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

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

Version: ~ [ linux-5.6 ] ~ [ linux-5.5.13 ] ~ [ linux-5.4.28 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.113 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.174 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.217 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.217 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.82 ] ~ [ 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-2.6.32.71 ] ~ [ 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 static void maybe_wait_bpf_programs(struct bpf_map *map)
687 {
688         /* Wait for any running BPF programs to complete so that
689          * userspace, when we return to it, knows that all programs
690          * that could be running use the new map value.
691          */
692         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
693             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
694                 synchronize_rcu();
695 }
696 
697 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
698 
699 static int map_update_elem(union bpf_attr *attr)
700 {
701         void __user *ukey = u64_to_user_ptr(attr->key);
702         void __user *uvalue = u64_to_user_ptr(attr->value);
703         int ufd = attr->map_fd;
704         struct bpf_map *map;
705         void *key, *value;
706         u32 value_size;
707         struct fd f;
708         int err;
709 
710         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
711                 return -EINVAL;
712 
713         f = fdget(ufd);
714         map = __bpf_map_get(f);
715         if (IS_ERR(map))
716                 return PTR_ERR(map);
717 
718         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
719                 err = -EPERM;
720                 goto err_put;
721         }
722 
723         key = memdup_user(ukey, map->key_size);
724         if (IS_ERR(key)) {
725                 err = PTR_ERR(key);
726                 goto err_put;
727         }
728 
729         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
730             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
731             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
732                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
733         else
734                 value_size = map->value_size;
735 
736         err = -ENOMEM;
737         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
738         if (!value)
739                 goto free_key;
740 
741         err = -EFAULT;
742         if (copy_from_user(value, uvalue, value_size) != 0)
743                 goto free_value;
744 
745         /* Need to create a kthread, thus must support schedule */
746         if (bpf_map_is_dev_bound(map)) {
747                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
748                 goto out;
749         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
750                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
751                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
752                 err = map->ops->map_update_elem(map, key, value, attr->flags);
753                 goto out;
754         }
755 
756         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
757          * inside bpf map update or delete otherwise deadlocks are possible
758          */
759         preempt_disable();
760         __this_cpu_inc(bpf_prog_active);
761         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
762             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
763                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
764         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
765                 err = bpf_percpu_array_update(map, key, value, attr->flags);
766         } else if (IS_FD_ARRAY(map)) {
767                 rcu_read_lock();
768                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
769                                                    attr->flags);
770                 rcu_read_unlock();
771         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
772                 rcu_read_lock();
773                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
774                                                   attr->flags);
775                 rcu_read_unlock();
776         } else {
777                 rcu_read_lock();
778                 err = map->ops->map_update_elem(map, key, value, attr->flags);
779                 rcu_read_unlock();
780         }
781         __this_cpu_dec(bpf_prog_active);
782         preempt_enable();
783         maybe_wait_bpf_programs(map);
784 out:
785 free_value:
786         kfree(value);
787 free_key:
788         kfree(key);
789 err_put:
790         fdput(f);
791         return err;
792 }
793 
794 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
795 
796 static int map_delete_elem(union bpf_attr *attr)
797 {
798         void __user *ukey = u64_to_user_ptr(attr->key);
799         int ufd = attr->map_fd;
800         struct bpf_map *map;
801         struct fd f;
802         void *key;
803         int err;
804 
805         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
806                 return -EINVAL;
807 
808         f = fdget(ufd);
809         map = __bpf_map_get(f);
810         if (IS_ERR(map))
811                 return PTR_ERR(map);
812 
813         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
814                 err = -EPERM;
815                 goto err_put;
816         }
817 
818         key = memdup_user(ukey, map->key_size);
819         if (IS_ERR(key)) {
820                 err = PTR_ERR(key);
821                 goto err_put;
822         }
823 
824         if (bpf_map_is_dev_bound(map)) {
825                 err = bpf_map_offload_delete_elem(map, key);
826                 goto out;
827         }
828 
829         preempt_disable();
830         __this_cpu_inc(bpf_prog_active);
831         rcu_read_lock();
832         err = map->ops->map_delete_elem(map, key);
833         rcu_read_unlock();
834         __this_cpu_dec(bpf_prog_active);
835         preempt_enable();
836         maybe_wait_bpf_programs(map);
837 out:
838         kfree(key);
839 err_put:
840         fdput(f);
841         return err;
842 }
843 
844 /* last field in 'union bpf_attr' used by this command */
845 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
846 
847 static int map_get_next_key(union bpf_attr *attr)
848 {
849         void __user *ukey = u64_to_user_ptr(attr->key);
850         void __user *unext_key = u64_to_user_ptr(attr->next_key);
851         int ufd = attr->map_fd;
852         struct bpf_map *map;
853         void *key, *next_key;
854         struct fd f;
855         int err;
856 
857         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
858                 return -EINVAL;
859 
860         f = fdget(ufd);
861         map = __bpf_map_get(f);
862         if (IS_ERR(map))
863                 return PTR_ERR(map);
864 
865         if (!(f.file->f_mode & FMODE_CAN_READ)) {
866                 err = -EPERM;
867                 goto err_put;
868         }
869 
870         if (ukey) {
871                 key = memdup_user(ukey, map->key_size);
872                 if (IS_ERR(key)) {
873                         err = PTR_ERR(key);
874                         goto err_put;
875                 }
876         } else {
877                 key = NULL;
878         }
879 
880         err = -ENOMEM;
881         next_key = kmalloc(map->key_size, GFP_USER);
882         if (!next_key)
883                 goto free_key;
884 
885         if (bpf_map_is_dev_bound(map)) {
886                 err = bpf_map_offload_get_next_key(map, key, next_key);
887                 goto out;
888         }
889 
890         rcu_read_lock();
