1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * binfmt_misc.c
4 *
5 * Copyright (C) 1997 Richard Günther
6 *
7 * binfmt_misc detects binaries via a magic or filename extension and invokes
8 * a specified wrapper. See Documentation/admin-guide/binfmt-misc.rst for more details.
9 */
10
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/hex.h>
16 #include <linux/init.h>
17 #include <linux/sched/mm.h>
18 #include <linux/magic.h>
19 #include <linux/binfmts.h>
20 #include <linux/slab.h>
21 #include <linux/ctype.h>
22 #include <linux/string_helpers.h>
23 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/namei.h>
26 #include <linux/mount.h>
27 #include <linux/fs_context.h>
28 #include <linux/syscalls.h>
29 #include <linux/fs.h>
30 #include <linux/uaccess.h>
31
32 #include "internal.h"
33
34 #ifdef DEBUG
35 # define USE_DEBUG 1
36 #else
37 # define USE_DEBUG 0
38 #endif
39
40 enum {
41 VERBOSE_STATUS = 1 /* make it zero to save 400 bytes kernel memory */
42 };
43
44 enum {Enabled, Magic};
45 #define MISC_FMT_PRESERVE_ARGV0 (1UL << 31)
46 #define MISC_FMT_OPEN_BINARY (1UL << 30)
47 #define MISC_FMT_CREDENTIALS (1UL << 29)
48 #define MISC_FMT_OPEN_FILE (1UL << 28)
49
50 typedef struct {
51 struct list_head list;
52 unsigned long flags; /* type, status, etc. */
53 int offset; /* offset of magic */
54 int size; /* size of magic/mask */
55 char *magic; /* magic or filename extension */
56 char *mask; /* mask, NULL for exact match */
57 const char *interpreter; /* filename of interpreter */
58 char *name;
59 struct dentry *dentry;
60 struct file *interp_file;
61 refcount_t users; /* sync removal with load_misc_binary() */
62 } Node;
63
64 static struct file_system_type bm_fs_type;
65
66 /*
67 * Max length of the register string. Determined by:
68 * - 7 delimiters
69 * - name: ~50 bytes
70 * - type: 1 byte
71 * - offset: 3 bytes (has to be smaller than BINPRM_BUF_SIZE)
72 * - magic: 128 bytes (512 in escaped form)
73 * - mask: 128 bytes (512 in escaped form)
74 * - interp: ~50 bytes
75 * - flags: 5 bytes
76 * Round that up a bit, and then back off to hold the internal data
77 * (like struct Node).
78 */
79 #define MAX_REGISTER_LENGTH 1920
80
81 /**
82 * search_binfmt_handler - search for a binary handler for @bprm
83 * @misc: handle to binfmt_misc instance
84 * @bprm: binary for which we are looking for a handler
85 *
86 * Search for a binary type handler for @bprm in the list of registered binary
87 * type handlers.
88 *
89 * Return: binary type list entry on success, NULL on failure
90 */
search_binfmt_handler(struct binfmt_misc * misc,struct linux_binprm * bprm)91 static Node *search_binfmt_handler(struct binfmt_misc *misc,
92 struct linux_binprm *bprm)
93 {
94 char *p = strrchr(bprm->interp, '.');
95 Node *e;
96
97 /* Walk all the registered handlers. */
98 list_for_each_entry(e, &misc->entries, list) {
99 char *s;
100 int j;
101
102 /* Make sure this one is currently enabled. */
103 if (!test_bit(Enabled, &e->flags))
104 continue;
105
106 /* Do matching based on extension if applicable. */
107 if (!test_bit(Magic, &e->flags)) {
108 if (p && !strcmp(e->magic, p + 1))
109 return e;
110 continue;
111 }
112
113 /* Do matching based on magic & mask. */
114 s = bprm->buf + e->offset;
115 if (e->mask) {
116 for (j = 0; j < e->size; j++)
117 if ((*s++ ^ e->magic[j]) & e->mask[j])
118 break;
119 } else {
120 for (j = 0; j < e->size; j++)
121 if ((*s++ ^ e->magic[j]))
122 break;
123 }
124 if (j == e->size)
125 return e;
126 }
127
128 return NULL;
129 }
130
131 /**
132 * get_binfmt_handler - try to find a binary type handler
133 * @misc: handle to binfmt_misc instance
134 * @bprm: binary for which we are looking for a handler
135 *
136 * Try to find a binfmt handler for the binary type. If one is found take a
137 * reference to protect against removal via bm_{entry,status}_write().
