Merge branch 'lru-map-fix'
[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                 if (map->ops->map_lookup_elem_sys_only)
812                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
813                 else
814                         ptr = map->ops->map_lookup_elem(map, key);
815                 if (IS_ERR(ptr)) {
816                         err = PTR_ERR(ptr);
817                 } else if (!ptr) {
818                         err = -ENOENT;
819                 } else {
820                         err = 0;
821                         if (attr->flags & BPF_F_LOCK)
822                                 /* lock 'ptr' and copy everything but lock */
823                                 copy_map_value_locked(map, value, ptr, true);
824                         else
825                                 copy_map_value(map, value, ptr);
826                         /* mask lock, since value wasn't zero inited */
827                         check_and_init_map_lock(map, value);
828                 }
829                 rcu_read_unlock();
830         }
831         this_cpu_dec(bpf_prog_active);
832         preempt_enable();
833
834 done:
835         if (err)
836                 goto free_value;
837
838         err = -EFAULT;
839         if (copy_to_user(uvalue, value, value_size) != 0)
840                 goto free_value;
841
842         err = 0;
843
844 free_value:
845         kfree(value);
846 free_key:
847         kfree(key);
848 err_put:
849         fdput(f);
850         return err;
851 }
852
853 static void maybe_wait_bpf_programs(struct bpf_map *map)
854 {
855         /* Wait for any running BPF programs to complete so that
856          * userspace, when we return to it, knows that all programs
857          * that could be running use the new map value.
858          */
859         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
860             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
861                 synchronize_rcu();
862 }
863
864 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
865
866 static int map_update_elem(union bpf_attr *attr)
867 {
868         void __user *ukey = u64_to_user_ptr(attr->key);
869         void __user *uvalue = u64_to_user_ptr(attr->value);
870         int ufd = attr->map_fd;
871         struct bpf_map *map;
872         void *key, *value;
873         u32 value_size;
874         struct fd f;
875         int err;
876
877         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
878                 return -EINVAL;
879
880         f = fdget(ufd);
881         map = __bpf_map_get(f);
882         if (IS_ERR(map))
883                 return PTR_ERR(map);
884         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
885                 err = -EPERM;
886                 goto err_put;
887         }
888
889         if ((attr->flags & BPF_F_LOCK) &&
890             !map_value_has_spin_lock(map)) {
891                 err = -EINVAL;
892                 goto err_put;
893         }
894
895         key = __bpf_copy_key(ukey, map->key_size);
896         if (IS_ERR(key)) {
897                 err = PTR_ERR(key);
898                 goto err_put;
899         }
900
901         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
902             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
903             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
904             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
905                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
906         else
907                 value_size = map->value_size;
908
909         err = -ENOMEM;
910         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
911         if (!value)
912                 goto free_key;
913
914         err = -EFAULT;
915         if (copy_from_user(value, uvalue, value_size) != 0)
916                 goto free_value;
917
918         /* Need to create a kthread, thus must support schedule */
919         if (bpf_map_is_dev_bound(map)) {
920                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
921                 goto out;
922         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
923                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
924                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
925                 err = map->ops->map_update_elem(map, key, value, attr->flags);
926                 goto out;
927         }
928
929         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
930          * inside bpf map update or delete otherwise deadlocks are possible
931          */
932         preempt_disable();
933         __this_cpu_inc(bpf_prog_active);
934         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
935             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
936                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
937         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
938                 err = bpf_percpu_array_update(map, key, value, attr->flags);
939         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
940                 err = bpf_percpu_cgroup_storage_update(map, key, value,
941                                                        attr->flags);
942         } else if (IS_FD_ARRAY(map)) {
943                 rcu_read_lock();
944                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
945                                                    attr->flags);
946                 rcu_read_unlock();
947         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
948                 rcu_read_lock();
949                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
950                                                   attr->flags);
951                 rcu_read_unlock();
952         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
953                 /* rcu_read_lock() is not needed */
954                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
955                                                          attr->flags);
956         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
957                    map->map_type == BPF_MAP_TYPE_STACK) {
958                 err = map->ops->map_push_elem(map, value, attr->flags);
959         } else {
960                 rcu_read_lock();
961                 err = map->ops->map_update_elem(map, key, value, attr->flags);
962                 rcu_read_unlock();
963         }
964         __this_cpu_dec(bpf_prog_active);
965         preempt_enable();
966         maybe_wait_bpf_programs(map);
967 out:
968 free_value:
969         kfree(value);
970 free_key:
971         kfree(key);
972 err_put:
973         fdput(f);
974         return err;
975 }
976
977 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
978
979 static int map_delete_elem(union bpf_attr *attr)
980 {
981         void __user *ukey = u64_to_user_ptr(attr->key);
982         int ufd = attr->map_fd;
983         struct bpf_map *map;
984         struct fd f;
985         void *key;
986         int err;
987
988         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
989                 return -EINVAL;
990
991         f = fdget(ufd);
992         map = __bpf_map_get(f);
993         if (IS_ERR(map))
994                 return PTR_ERR(map);
995         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
996                 err = -EPERM;
997                 goto err_put;
998         }
999
1000         key = __bpf_copy_key(ukey, map->key_size);
1001         if (IS_ERR(key)) {
1002                 err = PTR_ERR(key);
1003                 goto err_put;
1004         }
1005
1006         if (bpf_map_is_dev_bound(map)) {
1007                 err = bpf_map_offload_delete_elem(map, key);
1008                 goto out;
1009         }
1010
1011         preempt_disable();
1012         __this_cpu_inc(bpf_prog_active);
1013         rcu_read_lock();
1014         err = map->ops->map_delete_elem(map, key);
1015         rcu_read_unlock();
1016         __this_cpu_dec(bpf_prog_active);
1017         preempt_enable();
1018         maybe_wait_bpf_programs(map);
1019 out:
1020         kfree(key);
1021 err_put:
1022         fdput(f);
1023         return err;
1024 }
1025
1026 /* last field in 'union bpf_attr' used by this command */
1027 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1028
1029 static int map_get_next_key(union bpf_attr *attr)