891         err = map->ops->map_get_next_key(map, key, next_key);
892         rcu_read_unlock();
893 out:
894         if (err)
895                 goto free_next_key;
896 
897         err = -EFAULT;
898         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
899                 goto free_next_key;
900 
901         err = 0;
902 
903 free_next_key:
904         kfree(next_key);
905 free_key:
906         kfree(key);
907 err_put:
908         fdput(f);
909         return err;
910 }
911 
912 static const struct bpf_prog_ops * const bpf_prog_types[] = {
913 #define BPF_PROG_TYPE(_id, _name) \
914         [_id] = & _name ## _prog_ops,
915 #define BPF_MAP_TYPE(_id, _ops)
916 #include <linux/bpf_types.h>
917 #undef BPF_PROG_TYPE
918 #undef BPF_MAP_TYPE
919 };
920 
921 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
922 {
923         const struct bpf_prog_ops *ops;
924 
925         if (type >= ARRAY_SIZE(bpf_prog_types))
926                 return -EINVAL;
927         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
928         ops = bpf_prog_types[type];
929         if (!ops)
930                 return -EINVAL;
931 
932         if (!bpf_prog_is_dev_bound(prog->aux))
933                 prog->aux->ops = ops;
934         else
935                 prog->aux->ops = &bpf_offload_prog_ops;
936         prog->type = type;
937         return 0;
938 }
939 
940 /* drop refcnt on maps used by eBPF program and free auxilary data */
941 static void free_used_maps(struct bpf_prog_aux *aux)
942 {
943         int i;
944 
945         for (i = 0; i < aux->used_map_cnt; i++)
946                 bpf_map_put(aux->used_maps[i]);
947 
948         kfree(aux->used_maps);
949 }
950 
951 int __bpf_prog_charge(struct user_struct *user, u32 pages)
952 {
953         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
954         unsigned long user_bufs;
955 
956         if (user) {
957                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
958                 if (user_bufs > memlock_limit) {
959                         atomic_long_sub(pages, &user->locked_vm);
960                         return -EPERM;
961                 }
962         }
963 
964         return 0;
965 }
966 
967 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
968 {
969         if (user)
970                 atomic_long_sub(pages, &user->locked_vm);
971 }
972 
973 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
974 {
975         struct user_struct *user = get_current_user();
976         int ret;
977 
978         ret = __bpf_prog_charge(user, prog->pages);
979         if (ret) {
980                 free_uid(user);
981                 return ret;
982         }
983 
984         prog->aux->user = user;
985         return 0;
986 }
987 
988 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
989 {
990         struct user_struct *user = prog->aux->user;
991 
992         __bpf_prog_uncharge(user, prog->pages);
993         free_uid(user);
994 }
995 
996 static int bpf_prog_alloc_id(struct bpf_prog *prog)
997 {
998         int id;
999 
1000         idr_preload(GFP_KERNEL);
1001         spin_lock_bh(&prog_idr_lock);
1002         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1003         if (id > 0)
1004                 prog->aux->id = id;
1005         spin_unlock_bh(&prog_idr_lock);
1006         idr_preload_end();
1007 
1008         /* id is in [1, INT_MAX) */
1009         if (WARN_ON_ONCE(!id))
1010                 return -ENOSPC;
1011 
1012         return id > 0 ? 0 : id;
1013 }
1014 
1015 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1016 {
1017         /* cBPF to eBPF migrations are currently not in the idr store.
1018          * Offloaded programs are removed from the store when their device
1019          * disappears - even if someone grabs an fd to them they are unusable,
1020          * simply waiting for refcnt to drop to be freed.
1021          */
1022         if (!prog->aux->id)
1023                 return;
1024 
1025         if (do_idr_lock)
1026                 spin_lock_bh(&prog_idr_lock);
1027         else
1028                 __acquire(&prog_idr_lock);
1029 
1030         idr_remove(&prog_idr, prog->aux->id);
1031         prog->aux->id = 0;
1032 
1033         if (do_idr_lock)
1034                 spin_unlock_bh(&prog_idr_lock);
1035         else
1036                 __release(&prog_idr_lock);
1037 }
1038 
1039 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1040 {
1041         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1042 
1043         free_used_maps(aux);
1044         bpf_prog_uncharge_memlock(aux->prog);
1045         security_bpf_prog_free(aux);
1046         bpf_prog_free(aux->prog);
1047 }
1048 
1049 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1050 {
1051         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1052                 /* bpf_prog_free_id() must be called first */
1053                 bpf_prog_free_id(prog, do_idr_lock);
1054                 bpf_prog_kallsyms_del_all(prog);
1055 
1056                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1057         }
1058 }
1059 
1060 void bpf_prog_put(struct bpf_prog *prog)
1061 {
1062         __bpf_prog_put(prog, true);
1063 }
1064 EXPORT_SYMBOL_GPL(bpf_prog_put);
1065 
1066 static int bpf_prog_release(struct inode *inode, struct file *filp)
1067 {
1068         struct bpf_prog *prog = filp->private_data;
1069 
1070         bpf_prog_put(prog);
1071         return 0;
1072 }
1073 
1074 #ifdef CONFIG_PROC_FS
1075 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1076 {
1077         const struct bpf_prog *prog = filp->private_data;
1078         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1079 
1080         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1081         seq_printf(m,
1082                    "prog_type:\t%u\n"
1083                    "prog_jited:\t%u\n"
1084                    "prog_tag:\t%s\n"
1085                    "memlock:\t%llu\n"
1086                    "prog_id:\t%u\n",
1087                    prog->type,
1088                    prog->jited,
1089                    prog_tag,
1090                    prog->pages * 1ULL << PAGE_SHIFT,
1091                    prog->aux->id);
1092 }
1093 #endif
1094 
1095 const struct file_operations bpf_prog_fops = {
1096 #ifdef CONFIG_PROC_FS
1097         .show_fdinfo    = bpf_prog_show_fdinfo,
1098 #endif
1099         .release        = bpf_prog_release,
1100         .read           = bpf_dummy_read,
1101         .write          = bpf_dummy_write,
1102 };
1103 
1104 int bpf_prog_new_fd(struct bpf_prog *prog)
1105 {
1106         int ret;
1107 
1108         ret = security_bpf_prog(prog);
1109         if (ret < 0)
1110                 return ret;
1111 
1112         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1113                                 O_RDWR | O_CLOEXEC);
1114 }
1115 
1116 static struct bpf_prog *____bpf_prog_get(struct fd f)
1117 {
1118         if (!f.file)
1119                 return ERR_PTR(-EBADF);
1120         if (f.file->f_op != &bpf_prog_fops) {
1121                 fdput(f);
1122                 return ERR_PTR(-EINVAL);
1123         }
1124 
1125         return f.file->private_data;
1126 }
1127 
1128 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1129 {
1130         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1131                 atomic_sub(i, &prog->aux->refcnt);
1132                 return ERR_PTR(-EBUSY);
1133         }
1134         return prog;
1135 }
1136 EXPORT_SYMBOL_GPL(bpf_prog_add);
1137 
1138 void bpf_prog_sub(struct bpf_prog *prog, int i)
1139 {
1140         /* Only to be used for undoing previous bpf_prog_add() in some
1141          * error path. We still know that another entity in our call
1142          * path holds a reference to the program, thus atomic_sub() can
1143          * be safely used in such cases!
