1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/init.h>
3 #include <linux/async.h>
5 #include <linux/slab.h>
6 #include <linux/types.h>
7 #include <linux/fcntl.h>
8 #include <linux/delay.h>
9 #include <linux/string.h>
10 #include <linux/dirent.h>
11 #include <linux/syscalls.h>
12 #include <linux/utime.h>
13 #include <linux/file.h>
14 #include <linux/memblock.h>
16 #include <linux/namei.h>
17 #include <linux/init_syscalls.h>
19 static ssize_t __init xwrite(struct file *file, const char *p, size_t count,
24 /* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
26 ssize_t rv = kernel_write(file, p, count, pos);
29 if (rv == -EINTR || rv == -EAGAIN)
31 return out ? out : rv;
43 static __initdata char *message;
44 static void __init error(char *x)
50 static void panic_show_mem(const char *fmt, ...)
62 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
64 static __initdata struct hash {
65 int ino, minor, major;
68 char name[N_ALIGN(PATH_MAX)];
71 static inline int hash(int major, int minor, int ino)
73 unsigned long tmp = ino + minor + (major << 3);
78 static char __init *find_link(int major, int minor, int ino,
79 umode_t mode, char *name)
82 for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
85 if ((*p)->minor != minor)
87 if ((*p)->major != major)
89 if (((*p)->mode ^ mode) & S_IFMT)
93 q = kmalloc(sizeof(struct hash), GFP_KERNEL);
95 panic_show_mem("can't allocate link hash entry");
100 strcpy(q->name, name);
106 static void __init free_hash(void)
109 for (p = head; p < head + 32; p++) {
118 static long __init do_utime(char *filename, time64_t mtime)
120 struct timespec64 t[2];
126 return init_utimes(filename, t);
129 static __initdata LIST_HEAD(dir_list);
131 struct list_head list;
136 static void __init dir_add(const char *name, time64_t mtime)
138 struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
140 panic_show_mem("can't allocate dir_entry buffer");
141 INIT_LIST_HEAD(&de->list);
142 de->name = kstrdup(name, GFP_KERNEL);
144 list_add(&de->list, &dir_list);
147 static void __init dir_utime(void)
149 struct dir_entry *de, *tmp;
150 list_for_each_entry_safe(de, tmp, &dir_list, list) {
152 do_utime(de->name, de->mtime);
158 static __initdata time64_t mtime;
160 /* cpio header parsing */
162 static __initdata unsigned long ino, major, minor, nlink;
163 static __initdata umode_t mode;
164 static __initdata unsigned long body_len, name_len;
165 static __initdata uid_t uid;
166 static __initdata gid_t gid;
167 static __initdata unsigned rdev;
169 static void __init parse_header(char *s)
171 unsigned long parsed[12];
176 for (i = 0, s += 6; i < 12; i++, s += 8) {
178 parsed[i] = simple_strtoul(buf, NULL, 16);
185 mtime = parsed[5]; /* breaks in y2106 */
186 body_len = parsed[6];
189 rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
190 name_len = parsed[11];
195 static __initdata enum state {
206 static __initdata char *victim;
207 static unsigned long byte_count __initdata;
208 static __initdata loff_t this_header, next_header;
210 static inline void __init eat(unsigned n)
217 static __initdata char *collected;
218 static long remains __initdata;
219 static __initdata char *collect;
221 static void __init read_into(char *buf, unsigned size, enum state next)
223 if (byte_count >= size) {
228 collect = collected = buf;
235 static __initdata char *header_buf, *symlink_buf, *name_buf;
237 static int __init do_start(void)
239 read_into(header_buf, 110, GotHeader);
243 static int __init do_collect(void)
245 unsigned long n = remains;
248 memcpy(collect, victim, n);
251 if ((remains -= n) != 0)
257 static int __init do_header(void)
259 if (memcmp(collected, "070707", 6)==0) {
260 error("incorrect cpio method used: use -H newc option");
263 if (memcmp(collected, "070701", 6)) {
264 error("no cpio magic");
267 parse_header(collected);
268 next_header = this_header + N_ALIGN(name_len) + body_len;
269 next_header = (next_header + 3) & ~3;
271 if (name_len <= 0 || name_len > PATH_MAX)
274 if (body_len > PATH_MAX)
276 collect = collected = symlink_buf;
277 remains = N_ALIGN(name_len) + body_len;
278 next_state = GotSymlink;
282 if (S_ISREG(mode) || !body_len)
283 read_into(name_buf, N_ALIGN(name_len), GotName);
287 static int __init do_skip(void)
289 if (this_header + byte_count < next_header) {
293 eat(next_header - this_header);
299 static int __init do_reset(void)
301 while (byte_count && *victim == '\0')
303 if (byte_count && (this_header & 3))
304 error("broken padding");
308 static void __init clean_path(char *path, umode_t fmode)
312 if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
