x86/speculation/l1tf: Exempt zeroed PTEs from inversion
[muen/linux.git] / init / initramfs.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Many of the syscalls used in this file expect some of the arguments
4  * to be __user pointers not __kernel pointers.  To limit the sparse
5  * noise, turn off sparse checking for this file.
6  */
7 #ifdef __CHECKER__
8 #undef __CHECKER__
9 #warning "Sparse checking disabled for this file"
10 #endif
11
12 #include <linux/init.h>
13 #include <linux/fs.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/fcntl.h>
17 #include <linux/delay.h>
18 #include <linux/string.h>
19 #include <linux/dirent.h>
20 #include <linux/syscalls.h>
21 #include <linux/utime.h>
22 #include <linux/file.h>
23
24 static ssize_t __init xwrite(int fd, const char *p, size_t count)
25 {
26         ssize_t out = 0;
27
28         /* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
29         while (count) {
30                 ssize_t rv = ksys_write(fd, p, count);
31
32                 if (rv < 0) {
33                         if (rv == -EINTR || rv == -EAGAIN)
34                                 continue;
35                         return out ? out : rv;
36                 } else if (rv == 0)
37                         break;
38
39                 p += rv;
40                 out += rv;
41                 count -= rv;
42         }
43
44         return out;
45 }
46
47 static __initdata char *message;
48 static void __init error(char *x)
49 {
50         if (!message)
51                 message = x;
52 }
53
54 /* link hash */
55
56 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
57
58 static __initdata struct hash {
59         int ino, minor, major;
60         umode_t mode;
61         struct hash *next;
62         char name[N_ALIGN(PATH_MAX)];
63 } *head[32];
64
65 static inline int hash(int major, int minor, int ino)
66 {
67         unsigned long tmp = ino + minor + (major << 3);
68         tmp += tmp >> 5;
69         return tmp & 31;
70 }
71
72 static char __init *find_link(int major, int minor, int ino,
73                               umode_t mode, char *name)
74 {
75         struct hash **p, *q;
76         for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
77                 if ((*p)->ino != ino)
78                         continue;
79                 if ((*p)->minor != minor)
80                         continue;
81                 if ((*p)->major != major)
82                         continue;
83                 if (((*p)->mode ^ mode) & S_IFMT)
84                         continue;
85                 return (*p)->name;
86         }
87         q = kmalloc(sizeof(struct hash), GFP_KERNEL);
88         if (!q)
89                 panic("can't allocate link hash entry");
90         q->major = major;
91         q->minor = minor;
92         q->ino = ino;
93         q->mode = mode;
94         strcpy(q->name, name);
95         q->next = NULL;
96         *p = q;
97         return NULL;
98 }
99
100 static void __init free_hash(void)
101 {
102         struct hash **p, *q;
103         for (p = head; p < head + 32; p++) {
104                 while (*p) {
105                         q = *p;
106                         *p = q->next;
107                         kfree(q);
108                 }
109         }
110 }
111
112 static long __init do_utime(char *filename, time64_t mtime)
113 {
114         struct timespec64 t[2];
115
116         t[0].tv_sec = mtime;
117         t[0].tv_nsec = 0;
118         t[1].tv_sec = mtime;
119         t[1].tv_nsec = 0;
120
121         return do_utimes(AT_FDCWD, filename, t, AT_SYMLINK_NOFOLLOW);
122 }
123
124 static __initdata LIST_HEAD(dir_list);
125 struct dir_entry {
126         struct list_head list;
127         char *name;
128         time64_t mtime;
129 };
130
131 static void __init dir_add(const char *name, time64_t mtime)
132 {
133         struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
134         if (!de)
135                 panic("can't allocate dir_entry buffer");
136         INIT_LIST_HEAD(&de->list);
137         de->name = kstrdup(name, GFP_KERNEL);
138         de->mtime = mtime;
139         list_add(&de->list, &dir_list);
140 }
141
142 static void __init dir_utime(void)
143 {
144         struct dir_entry *de, *tmp;
145         list_for_each_entry_safe(de, tmp, &dir_list, list) {
146                 list_del(&de->list);
147                 do_utime(de->name, de->mtime);
148                 kfree(de->name);
149                 kfree(de);
