mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / binfmt_misc.c
blob8311e8ed76de37542d35dffc90c039452572f50e
1 /*
2 * binfmt_misc.c
4 * Copyright (C) 1997 Richard Günther
6 * binfmt_misc detects binaries via a magic or filename extension and invokes
7 * a specified wrapper. See Documentation/binfmt_misc.txt for more details.
8 */
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/sched/mm.h>
16 #include <linux/magic.h>
17 #include <linux/binfmts.h>
18 #include <linux/slab.h>
19 #include <linux/ctype.h>
20 #include <linux/string_helpers.h>
21 #include <linux/file.h>
22 #include <linux/pagemap.h>
23 #include <linux/namei.h>
24 #include <linux/mount.h>
25 #include <linux/syscalls.h>
26 #include <linux/fs.h>
27 #include <linux/uaccess.h>
29 #include "internal.h"
31 #ifdef DEBUG
32 # define USE_DEBUG 1
33 #else
34 # define USE_DEBUG 0
35 #endif
37 enum {
38 VERBOSE_STATUS = 1 /* make it zero to save 400 bytes kernel memory */
41 static LIST_HEAD(entries);
42 static int enabled = 1;
44 enum {Enabled, Magic};
45 #define MISC_FMT_PRESERVE_ARGV0 (1 << 31)
46 #define MISC_FMT_OPEN_BINARY (1 << 30)
47 #define MISC_FMT_CREDENTIALS (1 << 29)
48 #define MISC_FMT_OPEN_FILE (1 << 28)
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 } Node;
63 static DEFINE_RWLOCK(entries_lock);
64 static struct file_system_type bm_fs_type;
65 static struct vfsmount *bm_mnt;
66 static int entry_count;
69 * Max length of the register string. Determined by:
70 * - 7 delimiters
71 * - name: ~50 bytes
72 * - type: 1 byte
73 * - offset: 3 bytes (has to be smaller than BINPRM_BUF_SIZE)
74 * - magic: 128 bytes (512 in escaped form)
75 * - mask: 128 bytes (512 in escaped form)
76 * - interp: ~50 bytes
77 * - flags: 5 bytes
78 * Round that up a bit, and then back off to hold the internal data
79 * (like struct Node).
81 #define MAX_REGISTER_LENGTH 1920
84 * Check if we support the binfmt
85 * if we do, return the node, else NULL
86 * locking is done in load_misc_binary
88 static Node *check_file(struct linux_binprm *bprm)
90 char *p = strrchr(bprm->interp, '.');
91 struct list_head *l;
93 /* Walk all the registered handlers. */
94 list_for_each(l, &entries) {
95 Node *e = list_entry(l, Node, list);
96 char *s;
97 int j;
99 /* Make sure this one is currently enabled. */
100 if (!test_bit(Enabled, &e->flags))
101 continue;
103 /* Do matching based on extension if applicable. */
104 if (!test_bit(Magic, &e->flags)) {
105 if (p && !strcmp(e->magic, p + 1))
106 return e;
107 continue;
110 /* Do matching based on magic & mask. */
111 s = bprm->buf + e->offset;
112 if (e->mask) {
113 for (j = 0; j < e->size; j++)
114 if ((*s++ ^ e->magic[j]) & e->mask[j])
115 break;
116 } else {
117 for (j = 0; j < e->size; j++)
118 if ((*s++ ^ e->magic[j]))
119 break;
121 if (j == e->size)
122 return e;
124 return NULL;
128 * the loader itself
130 static int load_misc_binary(struct linux_binprm *bprm)
132 Node *fmt;
133 struct file *interp_file = NULL;
134 int retval;
135 int fd_binary = -1;
