
99 | case STT_FUNC: | |
100 | case STT_SECTION: | |
101 | case STT_FILE: | |
af65d648 | 102 | sym->st_value += VDSO_ADDR_ADJUST; |
d4f7a2c1 JF |
103 | } |
104 | } | |
105 | } | |
106 | ||
107 | static __init void reloc_dyn(Elf32_Ehdr *ehdr, unsigned offset) | |
108 | { | |
109 | Elf32_Dyn *dyn = (void *)ehdr + offset; | |
110 | ||
111 | for(; dyn->d_tag != DT_NULL; dyn++) | |
112 | switch(dyn->d_tag) { | |
113 | case DT_PLTGOT: | |
114 | case DT_HASH: | |
115 | case DT_STRTAB: | |
116 | case DT_SYMTAB: | |
117 | case DT_RELA: | |
118 | case DT_INIT: | |
119 | case DT_FINI: | |
120 | case DT_REL: | |
121 | case DT_DEBUG: | |
122 | case DT_JMPREL: | |
123 | case DT_VERSYM: | |
124 | case DT_VERDEF: | |
125 | case DT_VERNEED: | |
126 | case DT_ADDRRNGLO ... DT_ADDRRNGHI: | |
127 | /* definitely pointers needing relocation */ | |
af65d648 | 128 | dyn->d_un.d_ptr += VDSO_ADDR_ADJUST; |
d4f7a2c1 JF |
129 | break; |
130 | ||
131 | case DT_ENCODING ... OLD_DT_LOOS-1: | |
132 | case DT_LOOS ... DT_HIOS-1: | |
133 | /* Tags above DT_ENCODING are pointers if | |
134 | they're even */ | |
135 | if (dyn->d_tag >= DT_ENCODING && | |
136 | (dyn->d_tag & 1) == 0) | |
af65d648 | 137 | dyn->d_un.d_ptr += VDSO_ADDR_ADJUST; |
d4f7a2c1 JF |
138 | break; |
139 | ||
140 | case DT_VERDEFNUM: | |
141 | case DT_VERNEEDNUM: | |
142 | case DT_FLAGS_1: | |
143 | case DT_RELACOUNT: | |
144 | case DT_RELCOUNT: | |
145 | case DT_VALRNGLO ... DT_VALRNGHI: | |
146 | /* definitely not pointers */ | |
147 | break; | |
148 | ||
149 | case OLD_DT_LOOS ... DT_LOOS-1: | |
150 | case DT_HIOS ... DT_VALRNGLO-1: | |
151 | default: | |
152 | if (dyn->d_tag > DT_ENCODING) | |
153 | printk(KERN_INFO "VDSO: unexpected DT_tag %x\n", | |
154 | dyn->d_tag); | |
155 | break; | |
156 | } | |
157 | } | |
158 | ||
159 | static __init void relocate_vdso(Elf32_Ehdr *ehdr) | |
160 | { | |
161 | Elf32_Phdr *phdr; | |
162 | Elf32_Shdr *shdr; | |
163 | int i; | |
164 | ||
ecb783ea | 165 | BUG_ON(memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0 || |
af65d648 | 166 | !elf_check_arch_ia32(ehdr) || |
d4f7a2c1 JF |
167 | ehdr->e_type != ET_DYN); |
168 | ||
af65d648 | 169 | ehdr->e_entry += VDSO_ADDR_ADJUST; |
d4f7a2c1 JF |
170 | |
171 | /* rebase phdrs */ | |
172 | phdr = (void *)ehdr + ehdr->e_phoff; | |
173 | for (i = 0; i < ehdr->e_phnum; i++) { | |
af65d648 | 174 | phdr[i].p_vaddr += VDSO_ADDR_ADJUST; |
d4f7a2c1 JF |
175 | |
176 | /* relocate dynamic stuff */ | |
177 | if (phdr[i].p_type == PT_DYNAMIC) | |
178 | reloc_dyn(ehdr, phdr[i].p_offset); | |
179 | } | |
180 | ||
181 | /* rebase sections */ | |
182 | shdr = (void *)ehdr + ehdr->e_shoff; | |
183 | for(i = 0; i < ehdr->e_shnum; i++) { | |
184 | if (!(shdr[i].sh_flags & SHF_ALLOC)) | |
185 | continue; | |
186 | ||
af65d648 | 187 | shdr[i].sh_addr += VDSO_ADDR_ADJUST; |
d4f7a2c1 JF |
188 | |
189 | if (shdr[i].sh_type == SHT_SYMTAB || | |
190 | shdr[i].sh_type == SHT_DYNSYM) | |
191 | reloc_symtab(ehdr, shdr[i].sh_offset, | |
192 | shdr[i].sh_size); | |
193 | } | |
194 | } | |
d4f7a2c1 | 195 | |
af65d648 RM |
196 | /* |
197 | * These symbols are defined by vdso32.S to mark the bounds | |
198 | * of the ELF DSO images included therein. | |
199 | */ | |
200 | extern const char vdso32_default_start, vdso32_default_end; | |
201 | extern const char vdso32_sysenter_start, vdso32_sysenter_end; | |
202 | static struct page *vdso32_pages[1]; | |
203 | ||
204 | #ifdef CONFIG_X86_64 | |
205 | ||
b6ad92d4 | 206 | #define vdso32_sysenter() (boot_cpu_has(X86_FEATURE_SYSENTER32)) |
af65d648 RM |
207 | |
208 | /* May not be __init: called during resume */ | |
209 | void syscall32_cpu_init(void) | |
210 | { | |
af65d648 RM |
211 | /* Load these always in case some future AMD CPU supports |
212 | SYSENTER from compat mode too. */ | |
213 | checking_wrmsrl(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS); | |
214 | checking_wrmsrl(MSR_IA32_SYSENTER_ESP, 0ULL); | |
215 | checking_wrmsrl(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target); | |
216 | ||
217 | wrmsrl(MSR_CSTAR, ia32_cstar_target); | |
218 | } | |
219 | ||
220 | #define compat_uses_vma 1 | |
221 | ||
222 | static inline void map_compat_vdso(int map) | |
223 | { | |
224 | } | |
225 | ||
226 | #else /* CONFIG_X86_32 */ | |
227 | ||
228 | #define vdso32_sysenter() (boot_cpu_has(X86_FEATURE_SEP)) | |
229 | ||
6fe940d6 | 230 | void enable_sep_cpu(void) |
1da177e4 LT |
231 | { |
232 | int cpu = get_cpu(); | |
233 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
234 | ||
6fe940d6 LS |
235 | if (!boot_cpu_has(X86_FEATURE_SEP)) { |
236 | put_cpu(); | |
237 | return; | |
238 | } | |
239 | ||
a75c54f9 | 240 | tss->x86_tss.ss1 = __KERNEL_CS; |
faca6227 | 241 | tss->x86_tss.sp1 = sizeof(struct tss_struct) + (unsigned long) tss; |
1da177e4 | 242 | wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); |
faca6227 | 243 | wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.sp1, 0); |
0aa97fb2 | 244 | wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0); |
1da177e4 LT |
245 | put_cpu(); |
246 | } | |
247 | ||
1dbf527c JF |
248 | static struct vm_area_struct gate_vma; |
249 | ||
250 | static int __init gate_vma_init(void) | |
251 | { | |
252 | gate_vma.vm_mm = NULL; | |
253 | gate_vma.vm_start = FIXADDR_USER_START; | |
254 | gate_vma.vm_end = FIXADDR_USER_END; | |
255 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; | |
256 | gate_vma.vm_page_prot = __P101; | |
257 | /* | |
258 | * Make sure the vDSO gets into every core dump. | |
259 | * Dumping its contents makes post-mortem fully interpretable later | |
260 | * without matching up the same kernel and hardware config to see | |
261 | * what PC values meant. | |
262 | */ | |
263 | gate_vma.vm_flags |= VM_ALWAYSDUMP; | |
264 | return 0; | |
265 | } | |
266 | ||
af65d648 | 267 | #define compat_uses_vma 0 |
1da177e4 | 268 | |
1dbf527c JF |
269 | static void map_compat_vdso(int map) |
270 | { | |
271 | static int vdso_mapped; | |
272 | ||
273 | if (map == vdso_mapped) | |
274 | return; | |
275 | ||
276 | vdso_mapped = map; | |
277 | ||
af65d648 | 278 | __set_fixmap(FIX_VDSO, page_to_pfn(vdso32_pages[0]) << PAGE_SHIFT, |
1dbf527c JF |
279 | map ? PAGE_READONLY_EXEC : PAGE_NONE); |
280 | ||
281 | /* flush stray tlbs */ | |
282 | flush_tlb_all(); | |
283 | } | |
284 | ||
af65d648 RM |
285 | #endif /* CONFIG_X86_64 */ |
286 | ||
a6c4e076 | 287 | int __init sysenter_setup(void) |
1da177e4 | 288 | { |
7d91d531 | 289 | void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC); |
d4f7a2c1 JF |
290 | const void *vsyscall; |
291 | size_t vsyscall_len; | |
292 | ||
af65d648 | 293 | vdso32_pages[0] = virt_to_page(syscall_page); |
1da177e4 | 294 | |
af65d648 | 295 | #ifdef CONFIG_X86_32 |
1dbf527c | 296 | gate_vma_init(); |
af65d648 | 297 | #endif |
1da177e4 | 298 | |
af65d648 RM |
299 | if (!vdso32_sysenter()) { |
300 | vsyscall = &vdso32_default_start; | |
301 | vsyscall_len = &vdso32_default_end - &vdso32_default_start; | |
d4f7a2c1 | 302 | } else { |
af65d648 RM |
303 | vsyscall = &vdso32_sysenter_start; |
304 | vsyscall_len = &vdso32_sysenter_end - &vdso32_sysenter_start; | |
1da177e4 LT |
305 | } |
306 | ||
d4f7a2c1 JF |
307 | memcpy(syscall_page, vsyscall, vsyscall_len); |
308 | relocate_vdso(syscall_page); | |
1da177e4 | 309 | |
1da177e4 LT |
310 | return 0; |
311 | } | |
e6e5494c | 312 | |
e6e5494c IM |
313 | /* Setup a VMA at program startup for the vsyscall page */ |
314 | int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack) | |
315 | { | |
e6e5494c IM |
316 | struct mm_struct *mm = current->mm; |
317 | unsigned long addr; | |