150         }
151 }
152
153 static __initdata time64_t mtime;
154
155 /* cpio header parsing */
156
157 static __initdata unsigned long ino, major, minor, nlink;
158 static __initdata umode_t mode;
159 static __initdata unsigned long body_len, name_len;
160 static __initdata uid_t uid;
161 static __initdata gid_t gid;
162 static __initdata unsigned rdev;
163
164 static void __init parse_header(char *s)
165 {
166         unsigned long parsed[12];
167         char buf[9];
168         int i;
169
170         buf[8] = '\0';
171         for (i = 0, s += 6; i < 12; i++, s += 8) {
172                 memcpy(buf, s, 8);
173                 parsed[i] = simple_strtoul(buf, NULL, 16);
174         }
175         ino = parsed[0];
176         mode = parsed[1];
177         uid = parsed[2];
178         gid = parsed[3];
179         nlink = parsed[4];
180         mtime = parsed[5]; /* breaks in y2106 */
181         body_len = parsed[6];
182         major = parsed[7];
183         minor = parsed[8];
184         rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
185         name_len = parsed[11];
186 }
187
188 /* FSM */
189
190 static __initdata enum state {
191         Start,
192         Collect,
193         GotHeader,
194         SkipIt,
195         GotName,
196         CopyFile,
197         GotSymlink,
198         Reset
199 } state, next_state;
200
201 static __initdata char *victim;
202 static unsigned long byte_count __initdata;
203 static __initdata loff_t this_header, next_header;
204
205 static inline void __init eat(unsigned n)
206 {
207         victim += n;
208         this_header += n;
209         byte_count -= n;
210 }
211
212 static __initdata char *vcollected;
213 static __initdata char *collected;
214 static long remains __initdata;
215 static __initdata char *collect;
216
217 static void __init read_into(char *buf, unsigned size, enum state next)
218 {
219         if (byte_count >= size) {
220                 collected = victim;
221                 eat(size);
222                 state = next;
223         } else {
224                 collect = collected = buf;
225                 remains = size;
226                 next_state = next;
227                 state = Collect;
228         }
229 }
230
231 static __initdata char *header_buf, *symlink_buf, *name_buf;
232
233 static int __init do_start(void)
234 {
235         read_into(header_buf, 110, GotHeader);
236         return 0;
237 }
238
239 static int __init do_collect(void)
240 {
241         unsigned long n = remains;
242         if (byte_count < n)
243                 n = byte_count;
244         memcpy(collect, victim, n);
245         eat(n);
246         collect += n;
247         if ((remains -= n) != 0)
248                 return 1;
249         state = next_state;
250         return 0;
251 }
252
253 static int __init do_header(void)
254 {
255         if (memcmp(collected, "070707", 6)==0) {
256                 error("incorrect cpio method used: use -H newc option");
257                 return 1;
258         }
259         if (memcmp(collected, "070701", 6)) {
260                 error("no cpio magic");
261                 return 1;
262         }
263         parse_header(collected);
264         next_header = this_header + N_ALIGN(name_len) + body_len;
265         next_header = (next_header + 3) & ~3;
266         state = SkipIt;
267         if (name_len <= 0 || name_len > PATH_MAX)
268                 return 0;
269         if (S_ISLNK(mode)) {
270                 if (body_len > PATH_MAX)
271                         return 0;
272                 collect = collected = symlink_buf;
273                 remains = N_ALIGN(name_len) + body_len;
274                 next_state = GotSymlink;
275                 state = Collect;
276                 return 0;
277         }
278         if (S_ISREG(mode) || !body_len)
279                 read_into(name_buf, N_ALIGN(name_len), GotName);
280         return 0;
281 }
282
283 static int __init do_skip(void)
284 {
285         if (this_header + byte_count < next_header) {
286                 eat(byte_count);
287                 return 1;
288         } else {
289                 eat(next_header - this_header);
290                 state = next_state;
291                 return 0;
292         }
293 }
294
295 static int __init do_reset(void)
296 {
297         while (byte_count && *victim == '\0')
298                 eat(1);
299         if (byte_count && (this_header & 3))
300                 error("broken padding");