138 *
139 * Return: binary type list entry on success, NULL on failure
140 */
get_binfmt_handler(struct binfmt_misc * misc,struct linux_binprm * bprm)141 static Node *get_binfmt_handler(struct binfmt_misc *misc,
142 struct linux_binprm *bprm)
143 {
144 Node *e;
145
146 read_lock(&misc->entries_lock);
147 e = search_binfmt_handler(misc, bprm);
148 if (e)
149 refcount_inc(&e->users);
150 read_unlock(&misc->entries_lock);
151 return e;
152 }
153
154 /**
155 * put_binfmt_handler - put binary handler node
156 * @e: node to put
157 *
158 * Free node syncing with load_misc_binary() and defer final free to
159 * load_misc_binary() in case it is using the binary type handler we were
160 * requested to remove.
161 */
put_binfmt_handler(Node * e)162 static void put_binfmt_handler(Node *e)
163 {
164 if (refcount_dec_and_test(&e->users)) {
165 if (e->flags & MISC_FMT_OPEN_FILE)
166 filp_close(e->interp_file, NULL);
167 kfree(e);
168 }
169 }
170
171 /**
172 * load_binfmt_misc - load the binfmt_misc of the caller's user namespace
173 *
174 * To be called in load_misc_binary() to load the relevant struct binfmt_misc.
175 * If a user namespace doesn't have its own binfmt_misc mount it can make use
176 * of its ancestor's binfmt_misc handlers. This mimicks the behavior of
177 * pre-namespaced binfmt_misc where all registered binfmt_misc handlers where
178 * available to all user and user namespaces on the system.
179 *
180 * Return: the binfmt_misc instance of the caller's user namespace
181 */
load_binfmt_misc(void)182 static struct binfmt_misc *load_binfmt_misc(void)
183 {
184 const struct user_namespace *user_ns;
185 struct binfmt_misc *misc;
186
187 user_ns = current_user_ns();
188 while (user_ns) {
189 /* Pairs with smp_store_release() in bm_fill_super(). */
190 misc = smp_load_acquire(&user_ns->binfmt_misc);
191 if (misc)
192 return misc;
193
194 user_ns = user_ns->parent;
195 }
196
197 return &init_binfmt_misc;
198 }
199
200 /*
201 * the loader itself
202 */
load_misc_binary(struct linux_binprm * bprm)203 static int load_misc_binary(struct linux_binprm *bprm)
204 {
205 Node *fmt;
206 struct file *interp_file = NULL;
207 int retval = -ENOEXEC;
208 struct binfmt_misc *misc;
209
210 misc = load_binfmt_misc();
211 if (!misc->enabled)
212 return retval;
213
214 fmt = get_binfmt_handler(misc, bprm);
215 if (!fmt)
216 return retval;
217
218 /* Need to be able to load the file after exec */
219 retval = -ENOENT;
220 if (bprm->interp_flags & BINPRM_FLAGS_PATH_INACCESSIBLE)
221 goto ret;
222
223 if (fmt->flags & MISC_FMT_PRESERVE_ARGV0) {
224 bprm->interp_flags |= BINPRM_FLAGS_PRESERVE_ARGV0;
225 } else {
226 retval = remove_arg_zero(bprm);
227 if (retval)
228 goto ret;
229 }
230
231 if (fmt->flags & MISC_FMT_OPEN_BINARY)
232 bprm->have_execfd = 1;
233
234 /* make argv[1] be the path to the binary */
235 retval = copy_string_kernel(bprm->interp, bprm);
236 if (retval < 0)
237 goto ret;
238 bprm->argc++;
239
240 /* add the interp as argv[0] */
241 retval = copy_string_kernel(fmt->interpreter, bprm);
242 if (retval < 0)
243 goto ret;
244 bprm->argc++;
245
246 /* Update interp in case binfmt_script needs it. */
247 retval = bprm_change_interp(fmt->interpreter, bprm);
248 if (retval < 0)
249 goto ret;
250
251 if (fmt->flags & MISC_FMT_OPEN_FILE) {
252 interp_file = file_clone_open(fmt->interp_file);
253 if (!IS_ERR(interp_file))
254 deny_write_access(interp_file);
255 } else {
256 interp_file = open_exec(fmt->interpreter);
257 }
258 retval = PTR_ERR(interp_file);
259 if (IS_ERR(interp_file))
260 goto ret;
261
262 bprm->interpreter = interp_file;
263 if (fmt->flags & MISC_FMT_CREDENTIALS)
264 bprm->execfd_creds = 1;
265
266 retval = 0;
267 ret:
268
269 /*
270 * If we actually put the node here all concurrent calls to
271 * load_misc_binary() will have finished. We also know
272 * that for the refcount to be zero someone must have concurently
273 * removed the binary type handler from the list and it's our job to
274 * free it.
