vdso.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2
  3/*
  4 *    Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
  5 *			 <benh@kernel.crashing.org>
  6 */
  7
  8#include <linux/errno.h>
  9#include <linux/sched.h>
 10#include <linux/kernel.h>
 11#include <linux/mm.h>
 12#include <linux/smp.h>
 13#include <linux/stddef.h>
 14#include <linux/unistd.h>
 15#include <linux/slab.h>
 16#include <linux/user.h>
 17#include <linux/elf.h>
 18#include <linux/security.h>
 19#include <linux/syscalls.h>
 20#include <linux/vdso_datastore.h>
 21#include <vdso/datapage.h>
 22
 23#include <asm/syscall.h>
 24#include <asm/syscalls.h>
 25#include <asm/processor.h>
 26#include <asm/mmu.h>
 27#include <asm/mmu_context.h>
 28#include <asm/machdep.h>
 29#include <asm/cputable.h>
 30#include <asm/sections.h>
 31#include <asm/firmware.h>
 32#include <asm/vdso.h>
 33#include <asm/vdso_datapage.h>
 34#include <asm/setup.h>
 35
 36static_assert(__VDSO_PAGES == VDSO_NR_PAGES);
 37
 38/* The alignment of the vDSO */
 39#define VDSO_ALIGNMENT	(1 << 16)
 40
 41extern char vdso32_start, vdso32_end;
 42extern char vdso64_start, vdso64_end;
 43
 44static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma,
 45		       unsigned long text_size)
 46{
 47	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
 48
 49	if (new_size != text_size)
 50		return -EINVAL;
 51
 52	current->mm->context.vdso = (void __user *)new_vma->vm_start;
 53
 54	return 0;
 55}
 56
 57static int vdso32_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
 58{
 59	return vdso_mremap(sm, new_vma, &vdso32_end - &vdso32_start);
 60}
 61
 62static int vdso64_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
 63{
 64	return vdso_mremap(sm, new_vma, &vdso64_end - &vdso64_start);
 65}
 66
 67static void vdso_close(const struct vm_special_mapping *sm, struct vm_area_struct *vma)
 68{
 69	struct mm_struct *mm = vma->vm_mm;
 70
 71	/*
 72	 * close() is called for munmap() but also for mremap(). In the mremap()
 73	 * case the vdso pointer has already been updated by the mremap() hook
 74	 * above, so it must not be set to NULL here.
 75	 */
 76	if (vma->vm_start != (unsigned long)mm->context.vdso)
 77		return;
 78
 79	mm->context.vdso = NULL;
 80}
 81
 82static struct vm_special_mapping vdso32_spec __ro_after_init = {
 83	.name = "[vdso]",
 84	.mremap = vdso32_mremap,
 85	.close = vdso_close,
 86};
 87
 88static struct vm_special_mapping vdso64_spec __ro_after_init = {
 89	.name = "[vdso]",
 90	.mremap = vdso64_mremap,
 91	.close = vdso_close,
 92};
 93
 94/*
 95 * This is called from binfmt_elf, we create the special vma for the
 96 * vDSO and insert it into the mm struct tree
 97 */
 98static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 99{
100	unsigned long vdso_size, vdso_base, mappings_size;
101	struct vm_special_mapping *vdso_spec;
102	unsigned long vvar_size = VDSO_NR_PAGES * PAGE_SIZE;
103	struct mm_struct *mm = current->mm;
104	struct vm_area_struct *vma;
105
106	if (is_32bit_task()) {
107		vdso_spec = &vdso32_spec;
108		vdso_size = &vdso32_end - &vdso32_start;
109	} else {
110		vdso_spec = &vdso64_spec;
111		vdso_size = &vdso64_end - &vdso64_start;
112	}
113
114	mappings_size = vdso_size + vvar_size;
115	mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK;
116
117	/*
118	 * Pick a base address for the vDSO in process space.
119	 * Add enough to the size so that the result can be aligned.
120	 */
121	vdso_base = get_unmapped_area(NULL, 0, mappings_size, 0, 0);
122	if (IS_ERR_VALUE(vdso_base))
123		return vdso_base;
124
125	/* Add required alignment. */
126	vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
127
128	vma = vdso_install_vvar_mapping(mm, vdso_base);
129	if (IS_ERR(vma))
130		return PTR_ERR(vma);
131
132	/*
133	 * our vma flags don't have VM_WRITE so by default, the process isn't
134	 * allowed to write those pages.
