1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2007 Andi Kleen, SUSE Labs.
5 * This contains most of the x86 vDSO kernel-side code.
9 #include <linux/sched.h>
10 #include <linux/sched/task_stack.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/random.h>
14 #include <linux/elf.h>
15 #include <linux/cpu.h>
16 #include <linux/ptrace.h>
17 #include <linux/time_namespace.h>
19 #include <asm/pvclock.h>
20 #include <asm/vgtod.h>
21 #include <asm/proto.h>
27 #include <asm/cpufeature.h>
28 #include <clocksource/hyperv_timer.h>
30 #undef _ASM_X86_VVAR_H
31 #define EMIT_VVAR(name, offset) \
32 const size_t name ## _offset = offset;
35 struct vdso_data *arch_get_vdso_data(void *vvar_page)
37 return (struct vdso_data *)(vvar_page + _vdso_data_offset);
41 #if defined(CONFIG_X86_64)
42 unsigned int __read_mostly vdso64_enabled = 1;
45 void __init init_vdso_image(const struct vdso_image *image)
47 BUG_ON(image->size % PAGE_SIZE != 0);
49 apply_alternatives((struct alt_instr *)(image->data + image->alt),
50 (struct alt_instr *)(image->data + image->alt +
54 static const struct vm_special_mapping vvar_mapping;
57 static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
58 struct vm_area_struct *vma, struct vm_fault *vmf)
60 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
62 if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
63 return VM_FAULT_SIGBUS;
65 vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
70 static void vdso_fix_landing(const struct vdso_image *image,
71 struct vm_area_struct *new_vma)
73 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
74 if (in_ia32_syscall() && image == &vdso_image_32) {
75 struct pt_regs *regs = current_pt_regs();
76 unsigned long vdso_land = image->sym_int80_landing_pad;
77 unsigned long old_land_addr = vdso_land +
78 (unsigned long)current->mm->context.vdso;
80 /* Fixing userspace landing - look at do_fast_syscall_32 */
81 if (regs->ip == old_land_addr)
82 regs->ip = new_vma->vm_start + vdso_land;
87 static int vdso_mremap(const struct vm_special_mapping *sm,
88 struct vm_area_struct *new_vma)
90 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
91 const struct vdso_image *image = current->mm->context.vdso_image;
93 if (image->size != new_size)
96 vdso_fix_landing(image, new_vma);
97 current->mm->context.vdso = (void __user *)new_vma->vm_start;
102 static int vvar_mremap(const struct vm_special_mapping *sm,
103 struct vm_area_struct *new_vma)
105 const struct vdso_image *image = new_vma->vm_mm->context.vdso_image;
106 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
108 if (new_size != -image->sym_vvar_start)
114 #ifdef CONFIG_TIME_NS
115 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
117 if (likely(vma->vm_mm == current->mm))
118 return current->nsproxy->time_ns->vvar_page;
121 * VM_PFNMAP | VM_IO protect .fault() handler from being called
122 * through interfaces like /proc/$pid/mem or
123 * process_vm_{readv,writev}() as long as there's no .access()
124 * in special_mapping_vmops().
125 * For more details check_vma_flags() and __access_remote_vm()
128 WARN(1, "vvar_page accessed remotely");
134 * The vvar page layout depends on whether a task belongs to the root or
135 * non-root time namespace. Whenever a task changes its namespace, the VVAR
136 * page tables are cleared and then they will re-faulted with a
137 * corresponding layout.
138 * See also the comment near timens_setup_vdso_data() for details.
140 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
142 struct mm_struct *mm = task->mm;
143 struct vm_area_struct *vma;
145 if (down_write_killable(&mm->mmap_sem))
148 for (vma = mm->mmap; vma; vma = vma->vm_next) {
149 unsigned long size = vma->vm_end - vma->vm_start;
151 if (vma_is_special_mapping(vma, &vvar_mapping))
152 zap_page_range(vma, vma->vm_start, size);
155 up_write(&mm->mmap_sem);
159 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
165 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
166 struct vm_area_struct *vma, struct vm_fault *vmf)
168 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
173 return VM_FAULT_SIGBUS;
175 sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
176 image->sym_vvar_start;
179 * Sanity check: a symbol offset of zero means that the page
180 * does not exist for this vdso image, not that the page is at
181 * offset zero relative to the text mapping. This should be
182 * impossible here, because sym_offset should only be zero for
183 * the page past the end of the vvar mapping.
186 return VM_FAULT_SIGBUS;
188 if (sym_offset == image->sym_vvar_page) {
189 struct page *timens_page = find_timens_vvar_page(vma);
191 pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
194 * If a task belongs to a time namespace then a namespace
195 * specific VVAR is mapped with the sym_vvar_page offset and
196 * the real VVAR page is mapped with the sym_timens_page
198 * See also the comment near timens_setup_vdso_data().
205 * Optimization: inside time namespace pre-fault
206 * VVAR page too. As on timens page there are only
207 * offsets for clocks on VVAR, it'll be faulted
208 * shortly by VDSO code.