1144          */
1145         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1146 }
1147 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1148 
1149 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1150 {
1151         return bpf_prog_add(prog, 1);
1152 }
1153 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1154 
1155 /* prog_idr_lock should have been held */
1156 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1157 {
1158         int refold;
1159 
1160         refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1161 
1162         if (refold >= BPF_MAX_REFCNT) {
1163                 __bpf_prog_put(prog, false);
1164                 return ERR_PTR(-EBUSY);
1165         }
1166 
1167         if (!refold)
1168                 return ERR_PTR(-ENOENT);
1169 
1170         return prog;
1171 }
1172 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1173 
1174 bool bpf_prog_get_ok(struct bpf_prog *prog,
1175                             enum bpf_prog_type *attach_type, bool attach_drv)
1176 {
1177         /* not an attachment, just a refcount inc, always allow */
1178         if (!attach_type)
1179                 return true;
1180 
1181         if (prog->type != *attach_type)
1182                 return false;
1183         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1184                 return false;
1185 
1186         return true;
1187 }
1188 
1189 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1190                                        bool attach_drv)
1191 {
1192         struct fd f = fdget(ufd);
1193         struct bpf_prog *prog;
1194 
1195         prog = ____bpf_prog_get(f);
1196         if (IS_ERR(prog))
1197                 return prog;
1198         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1199                 prog = ERR_PTR(-EINVAL);
1200                 goto out;
1201         }
1202 
1203         prog = bpf_prog_inc(prog);
1204 out:
1205         fdput(f);
1206         return prog;
1207 }
1208 
1209 struct bpf_prog *bpf_prog_get(u32 ufd)
1210 {
1211         return __bpf_prog_get(ufd, NULL, false);
1212 }
1213 
1214 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1215                                        bool attach_drv)
1216 {
1217         return __bpf_prog_get(ufd, &type, attach_drv);
1218 }
1219 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1220 
1221 /* Initially all BPF programs could be loaded w/o specifying
1222  * expected_attach_type. Later for some of them specifying expected_attach_type
1223  * at load time became required so that program could be validated properly.
1224  * Programs of types that are allowed to be loaded both w/ and w/o (for
1225  * backward compatibility) expected_attach_type, should have the default attach
1226  * type assigned to expected_attach_type for the latter case, so that it can be
1227  * validated later at attach time.
1228  *
1229  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1230  * prog type requires it but has some attach types that have to be backward
1231  * compatible.
1232  */
1233 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1234 {
1235         switch (attr->prog_type) {
1236         case BPF_PROG_TYPE_CGROUP_SOCK:
1237                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1238                  * exist so checking for non-zero is the way to go here.
1239                  */
1240                 if (!attr->expected_attach_type)
1241                         attr->expected_attach_type =
1242                                 BPF_CGROUP_INET_SOCK_CREATE;
1243                 break;
1244         }
1245 }
1246 
1247 static int
1248 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1249                                 enum bpf_attach_type expected_attach_type)
1250 {
1251         switch (prog_type) {
1252         case BPF_PROG_TYPE_CGROUP_SOCK:
1253                 switch (expected_attach_type) {
1254                 case BPF_CGROUP_INET_SOCK_CREATE:
1255                 case BPF_CGROUP_INET4_POST_BIND:
1256                 case BPF_CGROUP_INET6_POST_BIND:
1257                         return 0;
1258                 default:
1259                         return -EINVAL;
1260                 }
1261         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1262                 switch (expected_attach_type) {
1263                 case BPF_CGROUP_INET4_BIND:
1264                 case BPF_CGROUP_INET6_BIND:
1265                 case BPF_CGROUP_INET4_CONNECT:
1266                 case BPF_CGROUP_INET6_CONNECT:
1267                 case BPF_CGROUP_UDP4_SENDMSG:
1268                 case BPF_CGROUP_UDP6_SENDMSG:
1269                         return 0;
1270                 default:
1271                         return -EINVAL;
1272                 }
1273         default:
1274                 return 0;
1275         }
1276 }
1277 
1278 /* last field in 'union bpf_attr' used by this command */
1279 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1280 
1281 static int bpf_prog_load(union bpf_attr *attr)
1282 {
1283         enum bpf_prog_type type = attr->prog_type;
1284         struct bpf_prog *prog;
1285         int err;
1286         char license[128];
1287         bool is_gpl;
1288 
1289         if (CHECK_ATTR(BPF_PROG_LOAD))
1290                 return -EINVAL;
1291 
1292         if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1293                 return -EINVAL;
1294 
1295         /* copy eBPF program license from user space */
1296         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1297                               sizeof(license) - 1) < 0)
1298                 return -EFAULT;
1299         license[sizeof(license) - 1] = 0;
1300 
1301         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1302         is_gpl = license_is_gpl_compatible(license);
1303 
1304         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1305                 return -E2BIG;
1306 
1307         if (type == BPF_PROG_TYPE_KPROBE &&
1308             attr->kern_version != LINUX_VERSION_CODE)
1309                 return -EINVAL;
1310 
1311         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1312             type != BPF_PROG_TYPE_CGROUP_SKB &&
1313             !capable(CAP_SYS_ADMIN))
1314                 return -EPERM;
1315 
1316         bpf_prog_load_fixup_attach_type(attr);
1317         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1318                 return -EINVAL;
1319 
1320         /* plain bpf_prog allocation */
1321         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1322         if (!prog)
1323                 return -ENOMEM;
1324 
1325         prog->expected_attach_type = attr->expected_attach_type;
1326 
1327         prog->aux->offload_requested = !!attr->prog_ifindex;
1328 
1329         err = security_bpf_prog_alloc(prog->aux);
1330         if (err)
1331                 goto free_prog_nouncharge;
1332 
1333         err = bpf_prog_charge_memlock(prog);
1334         if (err)
1335                 goto free_prog_sec;
1336 
1337         prog->len = attr->insn_cnt;
1338 
1339         err = -EFAULT;
1340         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1341                            bpf_prog_insn_size(prog)) != 0)
1342                 goto free_prog;
1343 
1344         prog->orig_prog = NULL;
1345         prog->jited = 0;
1346 
1347         atomic_set(&prog->aux->refcnt, 1);
1348         prog->gpl_compatible = is_gpl ? 1 : 0;
1349 
1350         if (bpf_prog_is_dev_bound(prog->aux)) {
1351                 err = bpf_prog_offload_init(prog, attr);
1352                 if (err)
1353                         goto free_prog;
1354         }
1355 
1356         /* find program type: socket_filter vs tracing_filter */
1357         err = find_prog_type(type, prog);
1358         if (err < 0)
1359                 goto free_prog;
1360 
1361         prog->aux->load_time = ktime_get_boot_ns();
1362         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1363         if (err)
1364                 goto free_prog;
1365 
1366         /* run eBPF verifier */
1367         err = bpf_check(&prog, attr);
1368         if (err < 0)
1369                 goto free_used_maps;
1370 
1371         prog = bpf_prog_select_runtime(prog, &err);
1372         if (err < 0)
1373                 goto free_used_maps;
1374 
1375         err = bpf_prog_alloc_id(prog);
1376         if (err)
1377                 goto free_used_maps;
1378 
1379         err = bpf_prog_new_fd(prog);
1380         if (err < 0) {
1381                 /* failed to allocate fd.