275 */
276 put_binfmt_handler(fmt);
277
278 return retval;
279 }
280
281 /* Command parsers */
282
283 /*
284 * parses and copies one argument enclosed in del from *sp to *dp,
285 * recognising the \x special.
286 * returns pointer to the copied argument or NULL in case of an
287 * error (and sets err) or null argument length.
288 */
scanarg(char * s,char del)289 static char *scanarg(char *s, char del)
290 {
291 char c;
292
293 while ((c = *s++) != del) {
294 if (c == '\\' && *s == 'x') {
295 s++;
296 if (!isxdigit(*s++))
297 return NULL;
298 if (!isxdigit(*s++))
299 return NULL;
300 }
301 }
302 s[-1] ='\0';
303 return s;
304 }
305
check_special_flags(char * sfs,Node * e)306 static char *check_special_flags(char *sfs, Node *e)
307 {
308 char *p = sfs;
309 int cont = 1;
310
311 /* special flags */
312 while (cont) {
313 switch (*p) {
314 case 'P':
315 pr_debug("register: flag: P (preserve argv0)\n");
316 p++;
317 e->flags |= MISC_FMT_PRESERVE_ARGV0;
318 break;
319 case 'O':
320 pr_debug("register: flag: O (open binary)\n");
321 p++;
322 e->flags |= MISC_FMT_OPEN_BINARY;
323 break;
324 case 'C':
325 pr_debug("register: flag: C (preserve creds)\n");
326 p++;
327 /* this flags also implies the
328 open-binary flag */
329 e->flags |= (MISC_FMT_CREDENTIALS |
330 MISC_FMT_OPEN_BINARY);
331 break;
332 case 'F':
333 pr_debug("register: flag: F: open interpreter file now\n");
334 p++;
335 e->flags |= MISC_FMT_OPEN_FILE;
336 break;
337 default:
338 cont = 0;
339 }
340 }
341
342 return p;
343 }
344
345 /*
346 * This registers a new binary format, it recognises the syntax
347 * ':name:type:offset:magic:mask:interpreter:flags'
348 * where the ':' is the IFS, that can be chosen with the first char
349 */
create_entry(const char __user * buffer,size_t count)350 static Node *create_entry(const char __user *buffer, size_t count)
351 {
352 Node *e;
353 int memsize, err;
354 char *buf, *p;
355 char del;
356
357 pr_debug("register: received %zu bytes\n", count);
358
359 /* some sanity checks */
360 err = -EINVAL;
361 if ((count < 11) || (count > MAX_REGISTER_LENGTH))
362 goto out;
363
364 err = -ENOMEM;
365 memsize = sizeof(Node) + count + 8;
366 e = kmalloc(memsize, GFP_KERNEL_ACCOUNT);
367 if (!e)
368 goto out;
369
370 p = buf = (char *)e + sizeof(Node);
371
372 memset(e, 0, sizeof(Node));
373 if (copy_from_user(buf, buffer, count))
374 goto efault;
375
376 del = *p++; /* delimeter */
377
378 pr_debug("register: delim: %#x {%c}\n", del, del);
379
380 /* Pad the buffer with the delim to simplify parsing below. */
381 memset(buf + count, del, 8);
382
383 /* Parse the 'name' field. */
384 e->name = p;
385 p = strchr(p, del);
386 if (!p)
387 goto einval;
388 *p++ = '\0';
389 if (!e->name[0] ||
390 !strcmp(e->name, ".") ||
391 !strcmp(e->name, "..") ||
392 strchr(e->name, '/'))
393 goto einval;
394
395 pr_debug("register: name: {%s}\n", e->name);
396
397 /* Parse the 'type' field. */
398 switch (*p++) {
399 case 'E':
400 pr_debug("register: type: E (extension)\n");
401 e->flags = 1 << Enabled;
402 break;
403 case 'M':
404 pr_debug("register: type: M (magic)\n");
405 e->flags = (1 << Enabled) | (1 << Magic);
406 break;
407 default:
408 goto einval;
409 }
410 if (*p++ != del)
411 goto einval;
412
413 if (test_bit(Magic, &e->flags)) {