137 retval = -ENOEXEC;
138 if (!enabled)
139 return retval;
141 /* to keep locking time low, we copy the interpreter string */
142 read_lock(&entries_lock);
143 fmt = check_file(bprm);
144 if (fmt)
145 dget(fmt->dentry);
146 read_unlock(&entries_lock);
147 if (!fmt)
148 return retval;
150 /* Need to be able to load the file after exec */
151 retval = -ENOENT;
152 if (bprm->interp_flags & BINPRM_FLAGS_PATH_INACCESSIBLE)
153 goto ret;
155 if (!(fmt->flags & MISC_FMT_PRESERVE_ARGV0)) {
156 retval = remove_arg_zero(bprm);
157 if (retval)
158 goto ret;
161 if (fmt->flags & MISC_FMT_OPEN_BINARY) {
163 /* if the binary should be opened on behalf of the
164 * interpreter than keep it open and assign descriptor
165 * to it
167 fd_binary = get_unused_fd_flags(0);
168 if (fd_binary < 0) {
169 retval = fd_binary;
170 goto ret;
172 fd_install(fd_binary, bprm->file);
174 /* if the binary is not readable than enforce mm->dumpable=0
175 regardless of the interpreter's permissions */
176 would_dump(bprm, bprm->file);
178 allow_write_access(bprm->file);
179 bprm->file = NULL;
181 /* mark the bprm that fd should be passed to interp */
182 bprm->interp_flags |= BINPRM_FLAGS_EXECFD;
183 bprm->interp_data = fd_binary;
185 } else {
186 allow_write_access(bprm->file);
187 fput(bprm->file);
188 bprm->file = NULL;
190 /* make argv[1] be the path to the binary */
191 retval = copy_strings_kernel(1, &bprm->interp, bprm);
192 if (retval < 0)
193 goto error;
194 bprm->argc++;
196 /* add the interp as argv[0] */
197 retval = copy_strings_kernel(1, &fmt->interpreter, bprm);
198 if (retval < 0)
199 goto error;
200 bprm->argc++;
202 /* Update interp in case binfmt_script needs it. */
203 retval = bprm_change_interp(fmt->interpreter, bprm);
204 if (retval < 0)
205 goto error;
207 if (fmt->flags & MISC_FMT_OPEN_FILE) {
208 interp_file = filp_clone_open(fmt->interp_file);
209 if (!IS_ERR(interp_file))
210 deny_write_access(interp_file);
211 } else {
212 interp_file = open_exec(fmt->interpreter);
214 retval = PTR_ERR(interp_file);
215 if (IS_ERR(interp_file))
216 goto error;
218 bprm->file = interp_file;
219 if (fmt->flags & MISC_FMT_CREDENTIALS) {
220 loff_t pos = 0;
223 * No need to call prepare_binprm(), it's already been
224 * done. bprm->buf is stale, update from interp_file.
226 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
227 retval = kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE,
228 &pos);
229 } else
230 retval = prepare_binprm(bprm);
232 if (retval < 0)
233 goto error;
235 retval = search_binary_handler(bprm);
236 if (retval < 0)
237 goto error;
239 ret:
240 dput(fmt->dentry);
241 return retval;
242 error:
243 if (fd_binary > 0)
244 sys_close(fd_binary);
245 bprm->interp_flags = 0;
246 bprm->interp_data = 0;
247 goto ret;
250 /* Command parsers */
253 * parses and copies one argument enclosed in del from *sp to *dp,
254 * recognising the \x special.
255 * returns pointer to the copied argument or NULL in case of an
256 * error (and sets err) or null argument length.
