lockdep: shrink held_lock structure
[linux-2.6] / security / commoncap.c
1 /* Common capabilities, needed by capability.o and root_plug.o
2  *
3  *      This program is free software; you can redistribute it and/or modify
4  *      it under the terms of the GNU General Public License as published by
5  *      the Free Software Foundation; either version 2 of the License, or
6  *      (at your option) any later version.
7  *
8  */
9
10 #include <linux/capability.h>
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/security.h>
15 #include <linux/file.h>
16 #include <linux/mm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/skbuff.h>
21 #include <linux/netlink.h>
22 #include <linux/ptrace.h>
23 #include <linux/xattr.h>
24 #include <linux/hugetlb.h>
25 #include <linux/mount.h>
26 #include <linux/sched.h>
27 #include <linux/prctl.h>
28 #include <linux/securebits.h>
29
30 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
31 {
32         NETLINK_CB(skb).eff_cap = current->cap_effective;
33         return 0;
34 }
35
36 int cap_netlink_recv(struct sk_buff *skb, int cap)
37 {
38         if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
39                 return -EPERM;
40         return 0;
41 }
42
43 EXPORT_SYMBOL(cap_netlink_recv);
44
45 /*
46  * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
47  * function.  That is, it has the reverse semantics: cap_capable()
48  * returns 0 when a task has a capability, but the kernel's capable()
49  * returns 1 for this case.
50  */
51 int cap_capable (struct task_struct *tsk, int cap)
52 {
53         /* Derived from include/linux/sched.h:capable. */
54         if (cap_raised(tsk->cap_effective, cap))
55                 return 0;
56         return -EPERM;
57 }
58
59 int cap_settime(struct timespec *ts, struct timezone *tz)
60 {
61         if (!capable(CAP_SYS_TIME))
62                 return -EPERM;
63         return 0;
64 }
65
66 int cap_ptrace (struct task_struct *parent, struct task_struct *child,
67                 unsigned int mode)
68 {
69         /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
70         if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
71             !__capable(parent, CAP_SYS_PTRACE))
72                 return -EPERM;
73         return 0;
74 }
75
76 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
77                 kernel_cap_t *inheritable, kernel_cap_t *permitted)
78 {
79         /* Derived from kernel/capability.c:sys_capget. */
80         *effective = target->cap_effective;
81         *inheritable = target->cap_inheritable;
82         *permitted = target->cap_permitted;
83         return 0;
84 }
85
86 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
87
88 static inline int cap_block_setpcap(struct task_struct *target)
89 {
90         /*
91          * No support for remote process capability manipulation with
92          * filesystem capability support.
93          */
94         return (target != current);
95 }
96
97 static inline int cap_inh_is_capped(void)
98 {
99         /*
100          * Return 1 if changes to the inheritable set are limited
101          * to the old permitted set. That is, if the current task
102          * does *not* possess the CAP_SETPCAP capability.
103          */
104         return (cap_capable(current, CAP_SETPCAP) != 0);
105 }
106
107 static inline int cap_limit_ptraced_target(void) { return 1; }
108
109 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
110
111 static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
112 static inline int cap_inh_is_capped(void) { return 1; }
113 static inline int cap_limit_ptraced_target(void)
114 {
115         return !capable(CAP_SETPCAP);
116 }
117
118 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
119
120 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
121                       kernel_cap_t *inheritable, kernel_cap_t *permitted)
122 {
123         if (cap_block_setpcap(target)) {
124                 return -EPERM;
125         }
126         if (cap_inh_is_capped()
127             && !cap_issubset(*inheritable,
128                              cap_combine(target->cap_inheritable,
129                                          current->cap_permitted))) {
130                 /* incapable of using this inheritable set */
131                 return -EPERM;
132         }
133         if (!cap_issubset(*inheritable,
134                            cap_combine(target->cap_inheritable,
135                                        current->cap_bset))) {
136                 /* no new pI capabilities outside bounding set */
137                 return -EPERM;
138         }
139
140         /* verify restrictions on target's new Permitted set */
141         if (!cap_issubset (*permitted,
142                            cap_combine (target->cap_permitted,
143                                         current->cap_permitted))) {
144                 return -EPERM;
145         }
146
147         /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
148         if (!cap_issubset (*effective, *permitted)) {
149                 return -EPERM;
150         }
151
152         return 0;
153 }
154
155 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
156                      kernel_cap_t *inheritable, kernel_cap_t *permitted)
157 {
158         target->cap_effective = *effective;
159         target->cap_inheritable = *inheritable;
160         target->cap_permitted = *permitted;
161 }
162
163 static inline void bprm_clear_caps(struct linux_binprm *bprm)
164 {
165         cap_clear(bprm->cap_post_exec_permitted);
166         bprm->cap_effective = false;
167 }
168
169 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
170
171 int cap_inode_need_killpriv(struct dentry *dentry)
172 {
173         struct inode *inode = dentry->d_inode;
174         int error;
175
176         if (!inode->i_op || !inode->i_op->getxattr)
177                return 0;
178
179         error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
180         if (error <= 0)
181                 return 0;
182         return 1;
183 }
184
185 int cap_inode_killpriv(struct dentry *dentry)
186 {
187         struct inode *inode = dentry->d_inode;
188
189         if (!inode->i_op || !inode->i_op->removexattr)
190                return 0;
191
192         return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
193 }
194
195 static inline int cap_from_disk(struct vfs_cap_data *caps,
196                                 struct linux_binprm *bprm, unsigned size)
197 {
198         __u32 magic_etc;
199         unsigned tocopy, i;
200         int ret;
201
202         if (size < sizeof(magic_etc))
203                 return -EINVAL;
204
205         magic_etc = le32_to_cpu(caps->magic_etc);
206
207         switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
208         case VFS_CAP_REVISION_1:
209                 if (size != XATTR_CAPS_SZ_1)
210                         return -EINVAL;
211                 tocopy = VFS_CAP_U32_1;
212                 break;
213         case VFS_CAP_REVISION_2:
214                 if (size != XATTR_CAPS_SZ_2)
215                         return -EINVAL;
216                 tocopy = VFS_CAP_U32_2;
217                 break;
218         default:
219                 return -EINVAL;
220         }
221
222         if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE) {
223                 bprm->cap_effective = true;
224         } else {
225                 bprm->cap_effective = false;
226         }
227
228         ret = 0;
229
230         CAP_FOR_EACH_U32(i) {
231                 __u32 value_cpu;
232
233                 if (i >= tocopy) {
234                         /*
235                          * Legacy capability sets have no upper bits
236                          */
237                         bprm->cap_post_exec_permitted.cap[i] = 0;
238                         continue;
239                 }
240                 /*
241                  * pP' = (X & fP) | (pI & fI)
242                  */
243                 value_cpu = le32_to_cpu(caps->data[i].permitted);
244                 bprm->cap_post_exec_permitted.cap[i] =
245                         (current->cap_bset.cap[i] & value_cpu) |
246                         (current->cap_inheritable.cap[i] &
247                                 le32_to_cpu(caps->data[i].inheritable));
248                 if (value_cpu & ~bprm->cap_post_exec_permitted.cap[i]) {
249                         /*
250                          * insufficient to execute correctly
251                          */
252                         ret = -EPERM;
253                 }
254         }
255
256         /*
257          * For legacy apps, with no internal support for recognizing they
258          * do not have enough capabilities, we return an error if they are
259          * missing some "forced" (aka file-permitted) capabilities.