414 /* Handle the 'M' (magic) format. */
415 char *s;
416
417 /* Parse the 'offset' field. */
418 s = strchr(p, del);
419 if (!s)
420 goto einval;
421 *s = '\0';
422 if (p != s) {
423 int r = kstrtoint(p, 10, &e->offset);
424 if (r != 0 || e->offset < 0)
425 goto einval;
426 }
427 p = s;
428 if (*p++)
429 goto einval;
430 pr_debug("register: offset: %#x\n", e->offset);
431
432 /* Parse the 'magic' field. */
433 e->magic = p;
434 p = scanarg(p, del);
435 if (!p)
436 goto einval;
437 if (!e->magic[0])
438 goto einval;
439 if (USE_DEBUG)
440 print_hex_dump_bytes(
441 KBUILD_MODNAME ": register: magic[raw]: ",
442 DUMP_PREFIX_NONE, e->magic, p - e->magic);
443
444 /* Parse the 'mask' field. */
445 e->mask = p;
446 p = scanarg(p, del);
447 if (!p)
448 goto einval;
449 if (!e->mask[0]) {
450 e->mask = NULL;
451 pr_debug("register: mask[raw]: none\n");
452 } else if (USE_DEBUG)
453 print_hex_dump_bytes(
454 KBUILD_MODNAME ": register: mask[raw]: ",
455 DUMP_PREFIX_NONE, e->mask, p - e->mask);
456
457 /*
458 * Decode the magic & mask fields.
459 * Note: while we might have accepted embedded NUL bytes from
460 * above, the unescape helpers here will stop at the first one
461 * it encounters.
462 */
463 e->size = string_unescape_inplace(e->magic, UNESCAPE_HEX);
464 if (e->mask &&
465 string_unescape_inplace(e->mask, UNESCAPE_HEX) != e->size)
466 goto einval;
467 if (e->size > BINPRM_BUF_SIZE ||
468 BINPRM_BUF_SIZE - e->size < e->offset)
469 goto einval;
470 pr_debug("register: magic/mask length: %i\n", e->size);
471 if (USE_DEBUG) {
472 print_hex_dump_bytes(
473 KBUILD_MODNAME ": register: magic[decoded]: ",
474 DUMP_PREFIX_NONE, e->magic, e->size);
475
476 if (e->mask) {
477 int i;
478 char *masked = kmalloc(e->size, GFP_KERNEL_ACCOUNT);
479
480 print_hex_dump_bytes(
481 KBUILD_MODNAME ": register: mask[decoded]: ",
482 DUMP_PREFIX_NONE, e->mask, e->size);
483
484 if (masked) {
485 for (i = 0; i < e->size; ++i)
486 masked[i] = e->magic[i] & e->mask[i];
487 print_hex_dump_bytes(
488 KBUILD_MODNAME ": register: magic[masked]: ",
489 DUMP_PREFIX_NONE, masked, e->size);
490
491 kfree(masked);
492 }
493 }
494 }
495 } else {
496 /* Handle the 'E' (extension) format. */
497
498 /* Skip the 'offset' field. */
499 p = strchr(p, del);
500 if (!p)
501 goto einval;
502 *p++ = '\0';
503
504 /* Parse the 'magic' field. */
505 e->magic = p;
506 p = strchr(p, del);
507 if (!p)
508 goto einval;
509 *p++ = '\0';
510 if (!e->magic[0] || strchr(e->magic, '/'))
511 goto einval;
512 pr_debug("register: extension: {%s}\n", e->magic);
513
514 /* Skip the 'mask' field. */
515 p = strchr(p, del);
516 if (!p)
517 goto einval;
518 *p++ = '\0';
519 }
520
521 /* Parse the 'interpreter' field. */
522 e->interpreter = p;
523 p = strchr(p, del);
524 if (!p)
525 goto einval;
526 *p++ = '\0';
527 if (!e->interpreter[0])
528 goto einval;
529 pr_debug("register: interpreter: {%s}\n", e->interpreter);
530
531 /* Parse the 'flags' field. */
532 p = check_special_flags(p, e);
533 if (*p == '\n')
534 p++;
535 if (p != buf + count)
536 goto einval;
537
538 return e;
539
540 out:
541 return ERR_PTR(err);
542
543 efault:
544 kfree(e);
545 return ERR_PTR(-EFAULT);
546 einval:
547 kfree(e);
548 return ERR_PTR(-EINVAL);
549 }
550
551 /*
