~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/bpf/syscall.c

Version: ~ [ linux-4.18-rc1 ] ~ [ linux-4.17.2 ] ~ [ linux-4.16.16 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.50 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.109 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.138 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.113 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.57 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp