tools/bpf: Sync kernel btf.h header
[muen/linux.git] / kernel / bpf / syscall.c
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of version 2 of the GNU General Public
5  * License as published by the Free Software Foundation.
6  *
7  * This program is distributed in the hope that it will be useful, but
8  * WITHOUT ANY WARRANTY; without even the implied warranty of
9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10  * General Public License for more details.
11  */
12 #include <linux/bpf.h>
13 #include <linux/bpf_trace.h>
14 #include <linux/bpf_lirc.h>
15 #include <linux/btf.h>
16 #include <linux/syscalls.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/vmalloc.h>
20 #include <linux/mmzone.h>
21 #include <linux/anon_inodes.h>
22 #include <linux/fdtable.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/license.h>
26 #include <linux/filter.h>
27 #include <linux/version.h>
28 #include <linux/kernel.h>
29 #include <linux/idr.h>
30 #include <linux/cred.h>
31 #include <linux/timekeeping.h>
32 #include <linux/ctype.h>
33 #include <linux/nospec.h>
34
35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
36                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
37                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
38                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
41
42 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
43
44 DEFINE_PER_CPU(int, bpf_prog_active);
45 static DEFINE_IDR(prog_idr);
46 static DEFINE_SPINLOCK(prog_idr_lock);
47 static DEFINE_IDR(map_idr);
48 static DEFINE_SPINLOCK(map_idr_lock);
49
50 int sysctl_unprivileged_bpf_disabled __read_mostly;
51
52 static const struct bpf_map_ops * const bpf_map_types[] = {
53 #define BPF_PROG_TYPE(_id, _ops)
54 #define BPF_MAP_TYPE(_id, _ops) \
55         [_id] = &_ops,
56 #include <linux/bpf_types.h>
57 #undef BPF_PROG_TYPE
58 #undef BPF_MAP_TYPE
59 };
60
61 /*
62  * If we're handed a bigger struct than we know of, ensure all the unknown bits
63  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
64  * we don't know about yet.
65  *
66  * There is a ToCToU between this function call and the following
67  * copy_from_user() call. However, this is not a concern since this function is
68  * meant to be a future-proofing of bits.
69  */
70 int bpf_check_uarg_tail_zero(void __user *uaddr,
71                              size_t expected_size,
72                              size_t actual_size)
73 {
74         unsigned char __user *addr;
75         unsigned char __user *end;
76         unsigned char val;
77         int err;
78
79         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
80                 return -E2BIG;
81
82         if (unlikely(!access_ok(uaddr, actual_size)))
83                 return -EFAULT;
84
85         if (actual_size <= expected_size)
86                 return 0;
87
88         addr = uaddr + expected_size;
89         end  = uaddr + actual_size;
90
91         for (; addr < end; addr++) {
92                 err = get_user(val, addr);
93                 if (err)
94                         return err;
95                 if (val)
96                         return -E2BIG;
97         }
98
99         return 0;
100 }
101
102 const struct bpf_map_ops bpf_map_offload_ops = {
103         .map_alloc = bpf_map_offload_map_alloc,
104         .map_free = bpf_map_offload_map_free,
105         .map_check_btf = map_check_no_btf,
106 };
107
108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
109 {
110         const struct bpf_map_ops *ops;
111         u32 type = attr->map_type;
112         struct bpf_map *map;
113         int err;
114
115         if (type >= ARRAY_SIZE(bpf_map_types))
116                 return ERR_PTR(-EINVAL);
117         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
118         ops = bpf_map_types[type];
119         if (!ops)
120                 return ERR_PTR(-EINVAL);
121
122         if (ops->map_alloc_check) {
123                 err = ops->map_alloc_check(attr);
124                 if (err)
125                         return ERR_PTR(err);
126         }
127         if (attr->map_ifindex)
128                 ops = &bpf_map_offload_ops;
129         map = ops->map_alloc(attr);
130         if (IS_ERR(map))
131                 return map;
132         map->ops = ops;
133         map->map_type = type;
134         return map;
135 }
136
137 void *bpf_map_area_alloc(size_t size, int numa_node)
138 {
139         /* We really just want to fail instead of triggering OOM killer
140          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
141          * which is used for lower order allocation requests.
142          *
143          * It has been observed that higher order allocation requests done by
144          * vmalloc with __GFP_NORETRY being set might fail due to not trying
145          * to reclaim memory from the page cache, thus we set
146          * __GFP_RETRY_MAYFAIL to avoid such situations.
147          */
148
149         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
150         void *area;
151
152         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
153                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
154                                     numa_node);
155                 if (area != NULL)
156                         return area;
157         }
158
159         return __vmalloc_node_flags_caller(size, numa_node,
160                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
161                                            flags, __builtin_return_address(0));
162 }
163
164 void bpf_map_area_free(void *area)
165 {
166         kvfree(area);
167 }
168
169 static u32 bpf_map_flags_retain_permanent(u32 flags)
170 {
171         /* Some map creation flags are not tied to the map object but
172          * rather to the map fd instead, so they have no meaning upon
173          * map object inspection since multiple file descriptors with
174          * different (access) properties can exist here. Thus, given
175          * this has zero meaning for the map itself, lets clear these
176          * from here.
177          */
178         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
179 }
180
181 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
182 {
183         map->map_type = attr->map_type;
184         map->key_size = attr->key_size;
185         map->value_size = attr->value_size;
186         map->max_entries = attr->max_entries;
187         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
188         map->numa_node = bpf_map_attr_numa_node(attr);
189 }
190
191 int bpf_map_precharge_memlock(u32 pages)
192 {
193         struct user_struct *user = get_current_user();
194         unsigned long memlock_limit, cur;
195
196         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
197         cur = atomic_long_read(&user->locked_vm);
198         free_uid(user);
199         if (cur + pages > memlock_limit)
200                 return -EPERM;
201         return 0;
202 }
203
204 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
205 {
206         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
207
208         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
209                 atomic_long_sub(pages, &user->locked_vm);
210                 return -EPERM;
211         }
212         return 0;
213 }
214
215 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
216 {
217         atomic_long_sub(pages, &user->locked_vm);
218 }
219
220 static int bpf_map_init_memlock(struct bpf_map *map)
221 {
222         struct user_struct *user = get_current_user();
223         int ret;
224
225         ret = bpf_charge_memlock(user, map->pages);
226         if (ret) {
227                 free_uid(user);
228                 return ret;
229         }
230         map->user = user;
231         return ret;
232 }
233
234 static void bpf_map_release_memlock(struct bpf_map *map)
235 {
236         struct user_struct *user = map->user;
237         bpf_uncharge_memlock(user, map->pages);
238         free_uid(user);
239 }
240
241 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
242 {
243         int ret;
244
245         ret = bpf_charge_memlock(map->user, pages);
246         if (ret)
247                 return ret;
248         map->pages += pages;
249         return ret;
250 }
251
252 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
253 {
254         bpf_uncharge_memlock(map->user, pages);
255         map->pages -= pages;
256 }
257
258 static int bpf_map_alloc_id(struct bpf_map *map)
259 {
260         int id;
261
262         idr_preload(GFP_KERNEL);
263         spin_lock_bh(&map_idr_lock);
264         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
265         if (id > 0)
266                 map->id = id;
267         spin_unlock_bh(&map_idr_lock);
268         idr_preload_end();
269
270         if (WARN_ON_ONCE(!id))
271                 return -ENOSPC;
272
273         return id > 0 ? 0 : id;
274 }
275
276 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
277 {
278         unsigned long flags;
279
280         /* Offloaded maps are removed from the IDR store when their device
281          * disappears - even if someone holds an fd to them they are unusable,
282          * the memory is gone, all ops will fail; they are simply waiting for
283          * refcnt to drop to be freed.
284          */
285         if (!map->id)
286                 return;
287
288         if (do_idr_lock)
289                 spin_lock_irqsave(&map_idr_lock, flags);
290         else
291                 __acquire(&map_idr_lock);
292
293         idr_remove(&map_idr, map->id);
294         map->id = 0;
295
296         if (do_idr_lock)
297                 spin_unlock_irqrestore(&map_idr_lock, flags);
298         else
299                 __release(&map_idr_lock);
300 }
301
302 /* called from workqueue */
303 static void bpf_map_free_deferred(struct work_struct *work)
304 {
305         struct bpf_map *map = container_of(work, struct bpf_map, work);
306
307         bpf_map_release_memlock(map);
308         security_bpf_map_free(map);
309         /* implementation dependent freeing */
310         map->ops->map_free(map);
311 }
312
313 static void bpf_map_put_uref(struct bpf_map *map)
314 {
315         if (atomic_dec_and_test(&map->usercnt)) {
316                 if (map->ops->map_release_uref)
317                         map->ops->map_release_uref(map);
318         }
319 }
320
321 /* decrement map refcnt and schedule it for freeing via workqueue
322  * (unrelying map implementation ops->map_free() might sleep)
323  */
324 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
325 {
326         if (atomic_dec_and_test(&map->refcnt)) {
327                 /* bpf_map_free_id() must be called first */
328                 bpf_map_free_id(map, do_idr_lock);
329                 btf_put(map->btf);
330                 INIT_WORK(&map->work, bpf_map_free_deferred);
331                 schedule_work(&map->work);
332         }
333 }
334
335 void bpf_map_put(struct bpf_map *map)
336 {
337         __bpf_map_put(map, true);
338 }
339 EXPORT_SYMBOL_GPL(bpf_map_put);
340
341 void bpf_map_put_with_uref(struct bpf_map *map)
342 {
343         bpf_map_put_uref(map);
344         bpf_map_put(map);
345 }
346
347 static int bpf_map_release(struct inode *inode, struct file *filp)
348 {
349         struct bpf_map *map = filp->private_data;
350
351         if (map->ops->map_release)
352                 map->ops->map_release(map, filp);
353
354         bpf_map_put_with_uref(map);
355         return 0;
356 }
357
358 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
359 {
360         fmode_t mode = f.file->f_mode;
361
362         /* Our file permissions may have been overridden by global
363          * map permissions facing syscall side.