1030 {
1031         void __user *ukey = u64_to_user_ptr(attr->key);
1032         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1033         int ufd = attr->map_fd;
1034         struct bpf_map *map;
1035         void *key, *next_key;
1036         struct fd f;
1037         int err;
1038
1039         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1040                 return -EINVAL;
1041
1042         f = fdget(ufd);
1043         map = __bpf_map_get(f);
1044         if (IS_ERR(map))
1045                 return PTR_ERR(map);
1046         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1047                 err = -EPERM;
1048                 goto err_put;
1049         }
1050
1051         if (ukey) {
1052                 key = __bpf_copy_key(ukey, map->key_size);
1053                 if (IS_ERR(key)) {
1054                         err = PTR_ERR(key);
1055                         goto err_put;
1056                 }
1057         } else {
1058                 key = NULL;
1059         }
1060
1061         err = -ENOMEM;
1062         next_key = kmalloc(map->key_size, GFP_USER);
1063         if (!next_key)
1064                 goto free_key;
1065
1066         if (bpf_map_is_dev_bound(map)) {
1067                 err = bpf_map_offload_get_next_key(map, key, next_key);
1068                 goto out;
1069         }
1070
1071         rcu_read_lock();
1072         err = map->ops->map_get_next_key(map, key, next_key);
1073         rcu_read_unlock();
1074 out:
1075         if (err)
1076                 goto free_next_key;
1077
1078         err = -EFAULT;
1079         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1080                 goto free_next_key;
1081
1082         err = 0;
1083
1084 free_next_key:
1085         kfree(next_key);
1086 free_key:
1087         kfree(key);
1088 err_put:
1089         fdput(f);
1090         return err;
1091 }
1092
1093 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1094
1095 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1096 {
1097         void __user *ukey = u64_to_user_ptr(attr->key);
1098         void __user *uvalue = u64_to_user_ptr(attr->value);
1099         int ufd = attr->map_fd;
1100         struct bpf_map *map;
1101         void *key, *value;
1102         u32 value_size;
1103         struct fd f;
1104         int err;
1105
1106         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1107                 return -EINVAL;
1108
1109         f = fdget(ufd);
1110         map = __bpf_map_get(f);
1111         if (IS_ERR(map))
1112                 return PTR_ERR(map);
1113         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1114                 err = -EPERM;
1115                 goto err_put;
1116         }
1117
1118         key = __bpf_copy_key(ukey, map->key_size);
1119         if (IS_ERR(key)) {
1120                 err = PTR_ERR(key);
1121                 goto err_put;
1122         }
1123
1124         value_size = map->value_size;
1125
1126         err = -ENOMEM;
1127         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1128         if (!value)
1129                 goto free_key;
1130
1131         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1132             map->map_type == BPF_MAP_TYPE_STACK) {
1133                 err = map->ops->map_pop_elem(map, value);
1134         } else {
1135                 err = -ENOTSUPP;
1136         }
1137
1138         if (err)
1139                 goto free_value;
1140
1141         if (copy_to_user(uvalue, value, value_size) != 0)
1142                 goto free_value;
1143
1144         err = 0;
1145
1146 free_value:
1147         kfree(value);
1148 free_key:
1149         kfree(key);
1150 err_put:
1151         fdput(f);
1152         return err;
1153 }
1154
1155 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1156
1157 static int map_freeze(const union bpf_attr *attr)
1158 {
1159         int err = 0, ufd = attr->map_fd;
1160         struct bpf_map *map;
1161         struct fd f;
1162
1163         if (CHECK_ATTR(BPF_MAP_FREEZE))
1164                 return -EINVAL;
1165
1166         f = fdget(ufd);
1167         map = __bpf_map_get(f);
1168         if (IS_ERR(map))
1169                 return PTR_ERR(map);
1170         if (READ_ONCE(map->frozen)) {
1171                 err = -EBUSY;
1172                 goto err_put;
1173         }
1174         if (!capable(CAP_SYS_ADMIN)) {
1175                 err = -EPERM;
1176                 goto err_put;
1177         }
1178
1179         WRITE_ONCE(map->frozen, true);
1180 err_put:
1181         fdput(f);
1182         return err;
1183 }
1184
1185 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1186 #define BPF_PROG_TYPE(_id, _name) \
1187         [_id] = & _name ## _prog_ops,
1188 #define BPF_MAP_TYPE(_id, _ops)
1189 #include <linux/bpf_types.h>
1190 #undef BPF_PROG_TYPE
1191 #undef BPF_MAP_TYPE
1192 };
1193
1194 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1195 {
1196         const struct bpf_prog_ops *ops;
1197
1198         if (type >= ARRAY_SIZE(bpf_prog_types))
1199                 return -EINVAL;
1200         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1201         ops = bpf_prog_types[type];
1202         if (!ops)
1203                 return -EINVAL;
1204
1205         if (!bpf_prog_is_dev_bound(prog->aux))
1206                 prog->aux->ops = ops;
1207         else
1208                 prog->aux->ops = &bpf_offload_prog_ops;
1209         prog->type = type;
1210         return 0;
1211 }
1212
1213 /* drop refcnt on maps used by eBPF program and free auxilary data */
1214 static void free_used_maps(struct bpf_prog_aux *aux)
1215 {
1216         enum bpf_cgroup_storage_type stype;
1217         int i;
1218
1219         for_each_cgroup_storage_type(stype) {
1220                 if (!aux->cgroup_storage[stype])
1221                         continue;
1222                 bpf_cgroup_storage_release(aux->prog,
1223                                            aux->cgroup_storage[stype]);
1224         }
1225
1226         for (i = 0; i < aux->used_map_cnt; i++)
1227                 bpf_map_put(aux->used_maps[i]);
1228
1229         kfree(aux->used_maps);
1230 }
1231
1232 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1233 {
1234         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1235         unsigned long user_bufs;
1236
1237         if (user) {
1238                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1239                 if (user_bufs > memlock_limit) {
1240                         atomic_long_sub(pages, &user->locked_vm);
1241                         return -EPERM;
1242                 }
1243         }
1244
1245         return 0;
1246 }
1247
1248 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1249 {
1250         if (user)
1251                 atomic_long_sub(pages, &user->locked_vm);
1252 }
1253
1254 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1255 {
1256         struct user_struct *user = get_current_user();
1257         int ret;
1258
1259         ret = __bpf_prog_charge(user, prog->pages);
1260         if (ret) {
1261                 free_uid(user);
1262                 return ret;
1263         }
1264
1265         prog->aux->user = user;
1266         return 0;
1267 }
1268
1269 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1270 {
1271         struct user_struct *user = prog->aux->user;
1272
1273         __bpf_prog_uncharge(user, prog->pages);
1274         free_uid(user);
1275 }
1276
1277 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1278 {
1279         int id;
1280
1281         idr_preload(GFP_KERNEL);
1282         spin_lock_bh(&prog_idr_lock);
1283         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1284         if (id > 0)
1285                 prog->aux->id = id;
1286         spin_unlock_bh(&prog_idr_lock);
1287         idr_preload_end();
1288
1289         /* id is in [1, INT_MAX) */
1290         if (WARN_ON_ONCE(!id))
1291                 return -ENOSPC;
1292
1293         return id > 0 ? 0 : id;
1294 }
1295
1296 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1297 {
1298         /* cBPF to eBPF migrations are currently not in the idr store.
1299          * Offloaded programs are removed from the store when their device
1300          * disappears - even if someone grabs an fd to them they are unusable,
1301          * simply waiting for refcnt to drop to be freed.
