kernel/kexec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * kexec.c - kexec_load system call
   4  * Copyright (C) 2002-2004 Eric Biederman  <ebiederm@xmission.com>
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/capability.h>
  10 #include <linux/mm.h>
  11 #include <linux/file.h>
  12 #include <linux/security.h>
  13 #include <linux/kexec.h>
  14 #include <linux/mutex.h>
  15 #include <linux/list.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/slab.h>
  19
  20 #include "kexec_internal.h"
  21
  22 static int copy_user_segment_list(struct kimage *image,
  23                                   unsigned long nr_segments,
  24                                   struct kexec_segment __user *segments)
  25 {
  26         int ret;
  27         size_t segment_bytes;
  28
  29         /* Read in the segments */
  30         image->nr_segments = nr_segments;
  31         segment_bytes = nr_segments * sizeof(*segments);
  32         ret = copy_from_user(image->segment, segments, segment_bytes);
  33         if (ret)
  34                 ret = -EFAULT;
  35
  36         return ret;
  37 }
  38
  39 static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
  40                              unsigned long nr_segments,
  41                              struct kexec_segment __user *segments,
  42                              unsigned long flags)
  43 {
  44         int ret;
  45         struct kimage *image;
  46         bool kexec_on_panic = flags & KEXEC_ON_CRASH;
  47
  48         if (kexec_on_panic) {
  49                 /* Verify we have a valid entry point */
  50                 if ((entry < phys_to_boot_phys(crashk_res.start)) ||
  51                     (entry > phys_to_boot_phys(crashk_res.end)))
  52                         return -EADDRNOTAVAIL;
  53         }
  54
  55         /* Allocate and initialize a controlling structure */
  56         image = do_kimage_alloc_init();
  57         if (!image)
  58                 return -ENOMEM;
  59
  60         image->start = entry;
  61
  62         ret = copy_user_segment_list(image, nr_segments, segments);
  63         if (ret)
  64                 goto out_free_image;
  65
  66         if (kexec_on_panic) {
  67                 /* Enable special crash kernel control page alloc policy. */
  68                 image->control_page = crashk_res.start;
  69                 image->type = KEXEC_TYPE_CRASH;
  70         }
  71
  72         ret = sanity_check_segment_list(image);
  73         if (ret)
  74                 goto out_free_image;
  75
  76         /*
  77          * Find a location for the control code buffer, and add it
  78          * the vector of segments so that it's pages will also be
  79          * counted as destination pages.
  80          */
  81         ret = -ENOMEM;
  82         image->control_code_page = kimage_alloc_control_pages(image,
  83                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
  84         if (!image->control_code_page) {
  85                 pr_err("Could not allocate control_code_buffer\n");
  86                 goto out_free_image;
  87         }
  88
  89         if (!kexec_on_panic) {
  90                 image->swap_page = kimage_alloc_control_pages(image, 0);
  91                 if (!image->swap_page) {
  92                         pr_err("Could not allocate swap buffer\n");
  93                         goto out_free_control_pages;
  94                 }
  95         }
  96
  97         *rimage = image;
  98         return 0;
  99 out_free_control_pages:
 100         kimage_free_page_list(&image->control_pages);
 101 out_free_image:
 102         kfree(image);
 103         return ret;
 104 }
 105
 106 static int do_kexec_load(unsigned long entry, unsigned long nr_segments,
 107                 struct kexec_segment __user *segments, unsigned long flags)
 108 {
 109         struct kimage **dest_image, *image;
 110         unsigned long i;
 111         int ret;
 112
 113         /*
 114          * Because we write directly to the reserved memory region when loading
 115          * crash kernels we need a serialization here to prevent multiple crash
 116          * kernels from attempting to load simultaneously.
 117          */
 118         if (!kexec_trylock())
 119                 return -EBUSY;
 120
 121         if (flags & KEXEC_ON_CRASH) {
 122                 dest_image = &kexec_crash_image;
 123                 if (kexec_crash_image)
 124                         arch_kexec_unprotect_crashkres();
 125         } else {
 126                 dest_image = &kexec_image;
 127         }
 128
 129         if (nr_segments == 0) {
 130                 /* Uninstall image */
 131                 kimage_free(xchg(dest_image, NULL));
 132                 ret = 0;
 133                 goto out_unlock;
 134         }
 135         if (flags & KEXEC_ON_CRASH) {
 136                 /*
 137                  * Loading another kernel to switch to if this one
 138                  * crashes.  Free any current crash dump kernel before
 139                  * we corrupt it.