260          */
261         return bprm->cap_effective ? ret : 0;
262 }
263
264 /* Locate any VFS capabilities: */
265 static int get_file_caps(struct linux_binprm *bprm)
266 {
267         struct dentry *dentry;
268         int rc = 0;
269         struct vfs_cap_data vcaps;
270         struct inode *inode;
271
272         if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
273                 bprm_clear_caps(bprm);
274                 return 0;
275         }
276
277         dentry = dget(bprm->file->f_dentry);
278         inode = dentry->d_inode;
279         if (!inode->i_op || !inode->i_op->getxattr)
280                 goto out;
281
282         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &vcaps,
283                                    XATTR_CAPS_SZ);
284         if (rc == -ENODATA || rc == -EOPNOTSUPP) {
285                 /* no data, that's ok */
286                 rc = 0;
287                 goto out;
288         }
289         if (rc < 0)
290                 goto out;
291
292         rc = cap_from_disk(&vcaps, bprm, rc);
293         if (rc == -EINVAL)
294                 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
295                        __func__, rc, bprm->filename);
296
297 out:
298         dput(dentry);
299         if (rc)
300                 bprm_clear_caps(bprm);
301
302         return rc;
303 }
304
305 #else
306 int cap_inode_need_killpriv(struct dentry *dentry)
307 {
308         return 0;
309 }
310
311 int cap_inode_killpriv(struct dentry *dentry)
312 {
313         return 0;
314 }
315
316 static inline int get_file_caps(struct linux_binprm *bprm)
317 {
318         bprm_clear_caps(bprm);
319         return 0;
320 }
321 #endif
322
323 int cap_bprm_set_security (struct linux_binprm *bprm)
324 {
325         int ret;
326
327         ret = get_file_caps(bprm);
328
329         if (!issecure(SECURE_NOROOT)) {
330                 /*
331                  * To support inheritance of root-permissions and suid-root
332                  * executables under compatibility mode, we override the
333                  * capability sets for the file.
334                  *
335                  * If only the real uid is 0, we do not set the effective
336                  * bit.
337                  */
338                 if (bprm->e_uid == 0 || current->uid == 0) {
339                         /* pP' = (cap_bset & ~0) | (pI & ~0) */
340                         bprm->cap_post_exec_permitted = cap_combine(
341                                 current->cap_bset, current->cap_inheritable
342                                 );
343                         bprm->cap_effective = (bprm->e_uid == 0);
344                         ret = 0;
345                 }
346         }
347
348         return ret;
349 }
350
351 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
352 {
353         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
354             !cap_issubset(bprm->cap_post_exec_permitted,
355                           current->cap_permitted)) {
356                 set_dumpable(current->mm, suid_dumpable);
357                 current->pdeath_signal = 0;
358
359                 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
360                         if (!capable(CAP_SETUID)) {
361                                 bprm->e_uid = current->uid;
362                                 bprm->e_gid = current->gid;
363                         }
364                         if (cap_limit_ptraced_target()) {
365                                 bprm->cap_post_exec_permitted = cap_intersect(
366                                         bprm->cap_post_exec_permitted,
367                                         current->cap_permitted);
368                         }
369                 }
370         }
371
372         current->suid = current->euid = current->fsuid = bprm->e_uid;
373         current->sgid = current->egid = current->fsgid = bprm->e_gid;
374
375         /* For init, we want to retain the capabilities set
376          * in the init_task struct. Thus we skip the usual
377          * capability rules */
378         if (!is_global_init(current)) {
379                 current->cap_permitted = bprm->cap_post_exec_permitted;
380                 if (bprm->cap_effective)
381                         current->cap_effective = bprm->cap_post_exec_permitted;
382                 else
383                         cap_clear(current->cap_effective);
384         }
385
386         /* AUD: Audit candidate if current->cap_effective is set */
387
388         current->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
389 }
390
391 int cap_bprm_secureexec (struct linux_binprm *bprm)
392 {
393         if (current->uid != 0) {
394                 if (bprm->cap_effective)
395                         return 1;
396                 if (!cap_isclear(bprm->cap_post_exec_permitted))
397                         return 1;
398         }
399
400         return (current->euid != current->uid ||
401                 current->egid != current->gid);
402 }
403
404 int cap_inode_setxattr(struct dentry *dentry, const char *name,
405                        const void *value, size_t size, int flags)
406 {
407         if (!strcmp(name, XATTR_NAME_CAPS)) {
408                 if (!capable(CAP_SETFCAP))
409                         return -EPERM;
410                 return 0;
411         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
412                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
413             !capable(CAP_SYS_ADMIN))
414                 return -EPERM;
415         return 0;
416 }
417
418 int cap_inode_removexattr(struct dentry *dentry, const char *name)
419 {
420         if (!strcmp(name, XATTR_NAME_CAPS)) {
421                 if (!capable(CAP_SETFCAP))
422                         return -EPERM;
423                 return 0;
424         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
425                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
426             !capable(CAP_SYS_ADMIN))
427                 return -EPERM;
428         return 0;
429 }
430
431 /* moved from kernel/sys.c. */
432 /* 
433  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
434  * a process after a call to setuid, setreuid, or setresuid.