301         return 1;
302 }
303
304 static int __init maybe_link(void)
305 {
306         if (nlink >= 2) {
307                 char *old = find_link(major, minor, ino, mode, collected);
308                 if (old)
309                         return (ksys_link(old, collected) < 0) ? -1 : 1;
310         }
311         return 0;
312 }
313
314 static void __init clean_path(char *path, umode_t fmode)
315 {
316         struct kstat st;
317
318         if (!vfs_lstat(path, &st) && (st.mode ^ fmode) & S_IFMT) {
319                 if (S_ISDIR(st.mode))
320                         ksys_rmdir(path);
321                 else
322                         ksys_unlink(path);
323         }
324 }
325
326 static __initdata int wfd;
327
328 static int __init do_name(void)
329 {
330         state = SkipIt;
331         next_state = Reset;
332         if (strcmp(collected, "TRAILER!!!") == 0) {
333                 free_hash();
334                 return 0;
335         }
336         clean_path(collected, mode);
337         if (S_ISREG(mode)) {
338                 int ml = maybe_link();
339                 if (ml >= 0) {
340                         int openflags = O_WRONLY|O_CREAT;
341                         if (ml != 1)
342                                 openflags |= O_TRUNC;
343                         wfd = ksys_open(collected, openflags, mode);
344
345                         if (wfd >= 0) {
346                                 ksys_fchown(wfd, uid, gid);
347                                 ksys_fchmod(wfd, mode);
348                                 if (body_len)
349                                         ksys_ftruncate(wfd, body_len);
350                                 vcollected = kstrdup(collected, GFP_KERNEL);
351                                 state = CopyFile;
352                         }
353                 }
354         } else if (S_ISDIR(mode)) {
355                 ksys_mkdir(collected, mode);
356                 ksys_chown(collected, uid, gid);
357                 ksys_chmod(collected, mode);
358                 dir_add(collected, mtime);
359         } else if (S_ISBLK(mode) || S_ISCHR(mode) ||
360                    S_ISFIFO(mode) || S_ISSOCK(mode)) {
361                 if (maybe_link() == 0) {
362                         ksys_mknod(collected, mode, rdev);
363                         ksys_chown(collected, uid, gid);
364                         ksys_chmod(collected, mode);
365                         do_utime(collected, mtime);
366                 }
367         }
368         return 0;
369 }
370
371 static int __init do_copy(void)
372 {
373         if (byte_count >= body_len) {
374                 if (xwrite(wfd, victim, body_len) != body_len)
375                         error("write error");
376                 ksys_close(wfd);
377                 do_utime(vcollected, mtime);
378                 kfree(vcollected);
379                 eat(body_len);
380                 state = SkipIt;
381                 return 0;
382         } else {
383                 if (xwrite(wfd, victim, byte_count) != byte_count)
384                         error("write error");
385                 body_len -= byte_count;
386                 eat(byte_count);
387                 return 1;
388         }
389 }
390
391 static int __init do_symlink(void)
392 {
393         collected[N_ALIGN(name_len) + body_len] = '\0';
394         clean_path(collected, 0);
395         ksys_symlink(collected + N_ALIGN(name_len), collected);
396         ksys_lchown(collected, uid, gid);
397         do_utime(collected, mtime);
398         state = SkipIt;
399         next_state = Reset;
400         return 0;
401 }
402
403 static __initdata int (*actions[])(void) = {
404         [Start]         = do_start,
405         [Collect]       = do_collect,
406         [GotHeader]     = do_header,
407         [SkipIt]        = do_skip,
408         [GotName]       = do_name,
409         [CopyFile]      = do_copy,
410         [GotSymlink]    = do_symlink,
411         [Reset]         = do_reset,
412 };
413
414 static long __init write_buffer(char *buf, unsigned long len)
415 {
416         byte_count = len;
417         victim = buf;
418
419         while (!actions[state]())
420                 ;
421         return len - byte_count;
422 }
423
424 static long __init flush_buffer(void *bufv, unsigned long len)
425 {
426         char *buf = (char *) bufv;
427         long written;
428         long origLen = len;
429         if (message)
