xref: /linux/arch/powerpc/kernel/vdso.c (revision fbf46565c67c626849c7ce2a326972d3008d2a91)
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/memblock.h>
20 #include <linux/syscalls.h>
21 #include <linux/time_namespace.h>
22 #include <vdso/datapage.h>
23 
24 #include <asm/syscall.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 
36 /* The alignment of the vDSO */
37 #define VDSO_ALIGNMENT	(1 << 16)
38 
39 extern char vdso32_start, vdso32_end;
40 extern char vdso64_start, vdso64_end;
41 
42 long sys_ni_syscall(void);
43 
44 /*
45  * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
46  * Once the early boot kernel code no longer needs to muck around
47  * with it, it will become dynamically allocated
48  */
49 static union {
50 	struct vdso_arch_data	data;
51 	u8			page[PAGE_SIZE];
52 } vdso_data_store __page_aligned_data;
53 struct vdso_arch_data *vdso_data = &vdso_data_store.data;
54 
55 enum vvar_pages {
56 	VVAR_DATA_PAGE_OFFSET,
57 	VVAR_TIMENS_PAGE_OFFSET,
58 	VVAR_NR_PAGES,
59 };
60 
61 static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma,
62 		       unsigned long text_size)
63 {
64 	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
65 
66 	if (new_size != text_size)
67 		return -EINVAL;
68 
69 	current->mm->context.vdso = (void __user *)new_vma->vm_start;
70 
71 	return 0;
72 }
73 
74 static int vdso32_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
75 {
76 	return vdso_mremap(sm, new_vma, &vdso32_end - &vdso32_start);
77 }
78 
79 static int vdso64_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
80 {
81 	return vdso_mremap(sm, new_vma, &vdso64_end - &vdso64_start);
82 }
83 
84 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
85 			     struct vm_area_struct *vma, struct vm_fault *vmf);
86 
87 static struct vm_special_mapping vvar_spec __ro_after_init = {
88 	.name = "[vvar]",
89 	.fault = vvar_fault,
90 };
91 
92 static struct vm_special_mapping vdso32_spec __ro_after_init = {
93 	.name = "[vdso]",
94 	.mremap = vdso32_mremap,
95 };
96 
97 static struct vm_special_mapping vdso64_spec __ro_after_init = {
98 	.name = "[vdso]",
99 	.mremap = vdso64_mremap,
100 };
101 
102 #ifdef CONFIG_TIME_NS
103 struct vdso_data *arch_get_vdso_data(void *vvar_page)
104 {
105 	return ((struct vdso_arch_data *)vvar_page)->data;
106 }
107 
108 /*
109  * The vvar mapping contains data for a specific time namespace, so when a task
110  * changes namespace we must unmap its vvar data for the old namespace.
111  * Subsequent faults will map in data for the new namespace.
112  *
113  * For more details see timens_setup_vdso_data().
114  */
115 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
116 {
117 	struct mm_struct *mm = task->mm;
118 	VMA_ITERATOR(vmi, mm, 0);
119 	struct vm_area_struct *vma;
120 
121 	mmap_read_lock(mm);
122 	for_each_vma(vmi, vma) {
123 		unsigned long size = vma->vm_end - vma->vm_start;
124 
125 		if (vma_is_special_mapping(vma, &vvar_spec))
126 			zap_page_range(vma, vma->vm_start, size);
127 	}
128 	mmap_read_unlock(mm);
129 
130 	return 0;
131 }
132 #endif
133 
134 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
135 			     struct vm_area_struct *vma, struct vm_fault *vmf)
136 {
137 	struct page *timens_page = find_timens_vvar_page(vma);
138 	unsigned long pfn;
139 
140 	switch (vmf->pgoff) {
141 	case VVAR_DATA_PAGE_OFFSET:
142 		if (timens_page)
143 			pfn = page_to_pfn(timens_page);
144 		else
145 			pfn = virt_to_pfn(vdso_data);
146 		break;
147 #ifdef CONFIG_TIME_NS
148 	case VVAR_TIMENS_PAGE_OFFSET:
149 		/*
150 		 * If a task belongs to a time namespace then a namespace
151 		 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
152 		 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
153 		 * offset.