552 * Set status of entry/binfmt_misc:
553 * '1' enables, '0' disables and '-1' clears entry/binfmt_misc
554 */
parse_command(const char __user * buffer,size_t count)555 static int parse_command(const char __user *buffer, size_t count)
556 {
557 char s[4];
558
559 if (count > 3)
560 return -EINVAL;
561 if (copy_from_user(s, buffer, count))
562 return -EFAULT;
563 if (!count)
564 return 0;
565 if (s[count - 1] == '\n')
566 count--;
567 if (count == 1 && s[0] == '0')
568 return 1;
569 if (count == 1 && s[0] == '1')
570 return 2;
571 if (count == 2 && s[0] == '-' && s[1] == '1')
572 return 3;
573 return -EINVAL;
574 }
575
576 /* generic stuff */
577
entry_status(Node * e,char * page)578 static void entry_status(Node *e, char *page)
579 {
580 char *dp = page;
581 const char *status = "disabled";
582
583 if (test_bit(Enabled, &e->flags))
584 status = "enabled";
585
586 if (!VERBOSE_STATUS) {
587 sprintf(page, "%s\n", status);
588 return;
589 }
590
591 dp += sprintf(dp, "%s\ninterpreter %s\n", status, e->interpreter);
592
593 /* print the special flags */
594 dp += sprintf(dp, "flags: ");
595 if (e->flags & MISC_FMT_PRESERVE_ARGV0)
596 *dp++ = 'P';
597 if (e->flags & MISC_FMT_OPEN_BINARY)
598 *dp++ = 'O';
599 if (e->flags & MISC_FMT_CREDENTIALS)
600 *dp++ = 'C';
601 if (e->flags & MISC_FMT_OPEN_FILE)
602 *dp++ = 'F';
603 *dp++ = '\n';
604
605 if (!test_bit(Magic, &e->flags)) {
606 sprintf(dp, "extension .%s\n", e->magic);
607 } else {
608 dp += sprintf(dp, "offset %i\nmagic ", e->offset);
609 dp = bin2hex(dp, e->magic, e->size);
610 if (e->mask) {
611 dp += sprintf(dp, "\nmask ");
612 dp = bin2hex(dp, e->mask, e->size);
613 }
614 *dp++ = '\n';
615 *dp = '\0';
616 }
617 }
618
bm_get_inode(struct super_block * sb,int mode)619 static struct inode *bm_get_inode(struct super_block *sb, int mode)
620 {
621 struct inode *inode = new_inode(sb);
622
623 if (inode) {
624 inode->i_ino = get_next_ino();
625 inode->i_mode = mode;
626 simple_inode_init_ts(inode);
627 }
628 return inode;
629 }
630
631 /**
632 * i_binfmt_misc - retrieve struct binfmt_misc from a binfmt_misc inode
633 * @inode: inode of the relevant binfmt_misc instance
634 *
635 * This helper retrieves struct binfmt_misc from a binfmt_misc inode. This can
636 * be done without any memory barriers because we are guaranteed that
637 * user_ns->binfmt_misc is fully initialized. It was fully initialized when the
638 * binfmt_misc mount was first created.
639 *
640 * Return: struct binfmt_misc of the relevant binfmt_misc instance
641 */
i_binfmt_misc(struct inode * inode)642 static struct binfmt_misc *i_binfmt_misc(struct inode *inode)
643 {
644 return inode->i_sb->s_user_ns->binfmt_misc;
645 }
646
647 /**
648 * bm_evict_inode - cleanup data associated with @inode
649 * @inode: inode to which the data is attached
650 *
651 * Cleanup the binary type handler data associated with @inode if a binary type
652 * entry is removed or the filesystem is unmounted and the super block is
653 * shutdown.
654 *
655 * If the ->evict call was not caused by a super block shutdown but by a write
656 * to remove the entry or all entries via bm_{entry,status}_write() the entry
657 * will have already been removed from the list. We keep the list_empty() check
658 * to make that explicit.