435  *
436  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
437  *  {r,e,s}uid != 0, the permitted and effective capabilities are
438  *  cleared.
439  *
440  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
441  *  capabilities of the process are cleared.
442  *
443  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
444  *  capabilities are set to the permitted capabilities.
445  *
446  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should 
447  *  never happen.
448  *
449  *  -astor 
450  *
451  * cevans - New behaviour, Oct '99
452  * A process may, via prctl(), elect to keep its capabilities when it
453  * calls setuid() and switches away from uid==0. Both permitted and
454  * effective sets will be retained.
455  * Without this change, it was impossible for a daemon to drop only some
456  * of its privilege. The call to setuid(!=0) would drop all privileges!
457  * Keeping uid 0 is not an option because uid 0 owns too many vital
458  * files..
459  * Thanks to Olaf Kirch and Peter Benie for spotting this.
460  */
461 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
462                                         int old_suid)
463 {
464         if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
465             (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
466             !issecure(SECURE_KEEP_CAPS)) {
467                 cap_clear (current->cap_permitted);
468                 cap_clear (current->cap_effective);
469         }
470         if (old_euid == 0 && current->euid != 0) {
471                 cap_clear (current->cap_effective);
472         }
473         if (old_euid != 0 && current->euid == 0) {
474                 current->cap_effective = current->cap_permitted;
475         }
476 }
477
478 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
479                           int flags)
480 {
481         switch (flags) {
482         case LSM_SETID_RE:
483         case LSM_SETID_ID:
484         case LSM_SETID_RES:
485                 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
486                 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
487                         cap_emulate_setxuid (old_ruid, old_euid, old_suid);
488                 }
489                 break;
490         case LSM_SETID_FS:
491                 {
492                         uid_t old_fsuid = old_ruid;
493
494                         /* Copied from kernel/sys.c:setfsuid. */
495
496                         /*
497                          * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
498                          *          if not, we might be a bit too harsh here.
499                          */
500
501                         if (!issecure (SECURE_NO_SETUID_FIXUP)) {
502                                 if (old_fsuid == 0 && current->fsuid != 0) {
503                                         current->cap_effective =
504                                                 cap_drop_fs_set(
505                                                     current->cap_effective);
506                                 }
507                                 if (old_fsuid != 0 && current->fsuid == 0) {
508                                         current->cap_effective =
509                                                 cap_raise_fs_set(
510                                                     current->cap_effective,
511                                                     current->cap_permitted);
512                                 }
513                         }
514                         break;
515                 }
516         default:
517                 return -EINVAL;
518         }
519
520         return 0;
521 }
522
523 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
524 /*
525  * Rationale: code calling task_setscheduler, task_setioprio, and
526  * task_setnice, assumes that
527  *   . if capable(cap_sys_nice), then those actions should be allowed
528  *   . if not capable(cap_sys_nice), but acting on your own processes,
529  *      then those actions should be allowed
530  * This is insufficient now since you can call code without suid, but
531  * yet with increased caps.
532  * So we check for increased caps on the target process.
533  */
534 static inline int cap_safe_nice(struct task_struct *p)
535 {
536         if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
537             !__capable(current, CAP_SYS_NICE))
538                 return -EPERM;
539         return 0;
540 }
541
542 int cap_task_setscheduler (struct task_struct *p, int policy,
543                            struct sched_param *lp)
544 {
545         return cap_safe_nice(p);
546 }
547
548 int cap_task_setioprio (struct task_struct *p, int ioprio)
549 {
550         return cap_safe_nice(p);
551 }
552
553 int cap_task_setnice (struct task_struct *p, int nice)
554 {
555         return cap_safe_nice(p);
556 }
557
558 /*
559  * called from kernel/sys.c for prctl(PR_CABSET_DROP)
560  * done without task_capability_lock() because it introduces
561  * no new races - i.e. only another task doing capget() on
562  * this task could get inconsistent info.  There can be no
563  * racing writer bc a task can only change its own caps.
