arch/x86/kernel/ldt.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 1992 Krishna Balasubramanian and Linus Torvalds
   4  * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
   5  * Copyright (C) 2002 Andi Kleen
   6  *
   7  * This handles calls from both 32bit and 64bit mode.
   8  */
   9
  10 #include <linux/errno.h>
  11 #include <linux/gfp.h>
  12 #include <linux/sched.h>
  13 #include <linux/string.h>
  14 #include <linux/mm.h>
  15 #include <linux/smp.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/slab.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/uaccess.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 static void refresh_ldt_segments(void)
  27 {
  28 #ifdef CONFIG_X86_64
  29         unsigned short sel;
  30
  31         /*
  32          * Make sure that the cached DS and ES descriptors match the updated
  33          * LDT.
  34          */
  35         savesegment(ds, sel);
  36         if ((sel & SEGMENT_TI_MASK) == SEGMENT_LDT)
  37                 loadsegment(ds, sel);
  38
  39         savesegment(es, sel);
  40         if ((sel & SEGMENT_TI_MASK) == SEGMENT_LDT)
  41                 loadsegment(es, sel);
  42 #endif
  43 }
  44
  45 /* context.lock is held for us, so we don't need any locking. */
  46 static void flush_ldt(void *__mm)
  47 {
  48         struct mm_struct *mm = __mm;
  49         mm_context_t *pc;
  50
  51         if (this_cpu_read(cpu_tlbstate.loaded_mm) != mm)
  52                 return;
  53
  54         pc = &mm->context;
  55         set_ldt(pc->ldt->entries, pc->ldt->nr_entries);
  56
  57         refresh_ldt_segments();
  58 }
  59
  60 /* The caller must call finalize_ldt_struct on the result. LDT starts zeroed. */
  61 static struct ldt_struct *alloc_ldt_struct(unsigned int num_entries)
  62 {
  63         struct ldt_struct *new_ldt;
  64         unsigned int alloc_size;
  65
  66         if (num_entries > LDT_ENTRIES)
  67                 return NULL;
  68
  69         new_ldt = kmalloc(sizeof(struct ldt_struct), GFP_KERNEL);
  70         if (!new_ldt)
  71                 return NULL;
  72
  73         BUILD_BUG_ON(LDT_ENTRY_SIZE != sizeof(struct desc_struct));
  74         alloc_size = num_entries * LDT_ENTRY_SIZE;
  75
  76         /*
  77          * Xen is very picky: it requires a page-aligned LDT that has no
  78          * trailing nonzero bytes in any page that contains LDT descriptors.
  79          * Keep it simple: zero the whole allocation and never allocate less
  80          * than PAGE_SIZE.
  81          */
  82         if (alloc_size > PAGE_SIZE)
  83                 new_ldt->entries = vzalloc(alloc_size);
  84         else
  85                 new_ldt->entries = (void *)get_zeroed_page(GFP_KERNEL);
  86
  87         if (!new_ldt->entries) {
  88                 kfree(new_ldt);
  89                 return NULL;
  90         }
  91
  92         new_ldt->nr_entries = num_entries;
  93         return new_ldt;
  94 }
  95
  96 /* After calling this, the LDT is immutable. */
  97 static void finalize_ldt_struct(struct ldt_struct *ldt)
  98 {
  99         paravirt_alloc_ldt(ldt->entries, ldt->nr_entries);
 100 }
 101
 102 /* context.lock is held */
 103 static void install_ldt(struct mm_struct *current_mm,
 104                         struct ldt_struct *ldt)
 105 {
 106         /* Synchronizes with lockless_dereference in load_mm_ldt. */
 107         smp_store_release(&current_mm->context.ldt, ldt);
 108
 109         /* Activate the LDT for all CPUs using current_mm. */
 110         on_each_cpu_mask(mm_cpumask(current_mm), flush_ldt, current_mm, true);
 111 }
 112
 113 static void free_ldt_struct(struct ldt_struct *ldt)
 114 {
 115         if (likely(!ldt))
 116                 return;
 117
 118         paravirt_free_ldt(ldt->entries, ldt->nr_entries);
 119         if (ldt->nr_entries * LDT_ENTRY_SIZE > PAGE_SIZE)
 120                 vfree_atomic(ldt->entries);
 121         else
 122                 free_page((unsigned long)ldt->entries);
 123         kfree(ldt);
 124 }
 125
 126 /*
 127  * we do not have to muck with descriptors here, that is
 128  * done in switch_mm() as needed.