1302          */
1303         if (!prog->aux->id)
1304                 return;
1305
1306         if (do_idr_lock)
1307                 spin_lock_bh(&prog_idr_lock);
1308         else
1309                 __acquire(&prog_idr_lock);
1310
1311         idr_remove(&prog_idr, prog->aux->id);
1312         prog->aux->id = 0;
1313
1314         if (do_idr_lock)
1315                 spin_unlock_bh(&prog_idr_lock);
1316         else
1317                 __release(&prog_idr_lock);
1318 }
1319
1320 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1321 {
1322         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1323
1324         free_used_maps(aux);
1325         bpf_prog_uncharge_memlock(aux->prog);
1326         security_bpf_prog_free(aux);
1327         bpf_prog_free(aux->prog);
1328 }
1329
1330 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1331 {
1332         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1333                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1334                 /* bpf_prog_free_id() must be called first */
1335                 bpf_prog_free_id(prog, do_idr_lock);
1336                 bpf_prog_kallsyms_del_all(prog);
1337                 btf_put(prog->aux->btf);
1338                 kvfree(prog->aux->func_info);
1339                 bpf_prog_free_linfo(prog);
1340
1341                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1342         }
1343 }
1344
1345 void bpf_prog_put(struct bpf_prog *prog)
1346 {
1347         __bpf_prog_put(prog, true);
1348 }
1349 EXPORT_SYMBOL_GPL(bpf_prog_put);
1350
1351 static int bpf_prog_release(struct inode *inode, struct file *filp)
1352 {
1353         struct bpf_prog *prog = filp->private_data;
1354
1355         bpf_prog_put(prog);
1356         return 0;
1357 }
1358
1359 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1360                                struct bpf_prog_stats *stats)
1361 {
1362         u64 nsecs = 0, cnt = 0;
1363         int cpu;
1364
1365         for_each_possible_cpu(cpu) {
1366                 const struct bpf_prog_stats *st;
1367                 unsigned int start;
1368                 u64 tnsecs, tcnt;
1369
1370                 st = per_cpu_ptr(prog->aux->stats, cpu);
1371                 do {
1372                         start = u64_stats_fetch_begin_irq(&st->syncp);
1373                         tnsecs = st->nsecs;
1374                         tcnt = st->cnt;
1375                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1376                 nsecs += tnsecs;
1377                 cnt += tcnt;
1378         }
1379         stats->nsecs = nsecs;
1380         stats->cnt = cnt;
1381 }
1382
1383 #ifdef CONFIG_PROC_FS
1384 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1385 {
1386         const struct bpf_prog *prog = filp->private_data;
1387         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1388         struct bpf_prog_stats stats;
1389
1390         bpf_prog_get_stats(prog, &stats);
1391         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1392         seq_printf(m,
1393                    "prog_type:\t%u\n"
1394                    "prog_jited:\t%u\n"
1395                    "prog_tag:\t%s\n"
1396                    "memlock:\t%llu\n"
1397                    "prog_id:\t%u\n"
1398                    "run_time_ns:\t%llu\n"
1399                    "run_cnt:\t%llu\n",
1400                    prog->type,
1401                    prog->jited,
1402                    prog_tag,
1403                    prog->pages * 1ULL << PAGE_SHIFT,
1404                    prog->aux->id,
1405                    stats.nsecs,
1406                    stats.cnt);
1407 }
1408 #endif
1409
1410 const struct file_operations bpf_prog_fops = {
1411 #ifdef CONFIG_PROC_FS
1412         .show_fdinfo    = bpf_prog_show_fdinfo,
1413 #endif
1414         .release        = bpf_prog_release,
1415         .read           = bpf_dummy_read,
1416         .write          = bpf_dummy_write,
1417 };
1418
1419 int bpf_prog_new_fd(struct bpf_prog *prog)
1420 {
1421         int ret;
1422
1423         ret = security_bpf_prog(prog);
1424         if (ret < 0)
1425                 return ret;
1426
1427         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1428                                 O_RDWR | O_CLOEXEC);
1429 }
1430
1431 static struct bpf_prog *____bpf_prog_get(struct fd f)
1432 {
1433         if (!f.file)
1434                 return ERR_PTR(-EBADF);
1435         if (f.file->f_op != &bpf_prog_fops) {
1436                 fdput(f);
1437                 return ERR_PTR(-EINVAL);
1438         }
1439
1440         return f.file->private_data;
1441 }
1442
1443 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1444 {
1445         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1446                 atomic_sub(i, &prog->aux->refcnt);
1447                 return ERR_PTR(-EBUSY);
1448         }
1449         return prog;
1450 }
1451 EXPORT_SYMBOL_GPL(bpf_prog_add);
1452
1453 void bpf_prog_sub(struct bpf_prog *prog, int i)
1454 {
1455         /* Only to be used for undoing previous bpf_prog_add() in some
1456          * error path. We still know that another entity in our call
1457          * path holds a reference to the program, thus atomic_sub() can
1458          * be safely used in such cases!
1459          */
1460         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1461 }
1462 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1463
1464 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1465 {
1466         return bpf_prog_add(prog, 1);
1467 }
1468 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1469
1470 /* prog_idr_lock should have been held */
1471 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1472 {
1473         int refold;
1474
1475         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1476
1477         if (refold >= BPF_MAX_REFCNT) {
1478                 __bpf_prog_put(prog, false);
1479                 return ERR_PTR(-EBUSY);
1480         }
1481
1482         if (!refold)
1483                 return ERR_PTR(-ENOENT);
1484
1485         return prog;
1486 }
1487 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1488
1489 bool bpf_prog_get_ok(struct bpf_prog *prog,
1490                             enum bpf_prog_type *attach_type, bool attach_drv)
1491 {
1492         /* not an attachment, just a refcount inc, always allow */
1493         if (!attach_type)
1494                 return true;
1495
1496         if (prog->type != *attach_type)
1497                 return false;
1498         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1499                 return false;
1500
1501         return true;
1502 }
1503
1504 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1505                                        bool attach_drv)
1506 {
1507         struct fd f = fdget(ufd);
1508         struct bpf_prog *prog;
1509
1510         prog = ____bpf_prog_get(f);
1511         if (IS_ERR(prog))
1512                 return prog;
1513         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1514                 prog = ERR_PTR(-EINVAL);
1515                 goto out;
1516         }
1517
1518         prog = bpf_prog_inc(prog);
1519 out:
1520         fdput(f);
1521         return prog;
1522 }
1523
1524 struct bpf_prog *bpf_prog_get(u32 ufd)
1525 {
1526         return __bpf_prog_get(ufd, NULL, false);
1527 }
1528
1529 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1530                                        bool attach_drv)
1531 {
1532         return __bpf_prog_get(ufd, &type, attach_drv);
1533 }
1534 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1535
1536 /* Initially all BPF programs could be loaded w/o specifying
1537  * expected_attach_type. Later for some of them specifying expected_attach_type
1538  * at load time became required so that program could be validated properly.
1539  * Programs of types that are allowed to be loaded both w/ and w/o (for
1540  * backward compatibility) expected_attach_type, should have the default attach
1541  * type assigned to expected_attach_type for the latter case, so that it can be
1542  * validated later at attach time.
1543  *
1544  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1545  * prog type requires it but has some attach types that have to be backward
1546  * compatible.
1547  */
1548 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1549 {
1550         switch (attr->prog_type) {
1551         case BPF_PROG_TYPE_CGROUP_SOCK:
1552                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1553                  * exist so checking for non-zero is the way to go here.
1554                  */
1555                 if (!attr->expected_attach_type)
1556                         attr->expected_attach_type =
1557                                 BPF_CGROUP_INET_SOCK_CREATE;
1558                 break;
1559         }
1560 }
1561
1562 static int
1563 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1564                                 enum bpf_attach_type expected_attach_type)
1565 {
1566         switch (prog_type) {
1567         case BPF_PROG_TYPE_CGROUP_SOCK:
1568                 switch (expected_attach_type) {
1569                 case BPF_CGROUP_INET_SOCK_CREATE:
1570                 case BPF_CGROUP_INET4_POST_BIND:
1571                 case BPF_CGROUP_INET6_POST_BIND:
1572                         return 0;
1573                 default:
1574                         return -EINVAL;
1575                 }
1576         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1577                 switch (expected_attach_type) {
1578                 case BPF_CGROUP_INET4_BIND:
1579                 case BPF_CGROUP_INET6_BIND:
1580                 case BPF_CGROUP_INET4_CONNECT:
1581                 case BPF_CGROUP_INET6_CONNECT:
1582                 case BPF_CGROUP_UDP4_SENDMSG:
1583                 case BPF_CGROUP_UDP6_SENDMSG:
1584                         return 0;
1585                 default:
1586                         return -EINVAL;
1587                 }
1588         default:
1589                 return 0;
1590         }
1591 }
1592
1593 /* last field in 'union bpf_attr' used by this command */
1594 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1595
1596 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1597 {
1598         enum bpf_prog_type type = attr->prog_type;
1599         struct bpf_prog *prog;
1600         int err;
1601         char license[128];
1602         bool is_gpl;
1603
1604         if (CHECK_ATTR(BPF_PROG_LOAD))
1605                 return -EINVAL;
1606
1607         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1608                 return -EINVAL;
1609
1610         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1611             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1612             !capable(CAP_SYS_ADMIN))
1613                 return -EPERM;
1614
1615         /* copy eBPF program license from user space */
1616         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1617                               sizeof(license) - 1) < 0)
1618                 return -EFAULT;
1619         license[sizeof(license) - 1] = 0;
1620
1621         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1622         is_gpl = license_is_gpl_compatible(license);
1623
1624         if (attr->insn_cnt == 0 ||
1625             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1626                 return -E2BIG;
1627         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1628             type != BPF_PROG_TYPE_CGROUP_SKB &&
1629             !capable(CAP_SYS_ADMIN))
1630                 return -EPERM;
1631
1632         bpf_prog_load_fixup_attach_type(attr);
1633         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1634                 return -EINVAL;
1635
1636         /* plain bpf_prog allocation */
1637         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1638         if (!prog)
1639                 return -ENOMEM;
1640
1641         prog->expected_attach_type = attr->expected_attach_type;
1642
1643         prog->aux->offload_requested = !!attr->prog_ifindex;
1644
1645         err = security_bpf_prog_alloc(prog->aux);
1646         if (err)
1647                 goto free_prog_nouncharge;
1648
1649         err = bpf_prog_charge_memlock(prog);
1650         if (err)
1651                 goto free_prog_sec;
1652
1653         prog->len = attr->insn_cnt;
1654
1655         err = -EFAULT;
1656         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1657                            bpf_prog_insn_size(prog)) != 0)
1658                 goto free_prog;
1659
1660         prog->orig_prog = NULL;
1661         prog->jited = 0;
1662
1663         atomic_set(&prog->aux->refcnt, 1);
1664         prog->gpl_compatible = is_gpl ? 1 : 0;
1665
1666         if (bpf_prog_is_dev_bound(prog->aux)) {
1667                 err = bpf_prog_offload_init(prog, attr);
1668                 if (err)
1669                         goto free_prog;
1670         }
1671
1672         /* find program type: socket_filter vs tracing_filter */
1673         err = find_prog_type(type, prog);
1674         if (err < 0)
1675                 goto free_prog;
1676
1677         prog->aux->load_time = ktime_get_boot_ns();
1678         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1679         if (err)
1680                 goto free_prog;
1681
1682         /* run eBPF verifier */
1683         err = bpf_check(&prog, attr, uattr);
1684         if (err < 0)
1685                 goto free_used_maps;
1686
1687         prog = bpf_prog_select_runtime(prog, &err);
1688         if (err < 0)
1689                 goto free_used_maps;
1690
1691         err = bpf_prog_alloc_id(prog);
1692         if (err)
1693                 goto free_used_maps;
1694
1695         err = bpf_prog_new_fd(prog);
1696         if (err < 0) {
1697                 /* failed to allocate fd.