564  */
565 static long cap_prctl_drop(unsigned long cap)
566 {
567         if (!capable(CAP_SETPCAP))
568                 return -EPERM;
569         if (!cap_valid(cap))
570                 return -EINVAL;
571         cap_lower(current->cap_bset, cap);
572         return 0;
573 }
574
575 #else
576 int cap_task_setscheduler (struct task_struct *p, int policy,
577                            struct sched_param *lp)
578 {
579         return 0;
580 }
581 int cap_task_setioprio (struct task_struct *p, int ioprio)
582 {
583         return 0;
584 }
585 int cap_task_setnice (struct task_struct *p, int nice)
586 {
587         return 0;
588 }
589 #endif
590
591 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
592                    unsigned long arg4, unsigned long arg5, long *rc_p)
593 {
594         long error = 0;
595
596         switch (option) {
597         case PR_CAPBSET_READ:
598                 if (!cap_valid(arg2))
599                         error = -EINVAL;
600                 else
601                         error = !!cap_raised(current->cap_bset, arg2);
602                 break;
603 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
604         case PR_CAPBSET_DROP:
605                 error = cap_prctl_drop(arg2);
606                 break;
607
608         /*
609          * The next four prctl's remain to assist with transitioning a
610          * system from legacy UID=0 based privilege (when filesystem
611          * capabilities are not in use) to a system using filesystem
612          * capabilities only - as the POSIX.1e draft intended.
613          *
614          * Note:
615          *
616          *  PR_SET_SECUREBITS =
617          *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
618          *    | issecure_mask(SECURE_NOROOT)
619          *    | issecure_mask(SECURE_NOROOT_LOCKED)
620          *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
621          *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
622          *
623          * will ensure that the current process and all of its
624          * children will be locked into a pure
625          * capability-based-privilege environment.
626          */
627         case PR_SET_SECUREBITS:
628                 if ((((current->securebits & SECURE_ALL_LOCKS) >> 1)
629                      & (current->securebits ^ arg2))                  /*[1]*/
630                     || ((current->securebits & SECURE_ALL_LOCKS
631                          & ~arg2))                                    /*[2]*/
632                     || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
633                     || (cap_capable(current, CAP_SETPCAP) != 0)) {    /*[4]*/
634                         /*
635                          * [1] no changing of bits that are locked
636                          * [2] no unlocking of locks
637                          * [3] no setting of unsupported bits
638                          * [4] doing anything requires privilege (go read about
639                          *     the "sendmail capabilities bug")
640                          */
641                         error = -EPERM;  /* cannot change a locked bit */
642                 } else {
643                         current->securebits = arg2;
644                 }
645                 break;
646         case PR_GET_SECUREBITS:
647                 error = current->securebits;
648                 break;
649
650 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
651
652         case PR_GET_KEEPCAPS:
653                 if (issecure(SECURE_KEEP_CAPS))
654                         error = 1;
655                 break;
656         case PR_SET_KEEPCAPS:
657                 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
658                         error = -EINVAL;
659                 else if (issecure(SECURE_KEEP_CAPS_LOCKED))
660                         error = -EPERM;
661                 else if (arg2)
662                         current->securebits |= issecure_mask(SECURE_KEEP_CAPS);
663                 else
664                         current->securebits &=
665                                 ~issecure_mask(SECURE_KEEP_CAPS);
666                 break;
667
668         default:
669                 /* No functionality available - continue with default */
670                 return 0;
671         }
672
673         /* Functionality provided */
674         *rc_p = error;
675         return 1;
676 }
677
678 void cap_task_reparent_to_init (struct task_struct *p)
679 {
680         cap_set_init_eff(p->cap_effective);
681         cap_clear(p->cap_inheritable);
682         cap_set_full(p->cap_permitted);
683         p->securebits = SECUREBITS_DEFAULT;
684         return;
685 }
686
687 int cap_syslog (int type)
688 {
689         if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
690                 return -EPERM;
691         return 0;
692 }
693
694 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
695 {
696         int cap_sys_admin = 0;
697
698         if (cap_capable(current, CAP_SYS_ADMIN) == 0)
699                 cap_sys_admin = 1;
700         return __vm_enough_memory(mm, pages, cap_sys_admin);
701 }
702