258 static char *scanarg(char *s, char del)
260 char c;
262 while ((c = *s++) != del) {
263 if (c == '\\' && *s == 'x') {
264 s++;
265 if (!isxdigit(*s++))
266 return NULL;
267 if (!isxdigit(*s++))
268 return NULL;
271 s[-1] ='\0';
272 return s;
275 static char *check_special_flags(char *sfs, Node *e)
277 char *p = sfs;
278 int cont = 1;
280 /* special flags */
281 while (cont) {
282 switch (*p) {
283 case 'P':
284 pr_debug("register: flag: P (preserve argv0)\n");
285 p++;
286 e->flags |= MISC_FMT_PRESERVE_ARGV0;
287 break;
288 case 'O':
289 pr_debug("register: flag: O (open binary)\n");
290 p++;
291 e->flags |= MISC_FMT_OPEN_BINARY;
292 break;
293 case 'C':
294 pr_debug("register: flag: C (preserve creds)\n");
295 p++;
296 /* this flags also implies the
297 open-binary flag */
298 e->flags |= (MISC_FMT_CREDENTIALS |
299 MISC_FMT_OPEN_BINARY);
300 break;
301 case 'F':
302 pr_debug("register: flag: F: open interpreter file now\n");
303 p++;
304 e->flags |= MISC_FMT_OPEN_FILE;
305 break;
306 default:
307 cont = 0;
311 return p;
315 * This registers a new binary format, it recognises the syntax
316 * ':name:type:offset:magic:mask:interpreter:flags'
317 * where the ':' is the IFS, that can be chosen with the first char
319 static Node *create_entry(const char __user *buffer, size_t count)
321 Node *e;
322 int memsize, err;
323 char *buf, *p;
324 char del;
326 pr_debug("register: received %zu bytes\n", count);
328 /* some sanity checks */
329 err = -EINVAL;
330 if ((count < 11) || (count > MAX_REGISTER_LENGTH))
331 goto out;
333 err = -ENOMEM;
334 memsize = sizeof(Node) + count + 8;
335 e = kmalloc(memsize, GFP_KERNEL);
336 if (!e)
337 goto out;
339 p = buf = (char *)e + sizeof(Node);
341 memset(e, 0, sizeof(Node));
342 if (copy_from_user(buf, buffer, count))
343 goto efault;
345 del = *p++; /* delimeter */
347 pr_debug("register: delim: %#x {%c}\n", del, del);
349 /* Pad the buffer with the delim to simplify parsing below. */
350 memset(buf + count, del, 8);
352 /* Parse the 'name' field. */
353 e->name = p;
354 p = strchr(p, del);
355 if (!p)
356 goto einval;
357 *p++ = '\0';
358 if (!e->name[0] ||
359 !strcmp(e->name, ".") ||
360 !strcmp(e->name, "..") ||
361 strchr(e->name, '/'))
362 goto einval;
364 pr_debug("register: name: {%s}\n", e->name);
366 /* Parse the 'type' field. */
367 switch (*p++) {
368 case 'E':
369 pr_debug("register: type: E (extension)\n");
370 e->flags = 1 << Enabled;
371 break;
372 case 'M':
373 pr_debug("register: type: M (magic)\n");
374 e->flags = (1 << Enabled) | (1 << Magic);
375 break;
376 default:
377 goto einval;
379 if (*p++ != del)
380 goto einval;
382 if (test_bit(Magic, &e->flags)) {
383 /* Handle the 'M' (magic) format. */
384 char *s;
386 /* Parse the 'offset' field. */
387 s = strchr(p, del);
388 if (!s)
389 goto einval;
390 *s = '\0';
391 if (p != s) {
392 int r = kstrtoint(p, 10, &e->offset);
393 if (r != 0 || e->offset < 0)
394 goto einval;
396 p = s;
397 if (*p++)
398 goto einval;
399 pr_debug("register: offset: %#x\n", e->offset);
401 /* Parse the 'magic' field. */
402 e->magic = p;
403 p = scanarg(p, del);
404 if (!p)
405 goto einval;
406 if (!e->magic[0])
407 goto einval;
408 if (USE_DEBUG)
409 print_hex_dump_bytes(
410 KBUILD_MODNAME ": register: magic[raw]: ",
411 DUMP_PREFIX_NONE, e->magic, p - e->magic);
413 /* Parse the 'mask' field. */
414 e->mask = p;
415 p = scanarg(p, del);
416 if (!p)
417 goto einval;
418 if (!e->mask[0]) {
419 e->mask = NULL;
420 pr_debug("register: mask[raw]: none\n");
421 } else if (USE_DEBUG)
422 print_hex_dump_bytes(
423 KBUILD_MODNAME ": register: mask[raw]: ",
424 DUMP_PREFIX_NONE, e->mask, p - e->mask);
427 * Decode the magic & mask fields.
428 * Note: while we might have accepted embedded NUL bytes from
429 * above, the unescape helpers here will stop at the first one
430 * it encounters.