1698                  * bpf_prog_put() is needed because the above
1699                  * bpf_prog_alloc_id() has published the prog
1700                  * to the userspace and the userspace may
1701                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1702                  */
1703                 bpf_prog_put(prog);
1704                 return err;
1705         }
1706
1707         bpf_prog_kallsyms_add(prog);
1708         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1709         return err;
1710
1711 free_used_maps:
1712         bpf_prog_free_linfo(prog);
1713         kvfree(prog->aux->func_info);
1714         btf_put(prog->aux->btf);
1715         bpf_prog_kallsyms_del_subprogs(prog);
1716         free_used_maps(prog->aux);
1717 free_prog:
1718         bpf_prog_uncharge_memlock(prog);
1719 free_prog_sec:
1720         security_bpf_prog_free(prog->aux);
1721 free_prog_nouncharge:
1722         bpf_prog_free(prog);
1723         return err;
1724 }
1725
1726 #define BPF_OBJ_LAST_FIELD file_flags
1727
1728 static int bpf_obj_pin(const union bpf_attr *attr)
1729 {
1730         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1731                 return -EINVAL;
1732
1733         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1734 }
1735
1736 static int bpf_obj_get(const union bpf_attr *attr)
1737 {
1738         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1739             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1740                 return -EINVAL;
1741
1742         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1743                                 attr->file_flags);
1744 }
1745
1746 struct bpf_raw_tracepoint {
1747         struct bpf_raw_event_map *btp;
1748         struct bpf_prog *prog;
1749 };
1750
1751 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1752 {
1753         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1754
1755         if (raw_tp->prog) {
1756                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1757                 bpf_prog_put(raw_tp->prog);
1758         }
1759         bpf_put_raw_tracepoint(raw_tp->btp);
1760         kfree(raw_tp);
1761         return 0;
1762 }
1763
1764 static const struct file_operations bpf_raw_tp_fops = {
1765         .release        = bpf_raw_tracepoint_release,
1766         .read           = bpf_dummy_read,
1767         .write          = bpf_dummy_write,
1768 };
1769
1770 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1771
1772 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1773 {
1774         struct bpf_raw_tracepoint *raw_tp;
1775         struct bpf_raw_event_map *btp;
1776         struct bpf_prog *prog;
1777         char tp_name[128];
1778         int tp_fd, err;
1779
1780         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1781                               sizeof(tp_name) - 1) < 0)
1782                 return -EFAULT;
1783         tp_name[sizeof(tp_name) - 1] = 0;
1784
1785         btp = bpf_get_raw_tracepoint(tp_name);
1786         if (!btp)
1787                 return -ENOENT;
1788
1789         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1790         if (!raw_tp) {
1791                 err = -ENOMEM;
1792                 goto out_put_btp;
1793         }
1794         raw_tp->btp = btp;
1795
1796         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1797         if (IS_ERR(prog)) {
1798                 err = PTR_ERR(prog);
1799                 goto out_free_tp;
1800         }
1801         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1802             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1803                 err = -EINVAL;
1804                 goto out_put_prog;
1805         }
1806
1807         err = bpf_probe_register(raw_tp->btp, prog);
1808         if (err)
1809                 goto out_put_prog;
1810
1811         raw_tp->prog = prog;
1812         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1813                                  O_CLOEXEC);
1814         if (tp_fd < 0) {
1815                 bpf_probe_unregister(raw_tp->btp, prog);
1816                 err = tp_fd;
1817                 goto out_put_prog;
1818         }
1819         return tp_fd;
1820
1821 out_put_prog:
1822         bpf_prog_put(prog);
1823 out_free_tp:
1824         kfree(raw_tp);
1825 out_put_btp:
1826         bpf_put_raw_tracepoint(btp);
1827         return err;
1828 }
1829
1830 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1831                                              enum bpf_attach_type attach_type)
1832 {
1833         switch (prog->type) {
1834         case BPF_PROG_TYPE_CGROUP_SOCK:
1835         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1836                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1837         default:
1838                 return 0;
1839         }
1840 }
1841
1842 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1843
1844 #define BPF_F_ATTACH_MASK \
1845         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1846
1847 static int bpf_prog_attach(const union bpf_attr *attr)
1848 {
1849         enum bpf_prog_type ptype;
1850         struct bpf_prog *prog;
1851         int ret;
1852
1853         if (!capable(CAP_NET_ADMIN))
1854                 return -EPERM;
1855
1856         if (CHECK_ATTR(BPF_PROG_ATTACH))
1857                 return -EINVAL;
1858
1859         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1860                 return -EINVAL;
1861
1862         switch (attr->attach_type) {
1863         case BPF_CGROUP_INET_INGRESS:
1864         case BPF_CGROUP_INET_EGRESS:
1865                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1866                 break;
1867         case BPF_CGROUP_INET_SOCK_CREATE:
1868         case BPF_CGROUP_INET4_POST_BIND:
1869         case BPF_CGROUP_INET6_POST_BIND:
1870                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1871                 break;
1872         case BPF_CGROUP_INET4_BIND:
1873         case BPF_CGROUP_INET6_BIND:
1874         case BPF_CGROUP_INET4_CONNECT:
1875         case BPF_CGROUP_INET6_CONNECT:
1876         case BPF_CGROUP_UDP4_SENDMSG:
1877         case BPF_CGROUP_UDP6_SENDMSG:
1878                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1879                 break;
1880         case BPF_CGROUP_SOCK_OPS:
1881                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1882                 break;
1883         case BPF_CGROUP_DEVICE:
1884                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1885                 break;
1886         case BPF_SK_MSG_VERDICT:
1887                 ptype = BPF_PROG_TYPE_SK_MSG;
1888                 break;
1889         case BPF_SK_SKB_STREAM_PARSER:
1890         case BPF_SK_SKB_STREAM_VERDICT:
1891                 ptype = BPF_PROG_TYPE_SK_SKB;
1892                 break;
1893         case BPF_LIRC_MODE2:
1894                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1895                 break;
1896         case BPF_FLOW_DISSECTOR:
1897                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1898                 break;
1899         case BPF_CGROUP_SYSCTL:
1900                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1901                 break;
1902         default:
1903                 return -EINVAL;
1904         }
1905
1906         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1907         if (IS_ERR(prog))
1908                 return PTR_ERR(prog);
1909
1910         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1911                 bpf_prog_put(prog);
1912                 return -EINVAL;
1913         }
1914
1915         switch (ptype) {
1916         case BPF_PROG_TYPE_SK_SKB:
1917         case BPF_PROG_TYPE_SK_MSG:
1918                 ret = sock_map_get_from_fd(attr, prog);
1919                 break;
1920         case BPF_PROG_TYPE_LIRC_MODE2:
1921                 ret = lirc_prog_attach(attr, prog);
1922                 break;
1923         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1924                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1925                 break;
1926         default:
1927                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1928         }
1929
1930         if (ret)
1931                 bpf_prog_put(prog);
1932         return ret;
1933 }
1934
1935 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1936
1937 static int bpf_prog_detach(const union bpf_attr *attr)
1938 {
1939         enum bpf_prog_type ptype;
1940
1941         if (!capable(CAP_NET_ADMIN))
1942                 return -EPERM;
1943
1944         if (CHECK_ATTR(BPF_PROG_DETACH))
1945                 return -EINVAL;
1946
1947         switch (attr->attach_type) {
1948         case BPF_CGROUP_INET_INGRESS:
1949         case BPF_CGROUP_INET_EGRESS:
1950                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1951                 break;
1952         case BPF_CGROUP_INET_SOCK_CREATE:
1953         case BPF_CGROUP_INET4_POST_BIND:
1954         case BPF_CGROUP_INET6_POST_BIND:
1955                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1956                 break;
1957         case BPF_CGROUP_INET4_BIND:
1958         case BPF_CGROUP_INET6_BIND:
1959         case BPF_CGROUP_INET4_CONNECT:
1960         case BPF_CGROUP_INET6_CONNECT:
1961         case BPF_CGROUP_UDP4_SENDMSG:
1962         case BPF_CGROUP_UDP6_SENDMSG:
1963                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1964                 break;
1965         case BPF_CGROUP_SOCK_OPS:
1966                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1967                 break;
1968         case BPF_CGROUP_DEVICE:
1969                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1970                 break;
1971         case BPF_SK_MSG_VERDICT:
1972                 return sock_map_get_from_fd(attr, NULL);
1973         case BPF_SK_SKB_STREAM_PARSER:
1974         case BPF_SK_SKB_STREAM_VERDICT:
1975                 return sock_map_get_from_fd(attr, NULL);
1976         case BPF_LIRC_MODE2:
1977                 return lirc_prog_detach(attr);
1978         case BPF_FLOW_DISSECTOR:
1979                 return skb_flow_dissector_bpf_prog_detach(attr);
1980         case BPF_CGROUP_SYSCTL:
1981                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1982                 break;
1983         default:
1984                 return -EINVAL;
1985         }
1986
1987         return cgroup_bpf_prog_detach(attr, ptype);
1988 }
1989
1990 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1991
1992 static int bpf_prog_query(const union bpf_attr *attr,
1993                           union bpf_attr __user *uattr)
1994 {
1995         if (!capable(CAP_NET_ADMIN))
1996                 return -EPERM;
1997         if (CHECK_ATTR(BPF_PROG_QUERY))
1998                 return -EINVAL;
1999         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2000                 return -EINVAL;
2001
2002         switch (attr->query.attach_type) {
2003         case BPF_CGROUP_INET_INGRESS:
2004         case BPF_CGROUP_INET_EGRESS:
2005         case BPF_CGROUP_INET_SOCK_CREATE:
2006         case BPF_CGROUP_INET4_BIND:
2007         case BPF_CGROUP_INET6_BIND:
2008         case BPF_CGROUP_INET4_POST_BIND:
2009         case BPF_CGROUP_INET6_POST_BIND:
2010         case BPF_CGROUP_INET4_CONNECT:
2011         case BPF_CGROUP_INET6_CONNECT:
2012         case BPF_CGROUP_UDP4_SENDMSG:
2013         case BPF_CGROUP_UDP6_SENDMSG:
2014         case BPF_CGROUP_SOCK_OPS:
2015         case BPF_CGROUP_DEVICE:
2016         case BPF_CGROUP_SYSCTL:
2017                 break;
2018         case BPF_LIRC_MODE2:
2019                 return lirc_prog_query(attr, uattr);
2020         case BPF_FLOW_DISSECTOR:
2021                 return skb_flow_dissector_prog_query(attr, uattr);
2022         default:
2023                 return -EINVAL;
2024         }
2025
2026         return cgroup_bpf_prog_query(attr, uattr);
2027 }
2028
2029 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2030
2031 static int bpf_prog_test_run(const union bpf_attr *attr,
2032                              union bpf_attr __user *uattr)
2033 {
2034         struct bpf_prog *prog;
2035         int ret = -ENOTSUPP;
2036
2037         if (!capable(CAP_SYS_ADMIN))
2038                 return -EPERM;
2039         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2040                 return -EINVAL;
2041
2042         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2043             (!attr->test.ctx_size_in && attr->test.ctx_in))
2044                 return -EINVAL;
2045
2046         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2047             (!attr->test.ctx_size_out && attr->test.ctx_out))
2048                 return -EINVAL;
2049
2050         prog = bpf_prog_get(attr->test.prog_fd);
2051         if (IS_ERR(prog))
2052                 return PTR_ERR(prog);
2053
2054         if (prog->aux->ops->test_run)
2055                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2056
2057         bpf_prog_put(prog);
2058         return ret;
2059 }
2060
2061 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2062
2063 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2064                                union bpf_attr __user *uattr,
2065                                struct idr *idr,
2066                                spinlock_t *lock)
2067 {
2068         u32 next_id = attr->start_id;
2069         int err = 0;
2070
2071         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2072                 return -EINVAL;
2073
2074         if (!capable(CAP_SYS_ADMIN))
2075                 return -EPERM;
2076
2077         next_id++;
2078         spin_lock_bh(lock);
2079         if (!idr_get_next(idr, &next_id))
2080                 err = -ENOENT;
2081         spin_unlock_bh(lock);
2082
2083         if (!err)
2084                 err = put_user(next_id, &uattr->next_id);
2085
2086         return err;
2087 }
2088
2089 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2090
2091 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2092 {
2093         struct bpf_prog *prog;
2094         u32 id = attr->prog_id;
2095         int fd;
2096
2097         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2098                 return -EINVAL;
2099
2100         if (!capable(CAP_SYS_ADMIN))
2101                 return -EPERM;
2102
2103         spin_lock_bh(&prog_idr_lock);
2104         prog = idr_find(&prog_idr, id);
2105         if (prog)
2106                 prog = bpf_prog_inc_not_zero(prog);
2107         else
2108                 prog = ERR_PTR(-ENOENT);
2109         spin_unlock_bh(&prog_idr_lock);
2110
2111         if (IS_ERR(prog))
2112                 return PTR_ERR(prog);
2113
2114         fd = bpf_prog_new_fd(prog);
2115         if (fd < 0)
2116                 bpf_prog_put(prog);
2117
2118         return fd;
2119 }
2120
2121 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2122
2123 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2124 {
2125         struct bpf_map *map;
2126         u32 id = attr->map_id;
2127         int f_flags;
2128         int fd;
2129
2130         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2131             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2132                 return -EINVAL;
2133
2134         if (!capable(CAP_SYS_ADMIN))
2135                 return -EPERM;
2136
2137         f_flags = bpf_get_file_flag(attr->open_flags);
2138         if (f_flags < 0)
2139                 return f_flags;
2140
2141         spin_lock_bh(&map_idr_lock);
2142         map = idr_find(&map_idr, id);
2143         if (map)
2144                 map = bpf_map_inc_not_zero(map, true);
2145         else
2146                 map = ERR_PTR(-ENOENT);
2147         spin_unlock_bh(&map_idr_lock);
