Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Functions to manage eBPF programs attached to cgroups
0004  *
0005  * Copyright (c) 2016 Daniel Mack
0006  */
0007 
0008 #include <linux/kernel.h>
0009 #include <linux/atomic.h>
0010 #include <linux/cgroup.h>
0011 #include <linux/filter.h>
0012 #include <linux/slab.h>
0013 #include <linux/sysctl.h>
0014 #include <linux/string.h>
0015 #include <linux/bpf.h>
0016 #include <linux/bpf-cgroup.h>
0017 #include <linux/bpf_lsm.h>
0018 #include <linux/bpf_verifier.h>
0019 #include <net/sock.h>
0020 #include <net/bpf_sk_storage.h>
0021 
0022 #include "../cgroup/cgroup-internal.h"
0023 
0024 DEFINE_STATIC_KEY_ARRAY_FALSE(cgroup_bpf_enabled_key, MAX_CGROUP_BPF_ATTACH_TYPE);
0025 EXPORT_SYMBOL(cgroup_bpf_enabled_key);
0026 
0027 /* __always_inline is necessary to prevent indirect call through run_prog
0028  * function pointer.
0029  */
0030 static __always_inline int
0031 bpf_prog_run_array_cg(const struct cgroup_bpf *cgrp,
0032               enum cgroup_bpf_attach_type atype,
0033               const void *ctx, bpf_prog_run_fn run_prog,
0034               int retval, u32 *ret_flags)
0035 {
0036     const struct bpf_prog_array_item *item;
0037     const struct bpf_prog *prog;
0038     const struct bpf_prog_array *array;
0039     struct bpf_run_ctx *old_run_ctx;
0040     struct bpf_cg_run_ctx run_ctx;
0041     u32 func_ret;
0042 
0043     run_ctx.retval = retval;
0044     migrate_disable();
0045     rcu_read_lock();
0046     array = rcu_dereference(cgrp->effective[atype]);
0047     item = &array->items[0];
0048     old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
0049     while ((prog = READ_ONCE(item->prog))) {
0050         run_ctx.prog_item = item;
0051         func_ret = run_prog(prog, ctx);
0052         if (ret_flags) {
0053             *(ret_flags) |= (func_ret >> 1);
0054             func_ret &= 1;
0055         }
0056         if (!func_ret && !IS_ERR_VALUE((long)run_ctx.retval))
0057             run_ctx.retval = -EPERM;
0058         item++;
0059     }
0060     bpf_reset_run_ctx(old_run_ctx);
0061     rcu_read_unlock();
0062     migrate_enable();
0063     return run_ctx.retval;
0064 }
0065 
0066 unsigned int __cgroup_bpf_run_lsm_sock(const void *ctx,
0067                        const struct bpf_insn *insn)
0068 {
0069     const struct bpf_prog *shim_prog;
0070     struct sock *sk;
0071     struct cgroup *cgrp;
0072     int ret = 0;
0073     u64 *args;
0074 
0075     args = (u64 *)ctx;
0076     sk = (void *)(unsigned long)args[0];
0077     /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
0078     shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
0079 
0080     cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
0081     if (likely(cgrp))
0082         ret = bpf_prog_run_array_cg(&cgrp->bpf,
0083                         shim_prog->aux->cgroup_atype,
0084                         ctx, bpf_prog_run, 0, NULL);
0085     return ret;
0086 }
0087 
0088 unsigned int __cgroup_bpf_run_lsm_socket(const void *ctx,
0089                      const struct bpf_insn *insn)
0090 {
0091     const struct bpf_prog *shim_prog;
0092     struct socket *sock;
0093     struct cgroup *cgrp;
0094     int ret = 0;
0095     u64 *args;
0096 
0097     args = (u64 *)ctx;
0098     sock = (void *)(unsigned long)args[0];
0099     /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
0100     shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
0101 
0102     cgrp = sock_cgroup_ptr(&sock->sk->sk_cgrp_data);
0103     if (likely(cgrp))
0104         ret = bpf_prog_run_array_cg(&cgrp->bpf,
0105                         shim_prog->aux->cgroup_atype,
0106                         ctx, bpf_prog_run, 0, NULL);
0107     return ret;
0108 }
0109 
0110 unsigned int __cgroup_bpf_run_lsm_current(const void *ctx,
0111                       const struct bpf_insn *insn)
0112 {
0113     const struct bpf_prog *shim_prog;
0114     struct cgroup *cgrp;
0115     int ret = 0;
0116 
0117     /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
0118     shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
0119 
0120     /* We rely on trampoline's __bpf_prog_enter_lsm_cgroup to grab RCU read lock. */
0121     cgrp = task_dfl_cgroup(current);
0122     if (likely(cgrp))
0123         ret = bpf_prog_run_array_cg(&cgrp->bpf,
0124                         shim_prog->aux->cgroup_atype,