432 e->size = string_unescape_inplace(e->magic, UNESCAPE_HEX);
433 if (e->mask &&
434 string_unescape_inplace(e->mask, UNESCAPE_HEX) != e->size)
435 goto einval;
436 if (e->size > BINPRM_BUF_SIZE ||
437 BINPRM_BUF_SIZE - e->size < e->offset)
438 goto einval;
439 pr_debug("register: magic/mask length: %i\n", e->size);
440 if (USE_DEBUG) {
441 print_hex_dump_bytes(
442 KBUILD_MODNAME ": register: magic[decoded]: ",
443 DUMP_PREFIX_NONE, e->magic, e->size);
445 if (e->mask) {
446 int i;
447 char *masked = kmalloc(e->size, GFP_KERNEL);
449 print_hex_dump_bytes(
450 KBUILD_MODNAME ": register: mask[decoded]: ",
451 DUMP_PREFIX_NONE, e->mask, e->size);
453 if (masked) {
454 for (i = 0; i < e->size; ++i)
455 masked[i] = e->magic[i] & e->mask[i];
456 print_hex_dump_bytes(
457 KBUILD_MODNAME ": register: magic[masked]: ",
458 DUMP_PREFIX_NONE, masked, e->size);
460 kfree(masked);
464 } else {
465 /* Handle the 'E' (extension) format. */
467 /* Skip the 'offset' field. */
468 p = strchr(p, del);
469 if (!p)
470 goto einval;
471 *p++ = '\0';
473 /* Parse the 'magic' field. */
474 e->magic = p;
475 p = strchr(p, del);
476 if (!p)
477 goto einval;
478 *p++ = '\0';
479 if (!e->magic[0] || strchr(e->magic, '/'))
480 goto einval;
481 pr_debug("register: extension: {%s}\n", e->magic);
483 /* Skip the 'mask' field. */
484 p = strchr(p, del);
485 if (!p)
486 goto einval;
487 *p++ = '\0';
490 /* Parse the 'interpreter' field. */
491 e->interpreter = p;
492 p = strchr(p, del);
493 if (!p)
494 goto einval;
495 *p++ = '\0';
496 if (!e->interpreter[0])
497 goto einval;
498 pr_debug("register: interpreter: {%s}\n", e->interpreter);
500 /* Parse the 'flags' field. */
501 p = check_special_flags(p, e);
502 if (*p == '\n')
503 p++;
504 if (p != buf + count)
505 goto einval;
507 return e;
509 out:
510 return ERR_PTR(err);
512 efault:
513 kfree(e);
514 return ERR_PTR(-EFAULT);
515 einval:
516 kfree(e);
517 return ERR_PTR(-EINVAL);
521 * Set status of entry/binfmt_misc:
522 * '1' enables, '0' disables and '-1' clears entry/binfmt_misc
524 static int parse_command(const char __user *buffer, size_t count)
526 char s[4];
528 if (count > 3)
529 return -EINVAL;
530 if (copy_from_user(s, buffer, count))
531 return -EFAULT;
532 if (!count)
533 return 0;
534 if (s[count - 1] == '\n')
535 count--;
536 if (count == 1 && s[0] == '0')
537 return 1;
538 if (count == 1 && s[0] == '1')
539 return 2;
540 if (count == 2 && s[0] == '-' && s[1] == '1')
541 return 3;
542 return -EINVAL;
545 /* generic stuff */
547 static void entry_status(Node *e, char *page)
549 char *dp = page;
550 const char *status = "disabled";
552 if (test_bit(Enabled, &e->flags))
553 status = "enabled";
555 if (!VERBOSE_STATUS) {
556 sprintf(page, "%s\n", status);
557 return;
560 dp += sprintf(dp, "%s\ninterpreter %s\n", status, e->interpreter);
562 /* print the special flags */
563 dp += sprintf(dp, "flags: ");
564 if (e->flags & MISC_FMT_PRESERVE_ARGV0)
565 *dp++ = 'P';
566 if (e->flags & MISC_FMT_OPEN_BINARY)
567 *dp++ = 'O';
568 if (e->flags & MISC_FMT_CREDENTIALS)
569 *dp++ = 'C';
570 if (e->flags & MISC_FMT_OPEN_FILE)
571 *dp++ = 'F';
572 *dp++ = '\n';
574 if (!test_bit(Magic, &e->flags)) {