2148
2149         if (IS_ERR(map))
2150                 return PTR_ERR(map);
2151
2152         fd = bpf_map_new_fd(map, f_flags);
2153         if (fd < 0)
2154                 bpf_map_put_with_uref(map);
2155
2156         return fd;
2157 }
2158
2159 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2160                                               unsigned long addr, u32 *off,
2161                                               u32 *type)
2162 {
2163         const struct bpf_map *map;
2164         int i;
2165
2166         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2167                 map = prog->aux->used_maps[i];
2168                 if (map == (void *)addr) {
2169                         *type = BPF_PSEUDO_MAP_FD;
2170                         return map;
2171                 }
2172                 if (!map->ops->map_direct_value_meta)
2173                         continue;
2174                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2175                         *type = BPF_PSEUDO_MAP_VALUE;
2176                         return map;
2177                 }
2178         }
2179
2180         return NULL;
2181 }
2182
2183 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2184 {
2185         const struct bpf_map *map;
2186         struct bpf_insn *insns;
2187         u32 off, type;
2188         u64 imm;
2189         int i;
2190
2191         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2192                         GFP_USER);
2193         if (!insns)
2194                 return insns;
2195
2196         for (i = 0; i < prog->len; i++) {
2197                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2198                         insns[i].code = BPF_JMP | BPF_CALL;
2199                         insns[i].imm = BPF_FUNC_tail_call;
2200                         /* fall-through */
2201                 }
2202                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2203                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2204                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2205                                 insns[i].code = BPF_JMP | BPF_CALL;
2206                         if (!bpf_dump_raw_ok())
2207                                 insns[i].imm = 0;
2208                         continue;
2209                 }
2210
2211                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2212                         continue;
2213
2214                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2215                 map = bpf_map_from_imm(prog, imm, &off, &type);
2216                 if (map) {
2217                         insns[i].src_reg = type;
2218                         insns[i].imm = map->id;
2219                         insns[i + 1].imm = off;
2220                         continue;
2221                 }
2222         }
2223
2224         return insns;
2225 }
2226
2227 static int set_info_rec_size(struct bpf_prog_info *info)
2228 {
2229         /*
2230          * Ensure info.*_rec_size is the same as kernel expected size
2231          *
2232          * or
2233          *
2234          * Only allow zero *_rec_size if both _rec_size and _cnt are
2235          * zero.  In this case, the kernel will set the expected
2236          * _rec_size back to the info.
2237          */
2238
2239         if ((info->nr_func_info || info->func_info_rec_size) &&
2240             info->func_info_rec_size != sizeof(struct bpf_func_info))
2241                 return -EINVAL;
2242
2243         if ((info->nr_line_info || info->line_info_rec_size) &&
2244             info->line_info_rec_size != sizeof(struct bpf_line_info))
2245                 return -EINVAL;
2246
2247         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2248             info->jited_line_info_rec_size != sizeof(__u64))
2249                 return -EINVAL;
2250
2251         info->func_info_rec_size = sizeof(struct bpf_func_info);
2252         info->line_info_rec_size = sizeof(struct bpf_line_info);
2253         info->jited_line_info_rec_size = sizeof(__u64);
2254
2255         return 0;
2256 }
2257
2258 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2259                                    const union bpf_attr *attr,
2260                                    union bpf_attr __user *uattr)
2261 {
2262         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2263         struct bpf_prog_info info = {};
2264         u32 info_len = attr->info.info_len;
2265         struct bpf_prog_stats stats;
2266         char __user *uinsns;
2267         u32 ulen;
2268         int err;
2269
2270         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2271         if (err)
2272                 return err;
2273         info_len = min_t(u32, sizeof(info), info_len);
2274
2275         if (copy_from_user(&info, uinfo, info_len))
2276                 return -EFAULT;
2277
2278         info.type = prog->type;
2279         info.id = prog->aux->id;
2280         info.load_time = prog->aux->load_time;
2281         info.created_by_uid = from_kuid_munged(current_user_ns(),
2282                                                prog->aux->user->uid);
2283         info.gpl_compatible = prog->gpl_compatible;
2284
2285         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2286         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2287
2288         ulen = info.nr_map_ids;
2289         info.nr_map_ids = prog->aux->used_map_cnt;
2290         ulen = min_t(u32, info.nr_map_ids, ulen);
2291         if (ulen) {
2292                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2293                 u32 i;
2294
2295                 for (i = 0; i < ulen; i++)
2296                         if (put_user(prog->aux->used_maps[i]->id,
2297                                      &user_map_ids[i]))
2298                                 return -EFAULT;
2299         }
2300
2301         err = set_info_rec_size(&info);
2302         if (err)
2303                 return err;
2304
2305         bpf_prog_get_stats(prog, &stats);
2306         info.run_time_ns = stats.nsecs;
2307         info.run_cnt = stats.cnt;
2308
2309         if (!capable(CAP_SYS_ADMIN)) {
2310                 info.jited_prog_len = 0;
2311                 info.xlated_prog_len = 0;
2312                 info.nr_jited_ksyms = 0;
2313                 info.nr_jited_func_lens = 0;
2314                 info.nr_func_info = 0;
2315                 info.nr_line_info = 0;
2316                 info.nr_jited_line_info = 0;
2317                 goto done;
2318         }
2319
2320         ulen = info.xlated_prog_len;
2321         info.xlated_prog_len = bpf_prog_insn_size(prog);
2322         if (info.xlated_prog_len && ulen) {
2323                 struct bpf_insn *insns_sanitized;
2324                 bool fault;
2325
2326                 if (prog->blinded && !bpf_dump_raw_ok()) {
2327                         info.xlated_prog_insns = 0;
2328                         goto done;
2329                 }
2330                 insns_sanitized = bpf_insn_prepare_dump(prog);
2331                 if (!insns_sanitized)
2332                         return -ENOMEM;
2333                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2334                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2335                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2336                 kfree(insns_sanitized);
2337                 if (fault)
2338                         return -EFAULT;
2339         }
2340
2341         if (bpf_prog_is_dev_bound(prog->aux)) {
2342                 err = bpf_prog_offload_info_fill(&info, prog);
2343                 if (err)
2344                         return err;
2345                 goto done;
2346         }
2347
2348         /* NOTE: the following code is supposed to be skipped for offload.
2349          * bpf_prog_offload_info_fill() is the place to fill similar fields
2350          * for offload.