659 */
bm_evict_inode(struct inode * inode)660 static void bm_evict_inode(struct inode *inode)
661 {
662 Node *e = inode->i_private;
663
664 clear_inode(inode);
665
666 if (e) {
667 struct binfmt_misc *misc;
668
669 misc = i_binfmt_misc(inode);
670 write_lock(&misc->entries_lock);
671 if (!list_empty(&e->list))
672 list_del_init(&e->list);
673 write_unlock(&misc->entries_lock);
674 put_binfmt_handler(e);
675 }
676 }
677
678 /**
679 * remove_binfmt_handler - remove a binary type handler
680 * @misc: handle to binfmt_misc instance
681 * @e: binary type handler to remove
682 *
683 * Remove a binary type handler from the list of binary type handlers and
684 * remove its associated dentry. This is called from
685 * binfmt_{entry,status}_write(). In the future, we might want to think about
686 * adding a proper ->unlink() method to binfmt_misc instead of forcing caller's
687 * to use writes to files in order to delete binary type handlers. But it has
688 * worked for so long that it's not a pressing issue.
689 */
remove_binfmt_handler(struct binfmt_misc * misc,Node * e)690 static void remove_binfmt_handler(struct binfmt_misc *misc, Node *e)
691 {
692 write_lock(&misc->entries_lock);
693 list_del_init(&e->list);
694 write_unlock(&misc->entries_lock);
695 locked_recursive_removal(e->dentry, NULL);
696 }
697
698 /* /<entry> */
699
700 static ssize_t
bm_entry_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)701 bm_entry_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
702 {
703 Node *e = file_inode(file)->i_private;
704 ssize_t res;
705 char *page;
706
707 page = (char *) __get_free_page(GFP_KERNEL);
708 if (!page)
709 return -ENOMEM;
710
711 entry_status(e, page);
712
713 res = simple_read_from_buffer(buf, nbytes, ppos, page, strlen(page));
714
715 free_page((unsigned long) page);
716 return res;
717 }
718
bm_entry_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)719 static ssize_t bm_entry_write(struct file *file, const char __user *buffer,
720 size_t count, loff_t *ppos)
721 {
722 struct inode *inode = file_inode(file);
723 Node *e = inode->i_private;
724 int res = parse_command(buffer, count);
725
726 switch (res) {
727 case 1:
728 /* Disable this handler. */
729 clear_bit(Enabled, &e->flags);
730 break;
731 case 2:
732 /* Enable this handler. */
733 set_bit(Enabled, &e->flags);
734 break;
735 case 3:
736 /* Delete this handler. */
737 inode = d_inode(inode->i_sb->s_root);
738 inode_lock_nested(inode, I_MUTEX_PARENT);
739
740 /*
741 * In order to add new element or remove elements from the list
742 * via bm_{entry,register,status}_write() inode_lock() on the
743 * root inode must be held.
744 * The lock is exclusive ensuring that the list can't be
745 * modified. Only load_misc_binary() can access but does so
746 * read-only. So we only need to take the write lock when we
747 * actually remove the entry from the list.
748 */
749 if (!list_empty(&e->list))
750 remove_binfmt_handler(i_binfmt_misc(inode), e);
751
752 inode_unlock(inode);
753 break;
754 default:
755 return res;
756 }
757
758 return count;
759 }
760
761 static const struct file_operations bm_entry_operations = {
762 .read = bm_entry_read,
763 .write = bm_entry_write,
764 .llseek = default_llseek,
765 };
766
767 /* /register */
768
769 /* add to filesystem */
add_entry(Node * e,struct super_block * sb)770 static int add_entry(Node *e, struct super_block *sb)
771 {
772 struct dentry *dentry = simple_start_creating(sb->s_root, e->name);
773 struct inode *inode;
774 struct binfmt_misc *misc;
775
776 if (IS_ERR(dentry))
777 return PTR_ERR(dentry);
778
779 inode = bm_get_inode(sb, S_IFREG | 0644);
780 if (unlikely(!inode)) {
781 simple_done_creating(dentry);
782 return -ENOMEM;
783 }
784
785 refcount_set(&e->users, 1);
786 e->dentry = dentry;
787 inode->i_private = e;
788 inode->i_fop = &bm_entry_operations;
789
790 d_make_persistent(dentry, inode);
791 misc = i_binfmt_misc(inode);
792 write_lock(&misc->entries_lock);
793 list_add(&e->list, &misc->entries);
794 write_unlock(&misc->entries_lock);
795 simple_done_creating(dentry);
796 return 0;
797 }
798
bm_register_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)799 static ssize_t bm_register_write(struct file *file, const char __user *buffer,
800 size_t count, loff_t *ppos)
801 {
802 Node *e;
803 struct super_block *sb = file_inode(file)->i_sb;
804 int err = 0;
805 struct file *f = NULL;
806
807 e = create_entry(buffer, count);
808
809 if (IS_ERR(e))
810 return PTR_ERR(e);
811
812 if (e->flags & MISC_FMT_OPEN_FILE) {
813 /*
814 * Now that we support unprivileged binfmt_misc mounts make
815 * sure we use the credentials that the register @file was
816 * opened with to also open the interpreter. Before that this
817 * didn't matter much as only a privileged process could open
818 * the register file.