154 		 * See also the comment near timens_setup_vdso_data().
155 		 */
156 		if (!timens_page)
157 			return VM_FAULT_SIGBUS;
158 		pfn = virt_to_pfn(vdso_data);
159 		break;
160 #endif /* CONFIG_TIME_NS */
161 	default:
162 		return VM_FAULT_SIGBUS;
163 	}
164 
165 	return vmf_insert_pfn(vma, vmf->address, pfn);
166 }
167 
168 /*
169  * This is called from binfmt_elf, we create the special vma for the
170  * vDSO and insert it into the mm struct tree
171  */
172 static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
173 {
174 	unsigned long vdso_size, vdso_base, mappings_size;
175 	struct vm_special_mapping *vdso_spec;
176 	unsigned long vvar_size = VVAR_NR_PAGES * PAGE_SIZE;
177 	struct mm_struct *mm = current->mm;
178 	struct vm_area_struct *vma;
179 
180 	if (is_32bit_task()) {
181 		vdso_spec = &vdso32_spec;
182 		vdso_size = &vdso32_end - &vdso32_start;
183 	} else {
184 		vdso_spec = &vdso64_spec;
185 		vdso_size = &vdso64_end - &vdso64_start;
186 	}
187 
188 	mappings_size = vdso_size + vvar_size;
189 	mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK;
190 
191 	/*
192 	 * Pick a base address for the vDSO in process space.
193 	 * Add enough to the size so that the result can be aligned.
194 	 */
195 	vdso_base = get_unmapped_area(NULL, 0, mappings_size, 0, 0);
196 	if (IS_ERR_VALUE(vdso_base))
197 		return vdso_base;
198 
199 	/* Add required alignment. */
200 	vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
201 
202 	/*
203 	 * Put vDSO base into mm struct. We need to do this before calling
204 	 * install_special_mapping or the perf counter mmap tracking code
205 	 * will fail to recognise it as a vDSO.
206 	 */
207 	mm->context.vdso = (void __user *)vdso_base + vvar_size;
208 
209 	vma = _install_special_mapping(mm, vdso_base, vvar_size,
210 				       VM_READ | VM_MAYREAD | VM_IO |
211 				       VM_DONTDUMP | VM_PFNMAP, &vvar_spec);
212 	if (IS_ERR(vma))
213 		return PTR_ERR(vma);
214 
215 	/*
216 	 * our vma flags don't have VM_WRITE so by default, the process isn't
217 	 * allowed to write those pages.
218 	 * gdb can break that with ptrace interface, and thus trigger COW on
219 	 * those pages but it's then your responsibility to never do that on
220 	 * the "data" page of the vDSO or you'll stop getting kernel updates
221 	 * and your nice userland gettimeofday will be totally dead.
222 	 * It's fine to use that for setting breakpoints in the vDSO code
223 	 * pages though.
