Loading...
1/*
2 * Copyright 2007 Andi Kleen, SUSE Labs.
3 * Subject to the GPL, v.2
4 *
5 * This contains most of the x86 vDSO kernel-side code.
6 */
7#include <linux/mm.h>
8#include <linux/err.h>
9#include <linux/sched.h>
10#include <linux/slab.h>
11#include <linux/init.h>
12#include <linux/random.h>
13#include <linux/elf.h>
14#include <linux/cpu.h>
15#include <asm/pvclock.h>
16#include <asm/vgtod.h>
17#include <asm/proto.h>
18#include <asm/vdso.h>
19#include <asm/vvar.h>
20#include <asm/page.h>
21#include <asm/hpet.h>
22#include <asm/desc.h>
23#include <asm/cpufeature.h>
24
25#if defined(CONFIG_X86_64)
26unsigned int __read_mostly vdso64_enabled = 1;
27#endif
28
29void __init init_vdso_image(const struct vdso_image *image)
30{
31 BUG_ON(image->size % PAGE_SIZE != 0);
32
33 apply_alternatives((struct alt_instr *)(image->data + image->alt),
34 (struct alt_instr *)(image->data + image->alt +
35 image->alt_len));
36}
37
38struct linux_binprm;
39
40/*
41 * Put the vdso above the (randomized) stack with another randomized
42 * offset. This way there is no hole in the middle of address space.
43 * To save memory make sure it is still in the same PTE as the stack
44 * top. This doesn't give that many random bits.
45 *
46 * Note that this algorithm is imperfect: the distribution of the vdso
47 * start address within a PMD is biased toward the end.
48 *
49 * Only used for the 64-bit and x32 vdsos.
50 */
51static unsigned long vdso_addr(unsigned long start, unsigned len)
52{
53#ifdef CONFIG_X86_32
54 return 0;
55#else
56 unsigned long addr, end;
57 unsigned offset;
58
59 /*
60 * Round up the start address. It can start out unaligned as a result
61 * of stack start randomization.
62 */
63 start = PAGE_ALIGN(start);
64
65 /* Round the lowest possible end address up to a PMD boundary. */
66 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
67 if (end >= TASK_SIZE_MAX)
68 end = TASK_SIZE_MAX;
69 end -= len;
70
71 if (end > start) {
72 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
73 addr = start + (offset << PAGE_SHIFT);
74 } else {
75 addr = start;
76 }
77
78 /*
79 * Forcibly align the final address in case we have a hardware
80 * issue that requires alignment for performance reasons.
81 */
82 addr = align_vdso_addr(addr);
83
84 return addr;
85#endif
86}
87
88static int vdso_fault(const struct vm_special_mapping *sm,
89 struct vm_area_struct *vma, struct vm_fault *vmf)
90{
91 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
92
93 if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
94 return VM_FAULT_SIGBUS;
95
96 vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
97 get_page(vmf->page);
98 return 0;
99}
100
101static const struct vm_special_mapping text_mapping = {
102 .name = "[vdso]",
103 .fault = vdso_fault,
104};
105
106static int vvar_fault(const struct vm_special_mapping *sm,
107 struct vm_area_struct *vma, struct vm_fault *vmf)
108{
109 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
110 long sym_offset;
111 int ret = -EFAULT;
112
113 if (!image)
114 return VM_FAULT_SIGBUS;
115
116 sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
117 image->sym_vvar_start;
118
119 /*
120 * Sanity check: a symbol offset of zero means that the page
121 * does not exist for this vdso image, not that the page is at
122 * offset zero relative to the text mapping. This should be
123 * impossible here, because sym_offset should only be zero for
124 * the page past the end of the vvar mapping.
125 */
126 if (sym_offset == 0)
127 return VM_FAULT_SIGBUS;
128
129 if (sym_offset == image->sym_vvar_page) {
130 ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address,
131 __pa_symbol(&__vvar_page) >> PAGE_SHIFT);
132 } else if (sym_offset == image->sym_hpet_page) {
133#ifdef CONFIG_HPET_TIMER
134 if (hpet_address && vclock_was_used(VCLOCK_HPET)) {
135 ret = vm_insert_pfn_prot(
136 vma,
137 (unsigned long)vmf->virtual_address,
138 hpet_address >> PAGE_SHIFT,
139 pgprot_noncached(PAGE_READONLY));
140 }
141#endif
142 } else if (sym_offset == image->sym_pvclock_page) {
143 struct pvclock_vsyscall_time_info *pvti =
144 pvclock_pvti_cpu0_va();
145 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
146 ret = vm_insert_pfn(
147 vma,
148 (unsigned long)vmf->virtual_address,
149 __pa(pvti) >> PAGE_SHIFT);
150 }
151 }
152
153 if (ret == 0 || ret == -EBUSY)
154 return VM_FAULT_NOPAGE;
155
156 return VM_FAULT_SIGBUS;
157}
158
159static int map_vdso(const struct vdso_image *image, bool calculate_addr)
160{
161 struct mm_struct *mm = current->mm;
162 struct vm_area_struct *vma;
163 unsigned long addr, text_start;
164 int ret = 0;
165 static const struct vm_special_mapping vvar_mapping = {
166 .name = "[vvar]",
167 .fault = vvar_fault,
168 };
169
170 if (calculate_addr) {
171 addr = vdso_addr(current->mm->start_stack,
172 image->size - image->sym_vvar_start);
173 } else {
174 addr = 0;
175 }
176
177 down_write(&mm->mmap_sem);
178
179 addr = get_unmapped_area(NULL, addr,
180 image->size - image->sym_vvar_start, 0, 0);
181 if (IS_ERR_VALUE(addr)) {
182 ret = addr;
183 goto up_fail;
184 }
185
186 text_start = addr - image->sym_vvar_start;
187 current->mm->context.vdso = (void __user *)text_start;
188 current->mm->context.vdso_image = image;
189
190 /*
191 * MAYWRITE to allow gdb to COW and set breakpoints
192 */
193 vma = _install_special_mapping(mm,
194 text_start,
195 image->size,
196 VM_READ|VM_EXEC|
197 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
198 &text_mapping);
199
200 if (IS_ERR(vma)) {
201 ret = PTR_ERR(vma);
202 goto up_fail;
203 }
204
205 vma = _install_special_mapping(mm,
206 addr,
207 -image->sym_vvar_start,
208 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
209 VM_PFNMAP,
210 &vvar_mapping);
211
212 if (IS_ERR(vma)) {
213 ret = PTR_ERR(vma);
214 goto up_fail;
215 }
216
217up_fail:
218 if (ret)
219 current->mm->context.vdso = NULL;
220
221 up_write(&mm->mmap_sem);
222 return ret;
223}
224
225#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
226static int load_vdso32(void)
227{
228 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
229 return 0;
230
231 return map_vdso(&vdso_image_32, false);
232}
233#endif
234
235#ifdef CONFIG_X86_64
236int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
237{
238 if (!vdso64_enabled)
239 return 0;
240
241 return map_vdso(&vdso_image_64, true);
242}
243
244#ifdef CONFIG_COMPAT
245int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
246 int uses_interp)
247{
248#ifdef CONFIG_X86_X32_ABI
249 if (test_thread_flag(TIF_X32)) {
250 if (!vdso64_enabled)
251 return 0;
252
253 return map_vdso(&vdso_image_x32, true);
254 }
255#endif
256#ifdef CONFIG_IA32_EMULATION
257 return load_vdso32();
258#else
259 return 0;
260#endif
261}
262#endif
263#else
264int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
265{
266 return load_vdso32();
267}
268#endif
269
270#ifdef CONFIG_X86_64
271static __init int vdso_setup(char *s)
272{
273 vdso64_enabled = simple_strtoul(s, NULL, 0);
274 return 0;
275}
276__setup("vdso=", vdso_setup);
277#endif
278
279#ifdef CONFIG_X86_64
280static void vgetcpu_cpu_init(void *arg)
281{
282 int cpu = smp_processor_id();
283 struct desc_struct d = { };
284 unsigned long node = 0;
285#ifdef CONFIG_NUMA
286 node = cpu_to_node(cpu);
287#endif
288 if (static_cpu_has(X86_FEATURE_RDTSCP))
289 write_rdtscp_aux((node << 12) | cpu);
290
291 /*
292 * Store cpu number in limit so that it can be loaded
293 * quickly in user space in vgetcpu. (12 bits for the CPU
294 * and 8 bits for the node)
295 */
296 d.limit0 = cpu | ((node & 0xf) << 12);
297 d.limit = node >> 4;
298 d.type = 5; /* RO data, expand down, accessed */
299 d.dpl = 3; /* Visible to user code */
300 d.s = 1; /* Not a system segment */
301 d.p = 1; /* Present */
302 d.d = 1; /* 32-bit */
303
304 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
305}
306
307static int
308vgetcpu_cpu_notifier(struct notifier_block *n, unsigned long action, void *arg)
309{
310 long cpu = (long)arg;
311
312 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
313 smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
314
315 return NOTIFY_DONE;
316}
317
318static int __init init_vdso(void)
319{
320 init_vdso_image(&vdso_image_64);
321
322#ifdef CONFIG_X86_X32_ABI
323 init_vdso_image(&vdso_image_x32);
324#endif
325
326 cpu_notifier_register_begin();
327
328 on_each_cpu(vgetcpu_cpu_init, NULL, 1);
329 /* notifier priority > KVM */
330 __hotcpu_notifier(vgetcpu_cpu_notifier, 30);
331
332 cpu_notifier_register_done();
333
334 return 0;
335}
336subsys_initcall(init_vdso);
337#endif /* CONFIG_X86_64 */
1/*
2 * Copyright 2007 Andi Kleen, SUSE Labs.
3 * Subject to the GPL, v.2
4 *
5 * This contains most of the x86 vDSO kernel-side code.
6 */
7#include <linux/mm.h>
8#include <linux/err.h>
9#include <linux/sched.h>
10#include <linux/slab.h>
11#include <linux/init.h>
12#include <linux/random.h>
13#include <linux/elf.h>
14#include <linux/cpu.h>
15#include <linux/ptrace.h>
16#include <asm/pvclock.h>
17#include <asm/vgtod.h>
18#include <asm/proto.h>
19#include <asm/vdso.h>
20#include <asm/vvar.h>
21#include <asm/page.h>
22#include <asm/desc.h>
23#include <asm/cpufeature.h>
24
25#if defined(CONFIG_X86_64)
26unsigned int __read_mostly vdso64_enabled = 1;
27#endif
28
29void __init init_vdso_image(const struct vdso_image *image)
30{
31 BUG_ON(image->size % PAGE_SIZE != 0);
32
33 apply_alternatives((struct alt_instr *)(image->data + image->alt),
34 (struct alt_instr *)(image->data + image->alt +
35 image->alt_len));
36}
37
38struct linux_binprm;
39
40static int vdso_fault(const struct vm_special_mapping *sm,
41 struct vm_area_struct *vma, struct vm_fault *vmf)
42{
43 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
44
45 if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
46 return VM_FAULT_SIGBUS;
47
48 vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
49 get_page(vmf->page);
50 return 0;
51}
52
53static void vdso_fix_landing(const struct vdso_image *image,
54 struct vm_area_struct *new_vma)
55{
56#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
57 if (in_ia32_syscall() && image == &vdso_image_32) {
58 struct pt_regs *regs = current_pt_regs();
59 unsigned long vdso_land = image->sym_int80_landing_pad;
60 unsigned long old_land_addr = vdso_land +
61 (unsigned long)current->mm->context.vdso;
62
63 /* Fixing userspace landing - look at do_fast_syscall_32 */
64 if (regs->ip == old_land_addr)
65 regs->ip = new_vma->vm_start + vdso_land;
66 }
67#endif
68}
69
70static int vdso_mremap(const struct vm_special_mapping *sm,
71 struct vm_area_struct *new_vma)
72{
73 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
74 const struct vdso_image *image = current->mm->context.vdso_image;
75
76 if (image->size != new_size)
77 return -EINVAL;
78
79 if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
80 return -EFAULT;
81
82 vdso_fix_landing(image, new_vma);
83 current->mm->context.vdso = (void __user *)new_vma->vm_start;
84
85 return 0;
86}
87
88static int vvar_fault(const struct vm_special_mapping *sm,
89 struct vm_area_struct *vma, struct vm_fault *vmf)
90{
91 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
92 long sym_offset;
93 int ret = -EFAULT;
94
95 if (!image)
96 return VM_FAULT_SIGBUS;
97
98 sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
99 image->sym_vvar_start;
100
101 /*
102 * Sanity check: a symbol offset of zero means that the page
103 * does not exist for this vdso image, not that the page is at
104 * offset zero relative to the text mapping. This should be
105 * impossible here, because sym_offset should only be zero for
106 * the page past the end of the vvar mapping.
107 */
108 if (sym_offset == 0)
109 return VM_FAULT_SIGBUS;
110
111 if (sym_offset == image->sym_vvar_page) {
112 ret = vm_insert_pfn(vma, vmf->address,
113 __pa_symbol(&__vvar_page) >> PAGE_SHIFT);
114 } else if (sym_offset == image->sym_pvclock_page) {
115 struct pvclock_vsyscall_time_info *pvti =
116 pvclock_pvti_cpu0_va();
117 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
118 ret = vm_insert_pfn(
119 vma,
120 vmf->address,
121 __pa(pvti) >> PAGE_SHIFT);
122 }
123 }
124
125 if (ret == 0 || ret == -EBUSY)
126 return VM_FAULT_NOPAGE;
127
128 return VM_FAULT_SIGBUS;
129}
130
131static const struct vm_special_mapping vdso_mapping = {
132 .name = "[vdso]",
133 .fault = vdso_fault,
134 .mremap = vdso_mremap,
135};
136static const struct vm_special_mapping vvar_mapping = {
137 .name = "[vvar]",
138 .fault = vvar_fault,
139};
140
141/*
142 * Add vdso and vvar mappings to current process.
143 * @image - blob to map
144 * @addr - request a specific address (zero to map at free addr)
145 */
146static int map_vdso(const struct vdso_image *image, unsigned long addr)
147{
148 struct mm_struct *mm = current->mm;
149 struct vm_area_struct *vma;
150 unsigned long text_start;
151 int ret = 0;
152
153 if (down_write_killable(&mm->mmap_sem))
154 return -EINTR;
155
156 addr = get_unmapped_area(NULL, addr,
157 image->size - image->sym_vvar_start, 0, 0);
158 if (IS_ERR_VALUE(addr)) {
159 ret = addr;
160 goto up_fail;
161 }
162
163 text_start = addr - image->sym_vvar_start;
164
165 /*
166 * MAYWRITE to allow gdb to COW and set breakpoints
167 */
168 vma = _install_special_mapping(mm,
169 text_start,
170 image->size,
171 VM_READ|VM_EXEC|
172 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
173 &vdso_mapping);
174
175 if (IS_ERR(vma)) {
176 ret = PTR_ERR(vma);
177 goto up_fail;
178 }
179
180 vma = _install_special_mapping(mm,
181 addr,
182 -image->sym_vvar_start,
183 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
184 VM_PFNMAP,
185 &vvar_mapping);
186
187 if (IS_ERR(vma)) {
188 ret = PTR_ERR(vma);
189 do_munmap(mm, text_start, image->size);
190 } else {
191 current->mm->context.vdso = (void __user *)text_start;
192 current->mm->context.vdso_image = image;
193 }
194
195up_fail:
196 up_write(&mm->mmap_sem);
197 return ret;
198}
199
200#ifdef CONFIG_X86_64
201/*
202 * Put the vdso above the (randomized) stack with another randomized
203 * offset. This way there is no hole in the middle of address space.
204 * To save memory make sure it is still in the same PTE as the stack
205 * top. This doesn't give that many random bits.
206 *
207 * Note that this algorithm is imperfect: the distribution of the vdso
208 * start address within a PMD is biased toward the end.
209 *
210 * Only used for the 64-bit and x32 vdsos.
211 */
212static unsigned long vdso_addr(unsigned long start, unsigned len)
213{
214 unsigned long addr, end;
215 unsigned offset;
216
217 /*
218 * Round up the start address. It can start out unaligned as a result
219 * of stack start randomization.
220 */
221 start = PAGE_ALIGN(start);
222
223 /* Round the lowest possible end address up to a PMD boundary. */
224 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
225 if (end >= TASK_SIZE_MAX)
226 end = TASK_SIZE_MAX;
227 end -= len;
228
229 if (end > start) {
230 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
231 addr = start + (offset << PAGE_SHIFT);
232 } else {
233 addr = start;
234 }
235
236 /*
237 * Forcibly align the final address in case we have a hardware
238 * issue that requires alignment for performance reasons.
