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 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments, 196 struct kexec_segment __user *, segments, unsigned long, flags) 197 { 198 int result; 199 200 /* We only trust the superuser with rebooting the system. */ 201 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) 202 return -EPERM; 203 204 /* 205 * Verify we have a legal set of flags 206 * This leaves us room for future extensions. 207 */ 208 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK)) 209 return -EINVAL; 210 211 /* Verify we are on the appropriate architecture */ 212 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) && 213 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT)) 214 return -EINVAL; 215 216 /* Put an artificial cap on the number 217 * of segments passed to kexec_load. 218 */ 219 if (nr_segments > KEXEC_SEGMENT_MAX) 220 return -EINVAL; 221 222 /* Because we write directly to the reserved memory 223 * region when loading crash kernels we need a mutex here to 224 * prevent multiple crash kernels from attempting to load 225 * simultaneously, and to prevent a crash kernel from loading 226 * over the top of a in use crash kernel. 227 * 228 * KISS: always take the mutex. 229 */ 230 if (!mutex_trylock(&kexec_mutex)) 231 return -EBUSY; 232 233 result = do_kexec_load(entry, nr_segments, segments, flags); 234 235 mutex_unlock(&kexec_mutex); 236 237 return result; 238 } 239 240 #ifdef CONFIG_COMPAT 241 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry, 242 compat_ulong_t, nr_segments, 243 struct compat_kexec_segment __user *, segments, 244 compat_ulong_t, flags) 245 { 246 struct compat_kexec_segment in; 247 struct kexec_segment out, __user *ksegments; 248 unsigned long i, result; 249 250 /* Don't allow clients that don't understand the native 251 * architecture to do anything. 252 */ 253 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT) 254 return -EINVAL; 255 256 if (nr_segments > KEXEC_SEGMENT_MAX) 257 return -EINVAL; 258 259 ksegments = compat_alloc_user_space(nr_segments * sizeof(out)); 260 for (i = 0; i < nr_segments; i++) { 261 result = copy_from_user(&in, &segments[i], sizeof(in)); 262 if (result) 263 return -EFAULT; 264 265 out.buf = compat_ptr(in.buf); 266 out.bufsz = in.bufsz; 267 out.mem = in.mem; 268 out.memsz = in.memsz; 269 270 result = copy_to_user(&ksegments[i], &out, sizeof(out)); 271 if (result) 272 return -EFAULT; 273 } 274 275 return sys_kexec_load(entry, nr_segments, ksegments, flags); 276 } 277 #endif 278