819 */
820 scoped_with_creds(file->f_cred)
821 f = open_exec(e->interpreter);
822 if (IS_ERR(f)) {
823 pr_notice("register: failed to install interpreter file %s\n",
824 e->interpreter);
825 kfree(e);
826 return PTR_ERR(f);
827 }
828 e->interp_file = f;
829 }
830
831 err = add_entry(e, sb);
832 if (err) {
833 if (f) {
834 exe_file_allow_write_access(f);
835 filp_close(f, NULL);
836 }
837 kfree(e);
838 return err;
839 }
840 return count;
841 }
842
843 static const struct file_operations bm_register_operations = {
844 .write = bm_register_write,
845 .llseek = noop_llseek,
846 };
847
848 /* /status */
849
850 static ssize_t
bm_status_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)851 bm_status_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
852 {
853 struct binfmt_misc *misc;
854 char *s;
855
856 misc = i_binfmt_misc(file_inode(file));
857 s = misc->enabled ? "enabled\n" : "disabled\n";
858 return simple_read_from_buffer(buf, nbytes, ppos, s, strlen(s));
859 }
860
bm_status_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)861 static ssize_t bm_status_write(struct file *file, const char __user *buffer,
862 size_t count, loff_t *ppos)
863 {
864 struct binfmt_misc *misc;
865 int res = parse_command(buffer, count);
866 Node *e, *next;
867 struct inode *inode;
868
869 misc = i_binfmt_misc(file_inode(file));
870 switch (res) {
871 case 1:
872 /* Disable all handlers. */
873 misc->enabled = false;
874 break;
875 case 2:
876 /* Enable all handlers. */
877 misc->enabled = true;
878 break;
879 case 3:
880 /* Delete all handlers. */
881 inode = d_inode(file_inode(file)->i_sb->s_root);
882 inode_lock_nested(inode, I_MUTEX_PARENT);
883
884 /*
885 * In order to add new element or remove elements from the list
886 * via bm_{entry,register,status}_write() inode_lock() on the
887 * root inode must be held.
888 * The lock is exclusive ensuring that the list can't be
889 * modified. Only load_misc_binary() can access but does so
890 * read-only. So we only need to take the write lock when we
891 * actually remove the entry from the list.
892 */
893 list_for_each_entry_safe(e, next, &misc->entries, list)
894 remove_binfmt_handler(misc, e);
895
896 inode_unlock(inode);
897 break;
898 default:
899 return res;
900 }
901
902 return count;
903 }
904
905 static const struct file_operations bm_status_operations = {
906 .read = bm_status_read,
907 .write = bm_status_write,
908 .llseek = default_llseek,
909 };
910
911 /* Superblock handling */
912
bm_put_super(struct super_block * sb)913 static void bm_put_super(struct super_block *sb)
914 {
915 struct user_namespace *user_ns = sb->s_fs_info;
916
917 sb->s_fs_info = NULL;
918 put_user_ns(user_ns);
919 }
920
921 static const struct super_operations s_ops = {
922 .statfs = simple_statfs,
923 .evict_inode = bm_evict_inode,
924 .put_super = bm_put_super,
925 };
926
bm_fill_super(struct super_block * sb,struct fs_context * fc)927 static int bm_fill_super(struct super_block *sb, struct fs_context *fc)
928 {
929 int err;
930 struct user_namespace *user_ns = sb->s_user_ns;
931 struct binfmt_misc *misc;
932 static const struct tree_descr bm_files[] = {
933 [2] = {"status", &bm_status_operations, S_IWUSR|S_IRUGO},
934 [3] = {"register", &bm_register_operations, S_IWUSR},
935 /* last one */ {""}
936 };
937
938 if (WARN_ON(user_ns != current_user_ns()))
939 return -EINVAL;
940
941 /*
942 * Lazily allocate a new binfmt_misc instance for this namespace, i.e.