0125                         ctx, bpf_prog_run, 0, NULL);
0126     return ret;
0127 }
0128 
0129 #ifdef CONFIG_BPF_LSM
0130 struct cgroup_lsm_atype {
0131     u32 attach_btf_id;
0132     int refcnt;
0133 };
0134 
0135 static struct cgroup_lsm_atype cgroup_lsm_atype[CGROUP_LSM_NUM];
0136 
0137 static enum cgroup_bpf_attach_type
0138 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id)
0139 {
0140     int i;
0141 
0142     lockdep_assert_held(&cgroup_mutex);
0143 
0144     if (attach_type != BPF_LSM_CGROUP)
0145         return to_cgroup_bpf_attach_type(attach_type);
0146 
0147     for (i = 0; i < ARRAY_SIZE(cgroup_lsm_atype); i++)
0148         if (cgroup_lsm_atype[i].attach_btf_id == attach_btf_id)
0149             return CGROUP_LSM_START + i;
0150 
0151     for (i = 0; i < ARRAY_SIZE(cgroup_lsm_atype); i++)
0152         if (cgroup_lsm_atype[i].attach_btf_id == 0)
0153             return CGROUP_LSM_START + i;
0154 
0155     return -E2BIG;
0156 
0157 }
0158 
0159 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype)
0160 {
0161     int i = cgroup_atype - CGROUP_LSM_START;
0162 
0163     lockdep_assert_held(&cgroup_mutex);
0164 
0165     WARN_ON_ONCE(cgroup_lsm_atype[i].attach_btf_id &&
0166              cgroup_lsm_atype[i].attach_btf_id != attach_btf_id);
0167 
0168     cgroup_lsm_atype[i].attach_btf_id = attach_btf_id;
0169     cgroup_lsm_atype[i].refcnt++;
0170 }
0171 
0172 void bpf_cgroup_atype_put(int cgroup_atype)
0173 {
0174     int i = cgroup_atype - CGROUP_LSM_START;
0175 
0176     mutex_lock(&cgroup_mutex);
0177     if (--cgroup_lsm_atype[i].refcnt <= 0)
0178         cgroup_lsm_atype[i].attach_btf_id = 0;
0179     WARN_ON_ONCE(cgroup_lsm_atype[i].refcnt < 0);
0180     mutex_unlock(&cgroup_mutex);
0181 }
0182 #else
0183 static enum cgroup_bpf_attach_type
0184 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id)
0185 {
0186     if (attach_type != BPF_LSM_CGROUP)
0187         return to_cgroup_bpf_attach_type(attach_type);
0188     return -EOPNOTSUPP;
0189 }
0190 #endif /* CONFIG_BPF_LSM */
0191 
0192 void cgroup_bpf_offline(struct cgroup *cgrp)
0193 {
0194     cgroup_get(cgrp);
0195     percpu_ref_kill(&cgrp->bpf.refcnt);
0196 }
0197 
0198 static void bpf_cgroup_storages_free(struct bpf_cgroup_storage *storages[])
0199 {
0200     enum bpf_cgroup_storage_type stype;
0201 
0202     for_each_cgroup_storage_type(stype)
0203         bpf_cgroup_storage_free(storages[stype]);
0204 }
0205 
0206 static int bpf_cgroup_storages_alloc(struct bpf_cgroup_storage *storages[],
0207                      struct bpf_cgroup_storage *new_storages[],
0208                      enum bpf_attach_type type,
0209                      struct bpf_prog *prog,
0210                      struct cgroup *cgrp)
0211 {
0212     enum bpf_cgroup_storage_type stype;
0213     struct bpf_cgroup_storage_key key;
0214     struct bpf_map *map;
0215 
0216     key.cgroup_inode_id = cgroup_id(cgrp);
0217     key.attach_type = type;
0218 
0219     for_each_cgroup_storage_type(stype) {
0220         map = prog->aux->cgroup_storage[stype];
0221         if (!map)
0222             continue;
0223 
0224         storages[stype] = cgroup_storage_lookup((void *)map, &key, false);
0225         if (storages[stype])
0226             continue;
0227 
0228         storages[stype] = bpf_cgroup_storage_alloc(prog, stype);
0229         if (IS_ERR(storages[stype])) {
0230             bpf_cgroup_storages_free(new_storages);
0231             return -ENOMEM;
0232         }
0233 
0234         new_storages[stype] = storages[stype];
0235     }
0236 
0237     return 0;
0238 }
0239 
0240 static void bpf_cgroup_storages_assign(struct bpf_cgroup_storage *dst[],
0241                        struct bpf_cgroup_storage *src[])
0242 {
0243     enum bpf_cgroup_storage_type stype;
0244 
0245     for_each_cgroup_storage_type(stype)
0246         dst[stype] = src[stype];
0247 }
0248 
0249 static void bpf_cgroup_storages_link(struct bpf_cgroup_storage *storages[],
0250                      struct cgroup *cgrp,
0251                      enum bpf_attach_type attach_type)
0252 {
0253     enum bpf_cgroup_storage_type stype;
0254 
0255     for_each_cgroup_storage_type(stype)
0256         bpf_cgroup_storage_link(storages[stype], cgrp, attach_type);
0257 }
0258 
0259 /* Called when bpf_cgroup_link is auto-detached from dying cgroup.