364          */
365         if (READ_ONCE(map->frozen))
366                 mode &= ~FMODE_CAN_WRITE;
367         return mode;
368 }
369
370 #ifdef CONFIG_PROC_FS
371 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
372 {
373         const struct bpf_map *map = filp->private_data;
374         const struct bpf_array *array;
375         u32 owner_prog_type = 0;
376         u32 owner_jited = 0;
377
378         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
379                 array = container_of(map, struct bpf_array, map);
380                 owner_prog_type = array->owner_prog_type;
381                 owner_jited = array->owner_jited;
382         }
383
384         seq_printf(m,
385                    "map_type:\t%u\n"
386                    "key_size:\t%u\n"
387                    "value_size:\t%u\n"
388                    "max_entries:\t%u\n"
389                    "map_flags:\t%#x\n"
390                    "memlock:\t%llu\n"
391                    "map_id:\t%u\n"
392                    "frozen:\t%u\n",
393                    map->map_type,
394                    map->key_size,
395                    map->value_size,
396                    map->max_entries,
397                    map->map_flags,
398                    map->pages * 1ULL << PAGE_SHIFT,
399                    map->id,
400                    READ_ONCE(map->frozen));
401
402         if (owner_prog_type) {
403                 seq_printf(m, "owner_prog_type:\t%u\n",
404                            owner_prog_type);
405                 seq_printf(m, "owner_jited:\t%u\n",
406                            owner_jited);
407         }
408 }
409 #endif
410
411 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
412                               loff_t *ppos)
413 {
414         /* We need this handler such that alloc_file() enables
415          * f_mode with FMODE_CAN_READ.
416          */
417         return -EINVAL;
418 }
419
420 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
421                                size_t siz, loff_t *ppos)
422 {
423         /* We need this handler such that alloc_file() enables
424          * f_mode with FMODE_CAN_WRITE.
425          */
426         return -EINVAL;
427 }
428
429 const struct file_operations bpf_map_fops = {
430 #ifdef CONFIG_PROC_FS
431         .show_fdinfo    = bpf_map_show_fdinfo,
432 #endif
433         .release        = bpf_map_release,
434         .read           = bpf_dummy_read,
435         .write          = bpf_dummy_write,
436 };
437
438 int bpf_map_new_fd(struct bpf_map *map, int flags)
439 {
440         int ret;
441
442         ret = security_bpf_map(map, OPEN_FMODE(flags));
443         if (ret < 0)
444                 return ret;
445
446         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
447                                 flags | O_CLOEXEC);
448 }
449
450 int bpf_get_file_flag(int flags)
451 {
452         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
453                 return -EINVAL;
454         if (flags & BPF_F_RDONLY)
455                 return O_RDONLY;
456         if (flags & BPF_F_WRONLY)
457                 return O_WRONLY;
458         return O_RDWR;
459 }
460
461 /* helper macro to check that unused fields 'union bpf_attr' are zero */
462 #define CHECK_ATTR(CMD) \
463         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
464                    sizeof(attr->CMD##_LAST_FIELD), 0, \
465                    sizeof(*attr) - \
466                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
467                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
468
469 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
470  * Return 0 on success and < 0 on error.
471  */
472 static int bpf_obj_name_cpy(char *dst, const char *src)
473 {
474         const char *end = src + BPF_OBJ_NAME_LEN;
475
476         memset(dst, 0, BPF_OBJ_NAME_LEN);
477         /* Copy all isalnum(), '_' and '.' chars. */
478         while (src < end && *src) {
479                 if (!isalnum(*src) &&
480                     *src != '_' && *src != '.')
481                         return -EINVAL;
482                 *dst++ = *src++;
483         }
484
485         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
486         if (src == end)
487                 return -EINVAL;
488
489         return 0;
490 }
491
492 int map_check_no_btf(const struct bpf_map *map,
493                      const struct btf *btf,
494                      const struct btf_type *key_type,
495                      const struct btf_type *value_type)
496 {
497         return -ENOTSUPP;
498 }
499
500 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
501                          u32 btf_key_id, u32 btf_value_id)
502 {
503         const struct btf_type *key_type, *value_type;
504         u32 key_size, value_size;
505         int ret = 0;
506
507         /* Some maps allow key to be unspecified. */
508         if (btf_key_id) {
509                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
510                 if (!key_type || key_size != map->key_size)
511                         return -EINVAL;
512         } else {
513                 key_type = btf_type_by_id(btf, 0);
514                 if (!map->ops->map_check_btf)
515                         return -EINVAL;
516         }
517
518         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
519         if (!value_type || value_size != map->value_size)
520                 return -EINVAL;
521
522         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
523
524         if (map_value_has_spin_lock(map)) {
525                 if (map->map_flags & BPF_F_RDONLY_PROG)
526                         return -EACCES;
527                 if (map->map_type != BPF_MAP_TYPE_HASH &&
528                     map->map_type != BPF_MAP_TYPE_ARRAY &&
529                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
530                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
531                         return -ENOTSUPP;
532                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
533                     map->value_size) {
534                         WARN_ONCE(1,
535                                   "verifier bug spin_lock_off %d value_size %d\n",
536                                   map->spin_lock_off, map->value_size);
537                         return -EFAULT;
538                 }
539         }
540
541         if (map->ops->map_check_btf)
542                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
543
544         return ret;
545 }
546
547 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
548 /* called via syscall */
549 static int map_create(union bpf_attr *attr)
550 {
551         int numa_node = bpf_map_attr_numa_node(attr);
552         struct bpf_map *map;
553         int f_flags;
554         int err;
555
556         err = CHECK_ATTR(BPF_MAP_CREATE);
557         if (err)
558                 return -EINVAL;
559
560         f_flags = bpf_get_file_flag(attr->map_flags);
561         if (f_flags < 0)
562                 return f_flags;
563
564         if (numa_node != NUMA_NO_NODE &&
565             ((unsigned int)numa_node >= nr_node_ids ||
566              !node_online(numa_node)))
567                 return -EINVAL;
568
569         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
570         map = find_and_alloc_map(attr);
571         if (IS_ERR(map))
572                 return PTR_ERR(map);
573
574         err = bpf_obj_name_cpy(map->name, attr->map_name);
575         if (err)
576                 goto free_map_nouncharge;
577
578         atomic_set(&map->refcnt, 1);
579         atomic_set(&map->usercnt, 1);
580
581         if (attr->btf_key_type_id || attr->btf_value_type_id) {
582                 struct btf *btf;
583
584                 if (!attr->btf_value_type_id) {
585                         err = -EINVAL;
586                         goto free_map_nouncharge;
587                 }
588
589                 btf = btf_get_by_fd(attr->btf_fd);
590                 if (IS_ERR(btf)) {
591                         err = PTR_ERR(btf);
592                         goto free_map_nouncharge;
593                 }
594
595                 err = map_check_btf(map, btf, attr->btf_key_type_id,
596                                     attr->btf_value_type_id);
597                 if (err) {
598                         btf_put(btf);
599                         goto free_map_nouncharge;
600                 }
601
602                 map->btf = btf;
603                 map->btf_key_type_id = attr->btf_key_type_id;
604                 map->btf_value_type_id = attr->btf_value_type_id;
605         } else {
606                 map->spin_lock_off = -EINVAL;
607         }
608
609         err = security_bpf_map_alloc(map);
610         if (err)
611                 goto free_map_nouncharge;
612
613         err = bpf_map_init_memlock(map);
614         if (err)
615                 goto free_map_sec;
616
617         err = bpf_map_alloc_id(map);
618         if (err)
619                 goto free_map;
620
621         err = bpf_map_new_fd(map, f_flags);
622         if (err < 0) {
623                 /* failed to allocate fd.
624                  * bpf_map_put_with_uref() is needed because the above
625                  * bpf_map_alloc_id() has published the map
626                  * to the userspace and the userspace may
627                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
628                  */
629                 bpf_map_put_with_uref(map);
630                 return err;
631         }
632
633         return err;
634
635 free_map:
636         bpf_map_release_memlock(map);
637 free_map_sec:
638         security_bpf_map_free(map);
639 free_map_nouncharge:
640         btf_put(map->btf);
641         map->ops->map_free(map);
642         return err;
643 }
644
645 /* if error is returned, fd is released.
646  * On success caller should complete fd access with matching fdput()
647  */
648 struct bpf_map *__bpf_map_get(struct fd f)
649 {
650         if (!f.file)
651                 return ERR_PTR(-EBADF);
652         if (f.file->f_op != &bpf_map_fops) {
653                 fdput(f);
654                 return ERR_PTR(-EINVAL);
655         }
656
657         return f.file->private_data;
658 }
659
660 /* prog's and map's refcnt limit */
661 #define BPF_MAX_REFCNT 32768
662
663 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
664 {
665         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
666                 atomic_dec(&map->refcnt);
667                 return ERR_PTR(-EBUSY);
668         }
669         if (uref)
670                 atomic_inc(&map->usercnt);
671         return map;
672 }
673 EXPORT_SYMBOL_GPL(bpf_map_inc);
674
675 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
676 {
677         struct fd f = fdget(ufd);
678         struct bpf_map *map;
679
680         map = __bpf_map_get(f);
681         if (IS_ERR(map))
682                 return map;
683
684         map = bpf_map_inc(map, true);
685         fdput(f);
686
687         return map;
688 }
689
690 /* map_idr_lock should have been held */
691 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
692                                             bool uref)
693 {
694         int refold;
695
696         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
697
698         if (refold >= BPF_MAX_REFCNT) {
699                 __bpf_map_put(map, false);
700                 return ERR_PTR(-EBUSY);
701         }
702
703         if (!refold)
704                 return ERR_PTR(-ENOENT);
705
706         if (uref)
707                 atomic_inc(&map->usercnt);
708
709         return map;
710 }
711
712 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
713 {
714         return -ENOTSUPP;
715 }
716
717 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
718 {
719         if (key_size)
720                 return memdup_user(ukey, key_size);
721
722         if (ukey)
723                 return ERR_PTR(-EINVAL);
724
725         return NULL;
726 }
727
728 /* last field in 'union bpf_attr' used by this command */
729 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
730
731 static int map_lookup_elem(union bpf_attr *attr)
732 {
733         void __user *ukey = u64_to_user_ptr(attr->key);
734         void __user *uvalue = u64_to_user_ptr(attr->value);
735         int ufd = attr->map_fd;
736         struct bpf_map *map;
737         void *key, *value, *ptr;
738         u32 value_size;
739         struct fd f;
740         int err;
741
742         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
743                 return -EINVAL;
744
745         if (attr->flags & ~BPF_F_LOCK)
746                 return -EINVAL;
747
748         f = fdget(ufd);
749         map = __bpf_map_get(f);
750         if (IS_ERR(map))
751                 return PTR_ERR(map);
752         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
753                 err = -EPERM;
754                 goto err_put;
755         }
756
757         if ((attr->flags & BPF_F_LOCK) &&
758             !map_value_has_spin_lock(map)) {
759                 err = -EINVAL;
760                 goto err_put;
761         }
762
763         key = __bpf_copy_key(ukey, map->key_size);
764         if (IS_ERR(key)) {
765                 err = PTR_ERR(key);
766                 goto err_put;
767         }
768
769         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
770             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
771             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
772             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
773                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
774         else if (IS_FD_MAP(map))
775                 value_size = sizeof(u32);
776         else
777                 value_size = map->value_size;
778
779         err = -ENOMEM;
780         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
781         if (!value)
782                 goto free_key;
783
784         if (bpf_map_is_dev_bound(map)) {
785                 err = bpf_map_offload_lookup_elem(map, key, value);
786                 goto done;
787         }
788
789         preempt_disable();
790         this_cpu_inc(bpf_prog_active);
791         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
792             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
793                 err = bpf_percpu_hash_copy(map, key, value);
794         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
795                 err = bpf_percpu_array_copy(map, key, value);
796         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
797                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
798         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
799                 err = bpf_stackmap_copy(map, key, value);
800         } else if (IS_FD_ARRAY(map)) {
801                 err = bpf_fd_array_map_lookup_elem(map, key, value);
802         } else if (IS_FD_HASH(map)) {
803                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
804         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
805                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
806         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
807                    map->map_type == BPF_MAP_TYPE_STACK) {
808                 err = map->ops->map_peek_elem(map, value);
809         } else {
810                 rcu_read_lock();
811                 ptr = map->ops->map_lookup_elem(map, key);
812                 if (IS_ERR(ptr)) {
813                         err = PTR_ERR(ptr);
814                 } else if (!ptr) {
815                         err = -ENOENT;
816                 } else {
817                         err = 0;
818                         if (attr->flags & BPF_F_LOCK)
819                                 /* lock 'ptr' and copy everything but lock */
820                                 copy_map_value_locked(map, value, ptr, true);
821                         else
822                                 copy_map_value(map, value, ptr);
823                         /* mask lock, since value wasn't zero inited */
824                         check_and_init_map_lock(map, value);
825                 }
826                 rcu_read_unlock();
827         }
828         this_cpu_dec(bpf_prog_active);
829         preempt_enable();
830
831 done:
832         if (err)
833                 goto free_value;
834
835         err = -EFAULT;
836         if (copy_to_user(uvalue, value, value_size) != 0)
837                 goto free_value;
838
839         err = 0;
840
841 free_value:
842         kfree(value);
843 free_key:
844         kfree(key);
845 err_put:
846         fdput(f);
847         return err;
848 }
849
850 static void maybe_wait_bpf_programs(struct bpf_map *map)
851 {
852         /* Wait for any running BPF programs to complete so that
853          * userspace, when we return to it, knows that all programs
854          * that could be running use the new map value.