313 (st.mode ^ fmode) & S_IFMT) {
314 if (S_ISDIR(st.mode))
321 static int __init maybe_link(void)
324 char *old = find_link(major, minor, ino, mode, collected);
326 clean_path(collected, 0);
327 return (init_link(old, collected) < 0) ? -1 : 1;
333 static __initdata struct file *wfile;
334 static __initdata loff_t wfile_pos;
336 static int __init do_name(void)
340 if (strcmp(collected, "TRAILER!!!") == 0) {
344 clean_path(collected, mode);
346 int ml = maybe_link();
348 int openflags = O_WRONLY|O_CREAT;
350 openflags |= O_TRUNC;
351 wfile = filp_open(collected, openflags, mode);
356 vfs_fchown(wfile, uid, gid);
357 vfs_fchmod(wfile, mode);
359 vfs_truncate(&wfile->f_path, body_len);
362 } else if (S_ISDIR(mode)) {
363 init_mkdir(collected, mode);
364 init_chown(collected, uid, gid, 0);
365 init_chmod(collected, mode);
366 dir_add(collected, mtime);
367 } else if (S_ISBLK(mode) || S_ISCHR(mode) ||
368 S_ISFIFO(mode) || S_ISSOCK(mode)) {
369 if (maybe_link() == 0) {
370 init_mknod(collected, mode, rdev);
371 init_chown(collected, uid, gid, 0);
372 init_chmod(collected, mode);
373 do_utime(collected, mtime);
379 static int __init do_copy(void)
381 if (byte_count >= body_len) {
382 struct timespec64 t[2] = { };
383 if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
384 error("write error");
388 vfs_utimes(&wfile->f_path, t);
395 if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
396 error("write error");
397 body_len -= byte_count;
403 static int __init do_symlink(void)
405 collected[N_ALIGN(name_len) + body_len] = '\0';
406 clean_path(collected, 0);
407 init_symlink(collected + N_ALIGN(name_len), collected);
408 init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
409 do_utime(collected, mtime);
415 static __initdata int (*actions[])(void) = {
417 [Collect] = do_collect,
418 [GotHeader] = do_header,
421 [CopyFile] = do_copy,
422 [GotSymlink] = do_symlink,
426 static long __init write_buffer(char *buf, unsigned long len)
431 while (!actions[state]())
433 return len - byte_count;
436 static long __init flush_buffer(void *bufv, unsigned long len)
438 char *buf = (char *) bufv;
443 while ((written = write_buffer(buf, len)) < len && !message) {
444 char c = buf[written];
454 error("junk within compressed archive");
459 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
461 #include <linux/decompress/generic.h>
463 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
466 decompress_fn decompress;
467 const char *compress_name;
468 static __initdata char msg_buf[64];
470 header_buf = kmalloc(110, GFP_KERNEL);
471 symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
472 name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
474 if (!header_buf || !symlink_buf || !name_buf)
475 panic_show_mem("can't allocate buffers");
480 while (!message && len) {
481 loff_t saved_offset = this_header;
482 if (*buf == '0' && !(this_header & 3)) {
484 written = write_buffer(buf, len);
496 decompress = decompress_method(buf, len, &compress_name);
497 pr_debug("Detected %s compressed data\n", compress_name);
499 int res = decompress(buf, len, NULL, flush_buffer, NULL,
502 error("decompressor failed");
503 } else if (compress_name) {
505 snprintf(msg_buf, sizeof msg_buf,
506 "compression method %s not configured",
511 error("invalid magic at start of compressed archive");
513 error("junk at the end of compressed archive");
514 this_header = saved_offset + my_inptr;
525 static int __initdata do_retain_initrd;
527 static int __init retain_initrd_param(char *str)
531 do_retain_initrd = 1;
534 __setup("retain_initrd", retain_initrd_param);
536 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
537 static int __init keepinitrd_setup(char *__unused)
539 do_retain_initrd = 1;
542 __setup("keepinitrd", keepinitrd_setup);
545 static bool __initdata initramfs_async = true;
546 static int __init initramfs_async_setup(char *str)
548 strtobool(str, &initramfs_async);
551 __setup("initramfs_async=", initramfs_async_setup);
553 extern char __initramfs_start[];
554 extern unsigned long __initramfs_size;
555 #include <linux/initrd.h>
556 #include <linux/kexec.h>
558 void __init reserve_initrd_mem(void)
563 /* Ignore the virtul address computed during device tree parsing */
564 initrd_start = initrd_end = 0;
566 if (!phys_initrd_size)
569 * Round the memory region to page boundaries as per free_initrd_mem()
570 * This allows us to detect whether the pages overlapping the initrd
571 * are in use, but more importantly, reserves the entire set of pages
572 * as we don't want these pages allocated for other purposes.