2351          */
2352         ulen = info.jited_prog_len;
2353         if (prog->aux->func_cnt) {
2354                 u32 i;
2355
2356                 info.jited_prog_len = 0;
2357                 for (i = 0; i < prog->aux->func_cnt; i++)
2358                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2359         } else {
2360                 info.jited_prog_len = prog->jited_len;
2361         }
2362
2363         if (info.jited_prog_len && ulen) {
2364                 if (bpf_dump_raw_ok()) {
2365                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2366                         ulen = min_t(u32, info.jited_prog_len, ulen);
2367
2368                         /* for multi-function programs, copy the JITed
2369                          * instructions for all the functions
2370                          */
2371                         if (prog->aux->func_cnt) {
2372                                 u32 len, free, i;
2373                                 u8 *img;
2374
2375                                 free = ulen;
2376                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2377                                         len = prog->aux->func[i]->jited_len;
2378                                         len = min_t(u32, len, free);
2379                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2380                                         if (copy_to_user(uinsns, img, len))
2381                                                 return -EFAULT;
2382                                         uinsns += len;
2383                                         free -= len;
2384                                         if (!free)
2385                                                 break;
2386                                 }
2387                         } else {
2388                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2389                                         return -EFAULT;
2390                         }
2391                 } else {
2392                         info.jited_prog_insns = 0;
2393                 }
2394         }
2395
2396         ulen = info.nr_jited_ksyms;
2397         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2398         if (ulen) {
2399                 if (bpf_dump_raw_ok()) {
2400                         unsigned long ksym_addr;
2401                         u64 __user *user_ksyms;
2402                         u32 i;
2403
2404                         /* copy the address of the kernel symbol
2405                          * corresponding to each function
2406                          */
2407                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2408                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2409                         if (prog->aux->func_cnt) {
2410                                 for (i = 0; i < ulen; i++) {
2411                                         ksym_addr = (unsigned long)
2412                                                 prog->aux->func[i]->bpf_func;
2413                                         if (put_user((u64) ksym_addr,
2414                                                      &user_ksyms[i]))
2415                                                 return -EFAULT;
2416                                 }
2417                         } else {
2418                                 ksym_addr = (unsigned long) prog->bpf_func;
2419                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2420                                         return -EFAULT;
2421                         }
2422                 } else {
2423                         info.jited_ksyms = 0;
2424                 }
2425         }
2426
2427         ulen = info.nr_jited_func_lens;
2428         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2429         if (ulen) {
2430                 if (bpf_dump_raw_ok()) {
2431                         u32 __user *user_lens;
2432                         u32 func_len, i;
2433
2434                         /* copy the JITed image lengths for each function */
2435                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2436                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2437                         if (prog->aux->func_cnt) {
2438                                 for (i = 0; i < ulen; i++) {
2439                                         func_len =
2440                                                 prog->aux->func[i]->jited_len;
2441                                         if (put_user(func_len, &user_lens[i]))
2442                                                 return -EFAULT;
2443                                 }
2444                         } else {
2445                                 func_len = prog->jited_len;
2446                                 if (put_user(func_len, &user_lens[0]))
2447                                         return -EFAULT;
2448                         }
2449                 } else {
2450                         info.jited_func_lens = 0;
2451                 }
2452         }
2453
2454         if (prog->aux->btf)
2455                 info.btf_id = btf_id(prog->aux->btf);
2456
2457         ulen = info.nr_func_info;
2458         info.nr_func_info = prog->aux->func_info_cnt;
2459         if (info.nr_func_info && ulen) {
2460                 char __user *user_finfo;
2461
2462                 user_finfo = u64_to_user_ptr(info.func_info);
2463                 ulen = min_t(u32, info.nr_func_info, ulen);
2464                 if (copy_to_user(user_finfo, prog->aux->func_info,
2465                                  info.func_info_rec_size * ulen))
2466                         return -EFAULT;
2467         }
2468
2469         ulen = info.nr_line_info;
2470         info.nr_line_info = prog->aux->nr_linfo;
2471         if (info.nr_line_info && ulen) {
2472                 __u8 __user *user_linfo;
2473
2474                 user_linfo = u64_to_user_ptr(info.line_info);
2475                 ulen = min_t(u32, info.nr_line_info, ulen);
2476                 if (copy_to_user(user_linfo, prog->aux->linfo,
2477                                  info.line_info_rec_size * ulen))
2478                         return -EFAULT;
2479         }
2480
2481         ulen = info.nr_jited_line_info;
2482         if (prog->aux->jited_linfo)
2483                 info.nr_jited_line_info = prog->aux->nr_linfo;
2484         else
2485                 info.nr_jited_line_info = 0;
2486         if (info.nr_jited_line_info && ulen) {
2487                 if (bpf_dump_raw_ok()) {
2488                         __u64 __user *user_linfo;
2489                         u32 i;
2490
2491                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2492                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2493                         for (i = 0; i < ulen; i++) {
2494                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2495                                              &user_linfo[i]))
2496                                         return -EFAULT;
2497                         }
2498                 } else {
2499                         info.jited_line_info = 0;
2500                 }
2501         }
2502
2503         ulen = info.nr_prog_tags;
2504         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2505         if (ulen) {
2506                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2507                 u32 i;
2508
2509                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2510                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2511                 if (prog->aux->func_cnt) {
2512                         for (i = 0; i < ulen; i++) {
2513                                 if (copy_to_user(user_prog_tags[i],
2514                                                  prog->aux->func[i]->tag,
2515                                                  BPF_TAG_SIZE))
2516                                         return -EFAULT;
2517                         }
2518                 } else {
2519                         if (copy_to_user(user_prog_tags[0],
2520                                          prog->tag, BPF_TAG_SIZE))
2521                                 return -EFAULT;
2522                 }
2523         }
2524
2525 done:
2526         if (copy_to_user(uinfo, &info, info_len) ||
2527             put_user(info_len, &uattr->info.info_len))
2528                 return -EFAULT;
2529
2530         return 0;
2531 }
2532
2533 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2534                                   const union bpf_attr *attr,
2535                                   union bpf_attr __user *uattr)
2536 {
2537         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2538         struct bpf_map_info info = {};
2539         u32 info_len = attr->info.info_len;
2540         int err;
2541
2542         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2543         if (err)
2544                 return err;
2545         info_len = min_t(u32, sizeof(info), info_len);
2546
2547         info.type = map->map_type;
2548         info.id = map->id;
2549         info.key_size = map->key_size;
2550         info.value_size = map->value_size;
2551         info.max_entries = map->max_entries;
2552         info.map_flags = map->map_flags;
2553         memcpy(info.name, map->name, sizeof(map->name));
2554
2555         if (map->btf) {
2556                 info.btf_id = btf_id(map->btf);
2557                 info.btf_key_type_id = map->btf_key_type_id;
2558                 info.btf_value_type_id = map->btf_value_type_id;
2559         }
2560
2561         if (bpf_map_is_dev_bound(map)) {
2562                 err = bpf_map_offload_info_fill(&info, map);
2563                 if (err)
2564                         return err;
2565         }
2566
2567         if (copy_to_user(uinfo, &info, info_len) ||
2568             put_user(info_len, &uattr->info.info_len))
2569                 return -EFAULT;
2570
2571         return 0;
2572 }
2573
2574 static int bpf_btf_get_info_by_fd(struct btf *btf,
2575                                   const union bpf_attr *attr,
2576                                   union bpf_attr __user *uattr)
2577 {
2578         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2579         u32 info_len = attr->info.info_len;
2580         int err;
2581
2582         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2583         if (err)
2584                 return err;
2585
2586         return btf_get_info_by_fd(btf, attr, uattr);
2587 }
2588
2589 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2590
2591 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2592                                   union bpf_attr __user *uattr)
2593 {
2594         int ufd = attr->info.bpf_fd;
2595         struct fd f;
2596         int err;
2597
2598         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2599                 return -EINVAL;
2600
2601         f = fdget(ufd);
2602         if (!f.file)
2603                 return -EBADFD;
2604
2605         if (f.file->f_op == &bpf_prog_fops)
2606                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2607                                               uattr);
2608         else if (f.file->f_op == &bpf_map_fops)
2609                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2610                                              uattr);
2611         else if (f.file->f_op == &btf_fops)
2612                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2613         else
2614                 err = -EINVAL;
2615
2616         fdput(f);
2617         return err;
2618 }
2619
2620 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2621
2622 static int bpf_btf_load(const union bpf_attr *attr)
2623 {
2624         if (CHECK_ATTR(BPF_BTF_LOAD))
2625                 return -EINVAL;
2626
2627         if (!capable(CAP_SYS_ADMIN))
2628                 return -EPERM;
2629
2630         return btf_new_fd(attr);
2631 }
2632
2633 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2634
2635 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2636 {
2637         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2638                 return -EINVAL;
2639
2640         if (!capable(CAP_SYS_ADMIN))
2641                 return -EPERM;
2642
2643         return btf_get_fd_by_id(attr->btf_id);
2644 }
2645
2646 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2647                                     union bpf_attr __user *uattr,
2648                                     u32 prog_id, u32 fd_type,
2649                                     const char *buf, u64 probe_offset,
2650                                     u64 probe_addr)
2651 {
2652         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2653         u32 len = buf ? strlen(buf) : 0, input_len;
2654         int err = 0;
2655
2656         if (put_user(len, &uattr->task_fd_query.buf_len))
2657                 return -EFAULT;
2658         input_len = attr->task_fd_query.buf_len;
2659         if (input_len && ubuf) {
2660                 if (!len) {
2661                         /* nothing to copy, just make ubuf NULL terminated */
2662                         char zero = '\0';
2663
2664                         if (put_user(zero, ubuf))
2665                                 return -EFAULT;
2666                 } else if (input_len >= len + 1) {
2667                         /* ubuf can hold the string with NULL terminator */
2668                         if (copy_to_user(ubuf, buf, len + 1))
2669                                 return -EFAULT;
2670                 } else {
2671                         /* ubuf cannot hold the string with NULL terminator,
2672                          * do a partial copy with NULL terminator.