1382                  * bpf_prog_put() is needed because the above
1383                  * bpf_prog_alloc_id() has published the prog
1384                  * to the userspace and the userspace may
1385                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1386                  */
1387                 bpf_prog_put(prog);
1388                 return err;
1389         }
1390 
1391         bpf_prog_kallsyms_add(prog);
1392         return err;
1393 
1394 free_used_maps:
1395         bpf_prog_kallsyms_del_subprogs(prog);
1396         free_used_maps(prog->aux);
1397 free_prog:
1398         bpf_prog_uncharge_memlock(prog);
1399 free_prog_sec:
1400         security_bpf_prog_free(prog->aux);
1401 free_prog_nouncharge:
1402         bpf_prog_free(prog);
1403         return err;
1404 }
1405 
1406 #define BPF_OBJ_LAST_FIELD file_flags
1407 
1408 static int bpf_obj_pin(const union bpf_attr *attr)
1409 {
1410         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1411                 return -EINVAL;
1412 
1413         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1414 }
1415 
1416 static int bpf_obj_get(const union bpf_attr *attr)
1417 {
1418         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1419             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1420                 return -EINVAL;
1421 
1422         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1423                                 attr->file_flags);
1424 }
1425 
1426 struct bpf_raw_tracepoint {
1427         struct bpf_raw_event_map *btp;
1428         struct bpf_prog *prog;
1429 };
1430 
1431 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1432 {
1433         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1434 
1435         if (raw_tp->prog) {
1436                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1437                 bpf_prog_put(raw_tp->prog);
1438         }
1439         kfree(raw_tp);
1440         return 0;
1441 }
1442 
1443 static const struct file_operations bpf_raw_tp_fops = {
1444         .release        = bpf_raw_tracepoint_release,
1445         .read           = bpf_dummy_read,
1446         .write          = bpf_dummy_write,
1447 };
1448 
1449 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1450 
1451 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1452 {
1453         struct bpf_raw_tracepoint *raw_tp;
1454         struct bpf_raw_event_map *btp;
1455         struct bpf_prog *prog;
1456         char tp_name[128];
1457         int tp_fd, err;
1458 
1459         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1460                               sizeof(tp_name) - 1) < 0)
1461                 return -EFAULT;
1462         tp_name[sizeof(tp_name) - 1] = 0;
1463 
1464         btp = bpf_find_raw_tracepoint(tp_name);
1465         if (!btp)
1466                 return -ENOENT;
1467 
1468         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1469         if (!raw_tp)
1470                 return -ENOMEM;
1471         raw_tp->btp = btp;
1472 
1473         prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1474                                  BPF_PROG_TYPE_RAW_TRACEPOINT);
1475         if (IS_ERR(prog)) {
1476                 err = PTR_ERR(prog);
1477                 goto out_free_tp;
1478         }
1479 
1480         err = bpf_probe_register(raw_tp->btp, prog);
1481         if (err)
1482                 goto out_put_prog;
1483 
1484         raw_tp->prog = prog;
1485         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1486                                  O_CLOEXEC);
1487         if (tp_fd < 0) {
1488                 bpf_probe_unregister(raw_tp->btp, prog);
1489                 err = tp_fd;
1490                 goto out_put_prog;
1491         }
1492         return tp_fd;
1493 
1494 out_put_prog:
1495         bpf_prog_put(prog);
1496 out_free_tp:
1497         kfree(raw_tp);
1498         return err;
1499 }
1500 
1501 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1502                                              enum bpf_attach_type attach_type)
1503 {
1504         switch (prog->type) {
1505         case BPF_PROG_TYPE_CGROUP_SOCK:
1506         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1507                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1508         default:
1509                 return 0;
1510         }
1511 }
1512 
1513 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1514 
1515 #define BPF_F_ATTACH_MASK \
1516         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1517 
1518 static int bpf_prog_attach(const union bpf_attr *attr)
1519 {
1520         enum bpf_prog_type ptype;
1521         struct bpf_prog *prog;
1522         int ret;
1523 
1524         if (!capable(CAP_NET_ADMIN))
1525                 return -EPERM;
1526 
1527         if (CHECK_ATTR(BPF_PROG_ATTACH))
1528                 return -EINVAL;
1529 
1530         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1531                 return -EINVAL;
1532 
1533         switch (attr->attach_type) {
1534         case BPF_CGROUP_INET_INGRESS:
1535         case BPF_CGROUP_INET_EGRESS:
1536                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1537                 break;
1538         case BPF_CGROUP_INET_SOCK_CREATE:
1539         case BPF_CGROUP_INET4_POST_BIND:
1540         case BPF_CGROUP_INET6_POST_BIND:
1541                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1542                 break;
1543         case BPF_CGROUP_INET4_BIND:
1544         case BPF_CGROUP_INET6_BIND:
1545         case BPF_CGROUP_INET4_CONNECT:
1546         case BPF_CGROUP_INET6_CONNECT:
1547         case BPF_CGROUP_UDP4_SENDMSG:
1548         case BPF_CGROUP_UDP6_SENDMSG:
1549                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1550                 break;
1551         case BPF_CGROUP_SOCK_OPS:
1552                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1553                 break;
1554         case BPF_CGROUP_DEVICE:
1555                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1556                 break;
1557         case BPF_SK_MSG_VERDICT:
1558                 ptype = BPF_PROG_TYPE_SK_MSG;
1559                 break;
1560         case BPF_SK_SKB_STREAM_PARSER:
1561         case BPF_SK_SKB_STREAM_VERDICT:
1562                 ptype = BPF_PROG_TYPE_SK_SKB;
1563                 break;
1564         case BPF_LIRC_MODE2:
1565                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1566                 break;
1567         default:
1568                 return -EINVAL;
1569         }
1570 
1571         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1572         if (IS_ERR(prog))
1573                 return PTR_ERR(prog);
1574 
