kernel/kexec.c

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