574 start = round_down(phys_initrd_start, PAGE_SIZE);
575 size = phys_initrd_size + (phys_initrd_start - start);
576 size = round_up(size, PAGE_SIZE);
578 if (!memblock_is_region_memory(start, size)) {
579 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
584 if (memblock_is_region_reserved(start, size)) {
585 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
590 memblock_reserve(start, size);
591 /* Now convert initrd to virtual addresses */
592 initrd_start = (unsigned long)__va(phys_initrd_start);
593 initrd_end = initrd_start + phys_initrd_size;
594 initrd_below_start_ok = 1;
598 pr_cont(" - disabling initrd\n");
603 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
605 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
606 unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
607 unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
609 memblock_free(__pa(aligned_start), aligned_end - aligned_start);
612 free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
616 #ifdef CONFIG_KEXEC_CORE
617 static bool __init kexec_free_initrd(void)
619 unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
620 unsigned long crashk_end = (unsigned long)__va(crashk_res.end);
623 * If the initrd region is overlapped with crashkernel reserved region,
624 * free only memory that is not part of crashkernel region.
626 if (initrd_start >= crashk_end || initrd_end <= crashk_start)
630 * Initialize initrd memory region since the kexec boot does not do.
632 memset((void *)initrd_start, 0, initrd_end - initrd_start);
633 if (initrd_start < crashk_start)
634 free_initrd_mem(initrd_start, crashk_start);
635 if (initrd_end > crashk_end)
636 free_initrd_mem(crashk_end, initrd_end);
640 static inline bool kexec_free_initrd(void)
644 #endif /* CONFIG_KEXEC_CORE */
646 #ifdef CONFIG_BLK_DEV_RAM
647 static void __init populate_initrd_image(char *err)
653 unpack_to_rootfs(__initramfs_start, __initramfs_size);
655 printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
657 file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
661 written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
663 if (written != initrd_end - initrd_start)
664 pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
665 written, initrd_end - initrd_start);
668 #endif /* CONFIG_BLK_DEV_RAM */
670 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
672 /* Load the built in initramfs */
673 char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
675 panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
677 if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
680 if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
681 printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
683 printk(KERN_INFO "Unpacking initramfs...\n");
685 err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
687 #ifdef CONFIG_BLK_DEV_RAM
688 populate_initrd_image(err);
690 printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
696 * If the initrd region is overlapped with crashkernel reserved region,
697 * free only memory that is not part of crashkernel region.
699 if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
700 free_initrd_mem(initrd_start, initrd_end);
704 flush_delayed_fput();
707 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
708 static async_cookie_t initramfs_cookie;
710 void wait_for_initramfs(void)
712 if (!initramfs_cookie) {
714 * Something before rootfs_initcall wants to access
715 * the filesystem/initramfs. Probably a bug. Make a
716 * note, avoid deadlocking the machine, and let the
717 * caller's access fail as it used to.
719 pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
722 async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
724 EXPORT_SYMBOL_GPL(wait_for_initramfs);
726 static int __init populate_rootfs(void)
728 initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
730 if (!initramfs_async)
731 wait_for_initramfs();
734 rootfs_initcall(populate_rootfs);