210 addr = vmf->address + (image->sym_timens_page - sym_offset);
211 err = vmf_insert_pfn(vma, addr, pfn);
212 if (unlikely(err & VM_FAULT_ERROR))
215 pfn = page_to_pfn(timens_page);
218 return vmf_insert_pfn(vma, vmf->address, pfn);
219 } else if (sym_offset == image->sym_pvclock_page) {
220 struct pvclock_vsyscall_time_info *pvti =
221 pvclock_get_pvti_cpu0_va();
222 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
223 return vmf_insert_pfn_prot(vma, vmf->address,
224 __pa(pvti) >> PAGE_SHIFT,
225 pgprot_decrypted(vma->vm_page_prot));
227 } else if (sym_offset == image->sym_hvclock_page) {
228 struct ms_hyperv_tsc_page *tsc_pg = hv_get_tsc_page();
230 if (tsc_pg && vclock_was_used(VCLOCK_HVCLOCK))
231 return vmf_insert_pfn(vma, vmf->address,
232 virt_to_phys(tsc_pg) >> PAGE_SHIFT);
233 } else if (sym_offset == image->sym_timens_page) {
234 struct page *timens_page = find_timens_vvar_page(vma);
237 return VM_FAULT_SIGBUS;
239 pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
240 return vmf_insert_pfn(vma, vmf->address, pfn);
243 return VM_FAULT_SIGBUS;
246 static const struct vm_special_mapping vdso_mapping = {
249 .mremap = vdso_mremap,
251 static const struct vm_special_mapping vvar_mapping = {
254 .mremap = vvar_mremap,
258 * Add vdso and vvar mappings to current process.
259 * @image - blob to map
260 * @addr - request a specific address (zero to map at free addr)
262 static int map_vdso(const struct vdso_image *image, unsigned long addr)
264 struct mm_struct *mm = current->mm;
265 struct vm_area_struct *vma;
266 unsigned long text_start;
269 if (down_write_killable(&mm->mmap_sem))
272 addr = get_unmapped_area(NULL, addr,
273 image->size - image->sym_vvar_start, 0, 0);
274 if (IS_ERR_VALUE(addr)) {
279 text_start = addr - image->sym_vvar_start;
282 * MAYWRITE to allow gdb to COW and set breakpoints
284 vma = _install_special_mapping(mm,
288 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
296 vma = _install_special_mapping(mm,
298 -image->sym_vvar_start,
299 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
305 do_munmap(mm, text_start, image->size, NULL);
307 current->mm->context.vdso = (void __user *)text_start;
308 current->mm->context.vdso_image = image;
312 up_write(&mm->mmap_sem);
318 * Put the vdso above the (randomized) stack with another randomized
319 * offset. This way there is no hole in the middle of address space.
320 * To save memory make sure it is still in the same PTE as the stack
321 * top. This doesn't give that many random bits.
323 * Note that this algorithm is imperfect: the distribution of the vdso
324 * start address within a PMD is biased toward the end.
326 * Only used for the 64-bit and x32 vdsos.
328 static unsigned long vdso_addr(unsigned long start, unsigned len)
330 unsigned long addr, end;
334 * Round up the start address. It can start out unaligned as a result
335 * of stack start randomization.
337 start = PAGE_ALIGN(start);
339 /* Round the lowest possible end address up to a PMD boundary. */
340 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
341 if (end >= TASK_SIZE_MAX)
346 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
347 addr = start + (offset << PAGE_SHIFT);
353 * Forcibly align the final address in case we have a hardware
354 * issue that requires alignment for performance reasons.
356 addr = align_vdso_addr(addr);
361 static int map_vdso_randomized(const struct vdso_image *image)
363 unsigned long addr = vdso_addr(current->mm->start_stack, image->size-image->sym_vvar_start);
365 return map_vdso(image, addr);
369 int map_vdso_once(const struct vdso_image *image, unsigned long addr)
371 struct mm_struct *mm = current->mm;
372 struct vm_area_struct *vma;
374 down_write(&mm->mmap_sem);
376 * Check if we have already mapped vdso blob - fail to prevent
377 * abusing from userspace install_speciall_mapping, which may
378 * not do accounting and rlimit right.
379 * We could search vma near context.vdso, but it's a slowpath,
380 * so let's explicitly check all VMAs to be completely sure.
382 for (vma = mm->mmap; vma; vma = vma->vm_next) {
383 if (vma_is_special_mapping(vma, &vdso_mapping) ||
384 vma_is_special_mapping(vma, &vvar_mapping)) {
385 up_write(&mm->mmap_sem);
389 up_write(&mm->mmap_sem);
391 return map_vdso(image, addr);
394 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
395 static int load_vdso32(void)
397 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
400 return map_vdso(&vdso_image_32, 0);
405 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
410 return map_vdso_randomized(&vdso_image_64);
414 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
417 #ifdef CONFIG_X86_X32_ABI
418 if (test_thread_flag(TIF_X32)) {
421 return map_vdso_randomized(&vdso_image_x32);
424 #ifdef CONFIG_IA32_EMULATION
425 return load_vdso32();
432 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
434 return load_vdso32();
439 static __init int vdso_setup(char *s)
441 vdso64_enabled = simple_strtoul(s, NULL, 0);
444 __setup("vdso=", vdso_setup);
446 static int __init init_vdso(void)
448 init_vdso_image(&vdso_image_64);
450 #ifdef CONFIG_X86_X32_ABI
451 init_vdso_image(&vdso_image_x32);
456 subsys_initcall(init_vdso);
457 #endif /* CONFIG_X86_64 */