135	 * gdb can break that with ptrace interface, and thus trigger COW on
136	 * those pages but it's then your responsibility to never do that on
137	 * the "data" page of the vDSO or you'll stop getting kernel updates
138	 * and your nice userland gettimeofday will be totally dead.
139	 * It's fine to use that for setting breakpoints in the vDSO code
140	 * pages though.
141	 */
142	vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size,
143				       VM_READ | VM_EXEC | VM_MAYREAD |
144				       VM_MAYWRITE | VM_MAYEXEC, vdso_spec);
145	if (IS_ERR(vma)) {
146		do_munmap(mm, vdso_base, vvar_size, NULL);
147		return PTR_ERR(vma);
148	}
149
150	// Now that the mappings are in place, set the mm VDSO pointer
151	mm->context.vdso = (void __user *)vdso_base + vvar_size;
152
153	return 0;
154}
155
156int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
157{
158	struct mm_struct *mm = current->mm;
159	int rc;
160
161	mm->context.vdso = NULL;
162
163	if (mmap_write_lock_killable(mm))
164		return -EINTR;
165
166	rc = __arch_setup_additional_pages(bprm, uses_interp);
167
168	mmap_write_unlock(mm);
169	return rc;
170}
171
172#define VDSO_DO_FIXUPS(type, value, bits, sec) do {					\
173	void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start);	\
174	void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end);	\
175											\
176	do_##type##_fixups((value), __start, __end);					\
177} while (0)
178
179static void __init vdso_fixup_features(void)
180{
181#ifdef CONFIG_PPC64
182	VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup);
183	VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup);
184	VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup);
185	VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup);
186#endif /* CONFIG_PPC64 */
187
188#ifdef CONFIG_VDSO32
189	VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup);
190	VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup);
191#ifdef CONFIG_PPC64
192	VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup);
193#endif /* CONFIG_PPC64 */
194	VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup);
195#endif
196}
197
198/*
199 * Called from setup_arch to initialize the bitmap of available
200 * syscalls in the systemcfg page
201 */
202static void __init vdso_setup_syscall_map(void)
203{
204	unsigned int i;
205
206	for (i = 0; i < NR_syscalls; i++) {
207		if (sys_call_table[i] != (void *)&sys_ni_syscall)
208			vdso_k_arch_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
209		if (IS_ENABLED(CONFIG_COMPAT) &&
210		    compat_sys_call_table[i] != (void *)&sys_ni_syscall)
211			vdso_k_arch_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
212	}
213}
214
215#ifdef CONFIG_PPC64
216int vdso_getcpu_init(void)
217{
218	unsigned long cpu, node, val;
219
220	/*
221	 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
222	 * in the next 16 bits.  The VDSO uses this to implement getcpu().
223	 */
224	cpu = get_cpu();
225	WARN_ON_ONCE(cpu > 0xffff);
226
227	node = cpu_to_node(cpu);
228	WARN_ON_ONCE(node > 0xffff);
229
230	val = (cpu & 0xffff) | ((node & 0xffff) << 16);
231	mtspr(SPRN_SPRG_VDSO_WRITE, val);
232	get_paca()->sprg_vdso = val;
233
234	put_cpu();
235
236	return 0;
237}
238/* We need to call this before SMP init */
239early_initcall(vdso_getcpu_init);
240#endif
241
242static struct page ** __init vdso_setup_pages(void *start, void *end)
243{
244	int i;
245	struct page **pagelist;
246	int pages = (end - start) >> PAGE_SHIFT;
247
248	pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
249	if (!pagelist)
250		panic("%s: Cannot allocate page list for VDSO", __func__);
251
252	for (i = 0; i < pages; i++)
253		pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
254
255	return pagelist;
256}
257
258static int __init vdso_init(void)
259{
260#ifdef CONFIG_PPC64
261	vdso_k_arch_data->dcache_block_size = ppc64_caches.l1d.block_size;
262	vdso_k_arch_data->icache_block_size = ppc64_caches.l1i.block_size;
263	vdso_k_arch_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
264	vdso_k_arch_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
265#endif /* CONFIG_PPC64 */
266
267	vdso_setup_syscall_map();
268
269	vdso_fixup_features();
270
271	if (IS_ENABLED(CONFIG_VDSO32))
272		vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end);
273
274	if (IS_ENABLED(CONFIG_PPC64))
275		vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end);
276
277	smp_wmb();
278
279	return 0;
280}
281arch_initcall(vdso_init);