0260  * It drops cgroup and bpf_prog refcounts, and marks bpf_link as defunct. It
0261  * doesn't free link memory, which will eventually be done by bpf_link's
0262  * release() callback, when its last FD is closed.
0263  */
0264 static void bpf_cgroup_link_auto_detach(struct bpf_cgroup_link *link)
0265 {
0266     cgroup_put(link->cgroup);
0267     link->cgroup = NULL;
0268 }
0269 
0270 /**
0271  * cgroup_bpf_release() - put references of all bpf programs and
0272  *                        release all cgroup bpf data
0273  * @work: work structure embedded into the cgroup to modify
0274  */
0275 static void cgroup_bpf_release(struct work_struct *work)
0276 {
0277     struct cgroup *p, *cgrp = container_of(work, struct cgroup,
0278                            bpf.release_work);
0279     struct bpf_prog_array *old_array;
0280     struct list_head *storages = &cgrp->bpf.storages;
0281     struct bpf_cgroup_storage *storage, *stmp;
0282 
0283     unsigned int atype;
0284 
0285     mutex_lock(&cgroup_mutex);
0286 
0287     for (atype = 0; atype < ARRAY_SIZE(cgrp->bpf.progs); atype++) {
0288         struct hlist_head *progs = &cgrp->bpf.progs[atype];
0289         struct bpf_prog_list *pl;
0290         struct hlist_node *pltmp;
0291 
0292         hlist_for_each_entry_safe(pl, pltmp, progs, node) {
0293             hlist_del(&pl->node);
0294             if (pl->prog) {
0295                 if (pl->prog->expected_attach_type == BPF_LSM_CGROUP)
0296                     bpf_trampoline_unlink_cgroup_shim(pl->prog);
0297                 bpf_prog_put(pl->prog);
0298             }
0299             if (pl->link) {
0300                 if (pl->link->link.prog->expected_attach_type == BPF_LSM_CGROUP)
0301                     bpf_trampoline_unlink_cgroup_shim(pl->link->link.prog);
0302                 bpf_cgroup_link_auto_detach(pl->link);
0303             }
0304             kfree(pl);
0305             static_branch_dec(&cgroup_bpf_enabled_key[atype]);
0306         }
0307         old_array = rcu_dereference_protected(
0308                 cgrp->bpf.effective[atype],
0309                 lockdep_is_held(&cgroup_mutex));
0310         bpf_prog_array_free(old_array);
0311     }
0312 
0313     list_for_each_entry_safe(storage, stmp, storages, list_cg) {
0314         bpf_cgroup_storage_unlink(storage);
0315         bpf_cgroup_storage_free(storage);
0316     }
0317 
0318     mutex_unlock(&cgroup_mutex);
0319 
0320     for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
0321         cgroup_bpf_put(p);
0322 
0323     percpu_ref_exit(&cgrp->bpf.refcnt);
0324     cgroup_put(cgrp);
0325 }
0326 
0327 /**
0328  * cgroup_bpf_release_fn() - callback used to schedule releasing
0329  *                           of bpf cgroup data
0330  * @ref: percpu ref counter structure
0331  */
0332 static void cgroup_bpf_release_fn(struct percpu_ref *ref)
0333 {
0334     struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt);
0335 
0336     INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release);
0337     queue_work(system_wq, &cgrp->bpf.release_work);
0338 }
0339 
0340 /* Get underlying bpf_prog of bpf_prog_list entry, regardless if it's through
0341  * link or direct prog.