224 	 */
225 	vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size,
226 				       VM_READ | VM_EXEC | VM_MAYREAD |
227 				       VM_MAYWRITE | VM_MAYEXEC, vdso_spec);
228 	if (IS_ERR(vma))
229 		do_munmap(mm, vdso_base, vvar_size, NULL);
230 
231 	return PTR_ERR_OR_ZERO(vma);
232 }
233 
234 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
235 {
236 	struct mm_struct *mm = current->mm;
237 	int rc;
238 
239 	mm->context.vdso = NULL;
240 
241 	if (mmap_write_lock_killable(mm))
242 		return -EINTR;
243 
244 	rc = __arch_setup_additional_pages(bprm, uses_interp);
245 	if (rc)
246 		mm->context.vdso = NULL;
247 
248 	mmap_write_unlock(mm);
249 	return rc;
250 }
251 
252 #define VDSO_DO_FIXUPS(type, value, bits, sec) do {					\
253 	void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start);	\
254 	void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end);	\
255 											\
256 	do_##type##_fixups((value), __start, __end);					\
257 } while (0)
258 
259 static void __init vdso_fixup_features(void)
260 {
261 #ifdef CONFIG_PPC64
262 	VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup);
263 	VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup);
264 	VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup);
265 	VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup);
266 #endif /* CONFIG_PPC64 */
267 
268 #ifdef CONFIG_VDSO32
269 	VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup);
270 	VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup);
271 #ifdef CONFIG_PPC64
272 	VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup);
273 #endif /* CONFIG_PPC64 */
274 	VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup);
275 #endif
276 }
277 
278 /*
279  * Called from setup_arch to initialize the bitmap of available
280  * syscalls in the systemcfg page
281  */
282 static void __init vdso_setup_syscall_map(void)
283 {
284 	unsigned int i;
285 
286 	for (i = 0; i < NR_syscalls; i++) {
287 		if (sys_call_table[i] != (void *)&sys_ni_syscall)
288 			vdso_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
289 		if (IS_ENABLED(CONFIG_COMPAT) &&
290 		    compat_sys_call_table[i] != (void *)&sys_ni_syscall)
291 			vdso_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
292 	}
293 }
294 
295 #ifdef CONFIG_PPC64
296 int vdso_getcpu_init(void)
297 {
298 	unsigned long cpu, node, val;
299 
300 	/*
301 	 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
302 	 * in the next 16 bits.  The VDSO uses this to implement getcpu().
303 	 */
304 	cpu = get_cpu();
305 	WARN_ON_ONCE(cpu > 0xffff);
306 
307 	node = cpu_to_node(cpu);
308 	WARN_ON_ONCE(node > 0xffff);
309 
310 	val = (cpu & 0xffff) | ((node & 0xffff) << 16);
311 	mtspr(SPRN_SPRG_VDSO_WRITE, val);
312 	get_paca()->sprg_vdso = val;
313 
314 	put_cpu();
315 
316 	return 0;
317 }
318 /* We need to call this before SMP init */
319 early_initcall(vdso_getcpu_init);
320 #endif
321 
322 static struct page ** __init vdso_setup_pages(void *start, void *end)
323 {
324 	int i;
325 	struct page **pagelist;
326 	int pages = (end - start) >> PAGE_SHIFT;
327 
328 	pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
329 	if (!pagelist)
330 		panic("%s: Cannot allocate page list for VDSO", __func__);
331 
332 	for (i = 0; i < pages; i++)
333 		pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
334 
335 	return pagelist;
336 }
337 
338 static int __init vdso_init(void)
339 {
340 #ifdef CONFIG_PPC64
341 	/*
342 	 * Fill up the "systemcfg" stuff for backward compatibility
343 	 */
344 	strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
345 	vdso_data->version.major = SYSTEMCFG_MAJOR;
346 	vdso_data->version.minor = SYSTEMCFG_MINOR;
347 	vdso_data->processor = mfspr(SPRN_PVR);
348 	/*
349 	 * Fake the old platform number for pSeries and add
350 	 * in LPAR bit if necessary
351 	 */
352 	vdso_data->platform = 0x100;
353 	if (firmware_has_feature(FW_FEATURE_LPAR))
354 		vdso_data->platform |= 1;
355 	vdso_data->physicalMemorySize = memblock_phys_mem_size();
356 	vdso_data->dcache_size = ppc64_caches.l1d.size;
357 	vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
358 	vdso_data->icache_size = ppc64_caches.l1i.size;
359 	vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
360 	vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
361 	vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
362 	vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
363 	vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
364 #endif /* CONFIG_PPC64 */
365 
366 	vdso_setup_syscall_map();
367 
368 	vdso_fixup_features();
369 
370 	if (IS_ENABLED(CONFIG_VDSO32))
371 		vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end);
372 
373 	if (IS_ENABLED(CONFIG_PPC64))
374 		vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end);
375 
376 	smp_wmb();
377 
378 	return 0;
379 }
380 arch_initcall(vdso_init);
381