 129  */
 130 int init_new_context_ldt(struct task_struct *tsk, struct mm_struct *mm)
 131 {
 132         struct ldt_struct *new_ldt;
 133         struct mm_struct *old_mm;
 134         int retval = 0;
 135
 136         mutex_init(&mm->context.lock);
 137         old_mm = current->mm;
 138         if (!old_mm) {
 139                 mm->context.ldt = NULL;
 140                 return 0;
 141         }
 142
 143         mutex_lock(&old_mm->context.lock);
 144         if (!old_mm->context.ldt) {
 145                 mm->context.ldt = NULL;
 146                 goto out_unlock;
 147         }
 148
 149         new_ldt = alloc_ldt_struct(old_mm->context.ldt->nr_entries);
 150         if (!new_ldt) {
 151                 retval = -ENOMEM;
 152                 goto out_unlock;
 153         }
 154
 155         memcpy(new_ldt->entries, old_mm->context.ldt->entries,
 156                new_ldt->nr_entries * LDT_ENTRY_SIZE);
 157         finalize_ldt_struct(new_ldt);
 158
 159         mm->context.ldt = new_ldt;
 160
 161 out_unlock:
 162         mutex_unlock(&old_mm->context.lock);
 163         return retval;
 164 }
 165
 166 /*
 167  * No need to lock the MM as we are the last user
 168  *
 169  * 64bit: Don't touch the LDT register - we're already in the next thread.
 170  */
 171 void destroy_context_ldt(struct mm_struct *mm)
 172 {
 173         free_ldt_struct(mm->context.ldt);
 174         mm->context.ldt = NULL;
 175 }
 176
 177 static int read_ldt(void __user *ptr, unsigned long bytecount)
 178 {
 179         struct mm_struct *mm = current->mm;
 180         unsigned long entries_size;
 181         int retval;
 182
 183         mutex_lock(&mm->context.lock);
 184
 185         if (!mm->context.ldt) {
 186                 retval = 0;
 187                 goto out_unlock;
 188         }
 189
 190         if (bytecount > LDT_ENTRY_SIZE * LDT_ENTRIES)
 191                 bytecount = LDT_ENTRY_SIZE * LDT_ENTRIES;
 192
 193         entries_size = mm->context.ldt->nr_entries * LDT_ENTRY_SIZE;
 194         if (entries_size > bytecount)
 195                 entries_size = bytecount;
 196
 197         if (copy_to_user(ptr, mm->context.ldt->entries, entries_size)) {
 198                 retval = -EFAULT;
 199                 goto out_unlock;
 200         }
 201
 202         if (entries_size != bytecount) {
 203                 /* Zero-fill the rest and pretend we read bytecount bytes. */
 204                 if (clear_user(ptr + entries_size, bytecount - entries_size)) {
 205                         retval = -EFAULT;
 206                         goto out_unlock;
 207                 }
 208         }
 209         retval = bytecount;
 210
 211 out_unlock:
 212         mutex_unlock(&mm->context.lock);
 213         return retval;
 214 }
 215
 216 static int read_default_ldt(void __user *ptr, unsigned long bytecount)
 217 {
 218         /* CHECKME: Can we use _one_ random number ? */
 219 #ifdef CONFIG_X86_32
 220         unsigned long size = 5 * sizeof(struct desc_struct);
 221 #else
 222         unsigned long size = 128;
 223 #endif
 224         if (bytecount > size)
 225                 bytecount = size;
 226         if (clear_user(ptr, bytecount))
 227                 return -EFAULT;
 228         return bytecount;
 229 }
 230
 231 static int write_ldt(void __user *ptr, unsigned long bytecount, int oldmode)
 232 {
 233         struct mm_struct *mm = current->mm;