430                 return -1;
431         while ((written = write_buffer(buf, len)) < len && !message) {
432                 char c = buf[written];
433                 if (c == '0') {
434                         buf += written;
435                         len -= written;
436                         state = Start;
437                 } else if (c == 0) {
438                         buf += written;
439                         len -= written;
440                         state = Reset;
441                 } else
442                         error("junk in compressed archive");
443         }
444         return origLen;
445 }
446
447 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
448
449 #include <linux/decompress/generic.h>
450
451 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
452 {
453         long written;
454         decompress_fn decompress;
455         const char *compress_name;
456         static __initdata char msg_buf[64];
457
458         header_buf = kmalloc(110, GFP_KERNEL);
459         symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
460         name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
461
462         if (!header_buf || !symlink_buf || !name_buf)
463                 panic("can't allocate buffers");
464
465         state = Start;
466         this_header = 0;
467         message = NULL;
468         while (!message && len) {
469                 loff_t saved_offset = this_header;
470                 if (*buf == '0' && !(this_header & 3)) {
471                         state = Start;
472                         written = write_buffer(buf, len);
473                         buf += written;
474                         len -= written;
475                         continue;
476                 }
477                 if (!*buf) {
478                         buf++;
479                         len--;
480                         this_header++;
481                         continue;
482                 }
483                 this_header = 0;
484                 decompress = decompress_method(buf, len, &compress_name);
485                 pr_debug("Detected %s compressed data\n", compress_name);
486                 if (decompress) {
487                         int res = decompress(buf, len, NULL, flush_buffer, NULL,
488                                    &my_inptr, error);
489                         if (res)
490                                 error("decompressor failed");
491                 } else if (compress_name) {
492                         if (!message) {
493                                 snprintf(msg_buf, sizeof msg_buf,
494                                          "compression method %s not configured",
495                                          compress_name);
496                                 message = msg_buf;
497                         }
498                 } else
499                         error("junk in compressed archive");
500                 if (state != Reset)
501                         error("junk in compressed archive");
502                 this_header = saved_offset + my_inptr;
503                 buf += my_inptr;
504                 len -= my_inptr;
505         }
506         dir_utime();
507         kfree(name_buf);
508         kfree(symlink_buf);
509         kfree(header_buf);
510         return message;
511 }
512
513 static int __initdata do_retain_initrd;
514
515 static int __init retain_initrd_param(char *str)
516 {
517         if (*str)
518                 return 0;
519         do_retain_initrd = 1;
520         return 1;
521 }
522 __setup("retain_initrd", retain_initrd_param);
523
524 extern char __initramfs_start[];
525 extern unsigned long __initramfs_size;
526 #include <linux/initrd.h>
527 #include <linux/kexec.h>
528
529 static void __init free_initrd(void)
530 {
531 #ifdef CONFIG_KEXEC_CORE
532         unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
533         unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
534 #endif
535         if (do_retain_initrd)
536                 goto skip;
537
538 #ifdef CONFIG_KEXEC_CORE
539         /*
540          * If the initrd region is overlapped with crashkernel reserved region,
541          * free only memory that is not part of crashkernel region.
542          */
543         if (initrd_start < crashk_end && initrd_end > crashk_start) {
544                 /*
545                  * Initialize initrd memory region since the kexec boot does
546                  * not do.