2673                          */
2674                         char zero = '\0';
2675
2676                         err = -ENOSPC;
2677                         if (copy_to_user(ubuf, buf, input_len - 1))
2678                                 return -EFAULT;
2679                         if (put_user(zero, ubuf + input_len - 1))
2680                                 return -EFAULT;
2681                 }
2682         }
2683
2684         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2685             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2686             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2687             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2688                 return -EFAULT;
2689
2690         return err;
2691 }
2692
2693 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2694
2695 static int bpf_task_fd_query(const union bpf_attr *attr,
2696                              union bpf_attr __user *uattr)
2697 {
2698         pid_t pid = attr->task_fd_query.pid;
2699         u32 fd = attr->task_fd_query.fd;
2700         const struct perf_event *event;
2701         struct files_struct *files;
2702         struct task_struct *task;
2703         struct file *file;
2704         int err;
2705
2706         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2707                 return -EINVAL;
2708
2709         if (!capable(CAP_SYS_ADMIN))
2710                 return -EPERM;
2711
2712         if (attr->task_fd_query.flags != 0)
2713                 return -EINVAL;
2714
2715         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2716         if (!task)
2717                 return -ENOENT;
2718
2719         files = get_files_struct(task);
2720         put_task_struct(task);
2721         if (!files)
2722                 return -ENOENT;
2723
2724         err = 0;
2725         spin_lock(&files->file_lock);
2726         file = fcheck_files(files, fd);
2727         if (!file)
2728                 err = -EBADF;
2729         else
2730                 get_file(file);
2731         spin_unlock(&files->file_lock);
2732         put_files_struct(files);
2733
2734         if (err)
2735                 goto out;
2736
2737         if (file->f_op == &bpf_raw_tp_fops) {
2738                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2739                 struct bpf_raw_event_map *btp = raw_tp->btp;
2740
2741                 err = bpf_task_fd_query_copy(attr, uattr,
2742                                              raw_tp->prog->aux->id,
2743                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2744                                              btp->tp->name, 0, 0);
2745                 goto put_file;
2746         }
2747
2748         event = perf_get_event(file);
2749         if (!IS_ERR(event)) {
2750                 u64 probe_offset, probe_addr;
2751                 u32 prog_id, fd_type;
2752                 const char *buf;
2753
2754                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2755                                               &buf, &probe_offset,
2756                                               &probe_addr);
2757                 if (!err)
2758                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2759                                                      fd_type, buf,
2760                                                      probe_offset,
2761                                                      probe_addr);
2762                 goto put_file;
2763         }
2764
2765         err = -ENOTSUPP;
2766 put_file:
2767         fput(file);
2768 out:
2769         return err;
2770 }
2771
2772 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2773 {
2774         union bpf_attr attr = {};
2775         int err;
2776
2777         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2778                 return -EPERM;
2779
2780         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2781         if (err)
2782                 return err;
2783         size = min_t(u32, size, sizeof(attr));
2784
2785         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2786         if (copy_from_user(&attr, uattr, size) != 0)
2787                 return -EFAULT;
2788
2789         err = security_bpf(cmd, &attr, size);
2790         if (err < 0)
2791                 return err;
2792
2793         switch (cmd) {
2794         case BPF_MAP_CREATE:
2795                 err = map_create(&attr);
2796                 break;
2797         case BPF_MAP_LOOKUP_ELEM:
2798                 err = map_lookup_elem(&attr);
2799                 break;
2800         case BPF_MAP_UPDATE_ELEM:
2801                 err = map_update_elem(&attr);
2802                 break;
2803         case BPF_MAP_DELETE_ELEM:
2804                 err = map_delete_elem(&attr);
2805                 break;
2806         case BPF_MAP_GET_NEXT_KEY:
2807                 err = map_get_next_key(&attr);
2808                 break;
2809         case BPF_MAP_FREEZE:
2810                 err = map_freeze(&attr);
2811                 break;
2812         case BPF_PROG_LOAD:
2813                 err = bpf_prog_load(&attr, uattr);
2814                 break;
2815         case BPF_OBJ_PIN:
2816                 err = bpf_obj_pin(&attr);
2817                 break;
2818         case BPF_OBJ_GET:
2819                 err = bpf_obj_get(&attr);
2820                 break;
2821         case BPF_PROG_ATTACH:
2822                 err = bpf_prog_attach(&attr);
2823                 break;
2824         case BPF_PROG_DETACH:
2825                 err = bpf_prog_detach(&attr);
2826                 break;
2827         case BPF_PROG_QUERY:
2828                 err = bpf_prog_query(&attr, uattr);
2829                 break;
2830         case BPF_PROG_TEST_RUN:
2831                 err = bpf_prog_test_run(&attr, uattr);
2832                 break;
2833         case BPF_PROG_GET_NEXT_ID:
2834                 err = bpf_obj_get_next_id(&attr, uattr,
2835                                           &prog_idr, &prog_idr_lock);
2836                 break;
2837         case BPF_MAP_GET_NEXT_ID:
2838                 err = bpf_obj_get_next_id(&attr, uattr,
2839                                           &map_idr, &map_idr_lock);
2840                 break;
2841         case BPF_PROG_GET_FD_BY_ID:
2842                 err = bpf_prog_get_fd_by_id(&attr);
2843                 break;
2844         case BPF_MAP_GET_FD_BY_ID:
2845                 err = bpf_map_get_fd_by_id(&attr);
2846                 break;
2847         case BPF_OBJ_GET_INFO_BY_FD:
2848                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2849                 break;
2850         case BPF_RAW_TRACEPOINT_OPEN:
2851                 err = bpf_raw_tracepoint_open(&attr);
2852                 break;
2853         case BPF_BTF_LOAD:
2854                 err = bpf_btf_load(&attr);
2855                 break;
2856         case BPF_BTF_GET_FD_BY_ID:
2857                 err = bpf_btf_get_fd_by_id(&attr);
2858                 break;
2859         case BPF_TASK_FD_QUERY:
2860                 err = bpf_task_fd_query(&attr, uattr);
2861                 break;
2862         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2863                 err = map_lookup_and_delete_elem(&attr);
2864                 break;
2865         default:
2866                 err = -EINVAL;
2867                 break;
2868         }
2869
2870         return err;
2871 }