1575         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1576                 bpf_prog_put(prog);
1577                 return -EINVAL;
1578         }
1579 
1580         switch (ptype) {
1581         case BPF_PROG_TYPE_SK_SKB:
1582         case BPF_PROG_TYPE_SK_MSG:
1583                 ret = sockmap_get_from_fd(attr, ptype, prog);
1584                 break;
1585         case BPF_PROG_TYPE_LIRC_MODE2:
1586                 ret = lirc_prog_attach(attr, prog);
1587                 break;
1588         default:
1589                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1590         }
1591 
1592         if (ret)
1593                 bpf_prog_put(prog);
1594         return ret;
1595 }
1596 
1597 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1598 
1599 static int bpf_prog_detach(const union bpf_attr *attr)
1600 {
1601         enum bpf_prog_type ptype;
1602 
1603         if (!capable(CAP_NET_ADMIN))
1604                 return -EPERM;
1605 
1606         if (CHECK_ATTR(BPF_PROG_DETACH))
1607                 return -EINVAL;
1608 
1609         switch (attr->attach_type) {
1610         case BPF_CGROUP_INET_INGRESS:
1611         case BPF_CGROUP_INET_EGRESS:
1612                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1613                 break;
1614         case BPF_CGROUP_INET_SOCK_CREATE:
1615         case BPF_CGROUP_INET4_POST_BIND:
1616         case BPF_CGROUP_INET6_POST_BIND:
1617                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1618                 break;
1619         case BPF_CGROUP_INET4_BIND:
1620         case BPF_CGROUP_INET6_BIND:
1621         case BPF_CGROUP_INET4_CONNECT:
1622         case BPF_CGROUP_INET6_CONNECT:
1623         case BPF_CGROUP_UDP4_SENDMSG:
1624         case BPF_CGROUP_UDP6_SENDMSG:
1625                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1626                 break;
1627         case BPF_CGROUP_SOCK_OPS:
1628                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1629                 break;
1630         case BPF_CGROUP_DEVICE:
1631                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1632                 break;
1633         case BPF_SK_MSG_VERDICT:
1634                 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1635         case BPF_SK_SKB_STREAM_PARSER:
1636         case BPF_SK_SKB_STREAM_VERDICT:
1637                 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1638         case BPF_LIRC_MODE2:
1639                 return lirc_prog_detach(attr);
1640         default:
1641                 return -EINVAL;
1642         }
1643 
1644         return cgroup_bpf_prog_detach(attr, ptype);
1645 }
1646 
1647 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1648 
1649 static int bpf_prog_query(const union bpf_attr *attr,
1650                           union bpf_attr __user *uattr)
1651 {
1652         if (!capable(CAP_NET_ADMIN))
1653                 return -EPERM;
1654         if (CHECK_ATTR(BPF_PROG_QUERY))
1655                 return -EINVAL;
1656         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1657                 return -EINVAL;
1658 
1659         switch (attr->query.attach_type) {
1660         case BPF_CGROUP_INET_INGRESS:
1661         case BPF_CGROUP_INET_EGRESS:
1662         case BPF_CGROUP_INET_SOCK_CREATE:
1663         case BPF_CGROUP_INET4_BIND:
1664         case BPF_CGROUP_INET6_BIND:
1665         case BPF_CGROUP_INET4_POST_BIND:
1666         case BPF_CGROUP_INET6_POST_BIND:
1667         case BPF_CGROUP_INET4_CONNECT:
1668         case BPF_CGROUP_INET6_CONNECT:
1669         case BPF_CGROUP_UDP4_SENDMSG:
1670         case BPF_CGROUP_UDP6_SENDMSG:
1671         case BPF_CGROUP_SOCK_OPS:
1672         case BPF_CGROUP_DEVICE:
1673                 break;
1674         case BPF_LIRC_MODE2:
1675                 return lirc_prog_query(attr, uattr);
1676         default:
1677                 return -EINVAL;
1678         }
1679 
1680         return cgroup_bpf_prog_query(attr, uattr);
1681 }
1682 
1683 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1684 
1685 static int bpf_prog_test_run(const union bpf_attr *attr,
1686                              union bpf_attr __user *uattr)
1687 {
1688         struct bpf_prog *prog;
1689         int ret = -ENOTSUPP;
1690 
1691         if (!capable(CAP_SYS_ADMIN))
1692                 return -EPERM;
1693         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1694                 return -EINVAL;
1695 
1696         prog = bpf_prog_get(attr->test.prog_fd);
1697         if (IS_ERR(prog))
1698                 return PTR_ERR(prog);
1699 
1700         if (prog->aux->ops->test_run)
1701                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1702 
1703         bpf_prog_put(prog);
1704         return ret;
1705 }
1706 
1707 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1708 
1709 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1710                                union bpf_attr __user *uattr,
1711                                struct idr *idr,
1712                                spinlock_t *lock)
1713 {
1714         u32 next_id = attr->start_id;
1715         int err = 0;
1716 
1717         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1718                 return -EINVAL;
1719 
1720         if (!capable(CAP_SYS_ADMIN))
1721                 return -EPERM;
1722 
1723         next_id++;
1724         spin_lock_bh(lock);
1725         if (!idr_get_next(idr, &next_id))
1726                 err = -ENOENT;
1727         spin_unlock_bh(lock);
1728 
1729         if (!err)
1730                 err = put_user(next_id, &uattr->next_id);
1731 
1732         return err;
1733 }
1734 
1735 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1736 
1737 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1738 {
1739         struct bpf_prog *prog;
1740         u32 id = attr->prog_id;
1741         int fd;
1742 
1743         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1744                 return -EINVAL;
1745 
1746         if (!capable(CAP_SYS_ADMIN))
1747                 return -EPERM;
1748 
1749         spin_lock_bh(&prog_idr_lock);
1750         prog = idr_find(&prog_idr, id);
1751         if (prog)
1752                 prog = bpf_prog_inc_not_zero(prog);
1753         else
1754                 prog = ERR_PTR(-ENOENT);
1755         spin_unlock_bh(&prog_idr_lock);
1756 
1757         if (IS_ERR(prog))
1758                 return PTR_ERR(prog);
1759 
1760         fd = bpf_prog_new_fd(prog);
1761         if (fd < 0)
1762                 bpf_prog_put(prog);
1763 
1764         return fd;
1765 }
1766 
1767 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1768 
1769 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1770 {
1771         struct bpf_map *map;