0342  */
0343 static struct bpf_prog *prog_list_prog(struct bpf_prog_list *pl)
0344 {
0345     if (pl->prog)
0346         return pl->prog;
0347     if (pl->link)
0348         return pl->link->link.prog;
0349     return NULL;
0350 }
0351 
0352 /* count number of elements in the list.
0353  * it's slow but the list cannot be long
0354  */
0355 static u32 prog_list_length(struct hlist_head *head)
0356 {
0357     struct bpf_prog_list *pl;
0358     u32 cnt = 0;
0359 
0360     hlist_for_each_entry(pl, head, node) {
0361         if (!prog_list_prog(pl))
0362             continue;
0363         cnt++;
0364     }
0365     return cnt;
0366 }
0367 
0368 /* if parent has non-overridable prog attached,
0369  * disallow attaching new programs to the descendent cgroup.
0370  * if parent has overridable or multi-prog, allow attaching
0371  */
0372 static bool hierarchy_allows_attach(struct cgroup *cgrp,
0373                     enum cgroup_bpf_attach_type atype)
0374 {
0375     struct cgroup *p;
0376 
0377     p = cgroup_parent(cgrp);
0378     if (!p)
0379         return true;
0380     do {
0381         u32 flags = p->bpf.flags[atype];
0382         u32 cnt;
0383 
0384         if (flags & BPF_F_ALLOW_MULTI)
0385             return true;
0386         cnt = prog_list_length(&p->bpf.progs[atype]);
0387         WARN_ON_ONCE(cnt > 1);
0388         if (cnt == 1)
0389             return !!(flags & BPF_F_ALLOW_OVERRIDE);
0390         p = cgroup_parent(p);
0391     } while (p);
0392     return true;
0393 }
0394 
0395 /* compute a chain of effective programs for a given cgroup:
0396  * start from the list of programs in this cgroup and add
0397  * all parent programs.
0398  * Note that parent's F_ALLOW_OVERRIDE-type program is yielding
0399  * to programs in this cgroup
0400  */
0401 static int compute_effective_progs(struct cgroup *cgrp,
0402                    enum cgroup_bpf_attach_type atype,
0403                    struct bpf_prog_array **array)
0404 {
0405     struct bpf_prog_array_item *item;
0406     struct bpf_prog_array *progs;
0407     struct bpf_prog_list *pl;
0408     struct cgroup *p = cgrp;
0409     int cnt = 0;
0410 
0411     /* count number of effective programs by walking parents */
0412     do {
0413         if (cnt == 0 || (p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
0414             cnt += prog_list_length(&p->bpf.progs[atype]);
0415         p = cgroup_parent(p);
0416     } while (p);
0417 
0418     progs = bpf_prog_array_alloc(cnt, GFP_KERNEL);
0419     if (!progs)
0420         return -ENOMEM;
0421 
0422     /* populate the array with effective progs */
0423     cnt = 0;
0424     p = cgrp;
0425     do {
0426         if (cnt > 0 && !(p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
0427             continue;
0428 
0429         hlist_for_each_entry(pl, &p->bpf.progs[atype], node) {
0430             if (!prog_list_prog(pl))
0431                 continue;
0432 
0433             item = &progs->items[cnt];
0434             item->prog = prog_list_prog(pl);
0435             bpf_cgroup_storages_assign(item->cgroup_storage,
0436                            pl->storage);
0437             cnt++;
0438         }
0439     } while ((p = cgroup_parent(p)));
0440 
0441     *array = progs;
0442     return 0;
0443 }
0444 
0445 static void activate_effective_progs(struct cgroup *cgrp,
0446                      enum cgroup_bpf_attach_type atype,
0447                      struct bpf_prog_array *old_array)
0448 {
0449     old_array = rcu_replace_pointer(cgrp->bpf.effective[atype], old_array,
0450                     lockdep_is_held(&cgroup_mutex));
0451     /* free prog array after grace period, since __cgroup_bpf_run_*()
0452      * might be still walking the array
0453      */
0454     bpf_prog_array_free(old_array);
0455 }
0456 
0457 /**
0458  * cgroup_bpf_inherit() - inherit effective programs from parent
0459  * @cgrp: the cgroup to modify
0460  */
0461 int cgroup_bpf_inherit(struct cgroup *cgrp)
0462 {
0463 /* has to use marco instead of const int, since compiler thinks
0464  * that array below is variable length