 234         struct ldt_struct *new_ldt, *old_ldt;
 235         unsigned int old_nr_entries, new_nr_entries;
 236         struct user_desc ldt_info;
 237         struct desc_struct ldt;
 238         int error;
 239
 240         error = -EINVAL;
 241         if (bytecount != sizeof(ldt_info))
 242                 goto out;
 243         error = -EFAULT;
 244         if (copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
 245                 goto out;
 246
 247         error = -EINVAL;
 248         if (ldt_info.entry_number >= LDT_ENTRIES)
 249                 goto out;
 250         if (ldt_info.contents == 3) {
 251                 if (oldmode)
 252                         goto out;
 253                 if (ldt_info.seg_not_present == 0)
 254                         goto out;
 255         }
 256
 257         if ((oldmode && !ldt_info.base_addr && !ldt_info.limit) ||
 258             LDT_empty(&ldt_info)) {
 259                 /* The user wants to clear the entry. */
 260                 memset(&ldt, 0, sizeof(ldt));
 261         } else {
 262                 if (!IS_ENABLED(CONFIG_X86_16BIT) && !ldt_info.seg_32bit) {
 263                         error = -EINVAL;
 264                         goto out;
 265                 }
 266
 267                 fill_ldt(&ldt, &ldt_info);
 268                 if (oldmode)
 269                         ldt.avl = 0;
 270         }
 271
 272         mutex_lock(&mm->context.lock);
 273
 274         old_ldt       = mm->context.ldt;
 275         old_nr_entries = old_ldt ? old_ldt->nr_entries : 0;
 276         new_nr_entries = max(ldt_info.entry_number + 1, old_nr_entries);
 277
 278         error = -ENOMEM;
 279         new_ldt = alloc_ldt_struct(new_nr_entries);
 280         if (!new_ldt)
 281                 goto out_unlock;
 282
 283         if (old_ldt)
 284                 memcpy(new_ldt->entries, old_ldt->entries, old_nr_entries * LDT_ENTRY_SIZE);
 285
 286         new_ldt->entries[ldt_info.entry_number] = ldt;
 287         finalize_ldt_struct(new_ldt);
 288
 289         install_ldt(mm, new_ldt);
 290         free_ldt_struct(old_ldt);
 291         error = 0;
 292
 293 out_unlock:
 294         mutex_unlock(&mm->context.lock);
 295 out:
 296         return error;
 297 }
 298
 299 SYSCALL_DEFINE3(modify_ldt, int , func , void __user * , ptr ,
 300                 unsigned long , bytecount)
 301 {
 302         int ret = -ENOSYS;
 303
 304         switch (func) {
 305         case 0:
 306                 ret = read_ldt(ptr, bytecount);
 307                 break;
 308         case 1:
 309                 ret = write_ldt(ptr, bytecount, 1);
 310                 break;
 311         case 2:
 312                 ret = read_default_ldt(ptr, bytecount);
 313                 break;
 314         case 0x11:
 315                 ret = write_ldt(ptr, bytecount, 0);
 316                 break;
 317         }
 318         /*
 319          * The SYSCALL_DEFINE() macros give us an 'unsigned long'
 320          * return type, but tht ABI for sys_modify_ldt() expects
 321          * 'int'.  This cast gives us an int-sized value in %rax
 322          * for the return code.  The 'unsigned' is necessary so
 323          * the compiler does not try to sign-extend the negative
 324          * return codes into the high half of the register when
 325          * taking the value from int->long.
 326          */
 327         return (unsigned int)ret;
 328 }