239 */
240 addr = align_vdso_addr(addr);
241
242 return addr;
243}
244
245static int map_vdso_randomized(const struct vdso_image *image)
246{
247 unsigned long addr = vdso_addr(current->mm->start_stack, image->size-image->sym_vvar_start);
248
249 return map_vdso(image, addr);
250}
251#endif
252
253int map_vdso_once(const struct vdso_image *image, unsigned long addr)
254{
255 struct mm_struct *mm = current->mm;
256 struct vm_area_struct *vma;
257
258 down_write(&mm->mmap_sem);
259 /*
260 * Check if we have already mapped vdso blob - fail to prevent
261 * abusing from userspace install_speciall_mapping, which may
262 * not do accounting and rlimit right.
263 * We could search vma near context.vdso, but it's a slowpath,
264 * so let's explicitely check all VMAs to be completely sure.
265 */
266 for (vma = mm->mmap; vma; vma = vma->vm_next) {
267 if (vma_is_special_mapping(vma, &vdso_mapping) ||
268 vma_is_special_mapping(vma, &vvar_mapping)) {
269 up_write(&mm->mmap_sem);
270 return -EEXIST;
271 }
272 }
273 up_write(&mm->mmap_sem);
274
275 return map_vdso(image, addr);
276}
277
278#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
279static int load_vdso32(void)
280{
281 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
282 return 0;
283
284 return map_vdso(&vdso_image_32, 0);
285}
286#endif
287
288#ifdef CONFIG_X86_64
289int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
290{
291 if (!vdso64_enabled)
292 return 0;
293
294 return map_vdso_randomized(&vdso_image_64);
295}
296
297#ifdef CONFIG_COMPAT
298int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
299 int uses_interp)
300{
301#ifdef CONFIG_X86_X32_ABI
302 if (test_thread_flag(TIF_X32)) {
303 if (!vdso64_enabled)
304 return 0;
305 return map_vdso_randomized(&vdso_image_x32);
306 }
307#endif
308#ifdef CONFIG_IA32_EMULATION
309 return load_vdso32();
310#else
311 return 0;
312#endif
313}
314#endif
315#else
316int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
317{
318 return load_vdso32();
319}
320#endif
321
322#ifdef CONFIG_X86_64
323static __init int vdso_setup(char *s)
324{
325 vdso64_enabled = simple_strtoul(s, NULL, 0);
326 return 0;
327}
328__setup("vdso=", vdso_setup);
329#endif
330
331#ifdef CONFIG_X86_64
332static void vgetcpu_cpu_init(void *arg)
333{
334 int cpu = smp_processor_id();
335 struct desc_struct d = { };
336 unsigned long node = 0;
337#ifdef CONFIG_NUMA
338 node = cpu_to_node(cpu);
339#endif
340 if (static_cpu_has(X86_FEATURE_RDTSCP))
341 write_rdtscp_aux((node << 12) | cpu);
342
343 /*
344 * Store cpu number in limit so that it can be loaded
345 * quickly in user space in vgetcpu. (12 bits for the CPU
346 * and 8 bits for the node)
347 */
348 d.limit0 = cpu | ((node & 0xf) << 12);
349 d.limit = node >> 4;
350 d.type = 5; /* RO data, expand down, accessed */
351 d.dpl = 3; /* Visible to user code */
352 d.s = 1; /* Not a system segment */
353 d.p = 1; /* Present */
354 d.d = 1; /* 32-bit */
355
356 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
357}
358
359static int vgetcpu_online(unsigned int cpu)
360{
361 return smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
362}
363
364static int __init init_vdso(void)
365{
366 init_vdso_image(&vdso_image_64);
367
368#ifdef CONFIG_X86_X32_ABI
369 init_vdso_image(&vdso_image_x32);
370#endif
371
372 /* notifier priority > KVM */
373 return cpuhp_setup_state(CPUHP_AP_X86_VDSO_VMA_ONLINE,
374 "x86/vdso/vma:online", vgetcpu_online, NULL);
375}
376subsys_initcall(init_vdso);
377#endif /* CONFIG_X86_64 */