855          */
856         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
857             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
858                 synchronize_rcu();
859 }
860
861 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
862
863 static int map_update_elem(union bpf_attr *attr)
864 {
865         void __user *ukey = u64_to_user_ptr(attr->key);
866         void __user *uvalue = u64_to_user_ptr(attr->value);
867         int ufd = attr->map_fd;
868         struct bpf_map *map;
869         void *key, *value;
870         u32 value_size;
871         struct fd f;
872         int err;
873
874         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
875                 return -EINVAL;
876
877         f = fdget(ufd);
878         map = __bpf_map_get(f);
879         if (IS_ERR(map))
880                 return PTR_ERR(map);
881         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
882                 err = -EPERM;
883                 goto err_put;
884         }
885
886         if ((attr->flags & BPF_F_LOCK) &&
887             !map_value_has_spin_lock(map)) {
888                 err = -EINVAL;
889                 goto err_put;
890         }
891
892         key = __bpf_copy_key(ukey, map->key_size);
893         if (IS_ERR(key)) {
894                 err = PTR_ERR(key);
895                 goto err_put;
896         }
897
898         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
899             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
900             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
901             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
902                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
903         else
904                 value_size = map->value_size;
905
906         err = -ENOMEM;
907         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
908         if (!value)
909                 goto free_key;
910
911         err = -EFAULT;
912         if (copy_from_user(value, uvalue, value_size) != 0)
913                 goto free_value;
914
915         /* Need to create a kthread, thus must support schedule */
916         if (bpf_map_is_dev_bound(map)) {
917                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
918                 goto out;
919         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
920                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
921                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
922                 err = map->ops->map_update_elem(map, key, value, attr->flags);
923                 goto out;
924         }
925
926         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
927          * inside bpf map update or delete otherwise deadlocks are possible
928          */
929         preempt_disable();
930         __this_cpu_inc(bpf_prog_active);
931         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
932             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
933                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
934         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
935                 err = bpf_percpu_array_update(map, key, value, attr->flags);
936         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
937                 err = bpf_percpu_cgroup_storage_update(map, key, value,
938                                                        attr->flags);
939         } else if (IS_FD_ARRAY(map)) {
940                 rcu_read_lock();
941                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
942                                                    attr->flags);
943                 rcu_read_unlock();
944         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
945                 rcu_read_lock();
946                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
947                                                   attr->flags);
948                 rcu_read_unlock();
949         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
950                 /* rcu_read_lock() is not needed */
951                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
952                                                          attr->flags);
953         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
954                    map->map_type == BPF_MAP_TYPE_STACK) {
955                 err = map->ops->map_push_elem(map, value, attr->flags);
956         } else {
957                 rcu_read_lock();
958                 err = map->ops->map_update_elem(map, key, value, attr->flags);
959                 rcu_read_unlock();
960         }
961         __this_cpu_dec(bpf_prog_active);
962         preempt_enable();
963         maybe_wait_bpf_programs(map);
964 out:
965 free_value:
966         kfree(value);
967 free_key:
968         kfree(key);
969 err_put:
970         fdput(f);
971         return err;
972 }
973
974 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
975
976 static int map_delete_elem(union bpf_attr *attr)
977 {
978         void __user *ukey = u64_to_user_ptr(attr->key);
979         int ufd = attr->map_fd;
980         struct bpf_map *map;
981         struct fd f;
982         void *key;
983         int err;
984
985         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
986                 return -EINVAL;
987
988         f = fdget(ufd);
989         map = __bpf_map_get(f);
990         if (IS_ERR(map))
991                 return PTR_ERR(map);
992         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
993                 err = -EPERM;
994                 goto err_put;
995         }
996
997         key = __bpf_copy_key(ukey, map->key_size);
998         if (IS_ERR(key)) {
999                 err = PTR_ERR(key);
1000                 goto err_put;
1001         }
1002
1003         if (bpf_map_is_dev_bound(map)) {
1004                 err = bpf_map_offload_delete_elem(map, key);
1005                 goto out;
1006         }
1007
1008         preempt_disable();
1009         __this_cpu_inc(bpf_prog_active);
1010         rcu_read_lock();
1011         err = map->ops->map_delete_elem(map, key);
1012         rcu_read_unlock();
1013         __this_cpu_dec(bpf_prog_active);
1014         preempt_enable();
1015         maybe_wait_bpf_programs(map);
1016 out:
1017         kfree(key);
1018 err_put:
1019         fdput(f);
1020         return err;
1021 }
1022
1023 /* last field in 'union bpf_attr' used by this command */
1024 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1025
1026 static int map_get_next_key(union bpf_attr *attr)
1027 {
1028         void __user *ukey = u64_to_user_ptr(attr->key);
1029         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1030         int ufd = attr->map_fd;
1031         struct bpf_map *map;
1032         void *key, *next_key;
1033         struct fd f;
1034         int err;
1035
1036         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1037                 return -EINVAL;
1038
1039         f = fdget(ufd);
1040         map = __bpf_map_get(f);
1041         if (IS_ERR(map))
1042                 return PTR_ERR(map);
1043         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1044                 err = -EPERM;
1045                 goto err_put;
1046         }
1047
1048         if (ukey) {
1049                 key = __bpf_copy_key(ukey, map->key_size);
1050                 if (IS_ERR(key)) {
1051                         err = PTR_ERR(key);
1052                         goto err_put;
1053                 }
1054         } else {
1055                 key = NULL;
1056         }
1057
1058         err = -ENOMEM;
1059         next_key = kmalloc(map->key_size, GFP_USER);
1060         if (!next_key)
1061                 goto free_key;
1062
1063         if (bpf_map_is_dev_bound(map)) {
1064                 err = bpf_map_offload_get_next_key(map, key, next_key);
1065                 goto out;
1066         }
1067
1068         rcu_read_lock();
1069         err = map->ops->map_get_next_key(map, key, next_key);
1070         rcu_read_unlock();
1071 out:
1072         if (err)
1073                 goto free_next_key;
1074
1075         err = -EFAULT;
1076         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1077                 goto free_next_key;
1078
1079         err = 0;
1080
1081 free_next_key:
1082         kfree(next_key);
1083 free_key:
1084         kfree(key);
1085 err_put:
1086         fdput(f);
1087         return err;
1088 }
1089
1090 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1091
1092 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1093 {
1094         void __user *ukey = u64_to_user_ptr(attr->key);
1095         void __user *uvalue = u64_to_user_ptr(attr->value);
1096         int ufd = attr->map_fd;
1097         struct bpf_map *map;
1098         void *key, *value;
1099         u32 value_size;
1100         struct fd f;
1101         int err;
1102
1103         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1104                 return -EINVAL;
1105
1106         f = fdget(ufd);
1107         map = __bpf_map_get(f);
1108         if (IS_ERR(map))
1109                 return PTR_ERR(map);
1110         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1111                 err = -EPERM;
1112                 goto err_put;
1113         }
1114
1115         key = __bpf_copy_key(ukey, map->key_size);
1116         if (IS_ERR(key)) {
1117                 err = PTR_ERR(key);
1118                 goto err_put;
1119         }
1120
1121         value_size = map->value_size;
1122
1123         err = -ENOMEM;
1124         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1125         if (!value)
1126                 goto free_key;
1127
1128         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1129             map->map_type == BPF_MAP_TYPE_STACK) {
1130                 err = map->ops->map_pop_elem(map, value);
1131         } else {
1132                 err = -ENOTSUPP;
1133         }
1134
1135         if (err)
1136                 goto free_value;
1137
1138         if (copy_to_user(uvalue, value, value_size) != 0)
1139                 goto free_value;
1140
1141         err = 0;
1142
1143 free_value:
1144         kfree(value);
1145 free_key:
1146         kfree(key);
1147 err_put:
1148         fdput(f);
1149         return err;
1150 }
1151
1152 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1153
1154 static int map_freeze(const union bpf_attr *attr)
1155 {
1156         int err = 0, ufd = attr->map_fd;
1157         struct bpf_map *map;
1158         struct fd f;
1159
1160         if (CHECK_ATTR(BPF_MAP_FREEZE))
1161                 return -EINVAL;
1162
1163         f = fdget(ufd);
1164         map = __bpf_map_get(f);
1165         if (IS_ERR(map))
1166                 return PTR_ERR(map);
1167         if (READ_ONCE(map->frozen)) {
1168                 err = -EBUSY;
1169                 goto err_put;
1170         }
1171         if (!capable(CAP_SYS_ADMIN)) {
1172                 err = -EPERM;
1173                 goto err_put;
1174         }
1175
1176         WRITE_ONCE(map->frozen, true);
1177 err_put:
1178         fdput(f);
1179         return err;
1180 }
1181
1182 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1183 #define BPF_PROG_TYPE(_id, _name) \
1184         [_id] = & _name ## _prog_ops,
1185 #define BPF_MAP_TYPE(_id, _ops)
1186 #include <linux/bpf_types.h>
1187 #undef BPF_PROG_TYPE
1188 #undef BPF_MAP_TYPE
1189 };
1190
1191 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1192 {
1193         const struct bpf_prog_ops *ops;
1194
1195         if (type >= ARRAY_SIZE(bpf_prog_types))
1196                 return -EINVAL;
1197         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1198         ops = bpf_prog_types[type];
1199         if (!ops)
1200                 return -EINVAL;
1201
1202         if (!bpf_prog_is_dev_bound(prog->aux))
1203                 prog->aux->ops = ops;
1204         else
1205                 prog->aux->ops = &bpf_offload_prog_ops;
1206         prog->type = type;
1207         return 0;
1208 }
1209
1210 /* drop refcnt on maps used by eBPF program and free auxilary data */
1211 static void free_used_maps(struct bpf_prog_aux *aux)
1212 {
1213         enum bpf_cgroup_storage_type stype;
1214         int i;
1215
1216         for_each_cgroup_storage_type(stype) {
1217                 if (!aux->cgroup_storage[stype])
1218                         continue;
1219                 bpf_cgroup_storage_release(aux->prog,
1220                                            aux->cgroup_storage[stype]);
1221         }
1222
1223         for (i = 0; i < aux->used_map_cnt; i++)
1224                 bpf_map_put(aux->used_maps[i]);
1225
1226         kfree(aux->used_maps);
1227 }
1228
1229 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1230 {
1231         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1232         unsigned long user_bufs;
1233
1234         if (user) {
1235                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1236                 if (user_bufs > memlock_limit) {
1237                         atomic_long_sub(pages, &user->locked_vm);
1238                         return -EPERM;
1239                 }
1240         }
1241
1242         return 0;
1243 }
1244
1245 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1246 {
1247         if (user)
1248                 atomic_long_sub(pages, &user->locked_vm);
1249 }
1250
1251 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1252 {
1253         struct user_struct *user = get_current_user();
1254         int ret;
1255
1256         ret = __bpf_prog_charge(user, prog->pages);
1257         if (ret) {
1258                 free_uid(user);
1259                 return ret;
1260         }
1261
1262         prog->aux->user = user;
1263         return 0;
1264 }
1265
1266 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1267 {
1268         struct user_struct *user = prog->aux->user;
1269
1270         __bpf_prog_uncharge(user, prog->pages);
1271         free_uid(user);
1272 }
1273
1274 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1275 {
1276         int id;
1277
1278         idr_preload(GFP_KERNEL);
1279         spin_lock_bh(&prog_idr_lock);
1280         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1281         if (id > 0)
1282                 prog->aux->id = id;
1283         spin_unlock_bh(&prog_idr_lock);
1284         idr_preload_end();
1285
1286         /* id is in [1, INT_MAX) */
1287         if (WARN_ON_ONCE(!id))
1288                 return -ENOSPC;
1289
1290         return id > 0 ? 0 : id;
1291 }
1292
1293 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1294 {
1295         /* cBPF to eBPF migrations are currently not in the idr store.
1296          * Offloaded programs are removed from the store when their device
1297          * disappears - even if someone grabs an fd to them they are unusable,
1298          * simply waiting for refcnt to drop to be freed.
1299          */
1300         if (!prog->aux->id)
1301                 return;
1302
1303         if (do_idr_lock)
1304                 spin_lock_bh(&prog_idr_lock);
1305         else
1306                 __acquire(&prog_idr_lock);
1307
1308         idr_remove(&prog_idr, prog->aux->id);
1309         prog->aux->id = 0;
1310
1311         if (do_idr_lock)
1312                 spin_unlock_bh(&prog_idr_lock);
1313         else
1314                 __release(&prog_idr_lock);
1315 }
1316
1317 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1318 {
1319         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1320
1321         free_used_maps(aux);
1322         bpf_prog_uncharge_memlock(aux->prog);
1323         security_bpf_prog_free(aux);
1324         bpf_prog_free(aux->prog);
1325 }
1326
1327 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1328 {
1329         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1330                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1331                 /* bpf_prog_free_id() must be called first */
1332                 bpf_prog_free_id(prog, do_idr_lock);
1333                 bpf_prog_kallsyms_del_all(prog);
1334                 btf_put(prog->aux->btf);
1335                 kvfree(prog->aux->func_info);
1336                 bpf_prog_free_linfo(prog);
1337
1338                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1339         }
1340 }
1341
1342 void bpf_prog_put(struct bpf_prog *prog)
1343 {
1344         __bpf_prog_put(prog, true);
1345 }
1346 EXPORT_SYMBOL_GPL(bpf_prog_put);
1347
1348 static int bpf_prog_release(struct inode *inode, struct file *filp)
1349 {
1350         struct bpf_prog *prog = filp->private_data;
1351
1352         bpf_prog_put(prog);
1353         return 0;
1354 }
1355
1356 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1357                                struct bpf_prog_stats *stats)
1358 {
1359         u64 nsecs = 0, cnt = 0;
1360         int cpu;
1361
1362         for_each_possible_cpu(cpu) {
1363                 const struct bpf_prog_stats *st;
1364                 unsigned int start;
1365                 u64 tnsecs, tcnt;
1366
1367                 st = per_cpu_ptr(prog->aux->stats, cpu);
1368                 do {
1369                         start = u64_stats_fetch_begin_irq(&st->syncp);
1370                         tnsecs = st->nsecs;
1371                         tcnt = st->cnt;
1372                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1373                 nsecs += tnsecs;
1374                 cnt += tcnt;
1375         }
1376         stats->nsecs = nsecs;
1377         stats->cnt = cnt;
1378 }
1379
1380 #ifdef CONFIG_PROC_FS
1381 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1382 {
1383         const struct bpf_prog *prog = filp->private_data;
1384         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1385         struct bpf_prog_stats stats;
1386
1387         bpf_prog_get_stats(prog, &stats);
1388         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1389         seq_printf(m,
1390                    "prog_type:\t%u\n"
1391                    "prog_jited:\t%u\n"
1392                    "prog_tag:\t%s\n"
1393                    "memlock:\t%llu\n"
1394                    "prog_id:\t%u\n"
1395                    "run_time_ns:\t%llu\n"
1396                    "run_cnt:\t%llu\n",
1397                    prog->type,
1398                    prog->jited,
1399                    prog_tag,
1400                    prog->pages * 1ULL << PAGE_SHIFT,
1401                    prog->aux->id,
1402                    stats.nsecs,
1403                    stats.cnt);
1404 }
1405 #endif
1406
1407 const struct file_operations bpf_prog_fops = {
1408 #ifdef CONFIG_PROC_FS
1409         .show_fdinfo    = bpf_prog_show_fdinfo,
1410 #endif
1411         .release        = bpf_prog_release,
1412         .read           = bpf_dummy_read,
1413         .write          = bpf_dummy_write,
1414 };
1415
1416 int bpf_prog_new_fd(struct bpf_prog *prog)
1417 {
1418         int ret;
1419
1420         ret = security_bpf_prog(prog);
1421         if (ret < 0)
1422                 return ret;
1423
1424         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1425                                 O_RDWR | O_CLOEXEC);
1426 }
1427
1428 static struct bpf_prog *____bpf_prog_get(struct fd f)
1429 {
1430         if (!f.file)
1431                 return ERR_PTR(-EBADF);
1432         if (f.file->f_op != &bpf_prog_fops) {
1433                 fdput(f);
1434                 return ERR_PTR(-EINVAL);
1435         }
1436
1437         return f.file->private_data;
1438 }
1439
1440 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1441 {
1442         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1443                 atomic_sub(i, &prog->aux->refcnt);
1444                 return ERR_PTR(-EBUSY);
1445         }
1446         return prog;
1447 }
1448 EXPORT_SYMBOL_GPL(bpf_prog_add);
1449
1450 void bpf_prog_sub(struct bpf_prog *prog, int i)
1451 {
1452         /* Only to be used for undoing previous bpf_prog_add() in some
1453          * error path. We still know that another entity in our call
1454          * path holds a reference to the program, thus atomic_sub() can
1455          * be safely used in such cases!
1456          */
1457         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1458 }
1459 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1460
1461 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1462 {
1463         return bpf_prog_add(prog, 1);
1464 }
1465 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1466
1467 /* prog_idr_lock should have been held */
1468 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1469 {
1470         int refold;
1471
1472         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1473
1474         if (refold >= BPF_MAX_REFCNT) {
1475                 __bpf_prog_put(prog, false);
1476                 return ERR_PTR(-EBUSY);
1477         }
1478
1479         if (!refold)
1480                 return ERR_PTR(-ENOENT);
1481
1482         return prog;
1483 }
1484 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1485
1486 bool bpf_prog_get_ok(struct bpf_prog *prog,
1487                             enum bpf_prog_type *attach_type, bool attach_drv)
1488 {
1489         /* not an attachment, just a refcount inc, always allow */
1490         if (!attach_type)
1491                 return true;
1492
1493         if (prog->type != *attach_type)
1494                 return false;
1495         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1496                 return false;
1497
1498         return true;
1499 }
1500
1501 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1502                                        bool attach_drv)
1503 {
1504         struct fd f = fdget(ufd);
1505         struct bpf_prog *prog;
1506
1507         prog = ____bpf_prog_get(f);
1508         if (IS_ERR(prog))
1509                 return prog;
1510         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1511                 prog = ERR_PTR(-EINVAL);
1512                 goto out;
1513         }
1514
1515         prog = bpf_prog_inc(prog);
1516 out:
1517         fdput(f);
1518         return prog;
1519 }
1520
1521 struct bpf_prog *bpf_prog_get(u32 ufd)
1522 {
1523         return __bpf_prog_get(ufd, NULL, false);
1524 }
1525
1526 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1527                                        bool attach_drv)
1528 {
1529         return __bpf_prog_get(ufd, &type, attach_drv);
1530 }
1531 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1532
1533 /* Initially all BPF programs could be loaded w/o specifying
1534  * expected_attach_type. Later for some of them specifying expected_attach_type
1535  * at load time became required so that program could be validated properly.
1536  * Programs of types that are allowed to be loaded both w/ and w/o (for
1537  * backward compatibility) expected_attach_type, should have the default attach
1538  * type assigned to expected_attach_type for the latter case, so that it can be
1539  * validated later at attach time.
1540  *
1541  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1542  * prog type requires it but has some attach types that have to be backward
1543  * compatible.
1544  */
1545 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1546 {
1547         switch (attr->prog_type) {
1548         case BPF_PROG_TYPE_CGROUP_SOCK:
1549                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1550                  * exist so checking for non-zero is the way to go here.
1551                  */
1552                 if (!attr->expected_attach_type)
1553                         attr->expected_attach_type =
1554                                 BPF_CGROUP_INET_SOCK_CREATE;
1555                 break;
1556         }
1557 }
1558
1559 static int
1560 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1561                                 enum bpf_attach_type expected_attach_type)
1562 {
1563         switch (prog_type) {
1564         case BPF_PROG_TYPE_CGROUP_SOCK:
1565                 switch (expected_attach_type) {
1566                 case BPF_CGROUP_INET_SOCK_CREATE:
1567                 case BPF_CGROUP_INET4_POST_BIND:
1568                 case BPF_CGROUP_INET6_POST_BIND:
1569                         return 0;
1570                 default:
1571                         return -EINVAL;
1572                 }
1573         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1574                 switch (expected_attach_type) {
1575                 case BPF_CGROUP_INET4_BIND:
1576                 case BPF_CGROUP_INET6_BIND:
1577                 case BPF_CGROUP_INET4_CONNECT:
1578                 case BPF_CGROUP_INET6_CONNECT:
1579                 case BPF_CGROUP_UDP4_SENDMSG:
1580                 case BPF_CGROUP_UDP6_SENDMSG:
1581                         return 0;
1582                 default:
1583                         return -EINVAL;
1584                 }
1585         default:
1586                 return 0;
1587         }
1588 }
1589
1590 /* last field in 'union bpf_attr' used by this command */
1591 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1592
1593 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1594 {
1595         enum bpf_prog_type type = attr->prog_type;
1596         struct bpf_prog *prog;
1597         int err;
1598         char license[128];
1599         bool is_gpl;
1600
1601         if (CHECK_ATTR(BPF_PROG_LOAD))
1602                 return -EINVAL;
1603
1604         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1605                 return -EINVAL;
1606
1607         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1608             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1609             !capable(CAP_SYS_ADMIN))
1610                 return -EPERM;
1611
1612         /* copy eBPF program license from user space */
1613         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1614                               sizeof(license) - 1) < 0)
1615                 return -EFAULT;
1616         license[sizeof(license) - 1] = 0;
1617
1618         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1619         is_gpl = license_is_gpl_compatible(license);
1620
1621         if (attr->insn_cnt == 0 ||
1622             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1623                 return -E2BIG;
1624         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1625             type != BPF_PROG_TYPE_CGROUP_SKB &&
1626             !capable(CAP_SYS_ADMIN))
1627                 return -EPERM;
1628
1629         bpf_prog_load_fixup_attach_type(attr);
1630         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1631                 return -EINVAL;
1632
1633         /* plain bpf_prog allocation */
1634         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1635         if (!prog)
1636                 return -ENOMEM;
1637
1638         prog->expected_attach_type = attr->expected_attach_type;
1639
1640         prog->aux->offload_requested = !!attr->prog_ifindex;
1641
1642         err = security_bpf_prog_alloc(prog->aux);
1643         if (err)
1644                 goto free_prog_nouncharge;
1645
1646         err = bpf_prog_charge_memlock(prog);
1647         if (err)
1648                 goto free_prog_sec;
1649
1650         prog->len = attr->insn_cnt;
1651
1652         err = -EFAULT;
1653         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1654                            bpf_prog_insn_size(prog)) != 0)
1655                 goto free_prog;
1656
1657         prog->orig_prog = NULL;
1658         prog->jited = 0;
1659
1660         atomic_set(&prog->aux->refcnt, 1);
1661         prog->gpl_compatible = is_gpl ? 1 : 0;
1662
1663         if (bpf_prog_is_dev_bound(prog->aux)) {
1664                 err = bpf_prog_offload_init(prog, attr);
1665                 if (err)
1666                         goto free_prog;
1667         }
1668
1669         /* find program type: socket_filter vs tracing_filter */
1670         err = find_prog_type(type, prog);
1671         if (err < 0)
1672                 goto free_prog;
1673
1674         prog->aux->load_time = ktime_get_boot_ns();
1675         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1676         if (err)
1677                 goto free_prog;
1678
1679         /* run eBPF verifier */
1680         err = bpf_check(&prog, attr, uattr);
1681         if (err < 0)
1682                 goto free_used_maps;
1683
1684         prog = bpf_prog_select_runtime(prog, &err);
1685         if (err < 0)
1686                 goto free_used_maps;
1687
1688         err = bpf_prog_alloc_id(prog);
1689         if (err)
1690                 goto free_used_maps;
1691
1692         err = bpf_prog_new_fd(prog);
1693         if (err < 0) {
1694                 /* failed to allocate fd.
1695                  * bpf_prog_put() is needed because the above
1696                  * bpf_prog_alloc_id() has published the prog
1697                  * to the userspace and the userspace may
1698                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1699                  */
1700                 bpf_prog_put(prog);
1701                 return err;
1702         }
1703
1704         bpf_prog_kallsyms_add(prog);
1705         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1706         return err;
1707
1708 free_used_maps:
1709         bpf_prog_free_linfo(prog);
1710         kvfree(prog->aux->func_info);
1711         btf_put(prog->aux->btf);
1712         bpf_prog_kallsyms_del_subprogs(prog);
1713         free_used_maps(prog->aux);
1714 free_prog:
1715         bpf_prog_uncharge_memlock(prog);
1716 free_prog_sec:
1717         security_bpf_prog_free(prog->aux);
1718 free_prog_nouncharge:
1719         bpf_prog_free(prog);
1720         return err;
1721 }
1722
1723 #define BPF_OBJ_LAST_FIELD file_flags
1724
1725 static int bpf_obj_pin(const union bpf_attr *attr)
1726 {
1727         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1728                 return -EINVAL;
1729
1730         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1731 }
1732
1733 static int bpf_obj_get(const union bpf_attr *attr)
1734 {
1735         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1736             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1737                 return -EINVAL;
1738
1739         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1740                                 attr->file_flags);
1741 }
1742
1743 struct bpf_raw_tracepoint {
1744         struct bpf_raw_event_map *btp;
1745         struct bpf_prog *prog;
1746 };
1747
1748 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1749 {
1750         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1751
1752         if (raw_tp->prog) {
1753                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1754                 bpf_prog_put(raw_tp->prog);
1755         }
1756         bpf_put_raw_tracepoint(raw_tp->btp);
1757         kfree(raw_tp);
1758         return 0;
1759 }
1760
1761 static const struct file_operations bpf_raw_tp_fops = {
1762         .release        = bpf_raw_tracepoint_release,
1763         .read           = bpf_dummy_read,
1764         .write          = bpf_dummy_write,
1765 };
1766
1767 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1768
1769 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1770 {
1771         struct bpf_raw_tracepoint *raw_tp;
1772         struct bpf_raw_event_map *btp;
1773         struct bpf_prog *prog;
1774         char tp_name[128];
1775         int tp_fd, err;
1776
1777         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1778                               sizeof(tp_name) - 1) < 0)
1779                 return -EFAULT;
1780         tp_name[sizeof(tp_name) - 1] = 0;
1781
1782         btp = bpf_get_raw_tracepoint(tp_name);
1783         if (!btp)
1784                 return -ENOENT;
1785
1786         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1787         if (!raw_tp) {
1788                 err = -ENOMEM;
1789                 goto out_put_btp;
1790         }
1791         raw_tp->btp = btp;
1792
1793         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1794         if (IS_ERR(prog)) {
1795                 err = PTR_ERR(prog);
1796                 goto out_free_tp;
1797         }
1798         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1799             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1800                 err = -EINVAL;
1801                 goto out_put_prog;
1802         }
1803
1804         err = bpf_probe_register(raw_tp->btp, prog);
1805         if (err)
1806                 goto out_put_prog;
1807
1808         raw_tp->prog = prog;
1809         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1810                                  O_CLOEXEC);
1811         if (tp_fd < 0) {
1812                 bpf_probe_unregister(raw_tp->btp, prog);
1813                 err = tp_fd;
1814                 goto out_put_prog;
1815         }
1816         return tp_fd;
1817
1818 out_put_prog:
1819         bpf_prog_put(prog);
1820 out_free_tp:
1821         kfree(raw_tp);
1822 out_put_btp:
1823         bpf_put_raw_tracepoint(btp);
1824         return err;
1825 }
1826
1827 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1828                                              enum bpf_attach_type attach_type)
1829 {
1830         switch (prog->type) {
1831         case BPF_PROG_TYPE_CGROUP_SOCK:
1832         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1833                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1834         default:
1835                 return 0;
1836         }
1837 }
1838
1839 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1840
1841 #define BPF_F_ATTACH_MASK \
1842         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1843
1844 static int bpf_prog_attach(const union bpf_attr *attr)
1845 {
1846         enum bpf_prog_type ptype;
1847         struct bpf_prog *prog;
1848         int ret;
1849
1850         if (!capable(CAP_NET_ADMIN))
1851                 return -EPERM;
1852
1853         if (CHECK_ATTR(BPF_PROG_ATTACH))
1854                 return -EINVAL;
1855
1856         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1857                 return -EINVAL;
1858
1859         switch (attr->attach_type) {
1860         case BPF_CGROUP_INET_INGRESS:
1861         case BPF_CGROUP_INET_EGRESS:
1862                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1863                 break;
1864         case BPF_CGROUP_INET_SOCK_CREATE:
1865         case BPF_CGROUP_INET4_POST_BIND:
1866         case BPF_CGROUP_INET6_POST_BIND:
1867                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1868                 break;
1869         case BPF_CGROUP_INET4_BIND:
1870         case BPF_CGROUP_INET6_BIND:
1871         case BPF_CGROUP_INET4_CONNECT:
1872         case BPF_CGROUP_INET6_CONNECT:
1873         case BPF_CGROUP_UDP4_SENDMSG:
1874         case BPF_CGROUP_UDP6_SENDMSG:
1875                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1876                 break;
1877         case BPF_CGROUP_SOCK_OPS:
1878                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1879                 break;
1880         case BPF_CGROUP_DEVICE:
1881                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1882                 break;
1883         case BPF_SK_MSG_VERDICT:
1884                 ptype = BPF_PROG_TYPE_SK_MSG;
1885                 break;
1886         case BPF_SK_SKB_STREAM_PARSER:
1887         case BPF_SK_SKB_STREAM_VERDICT:
1888                 ptype = BPF_PROG_TYPE_SK_SKB;
1889                 break;
1890         case BPF_LIRC_MODE2:
1891                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1892                 break;
1893         case BPF_FLOW_DISSECTOR:
1894                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1895                 break;
1896         case BPF_CGROUP_SYSCTL:
1897                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1898                 break;
1899         default:
1900                 return -EINVAL;
1901         }
1902
1903         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1904         if (IS_ERR(prog))
1905                 return PTR_ERR(prog);
1906
1907         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1908                 bpf_prog_put(prog);
1909                 return -EINVAL;
1910         }
1911
1912         switch (ptype) {
1913         case BPF_PROG_TYPE_SK_SKB:
1914         case BPF_PROG_TYPE_SK_MSG:
1915                 ret = sock_map_get_from_fd(attr, prog);
1916                 break;
1917         case BPF_PROG_TYPE_LIRC_MODE2:
1918                 ret = lirc_prog_attach(attr, prog);
1919                 break;
1920         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1921                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1922                 break;
1923         default:
1924                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1925         }
1926
1927         if (ret)
1928                 bpf_prog_put(prog);
1929         return ret;
1930 }
1931
1932 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1933
1934 static int bpf_prog_detach(const union bpf_attr *attr)
1935 {
1936         enum bpf_prog_type ptype;
1937
1938         if (!capable(CAP_NET_ADMIN))
1939                 return -EPERM;
1940
1941         if (CHECK_ATTR(BPF_PROG_DETACH))
1942                 return -EINVAL;
1943
1944         switch (attr->attach_type) {
1945         case BPF_CGROUP_INET_INGRESS:
1946         case BPF_CGROUP_INET_EGRESS:
1947                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1948                 break;
1949         case BPF_CGROUP_INET_SOCK_CREATE:
1950         case BPF_CGROUP_INET4_POST_BIND:
1951         case BPF_CGROUP_INET6_POST_BIND:
1952                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1953                 break;
1954         case BPF_CGROUP_INET4_BIND:
1955         case BPF_CGROUP_INET6_BIND:
1956         case BPF_CGROUP_INET4_CONNECT:
1957         case BPF_CGROUP_INET6_CONNECT:
1958         case BPF_CGROUP_UDP4_SENDMSG:
1959         case BPF_CGROUP_UDP6_SENDMSG:
1960                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1961                 break;
1962         case BPF_CGROUP_SOCK_OPS:
1963                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1964                 break;
1965         case BPF_CGROUP_DEVICE:
1966                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1967                 break;
1968         case BPF_SK_MSG_VERDICT:
1969                 return sock_map_get_from_fd(attr, NULL);
1970         case BPF_SK_SKB_STREAM_PARSER:
1971         case BPF_SK_SKB_STREAM_VERDICT:
1972                 return sock_map_get_from_fd(attr, NULL);
1973         case BPF_LIRC_MODE2:
1974                 return lirc_prog_detach(attr);
1975         case BPF_FLOW_DISSECTOR:
1976                 return skb_flow_dissector_bpf_prog_detach(attr);
1977         case BPF_CGROUP_SYSCTL:
1978                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1979                 break;
1980         default:
1981                 return -EINVAL;
1982         }
1983
1984         return cgroup_bpf_prog_detach(attr, ptype);
1985 }
1986
1987 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1988
1989 static int bpf_prog_query(const union bpf_attr *attr,
1990                           union bpf_attr __user *uattr)
1991 {
1992         if (!capable(CAP_NET_ADMIN))
1993                 return -EPERM;
1994         if (CHECK_ATTR(BPF_PROG_QUERY))
1995                 return -EINVAL;
1996         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1997                 return -EINVAL;
1998
1999         switch (attr->query.attach_type) {
2000         case BPF_CGROUP_INET_INGRESS:
2001         case BPF_CGROUP_INET_EGRESS:
2002         case BPF_CGROUP_INET_SOCK_CREATE:
2003         case BPF_CGROUP_INET4_BIND:
2004         case BPF_CGROUP_INET6_BIND:
2005         case BPF_CGROUP_INET4_POST_BIND:
2006         case BPF_CGROUP_INET6_POST_BIND:
2007         case BPF_CGROUP_INET4_CONNECT:
2008         case BPF_CGROUP_INET6_CONNECT:
2009         case BPF_CGROUP_UDP4_SENDMSG:
2010         case BPF_CGROUP_UDP6_SENDMSG:
2011         case BPF_CGROUP_SOCK_OPS:
2012         case BPF_CGROUP_DEVICE:
2013         case BPF_CGROUP_SYSCTL:
2014                 break;
2015         case BPF_LIRC_MODE2:
2016                 return lirc_prog_query(attr, uattr);
2017         case BPF_FLOW_DISSECTOR:
2018                 return skb_flow_dissector_prog_query(attr, uattr);
2019         default:
2020                 return -EINVAL;
2021         }
2022
2023         return cgroup_bpf_prog_query(attr, uattr);
2024 }
2025
2026 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2027
2028 static int bpf_prog_test_run(const union bpf_attr *attr,
2029                              union bpf_attr __user *uattr)
2030 {
2031         struct bpf_prog *prog;
2032         int ret = -ENOTSUPP;
2033
2034         if (!capable(CAP_SYS_ADMIN))
2035                 return -EPERM;
2036         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2037                 return -EINVAL;
2038
2039         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2040             (!attr->test.ctx_size_in && attr->test.ctx_in))
2041                 return -EINVAL;
2042
2043         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2044             (!attr->test.ctx_size_out && attr->test.ctx_out))
2045                 return -EINVAL;
2046
2047         prog = bpf_prog_get(attr->test.prog_fd);
2048         if (IS_ERR(prog))
2049                 return PTR_ERR(prog);
2050
2051         if (prog->aux->ops->test_run)
2052                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2053
2054         bpf_prog_put(prog);
2055         return ret;
2056 }
2057
2058 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2059
2060 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2061                                union bpf_attr __user *uattr,
2062                                struct idr *idr,
2063                                spinlock_t *lock)
2064 {
2065         u32 next_id = attr->start_id;
2066         int err = 0;
2067
2068         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2069                 return -EINVAL;
2070
2071         if (!capable(CAP_SYS_ADMIN))
2072                 return -EPERM;
2073
2074         next_id++;
2075         spin_lock_bh(lock);
2076         if (!idr_get_next(idr, &next_id))
2077                 err = -ENOENT;
2078         spin_unlock_bh(lock);
2079
2080         if (!err)
2081                 err = put_user(next_id, &uattr->next_id);
2082
2083         return err;
2084 }
2085
2086 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2087
2088 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2089 {
2090         struct bpf_prog *prog;
2091         u32 id = attr->prog_id;
2092         int fd;
2093
2094         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2095                 return -EINVAL;
2096
2097         if (!capable(CAP_SYS_ADMIN))
2098                 return -EPERM;
2099
2100         spin_lock_bh(&prog_idr_lock);
2101         prog = idr_find(&prog_idr, id);
2102         if (prog)
2103                 prog = bpf_prog_inc_not_zero(prog);
2104         else
2105                 prog = ERR_PTR(-ENOENT);
2106         spin_unlock_bh(&prog_idr_lock);
2107
2108         if (IS_ERR(prog))
2109                 return PTR_ERR(prog);
2110
2111         fd = bpf_prog_new_fd(prog);
2112         if (fd < 0)
2113                 bpf_prog_put(prog);
2114
2115         return fd;
2116 }
2117
2118 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2119
2120 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2121 {
2122         struct bpf_map *map;
2123         u32 id = attr->map_id;
2124         int f_flags;
2125         int fd;
2126
2127         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2128             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2129                 return -EINVAL;
2130
2131         if (!capable(CAP_SYS_ADMIN))
2132                 return -EPERM;
2133
2134         f_flags = bpf_get_file_flag(attr->open_flags);
2135         if (f_flags < 0)
2136                 return f_flags;
2137
2138         spin_lock_bh(&map_idr_lock);
2139         map = idr_find(&map_idr, id);
2140         if (map)
2141                 map = bpf_map_inc_not_zero(map, true);
2142         else
2143                 map = ERR_PTR(-ENOENT);
2144         spin_unlock_bh(&map_idr_lock);
2145
2146         if (IS_ERR(map))
2147                 return PTR_ERR(map);
2148
2149         fd = bpf_map_new_fd(map, f_flags);
2150         if (fd < 0)
2151                 bpf_map_put_with_uref(map);
2152
2153         return fd;
2154 }
2155
2156 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2157                                               unsigned long addr, u32 *off,
2158                                               u32 *type)
2159 {
2160         const struct bpf_map *map;
2161         int i;
2162
2163         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2164                 map = prog->aux->used_maps[i];
2165                 if (map == (void *)addr) {
2166                         *type = BPF_PSEUDO_MAP_FD;
2167                         return map;
2168                 }
2169                 if (!map->ops->map_direct_value_meta)
2170                         continue;
2171                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2172                         *type = BPF_PSEUDO_MAP_VALUE;
2173                         return map;
2174                 }
2175         }
2176
2177         return NULL;
2178 }
2179
2180 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2181 {
2182         const struct bpf_map *map;
2183         struct bpf_insn *insns;
2184         u32 off, type;
2185         u64 imm;
2186         int i;
2187
2188         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2189                         GFP_USER);
2190         if (!insns)
2191                 return insns;
2192
2193         for (i = 0; i < prog->len; i++) {
2194                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2195                         insns[i].code = BPF_JMP | BPF_CALL;
2196                         insns[i].imm = BPF_FUNC_tail_call;
2197                         /* fall-through */
2198                 }
2199                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2200                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2201                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2202                                 insns[i].code = BPF_JMP | BPF_CALL;
2203                         if (!bpf_dump_raw_ok())
2204                                 insns[i].imm = 0;
2205                         continue;
2206                 }
2207
2208                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2209                         continue;
2210
2211                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2212                 map = bpf_map_from_imm(prog, imm, &off, &type);
2213                 if (map) {
2214                         insns[i].src_reg = type;
2215                         insns[i].imm = map->id;
2216                         insns[i + 1].imm = off;
2217                         continue;
2218                 }
2219         }
2220
2221         return insns;
2222 }
2223
2224 static int set_info_rec_size(struct bpf_prog_info *info)
2225 {
2226         /*
2227          * Ensure info.*_rec_size is the same as kernel expected size
2228          *
2229          * or
2230          *
2231          * Only allow zero *_rec_size if both _rec_size and _cnt are
2232          * zero.  In this case, the kernel will set the expected
2233          * _rec_size back to the info.
2234          */
2235
2236         if ((info->nr_func_info || info->func_info_rec_size) &&
2237             info->func_info_rec_size != sizeof(struct bpf_func_info))
2238                 return -EINVAL;
2239
2240         if ((info->nr_line_info || info->line_info_rec_size) &&
2241             info->line_info_rec_size != sizeof(struct bpf_line_info))
2242                 return -EINVAL;
2243
2244         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2245             info->jited_line_info_rec_size != sizeof(__u64))
2246                 return -EINVAL;
2247
2248         info->func_info_rec_size = sizeof(struct bpf_func_info);
2249         info->line_info_rec_size = sizeof(struct bpf_line_info);
2250         info->jited_line_info_rec_size = sizeof(__u64);
2251
2252         return 0;
2253 }
2254
2255 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2256                                    const union bpf_attr *attr,
2257                                    union bpf_attr __user *uattr)
2258 {
2259         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2260         struct bpf_prog_info info = {};
2261         u32 info_len = attr->info.info_len;
2262         struct bpf_prog_stats stats;
2263         char __user *uinsns;
2264         u32 ulen;
2265         int err;
2266
2267         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2268         if (err)
2269                 return err;
2270         info_len = min_t(u32, sizeof(info), info_len);
2271
2272         if (copy_from_user(&info, uinfo, info_len))
2273                 return -EFAULT;
2274
2275         info.type = prog->type;
2276         info.id = prog->aux->id;
2277         info.load_time = prog->aux->load_time;
2278         info.created_by_uid = from_kuid_munged(current_user_ns(),
2279                                                prog->aux->user->uid);
2280         info.gpl_compatible = prog->gpl_compatible;
2281
2282         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2283         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2284
2285         ulen = info.nr_map_ids;
2286         info.nr_map_ids = prog->aux->used_map_cnt;
2287         ulen = min_t(u32, info.nr_map_ids, ulen);
2288         if (ulen) {
2289                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2290                 u32 i;
2291
2292                 for (i = 0; i < ulen; i++)
2293                         if (put_user(prog->aux->used_maps[i]->id,
2294                                      &user_map_ids[i]))
2295                                 return -EFAULT;
2296         }
2297
2298         err = set_info_rec_size(&info);
2299         if (err)
2300                 return err;
2301
2302         bpf_prog_get_stats(prog, &stats);
2303         info.run_time_ns = stats.nsecs;
2304         info.run_cnt = stats.cnt;
2305
2306         if (!capable(CAP_SYS_ADMIN)) {
2307                 info.jited_prog_len = 0;
2308                 info.xlated_prog_len = 0;
2309                 info.nr_jited_ksyms = 0;
2310                 info.nr_jited_func_lens = 0;
2311                 info.nr_func_info = 0;
2312                 info.nr_line_info = 0;
2313                 info.nr_jited_line_info = 0;
2314                 goto done;
2315         }
2316
2317         ulen = info.xlated_prog_len;
2318         info.xlated_prog_len = bpf_prog_insn_size(prog);
2319         if (info.xlated_prog_len && ulen) {
2320                 struct bpf_insn *insns_sanitized;
2321                 bool fault;
2322
2323                 if (prog->blinded && !bpf_dump_raw_ok()) {
2324                         info.xlated_prog_insns = 0;
2325                         goto done;
2326                 }
2327                 insns_sanitized = bpf_insn_prepare_dump(prog);
2328                 if (!insns_sanitized)
2329                         return -ENOMEM;
2330                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2331                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2332                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2333                 kfree(insns_sanitized);
2334                 if (fault)
2335                         return -EFAULT;
2336         }
2337
2338         if (bpf_prog_is_dev_bound(prog->aux)) {
2339                 err = bpf_prog_offload_info_fill(&info, prog);
2340                 if (err)
2341                         return err;
2342                 goto done;
2343         }
2344
2345         /* NOTE: the following code is supposed to be skipped for offload.
2346          * bpf_prog_offload_info_fill() is the place to fill similar fields
2347          * for offload.
2348          */
2349         ulen = info.jited_prog_len;
2350         if (prog->aux->func_cnt) {
2351                 u32 i;
2352
2353                 info.jited_prog_len = 0;
2354                 for (i = 0; i < prog->aux->func_cnt; i++)
2355                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2356         } else {
2357                 info.jited_prog_len = prog->jited_len;
2358         }
2359
2360         if (info.jited_prog_len && ulen) {
2361                 if (bpf_dump_raw_ok()) {
2362                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2363                         ulen = min_t(u32, info.jited_prog_len, ulen);
2364
2365                         /* for multi-function programs, copy the JITed
2366                          * instructions for all the functions
2367                          */
2368                         if (prog->aux->func_cnt) {
2369                                 u32 len, free, i;
2370                                 u8 *img;
2371
2372                                 free = ulen;
2373                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2374                                         len = prog->aux->func[i]->jited_len;
2375                                         len = min_t(u32, len, free);
2376                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2377                                         if (copy_to_user(uinsns, img, len))
2378                                                 return -EFAULT;
2379                                         uinsns += len;
2380                                         free -= len;
2381                                         if (!free)
2382                                                 break;
2383                                 }
2384                         } else {
2385                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2386                                         return -EFAULT;
2387                         }
2388                 } else {
2389                         info.jited_prog_insns = 0;
2390                 }
2391         }
2392
2393         ulen = info.nr_jited_ksyms;
2394         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2395         if (ulen) {
2396                 if (bpf_dump_raw_ok()) {
2397                         unsigned long ksym_addr;
2398                         u64 __user *user_ksyms;
2399                         u32 i;
2400
2401                         /* copy the address of the kernel symbol
2402                          * corresponding to each function
2403                          */
2404                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2405                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2406                         if (prog->aux->func_cnt) {
2407                                 for (i = 0; i < ulen; i++) {
2408                                         ksym_addr = (unsigned long)
2409                                                 prog->aux->func[i]->bpf_func;
2410                                         if (put_user((u64) ksym_addr,
2411                                                      &user_ksyms[i]))
2412                                                 return -EFAULT;
2413                                 }
2414                         } else {
2415                                 ksym_addr = (unsigned long) prog->bpf_func;
2416                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2417                                         return -EFAULT;
2418                         }
2419                 } else {
2420                         info.jited_ksyms = 0;
2421                 }
2422         }
2423
2424         ulen = info.nr_jited_func_lens;
2425         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2426         if (ulen) {
2427                 if (bpf_dump_raw_ok()) {
2428                         u32 __user *user_lens;
2429                         u32 func_len, i;
2430
2431                         /* copy the JITed image lengths for each function */
2432                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2433                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2434                         if (prog->aux->func_cnt) {
2435                                 for (i = 0; i < ulen; i++) {
2436                                         func_len =
2437                                                 prog->aux->func[i]->jited_len;
2438                                         if (put_user(func_len, &user_lens[i]))
2439                                                 return -EFAULT;
2440                                 }
2441                         } else {
2442                                 func_len = prog->jited_len;
2443                                 if (put_user(func_len, &user_lens[0]))
2444                                         return -EFAULT;
2445                         }
2446                 } else {
2447                         info.jited_func_lens = 0;
2448                 }
2449         }
2450
2451         if (prog->aux->btf)
2452                 info.btf_id = btf_id(prog->aux->btf);
2453
2454         ulen = info.nr_func_info;
2455         info.nr_func_info = prog->aux->func_info_cnt;
2456         if (info.nr_func_info && ulen) {
2457                 char __user *user_finfo;
2458
2459                 user_finfo = u64_to_user_ptr(info.func_info);
2460                 ulen = min_t(u32, info.nr_func_info, ulen);
2461                 if (copy_to_user(user_finfo, prog->aux->func_info,
2462                                  info.func_info_rec_size * ulen))
2463                         return -EFAULT;
2464         }
2465
2466         ulen = info.nr_line_info;
2467         info.nr_line_info = prog->aux->nr_linfo;
2468         if (info.nr_line_info && ulen) {
2469                 __u8 __user *user_linfo;
2470
2471                 user_linfo = u64_to_user_ptr(info.line_info);
2472                 ulen = min_t(u32, info.nr_line_info, ulen);
2473                 if (copy_to_user(user_linfo, prog->aux->linfo,
2474                                  info.line_info_rec_size * ulen))
2475                         return -EFAULT;
2476         }
2477
2478         ulen = info.nr_jited_line_info;
2479         if (prog->aux->jited_linfo)
2480                 info.nr_jited_line_info = prog->aux->nr_linfo;
2481         else
2482                 info.nr_jited_line_info = 0;
2483         if (info.nr_jited_line_info && ulen) {
2484                 if (bpf_dump_raw_ok()) {
2485                         __u64 __user *user_linfo;
2486                         u32 i;
2487
2488                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2489                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2490                         for (i = 0; i < ulen; i++) {
2491                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2492                                              &user_linfo[i]))
2493                                         return -EFAULT;
2494                         }
2495                 } else {
2496                         info.jited_line_info = 0;
2497                 }
2498         }
2499
2500         ulen = info.nr_prog_tags;
2501         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2502         if (ulen) {
2503                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2504                 u32 i;
2505
2506                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2507                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2508                 if (prog->aux->func_cnt) {
2509                         for (i = 0; i < ulen; i++) {
2510                                 if (copy_to_user(user_prog_tags[i],
2511                                                  prog->aux->func[i]->tag,
2512                                                  BPF_TAG_SIZE))
2513                                         return -EFAULT;
2514                         }
2515                 } else {
2516                         if (copy_to_user(user_prog_tags[0],
2517                                          prog->tag, BPF_TAG_SIZE))
2518                                 return -EFAULT;
2519                 }
2520         }
2521
2522 done:
2523         if (copy_to_user(uinfo, &info, info_len) ||
2524             put_user(info_len, &uattr->info.info_len))
2525                 return -EFAULT;
2526
2527         return 0;
2528 }
2529
2530 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2531                                   const union bpf_attr *attr,
2532                                   union bpf_attr __user *uattr)
2533 {
2534         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2535         struct bpf_map_info info = {};
2536         u32 info_len = attr->info.info_len;
2537         int err;
2538
2539         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2540         if (err)
2541                 return err;
2542         info_len = min_t(u32, sizeof(info), info_len);
2543
2544         info.type = map->map_type;
2545         info.id = map->id;
2546         info.key_size = map->key_size;
2547         info.value_size = map->value_size;
2548         info.max_entries = map->max_entries;
2549         info.map_flags = map->map_flags;
2550         memcpy(info.name, map->name, sizeof(map->name));
2551
2552         if (map->btf) {
2553                 info.btf_id = btf_id(map->btf);
2554                 info.btf_key_type_id = map->btf_key_type_id;
2555                 info.btf_value_type_id = map->btf_value_type_id;
2556         }
2557
2558         if (bpf_map_is_dev_bound(map)) {
2559                 err = bpf_map_offload_info_fill(&info, map);
2560                 if (err)
2561                         return err;
2562         }
2563
2564         if (copy_to_user(uinfo, &info, info_len) ||
2565             put_user(info_len, &uattr->info.info_len))
2566                 return -EFAULT;
2567
2568         return 0;
2569 }
2570
2571 static int bpf_btf_get_info_by_fd(struct btf *btf,
2572                                   const union bpf_attr *attr,
2573                                   union bpf_attr __user *uattr)
2574 {
2575         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2576         u32 info_len = attr->info.info_len;
2577         int err;
2578
2579         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2580         if (err)
2581                 return err;
2582
2583         return btf_get_info_by_fd(btf, attr, uattr);
2584 }
2585
2586 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2587
2588 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2589                                   union bpf_attr __user *uattr)
2590 {
2591         int ufd = attr->info.bpf_fd;
2592         struct fd f;
2593         int err;
2594
2595         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2596                 return -EINVAL;
2597
2598         f = fdget(ufd);
2599         if (!f.file)
2600                 return -EBADFD;
2601
2602         if (f.file->f_op == &bpf_prog_fops)
2603                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2604                                               uattr);
2605         else if (f.file->f_op == &bpf_map_fops)
2606                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2607                                              uattr);
2608         else if (f.file->f_op == &btf_fops)
2609                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2610         else
2611                 err = -EINVAL;
2612
2613         fdput(f);
2614         return err;
2615 }
2616
2617 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2618
2619 static int bpf_btf_load(const union bpf_attr *attr)
2620 {
2621         if (CHECK_ATTR(BPF_BTF_LOAD))
2622                 return -EINVAL;
2623
2624         if (!capable(CAP_SYS_ADMIN))
2625                 return -EPERM;
2626
2627         return btf_new_fd(attr);
2628 }
2629
2630 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2631
2632 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2633 {
2634         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2635                 return -EINVAL;
2636
2637         if (!capable(CAP_SYS_ADMIN))
2638                 return -EPERM;
2639
2640         return btf_get_fd_by_id(attr->btf_id);
2641 }
2642
2643 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2644                                     union bpf_attr __user *uattr,
2645                                     u32 prog_id, u32 fd_type,
2646                                     const char *buf, u64 probe_offset,
2647                                     u64 probe_addr)
2648 {
2649         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2650         u32 len = buf ? strlen(buf) : 0, input_len;
2651         int err = 0;
2652
2653         if (put_user(len, &uattr->task_fd_query.buf_len))
2654                 return -EFAULT;
2655         input_len = attr->task_fd_query.buf_len;
2656         if (input_len && ubuf) {
2657                 if (!len) {
2658                         /* nothing to copy, just make ubuf NULL terminated */
2659                         char zero = '\0';
2660
2661                         if (put_user(zero, ubuf))
2662                                 return -EFAULT;
2663                 } else if (input_len >= len + 1) {
2664                         /* ubuf can hold the string with NULL terminator */
2665                         if (copy_to_user(ubuf, buf, len + 1))
2666                                 return -EFAULT;
2667                 } else {
2668                         /* ubuf cannot hold the string with NULL terminator,
2669                          * do a partial copy with NULL terminator.
2670                          */
2671                         char zero = '\0';
2672
2673                         err = -ENOSPC;
2674                         if (copy_to_user(ubuf, buf, input_len - 1))
2675                                 return -EFAULT;
2676                         if (put_user(zero, ubuf + input_len - 1))
2677                                 return -EFAULT;
2678                 }
2679         }
2680
2681         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2682             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2683             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2684             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2685                 return -EFAULT;
2686
2687         return err;
2688 }
2689
2690 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2691
2692 static int bpf_task_fd_query(const union bpf_attr *attr,
2693                              union bpf_attr __user *uattr)
2694 {
2695         pid_t pid = attr->task_fd_query.pid;
2696         u32 fd = attr->task_fd_query.fd;
2697         const struct perf_event *event;
2698         struct files_struct *files;
2699         struct task_struct *task;
2700         struct file *file;
2701         int err;
2702
2703         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2704                 return -EINVAL;
2705
2706         if (!capable(CAP_SYS_ADMIN))
2707                 return -EPERM;
2708
2709         if (attr->task_fd_query.flags != 0)
2710                 return -EINVAL;
2711
2712         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2713         if (!task)
2714                 return -ENOENT;
2715
2716         files = get_files_struct(task);
2717         put_task_struct(task);
2718         if (!files)
2719                 return -ENOENT;
2720
2721         err = 0;
2722         spin_lock(&files->file_lock);
2723         file = fcheck_files(files, fd);
2724         if (!file)
2725                 err = -EBADF;
2726         else
2727                 get_file(file);
2728         spin_unlock(&files->file_lock);
2729         put_files_struct(files);
2730
2731         if (err)
2732                 goto out;
2733
2734         if (file->f_op == &bpf_raw_tp_fops) {
2735                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2736                 struct bpf_raw_event_map *btp = raw_tp->btp;
2737
2738                 err = bpf_task_fd_query_copy(attr, uattr,
2739                                              raw_tp->prog->aux->id,
2740                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2741                                              btp->tp->name, 0, 0);
2742                 goto put_file;
2743         }
2744
2745         event = perf_get_event(file);
2746         if (!IS_ERR(event)) {
2747                 u64 probe_offset, probe_addr;
2748                 u32 prog_id, fd_type;
2749                 const char *buf;
2750
2751                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2752                                               &buf, &probe_offset,
2753                                               &probe_addr);
2754                 if (!err)
2755                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2756                                                      fd_type, buf,
2757                                                      probe_offset,
2758                                                      probe_addr);
2759                 goto put_file;
2760         }
2761
2762         err = -ENOTSUPP;
2763 put_file:
2764         fput(file);
2765 out:
2766         return err;
2767 }
2768
2769 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2770 {
2771         union bpf_attr attr = {};
2772         int err;
2773
2774         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2775                 return -EPERM;
2776
2777         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2778         if (err)
2779                 return err;
2780         size = min_t(u32, size, sizeof(attr));
2781
2782         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2783         if (copy_from_user(&attr, uattr, size) != 0)
2784                 return -EFAULT;
2785
2786         err = security_bpf(cmd, &attr, size);
2787         if (err < 0)
2788                 return err;
2789
2790         switch (cmd) {
2791         case BPF_MAP_CREATE:
2792                 err = map_create(&attr);
2793                 break;
2794         case BPF_MAP_LOOKUP_ELEM:
2795                 err = map_lookup_elem(&attr);
2796                 break;
2797         case BPF_MAP_UPDATE_ELEM:
2798                 err = map_update_elem(&attr);
2799                 break;
2800         case BPF_MAP_DELETE_ELEM:
2801                 err = map_delete_elem(&attr);
2802                 break;
2803         case BPF_MAP_GET_NEXT_KEY:
2804                 err = map_get_next_key(&attr);
2805                 break;
2806         case BPF_MAP_FREEZE:
2807                 err = map_freeze(&attr);
2808                 break;
2809         case BPF_PROG_LOAD:
2810                 err = bpf_prog_load(&attr, uattr);
2811                 break;
2812         case BPF_OBJ_PIN:
2813                 err = bpf_obj_pin(&attr);
2814                 break;
2815         case BPF_OBJ_GET:
2816                 err = bpf_obj_get(&attr);
2817                 break;
2818         case BPF_PROG_ATTACH:
2819                 err = bpf_prog_attach(&attr);
2820                 break;
2821         case BPF_PROG_DETACH:
2822                 err = bpf_prog_detach(&attr);
2823                 break;
2824         case BPF_PROG_QUERY:
2825                 err = bpf_prog_query(&attr, uattr);
2826                 break;
2827         case BPF_PROG_TEST_RUN:
2828                 err = bpf_prog_test_run(&attr, uattr);
2829                 break;
2830         case BPF_PROG_GET_NEXT_ID:
2831                 err = bpf_obj_get_next_id(&attr, uattr,
2832                                           &prog_idr, &prog_idr_lock);
2833                 break;
2834         case BPF_MAP_GET_NEXT_ID:
2835                 err = bpf_obj_get_next_id(&attr, uattr,
2836                                           &map_idr, &map_idr_lock);
2837                 break;
2838         case BPF_PROG_GET_FD_BY_ID:
2839                 err = bpf_prog_get_fd_by_id(&attr);
2840                 break;
2841         case BPF_MAP_GET_FD_BY_ID:
2842                 err = bpf_map_get_fd_by_id(&attr);
2843                 break;
2844         case BPF_OBJ_GET_INFO_BY_FD:
2845                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2846                 break;
2847         case BPF_RAW_TRACEPOINT_OPEN:
2848                 err = bpf_raw_tracepoint_open(&attr);
2849                 break;
2850         case BPF_BTF_LOAD:
2851                 err = bpf_btf_load(&attr);
2852                 break;
2853         case BPF_BTF_GET_FD_BY_ID:
2854                 err = bpf_btf_get_fd_by_id(&attr);
2855                 break;
2856         case BPF_TASK_FD_QUERY:
2857                 err = bpf_task_fd_query(&attr, uattr);
2858                 break;
2859         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2860                 err = map_lookup_and_delete_elem(&attr);
2861                 break;
2862         default:
2863                 err = -EINVAL;
2864                 break;
2865         }
2866
2867         return err;
2868 }