1772         u32 id = attr->map_id;
1773         int f_flags;
1774         int fd;
1775 
1776         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1777             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1778                 return -EINVAL;
1779 
1780         if (!capable(CAP_SYS_ADMIN))
1781                 return -EPERM;
1782 
1783         f_flags = bpf_get_file_flag(attr->open_flags);
1784         if (f_flags < 0)
1785                 return f_flags;
1786 
1787         spin_lock_bh(&map_idr_lock);
1788         map = idr_find(&map_idr, id);
1789         if (map)
1790                 map = bpf_map_inc_not_zero(map, true);
1791         else
1792                 map = ERR_PTR(-ENOENT);
1793         spin_unlock_bh(&map_idr_lock);
1794 
1795         if (IS_ERR(map))
1796                 return PTR_ERR(map);
1797 
1798         fd = bpf_map_new_fd(map, f_flags);
1799         if (fd < 0)
1800                 bpf_map_put(map);
1801 
1802         return fd;
1803 }
1804 
1805 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1806                                               unsigned long addr)
1807 {
1808         int i;
1809 
1810         for (i = 0; i < prog->aux->used_map_cnt; i++)
1811                 if (prog->aux->used_maps[i] == (void *)addr)
1812                         return prog->aux->used_maps[i];
1813         return NULL;
1814 }
1815 
1816 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1817 {
1818         const struct bpf_map *map;
1819         struct bpf_insn *insns;
1820         u64 imm;
1821         int i;
1822 
1823         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1824                         GFP_USER);
1825         if (!insns)
1826                 return insns;
1827 
1828         for (i = 0; i < prog->len; i++) {
1829                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1830                         insns[i].code = BPF_JMP | BPF_CALL;
1831                         insns[i].imm = BPF_FUNC_tail_call;
1832                         /* fall-through */
1833                 }
1834                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1835                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1836                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1837                                 insns[i].code = BPF_JMP | BPF_CALL;
1838                         if (!bpf_dump_raw_ok())
1839                                 insns[i].imm = 0;
1840                         continue;
1841                 }
1842 
1843                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1844                         continue;
1845 
1846                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1847                 map = bpf_map_from_imm(prog, imm);
1848                 if (map) {
1849                         insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1850                         insns[i].imm = map->id;
1851                         insns[i + 1].imm = 0;
1852                         continue;
1853                 }
1854 
1855                 if (!bpf_dump_raw_ok() &&
1856                     imm == (unsigned long)prog->aux) {
1857                         insns[i].imm = 0;
1858                         insns[i + 1].imm = 0;
1859                         continue;
1860                 }
1861         }
1862 
1863         return insns;
1864 }
1865 
1866 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1867                                    const union bpf_attr *attr,
1868                                    union bpf_attr __user *uattr)
1869 {
1870         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1871         struct bpf_prog_info info = {};
1872         u32 info_len = attr->info.info_len;
1873         char __user *uinsns;
1874         u32 ulen;
1875         int err;
1876 
1877         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1878         if (err)
1879                 return err;
1880         info_len = min_t(u32, sizeof(info), info_len);
1881 
1882         if (copy_from_user(&info, uinfo, info_len))
1883                 return -EFAULT;
1884 
1885         info.type = prog->type;
1886         info.id = prog->aux->id;
1887         info.load_time = prog->aux->load_time;
1888         info.created_by_uid = from_kuid_munged(current_user_ns(),
1889                                                prog->aux->user->uid);
1890         info.gpl_compatible = prog->gpl_compatible;
1891 
1892         memcpy(info.tag, prog->tag, sizeof(prog->tag));
1893         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1894 
1895         ulen = info.nr_map_ids;
1896         info.nr_map_ids = prog->aux->used_map_cnt;
1897         ulen = min_t(u32, info.nr_map_ids, ulen);
1898         if (ulen) {
1899                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1900                 u32 i;
1901 
1902                 for (i = 0; i < ulen; i++)
1903                         if (put_user(prog->aux->used_maps[i]->id,
1904                                      &user_map_ids[i]))
1905                                 return -EFAULT;
1906         }
1907 
1908         if (!capable(CAP_SYS_ADMIN)) {
1909                 info.jited_prog_len = 0;
1910                 info.xlated_prog_len = 0;
1911                 info.nr_jited_ksyms = 0;
1912                 goto done;
1913         }
1914 
1915         ulen = info.xlated_prog_len;
1916         info.xlated_prog_len = bpf_prog_insn_size(prog);
1917         if (info.xlated_prog_len && ulen) {
1918                 struct bpf_insn *insns_sanitized;
1919                 bool fault;
1920 
1921                 if (prog->blinded && !bpf_dump_raw_ok()) {
1922                         info.xlated_prog_insns = 0;
1923                         goto done;
1924                 }
1925                 insns_sanitized = bpf_insn_prepare_dump(prog);
1926                 if (!insns_sanitized)
1927                         return -ENOMEM;
1928                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1929                 ulen = min_t(u32, info.xlated_prog_len, ulen);
1930                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1931                 kfree(insns_sanitized);
1932                 if (fault)
1933                         return -EFAULT;
1934         }
1935 
1936         if (bpf_prog_is_dev_bound(prog->aux)) {
1937                 err = bpf_prog_offload_info_fill(&info, prog);
1938                 if (err)
1939                         return err;
1940                 goto done;
1941         }
1942 
1943         /* NOTE: the following code is supposed to be skipped for offload.
1944          * bpf_prog_offload_info_fill() is the place to fill similar fields
1945          * for offload.
1946          */
1947         ulen = info.jited_prog_len;
1948         if (prog->aux->func_cnt) {
1949                 u32 i;
1950 
1951                 info.jited_prog_len = 0;
1952                 for (i = 0; i < prog->aux->func_cnt; i++)
1953                         info.jited_prog_len += prog->aux->func[i]->jited_len;
1954         } else {
1955                 info.jited_prog_len = prog->jited_len;
1956         }
1957 
1958         if (info.jited_prog_len && ulen) {
1959                 if (bpf_dump_raw_ok()) {
1960                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
1961                         ulen = min_t(u32, info.jited_prog_len, ulen);
1962 
1963                         /* for multi-function programs, copy the JITed
1964                          * instructions for all the functions
1965                          */
1966                         if (prog->aux->func_cnt) {
1967                                 u32 len, free, i;
1968                                 u8 *img;
1969 
1970                                 free = ulen;
1971                                 for (i = 0; i < prog->aux->func_cnt; i++) {
1972                                         len = prog->aux->func[i]->jited_len;
1973                                         len = min_t(u32, len, free);
1974                                         img = (u8 *) prog->aux->func[i]->bpf_func;
1975                                         if (copy_to_user(uinsns, img, len))
1976                                                 return -EFAULT;
1977                                         uinsns += len;
1978                                         free -= len;
1979                                         if (!free)
1980                                                 break;
1981                                 }
1982                         } else {
1983                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1984                                         return -EFAULT;
1985                         }
1986                 } else {
1987                         info.jited_prog_insns = 0;
1988                 }
1989         }
1990 
1991         ulen = info.nr_jited_ksyms;
1992         info.nr_jited_ksyms = prog->aux->func_cnt;
1993         if (info.nr_jited_ksyms && ulen) {
1994                 if (bpf_dump_raw_ok()) {
1995                         u64 __user *user_ksyms;
1996                         ulong ksym_addr;
1997                         u32 i;
1998 
1999                         /* copy the address of the kernel symbol
2000                          * corresponding to each function
2001                          */
2002                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2003                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2004                         for (i = 0; i < ulen; i++) {
2005                                 ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2006                                 ksym_addr &= PAGE_MASK;
2007                                 if (put_user((u64) ksym_addr, &user_ksyms[i]))
2008                                         return -EFAULT;
2009                         }
2010                 } else {
2011                         info.jited_ksyms = 0;
2012                 }
2013         }
2014 
2015         ulen = info.nr_jited_func_lens;
2016         info.nr_jited_func_lens = prog->aux->func_cnt;
2017         if (info.nr_jited_func_lens && ulen) {
2018                 if (bpf_dump_raw_ok()) {
2019                         u32 __user *user_lens;
2020                         u32 func_len, i;
2021 
2022                         /* copy the JITed image lengths for each function */
2023                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2024                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2025                         for (i = 0; i < ulen; i++) {
2026                                 func_len = prog->aux->func[i]->jited_len;
2027                                 if (put_user(func_len, &user_lens[i]))
2028                                         return -EFAULT;
2029                         }
2030                 } else {
2031                         info.jited_func_lens = 0;
2032                 }
2033         }
2034 
2035 done:
2036         if (copy_to_user(uinfo, &info, info_len) ||
2037             put_user(info_len, &uattr->info.info_len))
2038                 return -EFAULT;
2039 
2040         return 0;
2041 }
2042 
2043 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2044                                   const union bpf_attr *attr,
2045                                   union bpf_attr __user *uattr)
2046 {
2047         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2048         struct bpf_map_info info = {};
2049         u32 info_len = attr->info.info_len;
2050         int err;
2051 
2052         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2053         if (err)
2054                 return err;
2055         info_len = min_t(u32, sizeof(info), info_len);
2056 
2057         info.type = map->map_type;
2058         info.id = map->id;
2059         info.key_size = map->key_size;
2060         info.value_size = map->value_size;
2061         info.max_entries = map->max_entries;
2062         info.map_flags = map->map_flags;
2063         memcpy(info.name, map->name, sizeof(map->name));
2064 
2065         if (map->btf) {
2066                 info.btf_id = btf_id(map->btf);
2067                 info.btf_key_type_id = map->btf_key_type_id;
2068                 info.btf_value_type_id = map->btf_value_type_id;
2069         }
2070 
2071         if (bpf_map_is_dev_bound(map)) {
2072                 err = bpf_map_offload_info_fill(&info, map);
2073                 if (err)
2074                         return err;
2075         }
2076 
2077         if (copy_to_user(uinfo, &info, info_len) ||
2078             put_user(info_len, &uattr->info.info_len))
2079                 return -EFAULT;
2080 
2081         return 0;
2082 }
2083 
2084 static int bpf_btf_get_info_by_fd(struct btf *btf,
2085                                   const union bpf_attr *attr,
2086                                   union bpf_attr __user *uattr)
2087 {
2088         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2089         u32 info_len = attr->info.info_len;
2090         int err;
2091 
2092         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2093         if (err)
2094                 return err;
2095 
2096         return btf_get_info_by_fd(btf, attr, uattr);
2097 }
2098 
2099 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2100 
2101 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2102                                   union bpf_attr __user *uattr)
2103 {
2104         int ufd = attr->info.bpf_fd;
2105         struct fd f;
2106         int err;
2107 
2108         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2109                 return -EINVAL;
2110 
2111         f = fdget(ufd);
2112         if (!f.file)
2113                 return -EBADFD;
2114 
2115         if (f.file->f_op == &bpf_prog_fops)
2116                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2117                                               uattr);
2118         else if (f.file->f_op == &bpf_map_fops)
2119                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2120                                              uattr);
2121         else if (f.file->f_op == &btf_fops)
2122                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2123         else
2124                 err = -EINVAL;
2125 
2126         fdput(f);
2127         return err;
2128 }
2129 
2130 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2131 
2132 static int bpf_btf_load(const union bpf_attr *attr)
2133 {
2134         if (CHECK_ATTR(BPF_BTF_LOAD))
2135                 return -EINVAL;
2136 
2137         if (!capable(CAP_SYS_ADMIN))
2138                 return -EPERM;
2139 
2140         return btf_new_fd(attr);
2141 }
2142 
2143 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2144 
2145 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2146 {
2147         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2148                 return -EINVAL;
2149 
2150         if (!capable(CAP_SYS_ADMIN))
2151                 return -EPERM;
2152 
2153         return btf_get_fd_by_id(attr->btf_id);
2154 }
2155 
2156 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2157                                     union bpf_attr __user *uattr,
2158                                     u32 prog_id, u32 fd_type,
2159                                     const char *buf, u64 probe_offset,
2160                                     u64 probe_addr)
2161 {
2162         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2163         u32 len = buf ? strlen(buf) : 0, input_len;
2164         int err = 0;
2165 
2166         if (put_user(len, &uattr->task_fd_query.buf_len))
2167                 return -EFAULT;
2168         input_len = attr->task_fd_query.buf_len;
2169         if (input_len && ubuf) {
2170                 if (!len) {
2171                         /* nothing to copy, just make ubuf NULL terminated */
2172                         char zero = '\0';
2173 
2174                         if (put_user(zero, ubuf))
2175                                 return -EFAULT;
2176                 } else if (input_len >= len + 1) {
2177                         /* ubuf can hold the string with NULL terminator */
2178                         if (copy_to_user(ubuf, buf, len + 1))
2179                                 return -EFAULT;
2180                 } else {
2181                         /* ubuf cannot hold the string with NULL terminator,
2182                          * do a partial copy with NULL terminator.