575 sprintf(dp, "extension .%s\n", e->magic);
576 } else {
577 dp += sprintf(dp, "offset %i\nmagic ", e->offset);
578 dp = bin2hex(dp, e->magic, e->size);
579 if (e->mask) {
580 dp += sprintf(dp, "\nmask ");
581 dp = bin2hex(dp, e->mask, e->size);
583 *dp++ = '\n';
584 *dp = '\0';
588 static struct inode *bm_get_inode(struct super_block *sb, int mode)
590 struct inode *inode = new_inode(sb);
592 if (inode) {
593 inode->i_ino = get_next_ino();
594 inode->i_mode = mode;
595 inode->i_atime = inode->i_mtime = inode->i_ctime =
596 current_time(inode);
598 return inode;
601 static void bm_evict_inode(struct inode *inode)
603 Node *e = inode->i_private;
605 if (e && e->flags & MISC_FMT_OPEN_FILE)
606 filp_close(e->interp_file, NULL);
608 clear_inode(inode);
609 kfree(e);
612 static void kill_node(Node *e)
614 struct dentry *dentry;
616 write_lock(&entries_lock);
617 list_del_init(&e->list);
618 write_unlock(&entries_lock);
620 dentry = e->dentry;
621 drop_nlink(d_inode(dentry));
622 d_drop(dentry);
623 dput(dentry);
624 simple_release_fs(&bm_mnt, &entry_count);
627 /* /<entry> */
629 static ssize_t
630 bm_entry_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
632 Node *e = file_inode(file)->i_private;
633 ssize_t res;
634 char *page;
636 page = (char *) __get_free_page(GFP_KERNEL);
637 if (!page)
638 return -ENOMEM;
640 entry_status(e, page);
642 res = simple_read_from_buffer(buf, nbytes, ppos, page, strlen(page));
644 free_page((unsigned long) page);
645 return res;
648 static ssize_t bm_entry_write(struct file *file, const char __user *buffer,
649 size_t count, loff_t *ppos)
651 struct dentry *root;
652 Node *e = file_inode(file)->i_private;
653 int res = parse_command(buffer, count);
655 switch (res) {
656 case 1:
657 /* Disable this handler. */
658 clear_bit(Enabled, &e->flags);
659 break;
660 case 2:
661 /* Enable this handler. */
662 set_bit(Enabled, &e->flags);
663 break;
664 case 3:
665 /* Delete this handler. */
666 root = file_inode(file)->i_sb->s_root;
667 inode_lock(d_inode(root));
669 if (!list_empty(&e->list))
670 kill_node(e);
672 inode_unlock(d_inode(root));
673 break;
674 default:
675 return res;
678 return count;
681 static const struct file_operations bm_entry_operations = {
682 .read = bm_entry_read,
683 .write = bm_entry_write,
684 .llseek = default_llseek,
687 /* /register */
689 static ssize_t bm_register_write(struct file *file, const char __user *buffer,
690 size_t count, loff_t *ppos)
692 Node *e;
693 struct inode *inode;
694 struct super_block *sb = file_inode(file)->i_sb;
695 struct dentry *root = sb->s_root, *dentry;
696 int err = 0;
698 e = create_entry(buffer, count);
700 if (IS_ERR(e))
701 return PTR_ERR(e);
703 inode_lock(d_inode(root));
704 dentry = lookup_one_len(e->name, root, strlen(e->name));
705 err = PTR_ERR(dentry);
706 if (IS_ERR(dentry))
707 goto out;
709 err = -EEXIST;
710 if (d_really_is_positive(dentry))
711 goto out2;
713 inode = bm_get_inode(sb, S_IFREG | 0644);
715 err = -ENOMEM;
716 if (!inode)
717 goto out2;
719 err = simple_pin_fs(&bm_fs_type, &bm_mnt, &entry_count);
720 if (err) {
721 iput(inode);
722 inode = NULL;
723 goto out2;
726 if (e->flags & MISC_FMT_OPEN_FILE) {
727 struct file *f;
729 f = open_exec(e->interpreter);
730 if (IS_ERR(f)) {
731 err = PTR_ERR(f);
732 pr_notice("register: failed to install interpreter file %s\n", e->interpreter);
733 simple_release_fs(&bm_mnt, &entry_count);
734 iput(inode);
735 inode = NULL;
736 goto out2;
738 e->interp_file = f;
741 e->dentry = dget(dentry);
742 inode->i_private = e;
743 inode->i_fop = &bm_entry_operations;
745 d_instantiate(dentry, inode);
746 write_lock(&entries_lock);
747 list_add(&e->list, &entries);
748 write_unlock(&entries_lock);
750 err = 0;
751 out2:
752 dput(dentry);
753 out:
754 inode_unlock(d_inode(root));
756 if (err) {
757 kfree(e);
758 return err;
760 return count;
763 static const struct file_operations bm_register_operations = {
764 .write = bm_register_write,
765 .llseek = noop_llseek,
768 /* /status */
770 static ssize_t
771 bm_status_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
773 char *s = enabled ? "enabled\n" : "disabled\n";
775 return simple_read_from_buffer(buf, nbytes, ppos, s, strlen(s));
778 static ssize_t bm_status_write(struct file *file, const char __user *buffer,
779 size_t count, loff_t *ppos)
781 int res = parse_command(buffer, count);
782 struct dentry *root;
784 switch (res) {
785 case 1:
786 /* Disable all handlers. */
787 enabled = 0;
788 break;
789 case 2:
790 /* Enable all handlers. */
791 enabled = 1;
792 break;
793 case 3:
794 /* Delete all handlers. */
795 root = file_inode(file)->i_sb->s_root;
796 inode_lock(d_inode(root));
798 while (!list_empty(&entries))
799 kill_node(list_first_entry(&entries, Node, list));
801 inode_unlock(d_inode(root));
802 break;
803 default:
804 return res;
807 return count;
810 static const struct file_operations bm_status_operations = {
811 .read = bm_status_read,
812 .write = bm_status_write,
813 .llseek = default_llseek,
816 /* Superblock handling */
818 static const struct super_operations s_ops = {
819 .statfs = simple_statfs,
820 .evict_inode = bm_evict_inode,
823 static int bm_fill_super(struct super_block *sb, void *data, int silent)
825 int err;
826 static const struct tree_descr bm_files[] = {
827 [2] = {"status", &bm_status_operations, S_IWUSR|S_IRUGO},
828 [3] = {"register", &bm_register_operations, S_IWUSR},
829 /* last one */ {""}
832 err = simple_fill_super(sb, BINFMTFS_MAGIC, bm_files);
833 if (!err)
834 sb->s_op = &s_ops;
835 return err;
838 static struct dentry *bm_mount(struct file_system_type *fs_type,
839 int flags, const char *dev_name, void *data)
841 return mount_single(fs_type, flags, data, bm_fill_super);
844 static struct linux_binfmt misc_format = {
845 .module = THIS_MODULE,
846 .load_binary = load_misc_binary,
849 static struct file_system_type bm_fs_type = {
850 .owner = THIS_MODULE,
851 .name = "binfmt_misc",
852 .mount = bm_mount,
853 .kill_sb = kill_litter_super,
855 MODULE_ALIAS_FS("binfmt_misc");
857 static int __init init_misc_binfmt(void)
859 int err = register_filesystem(&bm_fs_type);
860 if (!err)
861 insert_binfmt(&misc_format);
862 return err;
865 static void __exit exit_misc_binfmt(void)
867 unregister_binfmt(&misc_format);
868 unregister_filesystem(&bm_fs_type);
871 core_initcall(init_misc_binfmt);
872 module_exit(exit_misc_binfmt);
873 MODULE_LICENSE("GPL");