943 * do it here during the first mount of binfmt_misc. We don't need to
944 * waste memory for every user namespace allocation. It's likely much
945 * more common to not mount a separate binfmt_misc instance than it is
946 * to mount one.
947 *
948 * While multiple superblocks can exist they are keyed by userns in
949 * s_fs_info for binfmt_misc. Hence, the vfs guarantees that
950 * bm_fill_super() is called exactly once whenever a binfmt_misc
951 * superblock for a userns is created. This in turn lets us conclude
952 * that when a binfmt_misc superblock is created for the first time for
953 * a userns there's no one racing us. Therefore we don't need any
954 * barriers when we dereference binfmt_misc.
955 */
956 misc = user_ns->binfmt_misc;
957 if (!misc) {
958 /*
959 * If it turns out that most user namespaces actually want to
960 * register their own binary type handler and therefore all
961 * create their own separate binfmt_misc mounts we should
962 * consider turning this into a kmem cache.
963 */
964 misc = kzalloc_obj(struct binfmt_misc);
965 if (!misc)
966 return -ENOMEM;
967
968 INIT_LIST_HEAD(&misc->entries);
969 rwlock_init(&misc->entries_lock);
970
971 /* Pairs with smp_load_acquire() in load_binfmt_misc(). */
972 smp_store_release(&user_ns->binfmt_misc, misc);
973 }
974
975 /*
976 * When the binfmt_misc superblock for this userns is shutdown
977 * ->enabled might have been set to false and we don't reinitialize
978 * ->enabled again in put_super() as someone might already be mounting
979 * binfmt_misc again. It also would be pointless since by the time
980 * ->put_super() is called we know that the binary type list for this
981 * bintfmt_misc mount is empty making load_misc_binary() return
982 * -ENOEXEC independent of whether ->enabled is true. Instead, if
983 * someone mounts binfmt_misc for the first time or again we simply
984 * reset ->enabled to true.
985 */
986 misc->enabled = true;
987
988 err = simple_fill_super(sb, BINFMTFS_MAGIC, bm_files);
989 if (!err)
990 sb->s_op = &s_ops;
991 return err;
992 }
993
bm_free(struct fs_context * fc)994 static void bm_free(struct fs_context *fc)
995 {
996 if (fc->s_fs_info)
997 put_user_ns(fc->s_fs_info);
998 }
999
bm_get_tree(struct fs_context * fc)1000 static int bm_get_tree(struct fs_context *fc)
1001 {
1002 return get_tree_keyed(fc, bm_fill_super, get_user_ns(fc->user_ns));
1003 }
1004
1005 static const struct fs_context_operations bm_context_ops = {
1006 .free = bm_free,
1007 .get_tree = bm_get_tree,
1008 };
1009
bm_init_fs_context(struct fs_context * fc)1010 static int bm_init_fs_context(struct fs_context *fc)
1011 {
1012 fc->ops = &bm_context_ops;
1013 return 0;
1014 }
1015
1016 static struct linux_binfmt misc_format = {
1017 .module = THIS_MODULE,
1018 .load_binary = load_misc_binary,
1019 };
1020
1021 static struct file_system_type bm_fs_type = {
1022 .owner = THIS_MODULE,
1023 .name = "binfmt_misc",
1024 .init_fs_context = bm_init_fs_context,
1025 .fs_flags = FS_USERNS_MOUNT,
1026 .kill_sb = kill_anon_super,
1027 };
1028 MODULE_ALIAS_FS("binfmt_misc");
1029
init_misc_binfmt(void)1030 static int __init init_misc_binfmt(void)
1031 {
1032 int err = register_filesystem(&bm_fs_type);
1033 if (!err)
1034 insert_binfmt(&misc_format);
1035 return err;
1036 }
1037
exit_misc_binfmt(void)1038 static void __exit exit_misc_binfmt(void)
1039 {
1040 unregister_binfmt(&misc_format);
1041 unregister_filesystem(&bm_fs_type);
1042 }
1043
1044 core_initcall(init_misc_binfmt);
1045 module_exit(exit_misc_binfmt);
1046 MODULE_DESCRIPTION("Kernel support for miscellaneous binaries");
1047 MODULE_LICENSE("GPL");
1048