1 /*
2 * Copyright (C) 1992 Krishna Balasubramanian and Linus Torvalds
3 * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
4 * Copyright (C) 2002 Andi Kleen
5 *
6 * This handles calls from both 32bit and 64bit mode.
7 */
8
9 #include <linux/errno.h>
10 #include <linux/gfp.h>
11 #include <linux/sched.h>
12 #include <linux/string.h>
13 #include <linux/mm.h>
14 #include <linux/smp.h>
15 #include <linux/syscalls.h>
16 #include <linux/slab.h>
17 #include <linux/vmalloc.h>
18 #include <linux/uaccess.h>
19 #include <linux/kaiser.h>
20
21 #include <asm/ldt.h>
22 #include <asm/desc.h>
23 #include <asm/mmu_context.h>
24 #include <asm/syscalls.h>
25
26 /* context.lock is held for us, so we don't need any locking. */
flush_ldt(void * current_mm)27 static void flush_ldt(void *current_mm)
28 {
29 mm_context_t *pc;
30
31 if (current->active_mm != current_mm)
32 return;
33
34 pc = ¤t->active_mm->context;
35 set_ldt(pc->ldt->entries, pc->ldt->size);
36 }
37
__free_ldt_struct(struct ldt_struct * ldt)38 static void __free_ldt_struct(struct ldt_struct *ldt)
39 {
40 if (ldt->size * LDT_ENTRY_SIZE > PAGE_SIZE)
41 vfree(ldt->entries);
42 else
43 free_page((unsigned long)ldt->entries);
44 kfree(ldt);
45 }
46
47 /* The caller must call finalize_ldt_struct on the result. LDT starts zeroed. */
alloc_ldt_struct(int size)48 static struct ldt_struct *alloc_ldt_struct(int size)
49 {
50 struct ldt_struct *new_ldt;
51 int alloc_size;
52 int ret;
53
54 if (size > LDT_ENTRIES)
55 return NULL;
56
57 new_ldt = kmalloc(sizeof(struct ldt_struct), GFP_KERNEL);
58 if (!new_ldt)
59 return NULL;
60
61 BUILD_BUG_ON(LDT_ENTRY_SIZE != sizeof(struct desc_struct));
62 alloc_size = size * LDT_ENTRY_SIZE;
63
64 /*
65 * Xen is very picky: it requires a page-aligned LDT that has no
66 * trailing nonzero bytes in any page that contains LDT descriptors.
67 * Keep it simple: zero the whole allocation and never allocate less
68 * than PAGE_SIZE.
69 */
70 if (alloc_size > PAGE_SIZE)
71 new_ldt->entries = vzalloc(alloc_size);
72 else
73 new_ldt->entries = (void *)get_zeroed_page(GFP_KERNEL);
74
75 if (!new_ldt->entries) {
76 kfree(new_ldt);
77 return NULL;
78 }
79
80 ret = kaiser_add_mapping((unsigned long)new_ldt->entries, alloc_size,
81 __PAGE_KERNEL);
82 new_ldt->size = size;
83 if (ret) {
84 __free_ldt_struct(new_ldt);
85 return NULL;
86 }
87 return new_ldt;
88 }
89
90 /* After calling this, the LDT is immutable. */
finalize_ldt_struct(struct ldt_struct * ldt)91 static void finalize_ldt_struct(struct ldt_struct *ldt)
92 {
93 paravirt_alloc_ldt(ldt->entries, ldt->size);
94 }
95
96 /* context.lock is held */
install_ldt(struct mm_struct * current_mm,struct ldt_struct * ldt)97 static void install_ldt(struct mm_struct *current_mm,
98 struct ldt_struct *ldt)
99 {
100 /* Synchronizes with lockless_dereference in load_mm_ldt. */
101 smp_store_release(¤t_mm->context.ldt, ldt);
102
103 /* Activate the LDT for all CPUs using current_mm. */
104 on_each_cpu_mask(mm_cpumask(current_mm), flush_ldt, current_mm, true);
105 }
106
free_ldt_struct(struct ldt_struct * ldt)107 static void free_ldt_struct(struct ldt_struct *ldt)
108 {
109 if (likely(!ldt))
110 return;
111
112 kaiser_remove_mapping((unsigned long)ldt->entries,
113 ldt->size * LDT_ENTRY_SIZE);
114 paravirt_free_ldt(ldt->entries, ldt->size);
115 __free_ldt_struct(ldt);
116 }
117
118 /*
119 * we do not have to muck with descriptors here, that is
120 * done in switch_mm() as needed.
121 */
init_new_context_ldt(struct task_struct * tsk,struct mm_struct * mm)122 int init_new_context_ldt(struct task_struct *tsk, struct mm_struct *mm)
123 {
124 struct ldt_struct *new_ldt;
125 struct mm_struct *old_mm;
126 int retval = 0;
127
128 mutex_init(&mm->context.lock);
129 old_mm = current->mm;
130 if (!old_mm) {
131 mm->context.ldt = NULL;
132 return 0;
133 }
134
135 mutex_lock(&old_mm->context.lock);
136 if (!old_mm->context.ldt) {
137 mm->context.ldt = NULL;
138 goto out_unlock;
139 }
140
141 new_ldt = alloc_ldt_struct(old_mm->context.ldt->size);
142 if (!new_ldt) {
143 retval = -ENOMEM;
144 goto out_unlock;
145 }
146
147 memcpy(new_ldt->entries, old_mm->context.ldt->entries,
148 new_ldt->size * LDT_ENTRY_SIZE);
149 finalize_ldt_struct(new_ldt);
150
151 mm->context.ldt = new_ldt;
152
153 out_unlock:
154 mutex_unlock(&old_mm->context.lock);
155 return retval;
156 }
157
158 /*
159 * No need to lock the MM as we are the last user
160 *
161 * 64bit: Don't touch the LDT register - we're already in the next thread.
162 */
destroy_context_ldt(struct mm_struct * mm)163 void destroy_context_ldt(struct mm_struct *mm)
164 {
165 free_ldt_struct(mm->context.ldt);
166 mm->context.ldt = NULL;
167 }
168
read_ldt(void __user * ptr,unsigned long bytecount)169 static int read_ldt(void __user *ptr, unsigned long bytecount)
170 {
171 int retval;
172 unsigned long size;
173 struct mm_struct *mm = current->mm;
174
175 mutex_lock(&mm->context.lock);
176
177 if (!mm->context.ldt) {
178 retval = 0;
179 goto out_unlock;
180 }
181
182 if (bytecount > LDT_ENTRY_SIZE * LDT_ENTRIES)
183 bytecount = LDT_ENTRY_SIZE * LDT_ENTRIES;
184
185 size = mm->context.ldt->size * LDT_ENTRY_SIZE;
186 if (size > bytecount)
187 size = bytecount;
188
189 if (copy_to_user(ptr, mm->context.ldt->entries, size)) {
190 retval = -EFAULT;
191 goto out_unlock;
192 }
193
194 if (size != bytecount) {
195 /* Zero-fill the rest and pretend we read bytecount bytes. */
196 if (clear_user(ptr + size, bytecount - size)) {
197 retval = -EFAULT;
198 goto out_unlock;
199 }
200 }
201 retval = bytecount;
202
203 out_unlock:
204 mutex_unlock(&mm->context.lock);
205 return retval;
206 }
207
read_default_ldt(void __user * ptr,unsigned long bytecount)208 static int read_default_ldt(void __user *ptr, unsigned long bytecount)
209 {
210 /* CHECKME: Can we use _one_ random number ? */
211 #ifdef CONFIG_X86_32
212 unsigned long size = 5 * sizeof(struct desc_struct);
213 #else
214 unsigned long size = 128;
215 #endif
216 if (bytecount > size)
217 bytecount = size;
218 if (clear_user(ptr, bytecount))
219 return -EFAULT;
220 return bytecount;
221 }
222
write_ldt(void __user * ptr,unsigned long bytecount,int oldmode)223 static int write_ldt(void __user *ptr, unsigned long bytecount, int oldmode)
224 {
225 struct mm_struct *mm = current->mm;
226 struct desc_struct ldt;
227 int error;
228 struct user_desc ldt_info;
229 int oldsize, newsize;
230 struct ldt_struct *new_ldt, *old_ldt;
231
232 error = -EINVAL;
233 if (bytecount != sizeof(ldt_info))
234 goto out;
235 error = -EFAULT;
236 if (copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
237 goto out;
238
239 error = -EINVAL;
240 if (ldt_info.entry_number >= LDT_ENTRIES)
241 goto out;
242 if (ldt_info.contents == 3) {
243 if (oldmode)
244 goto out;
245 if (ldt_info.seg_not_present == 0)
246 goto out;
247 }
248
249 if ((oldmode && !ldt_info.base_addr && !ldt_info.limit) ||
250 LDT_empty(&ldt_info)) {
251 /* The user wants to clear the entry. */
252 memset(&ldt, 0, sizeof(ldt));
253 } else {
254 if (!IS_ENABLED(CONFIG_X86_16BIT) && !ldt_info.seg_32bit) {
255 error = -EINVAL;
256 goto out;
257 }
258
259 fill_ldt(&ldt, &ldt_info);
260 if (oldmode)
261 ldt.avl = 0;
262 }
263
264 mutex_lock(&mm->context.lock);
265
266 old_ldt = mm->context.ldt;
267 oldsize = old_ldt ? old_ldt->size : 0;
268 newsize = max((int)(ldt_info.entry_number + 1), oldsize);
269
270 error = -ENOMEM;
271 new_ldt = alloc_ldt_struct(newsize);
272 if (!new_ldt)
273 goto out_unlock;
274
275 if (old_ldt)
276 memcpy(new_ldt->entries, old_ldt->entries, oldsize * LDT_ENTRY_SIZE);
277 new_ldt->entries[ldt_info.entry_number] = ldt;
278 finalize_ldt_struct(new_ldt);
279
280 install_ldt(mm, new_ldt);
281 free_ldt_struct(old_ldt);
282 error = 0;
283
284 out_unlock:
285 mutex_unlock(&mm->context.lock);
286 out:
287 return error;
288 }
289
SYSCALL_DEFINE3(modify_ldt,int,func,void __user *,ptr,unsigned long,bytecount)290 SYSCALL_DEFINE3(modify_ldt, int , func , void __user * , ptr ,
291 unsigned long , bytecount)
292 {
293 int ret = -ENOSYS;
294
295 switch (func) {
296 case 0:
297 ret = read_ldt(ptr, bytecount);
298 break;
299 case 1:
300 ret = write_ldt(ptr, bytecount, 1);
301 break;
302 case 2:
303 ret = read_default_ldt(ptr, bytecount);
304 break;
305 case 0x11:
306 ret = write_ldt(ptr, bytecount, 0);
307 break;
308 }
309 /*
310 * The SYSCALL_DEFINE() macros give us an 'unsigned long'
311 * return type, but tht ABI for sys_modify_ldt() expects
312 * 'int'. This cast gives us an int-sized value in %rax
313 * for the return code. The 'unsigned' is necessary so
314 * the compiler does not try to sign-extend the negative
315 * return codes into the high half of the register when
316 * taking the value from int->long.
317 */
318 return (unsigned int)ret;
319 }
320