2183                          */
2184                         char zero = '\0';
2185 
2186                         err = -ENOSPC;
2187                         if (copy_to_user(ubuf, buf, input_len - 1))
2188                                 return -EFAULT;
2189                         if (put_user(zero, ubuf + input_len - 1))
2190                                 return -EFAULT;
2191                 }
2192         }
2193 
2194         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2195             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2196             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2197             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2198                 return -EFAULT;
2199 
2200         return err;
2201 }
2202 
2203 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2204 
2205 static int bpf_task_fd_query(const union bpf_attr *attr,
2206                              union bpf_attr __user *uattr)
2207 {
2208         pid_t pid = attr->task_fd_query.pid;
2209         u32 fd = attr->task_fd_query.fd;
2210         const struct perf_event *event;
2211         struct files_struct *files;
2212         struct task_struct *task;
2213         struct file *file;
2214         int err;
2215 
2216         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2217                 return -EINVAL;
2218 
2219         if (!capable(CAP_SYS_ADMIN))
2220                 return -EPERM;
2221 
2222         if (attr->task_fd_query.flags != 0)
2223                 return -EINVAL;
2224 
2225         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2226         if (!task)
2227                 return -ENOENT;
2228 
2229         files = get_files_struct(task);
2230         put_task_struct(task);
2231         if (!files)
2232                 return -ENOENT;
2233 
2234         err = 0;
2235         spin_lock(&files->file_lock);
2236         file = fcheck_files(files, fd);
2237         if (!file)
2238                 err = -EBADF;
2239         else
2240                 get_file(file);
2241         spin_unlock(&files->file_lock);
2242         put_files_struct(files);
2243 
2244         if (err)
2245                 goto out;
2246 
2247         if (file->f_op == &bpf_raw_tp_fops) {
2248                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2249                 struct bpf_raw_event_map *btp = raw_tp->btp;
2250 
2251                 err = bpf_task_fd_query_copy(attr, uattr,
2252                                              raw_tp->prog->aux->id,
2253                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2254                                              btp->tp->name, 0, 0);
2255                 goto put_file;
2256         }
2257 
2258         event = perf_get_event(file);
2259         if (!IS_ERR(event)) {
2260                 u64 probe_offset, probe_addr;
2261                 u32 prog_id, fd_type;
2262                 const char *buf;
2263 
2264                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2265                                               &buf, &probe_offset,
2266                                               &probe_addr);
2267                 if (!err)
2268                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2269                                                      fd_type, buf,
2270                                                      probe_offset,
2271                                                      probe_addr);
2272                 goto put_file;
2273         }
2274 
2275         err = -ENOTSUPP;
2276 put_file:
2277         fput(file);
2278 out:
2279         return err;
2280 }
2281 
2282 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2283 {
2284         union bpf_attr attr = {};
2285         int err;
2286 
2287         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2288                 return -EPERM;
2289 
2290         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2291         if (err)
2292                 return err;
2293         size = min_t(u32, size, sizeof(attr));
2294 
2295         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2296         if (copy_from_user(&attr, uattr, size) != 0)
2297                 return -EFAULT;
2298 
2299         err = security_bpf(cmd, &attr, size);
2300         if (err < 0)
2301                 return err;
2302 
2303         switch (cmd) {
2304         case BPF_MAP_CREATE:
2305                 err = map_create(&attr);
2306                 break;
2307         case BPF_MAP_LOOKUP_ELEM:
2308                 err = map_lookup_elem(&attr);
2309                 break;
2310         case BPF_MAP_UPDATE_ELEM:
2311                 err = map_update_elem(&attr);
2312                 break;
2313         case BPF_MAP_DELETE_ELEM:
2314                 err = map_delete_elem(&attr);
2315                 break;
2316         case BPF_MAP_GET_NEXT_KEY:
2317                 err = map_get_next_key(&attr);
2318                 break;
2319         case BPF_PROG_LOAD:
2320                 err = bpf_prog_load(&attr);
2321                 break;
2322         case BPF_OBJ_PIN:
2323                 err = bpf_obj_pin(&attr);
2324                 break;
2325         case BPF_OBJ_GET:
2326                 err = bpf_obj_get(&attr);
2327                 break;
2328         case BPF_PROG_ATTACH:
2329                 err = bpf_prog_attach(&attr);
2330                 break;
2331         case BPF_PROG_DETACH:
2332                 err = bpf_prog_detach(&attr);
2333                 break;
2334         case BPF_PROG_QUERY:
2335                 err = bpf_prog_query(&attr, uattr);
2336                 break;
2337         case BPF_PROG_TEST_RUN:
2338                 err = bpf_prog_test_run(&attr, uattr);
2339                 break;
2340         case BPF_PROG_GET_NEXT_ID:
2341                 err = bpf_obj_get_next_id(&attr, uattr,
2342                                           &prog_idr, &prog_idr_lock);
2343                 break;
2344         case BPF_MAP_GET_NEXT_ID:
2345                 err = bpf_obj_get_next_id(&attr, uattr,
2346                                           &map_idr, &map_idr_lock);
2347                 break;
2348         case BPF_PROG_GET_FD_BY_ID:
2349                 err = bpf_prog_get_fd_by_id(&attr);
2350                 break;
2351         case BPF_MAP_GET_FD_BY_ID:
2352                 err = bpf_map_get_fd_by_id(&attr);
2353                 break;
2354         case BPF_OBJ_GET_INFO_BY_FD:
2355                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2356                 break;
2357         case BPF_RAW_TRACEPOINT_OPEN:
2358                 err = bpf_raw_tracepoint_open(&attr);
2359                 break;
2360         case BPF_BTF_LOAD:
2361                 err = bpf_btf_load(&attr);
2362                 break;
2363         case BPF_BTF_GET_FD_BY_ID:
2364                 err = bpf_btf_get_fd_by_id(&attr);
2365                 break;
2366         case BPF_TASK_FD_QUERY:
2367                 err = bpf_task_fd_query(&attr, uattr);
2368                 break;
2369         default:
2370                 err = -EINVAL;
2371                 break;
2372         }
2373 
2374         return err;
2375 }
2376 

~ [ 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