547                  */
548                 memset((void *)initrd_start, 0, initrd_end - initrd_start);
549                 if (initrd_start < crashk_start)
550                         free_initrd_mem(initrd_start, crashk_start);
551                 if (initrd_end > crashk_end)
552                         free_initrd_mem(crashk_end, initrd_end);
553         } else
554 #endif
555                 free_initrd_mem(initrd_start, initrd_end);
556 skip:
557         initrd_start = 0;
558         initrd_end = 0;
559 }
560
561 #ifdef CONFIG_BLK_DEV_RAM
562 #define BUF_SIZE 1024
563 static void __init clean_rootfs(void)
564 {
565         int fd;
566         void *buf;
567         struct linux_dirent64 *dirp;
568         int num;
569
570         fd = ksys_open("/", O_RDONLY, 0);
571         WARN_ON(fd < 0);
572         if (fd < 0)
573                 return;
574         buf = kzalloc(BUF_SIZE, GFP_KERNEL);
575         WARN_ON(!buf);
576         if (!buf) {
577                 ksys_close(fd);
578                 return;
579         }
580
581         dirp = buf;
582         num = ksys_getdents64(fd, dirp, BUF_SIZE);
583         while (num > 0) {
584                 while (num > 0) {
585                         struct kstat st;
586                         int ret;
587
588                         ret = vfs_lstat(dirp->d_name, &st);
589                         WARN_ON_ONCE(ret);
590                         if (!ret) {
591                                 if (S_ISDIR(st.mode))
592                                         ksys_rmdir(dirp->d_name);
593                                 else
594                                         ksys_unlink(dirp->d_name);
595                         }
596
597                         num -= dirp->d_reclen;
598                         dirp = (void *)dirp + dirp->d_reclen;
599                 }
600                 dirp = buf;
601                 memset(buf, 0, BUF_SIZE);
602                 num = ksys_getdents64(fd, dirp, BUF_SIZE);
603         }
604
605         ksys_close(fd);
606         kfree(buf);
607 }
608 #endif
609
610 static int __init populate_rootfs(void)
611 {
612         /* Load the built in initramfs */
613         char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
614         if (err)
615                 panic("%s", err); /* Failed to decompress INTERNAL initramfs */
616         /* If available load the bootloader supplied initrd */
617         if (initrd_start && !IS_ENABLED(CONFIG_INITRAMFS_FORCE)) {
618 #ifdef CONFIG_BLK_DEV_RAM
619                 int fd;
620                 printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
621                 err = unpack_to_rootfs((char *)initrd_start,
622                         initrd_end - initrd_start);
623                 if (!err) {
624                         free_initrd();
625                         goto done;
626                 } else {
627                         clean_rootfs();
628                         unpack_to_rootfs(__initramfs_start, __initramfs_size);
629                 }
630                 printk(KERN_INFO "rootfs image is not initramfs (%s)"
631                                 "; looks like an initrd\n", err);
632                 fd = ksys_open("/initrd.image",
633                               O_WRONLY|O_CREAT, 0700);
634                 if (fd >= 0) {
635                         ssize_t written = xwrite(fd, (char *)initrd_start,
636                                                 initrd_end - initrd_start);
637
638                         if (written != initrd_end - initrd_start)
639                                 pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
640                                        written, initrd_end - initrd_start);
641
642                         ksys_close(fd);
643                         free_initrd();
644                 }
645         done:
646                 /* empty statement */;
647 #else
648                 printk(KERN_INFO "Unpacking initramfs...\n");
649                 err = unpack_to_rootfs((char *)initrd_start,
650                         initrd_end - initrd_start);
651                 if (err)
652                         printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
653                 free_initrd();
654 #endif
655         }
656         flush_delayed_fput();
657         /*
658          * Try loading default modules from initramfs.  This gives
659          * us a chance to load before device_initcalls.
660          */
661         load_default_modules();
662
663         return 0;
664 }
665 rootfs_initcall(populate_rootfs);