 140                  */
 141                 kimage_free(xchg(&kexec_crash_image, NULL));
 142         }
 143
 144         ret = kimage_alloc_init(&image, entry, nr_segments, segments, flags);
 145         if (ret)
 146                 goto out_unlock;
 147
 148         if (flags & KEXEC_PRESERVE_CONTEXT)
 149                 image->preserve_context = 1;
 150
 151         ret = machine_kexec_prepare(image);
 152         if (ret)
 153                 goto out;
 154
 155         /*
 156          * Some architecture(like S390) may touch the crash memory before
 157          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
 158          */
 159         ret = kimage_crash_copy_vmcoreinfo(image);
 160         if (ret)
 161                 goto out;
 162
 163         for (i = 0; i < nr_segments; i++) {
 164                 ret = kimage_load_segment(image, &image->segment[i]);
 165                 if (ret)
 166                         goto out;
 167         }
 168
 169         kimage_terminate(image);
 170
 171         ret = machine_kexec_post_load(image);
 172         if (ret)
 173                 goto out;
 174
 175         /* Install the new kernel and uninstall the old */
 176         image = xchg(dest_image, image);
 177
 178 out:
 179         if ((flags & KEXEC_ON_CRASH) && kexec_crash_image)
 180                 arch_kexec_protect_crashkres();
 181
 182         kimage_free(image);
 183 out_unlock:
 184         kexec_unlock();
 185         return ret;
 186 }
 187
 188 /*
 189  * Exec Kernel system call: for obvious reasons only root may call it.
 190  *
 191  * This call breaks up into three pieces.
 192  * - A generic part which loads the new kernel from the current
 193  *   address space, and very carefully places the data in the
 194  *   allocated pages.
 195  *
 196  * - A generic part that interacts with the kernel and tells all of
 197  *   the devices to shut down.  Preventing on-going dmas, and placing
 198  *   the devices in a consistent state so a later kernel can
 199  *   reinitialize them.
 200  *
 201  * - A machine specific part that includes the syscall number
 202  *   and then copies the image to it's final destination.  And
 203  *   jumps into the image at entry.
 204  *
 205  * kexec does not sync, or unmount filesystems so if you need
 206  * that to happen you need to do that yourself.
 207  */
 208
 209 static inline int kexec_load_check(unsigned long nr_segments,
 210                                    unsigned long flags)
 211 {
 212         int result;
 213
 214         /* We only trust the superuser with rebooting the system. */
 215         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
 216                 return -EPERM;
 217
 218         /* Permit LSMs and IMA to fail the kexec */
 219         result = security_kernel_load_data(LOADING_KEXEC_IMAGE, false);
 220         if (result < 0)
 221                 return result;
 222
 223         /*
 224          * kexec can be used to circumvent module loading restrictions, so
 225          * prevent loading in that case
 226          */
 227         result = security_locked_down(LOCKDOWN_KEXEC);
 228         if (result)
 229                 return result;
 230
 231         /*
 232          * Verify we have a legal set of flags
 233          * This leaves us room for future extensions.
 234          */
 235         if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
 236                 return -EINVAL;
 237
 238         /* Put an artificial cap on the number
 239          * of segments passed to kexec_load.
 240          */
 241         if (nr_segments > KEXEC_SEGMENT_MAX)
 242                 return -EINVAL;
 243
 244         return 0;
 245 }
 246
 247 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
 248                 struct kexec_segment __user *, segments, unsigned long, flags)
 249 {
 250         int result;
 251
 252         result = kexec_load_check(nr_segments, flags);
 253         if (result)
 254                 return result;
 255
 256         /* Verify we are on the appropriate architecture */
 257         if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
 258                 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
 259                 return -EINVAL;
 260
 261         result = do_kexec_load(entry, nr_segments, segments, flags);
 262
 263         return result;
 264 }
 265
 266 #ifdef CONFIG_COMPAT
 267 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
 268                        compat_ulong_t, nr_segments,
 269                        struct compat_kexec_segment __user *, segments,
 270                        compat_ulong_t, flags)
 271 {
 272         struct compat_kexec_segment in;
 273         struct kexec_segment out, __user *ksegments;
 274         unsigned long i, result;
 275
 276         result = kexec_load_check(nr_segments, flags);
 277         if (result)
 278                 return result;
 279
 280         /* Don't allow clients that don't understand the native
 281          * architecture to do anything.
 282          */
 283         if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
 284                 return -EINVAL;
 285
 286         ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
 287         for (i = 0; i < nr_segments; i++) {
 288                 result = copy_from_user(&in, &segments[i], sizeof(in));
 289                 if (result)
 290                         return -EFAULT;
 291
 292                 out.buf   = compat_ptr(in.buf);
 293                 out.bufsz = in.bufsz;
 294                 out.mem   = in.mem;
 295                 out.memsz = in.memsz;
 296
 297                 result = copy_to_user(&ksegments[i], &out, sizeof(out));
 298                 if (result)
 299                         return -EFAULT;
 300         }
 301
 302         result = do_kexec_load(entry, nr_segments, ksegments, flags);
 303
 304         return result;
 305 }
 306 #endif