752783c0 | 318 | int ret = 0; |
1dbf527c | 319 | bool compat; |
e6e5494c | 320 | |
5de253cc RM |
321 | if (vdso_enabled == VDSO_DISABLED) |
322 | return 0; | |
323 | ||
e6e5494c | 324 | down_write(&mm->mmap_sem); |
e6e5494c | 325 | |
1dbf527c JF |
326 | /* Test compat mode once here, in case someone |
327 | changes it via sysctl */ | |
328 | compat = (vdso_enabled == VDSO_COMPAT); | |
329 | ||
330 | map_compat_vdso(compat); | |
331 | ||
332 | if (compat) | |
333 | addr = VDSO_HIGH_BASE; | |
334 | else { | |
335 | addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0); | |
336 | if (IS_ERR_VALUE(addr)) { | |
337 | ret = addr; | |
338 | goto up_fail; | |
339 | } | |
af65d648 | 340 | } |
1dbf527c | 341 | |
af65d648 | 342 | if (compat_uses_vma || !compat) { |
1dbf527c JF |
343 | /* |
344 | * MAYWRITE to allow gdb to COW and set breakpoints | |
345 | * | |
346 | * Make sure the vDSO gets into every core dump. | |
347 | * Dumping its contents makes post-mortem fully | |
348 | * interpretable later without matching up the same | |
349 | * kernel and hardware config to see what PC values | |
350 | * meant. | |
351 | */ | |
352 | ret = install_special_mapping(mm, addr, PAGE_SIZE, | |
353 | VM_READ|VM_EXEC| | |
354 | VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC| | |
355 | VM_ALWAYSDUMP, | |
af65d648 | 356 | vdso32_pages); |
1dbf527c JF |
357 | |
358 | if (ret) | |
359 | goto up_fail; | |
360 | } | |
e6e5494c IM |
361 | |
362 | current->mm->context.vdso = (void *)addr; | |
363 | current_thread_info()->sysenter_return = | |
6c3652ef | 364 | VDSO32_SYMBOL(addr, SYSENTER_RETURN); |
1dbf527c JF |
365 | |
366 | up_fail: | |
e6e5494c | 367 | up_write(&mm->mmap_sem); |
1dbf527c | 368 | |
e6e5494c | 369 | return ret; |
e6e5494c IM |
370 | } |
371 | ||
af65d648 RM |
372 | #ifdef CONFIG_X86_64 |
373 | ||
374 | __initcall(sysenter_setup); | |
375 | ||
a97f52e6 RM |
376 | #ifdef CONFIG_SYSCTL |
377 | /* Register vsyscall32 into the ABI table */ | |
378 | #include <linux/sysctl.h> | |
379 | ||
380 | static ctl_table abi_table2[] = { | |
381 | { | |
382 | .procname = "vsyscall32", | |
383 | .data = &sysctl_vsyscall32, | |
384 | .maxlen = sizeof(int), | |
385 | .mode = 0644, | |
386 | .proc_handler = proc_dointvec | |
387 | }, | |
388 | {} | |
389 | }; | |
390 | ||
391 | static ctl_table abi_root_table2[] = { | |
392 | { | |
393 | .ctl_name = CTL_ABI, | |
394 | .procname = "abi", | |
395 | .mode = 0555, | |
396 | .child = abi_table2 | |
397 | }, | |
398 | {} | |
399 | }; | |
400 | ||
401 | static __init int ia32_binfmt_init(void) | |
402 | { | |
403 | register_sysctl_table(abi_root_table2); | |
404 | return 0; | |
405 | } | |
406 | __initcall(ia32_binfmt_init); | |
407 | #endif | |
408 | ||
af65d648 RM |
409 | #else /* CONFIG_X86_32 */ |
410 | ||
e6e5494c IM |
411 | const char *arch_vma_name(struct vm_area_struct *vma) |
412 | { | |
413 | if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) | |
414 | return "[vdso]"; | |
415 | return NULL; | |
416 | } | |
417 | ||
418 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) | |
419 | { | |
1dbf527c JF |
420 | struct mm_struct *mm = tsk->mm; |
421 | ||
422 | /* Check to see if this task was created in compat vdso mode */ | |
423 | if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE) | |
424 | return &gate_vma; | |
e6e5494c IM |
425 | return NULL; |
426 | } | |
427 | ||
428 | int in_gate_area(struct task_struct *task, unsigned long addr) | |
429 | { | |
2f30c007 JB |
430 | const struct vm_area_struct *vma = get_gate_vma(task); |
431 | ||
432 | return vma && addr >= vma->vm_start && addr < vma->vm_end; | |
e6e5494c IM |
433 | } |
434 | ||
435 | int in_gate_area_no_task(unsigned long addr) | |
436 | { | |
437 | return 0; | |
438 | } | |
af65d648 RM |
439 | |
440 | #endif /* CONFIG_X86_64 */ |