0465  */
0466 #define NR ARRAY_SIZE(cgrp->bpf.effective)
0467     struct bpf_prog_array *arrays[NR] = {};
0468     struct cgroup *p;
0469     int ret, i;
0470 
0471     ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0,
0472                   GFP_KERNEL);
0473     if (ret)
0474         return ret;
0475 
0476     for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
0477         cgroup_bpf_get(p);
0478 
0479     for (i = 0; i < NR; i++)
0480         INIT_HLIST_HEAD(&cgrp->bpf.progs[i]);
0481 
0482     INIT_LIST_HEAD(&cgrp->bpf.storages);
0483 
0484     for (i = 0; i < NR; i++)
0485         if (compute_effective_progs(cgrp, i, &arrays[i]))
0486             goto cleanup;
0487 
0488     for (i = 0; i < NR; i++)
0489         activate_effective_progs(cgrp, i, arrays[i]);
0490 
0491     return 0;
0492 cleanup:
0493     for (i = 0; i < NR; i++)
0494         bpf_prog_array_free(arrays[i]);
0495 
0496     for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
0497         cgroup_bpf_put(p);
0498 
0499     percpu_ref_exit(&cgrp->bpf.refcnt);
0500 
0501     return -ENOMEM;
0502 }
0503 
0504 static int update_effective_progs(struct cgroup *cgrp,
0505                   enum cgroup_bpf_attach_type atype)
0506 {
0507     struct cgroup_subsys_state *css;
0508     int err;
0509 
0510     /* allocate and recompute effective prog arrays */
0511     css_for_each_descendant_pre(css, &cgrp->self) {
0512         struct cgroup *desc = container_of(css, struct cgroup, self);
0513 
0514         if (percpu_ref_is_zero(&desc->bpf.refcnt))
0515             continue;
0516 
0517         err = compute_effective_progs(desc, atype, &desc->bpf.inactive);
0518         if (err)
0519             goto cleanup;
0520     }
0521 
0522     /* all allocations were successful. Activate all prog arrays */
0523     css_for_each_descendant_pre(css, &cgrp->self) {
0524         struct cgroup *desc = container_of(css, struct cgroup, self);
0525 
0526         if (percpu_ref_is_zero(&desc->bpf.refcnt)) {
0527             if (unlikely(desc->bpf.inactive)) {
0528                 bpf_prog_array_free(desc->bpf.inactive);
0529                 desc->bpf.inactive = NULL;
0530             }
0531             continue;
0532         }
0533 
0534         activate_effective_progs(desc, atype, desc->bpf.inactive);
0535         desc->bpf.inactive = NULL;
0536     }
0537 
0538     return 0;
0539 
0540 cleanup:
0541     /* oom while computing effective. Free all computed effective arrays
0542      * since they were not activated
0543      */
0544     css_for_each_descendant_pre(css, &cgrp->self) {
0545         struct cgroup *desc = container_of(css, struct cgroup, self);
0546 
0547         bpf_prog_array_free(desc->bpf.inactive);
0548         desc->bpf.inactive = NULL;
0549     }
0550 
0551     return err;
0552 }
0553 
0554 #define BPF_CGROUP_MAX_PROGS 64
0555 
0556 static struct bpf_prog_list *find_attach_entry(struct hlist_head *progs,
0557                            struct bpf_prog *prog,
0558                            struct bpf_cgroup_link *link,
0559                            struct bpf_prog *replace_prog,
0560                            bool allow_multi)
0561 {
0562     struct bpf_prog_list *pl;
0563 
0564     /* single-attach case */
0565     if (!allow_multi) {
0566         if (hlist_empty(progs))
0567             return NULL;
0568         return hlist_entry(progs->first, typeof(*pl), node);
0569     }
0570 
0571     hlist_for_each_entry(pl, progs, node) {
0572         if (prog && pl->prog == prog && prog != replace_prog)
0573             /* disallow attaching the same prog twice */
0574             return ERR_PTR(-EINVAL);
0575         if (link && pl->link == link)
0576             /* disallow attaching the same link twice */
0577             return ERR_PTR(-EINVAL);
0578     }
0579 
0580     /* direct prog multi-attach w/ replacement case */
0581     if (replace_prog) {
0582         hlist_for_each_entry(pl, progs, node) {
0583             if (pl->prog == replace_prog)
0584                 /* a match found */
0585                 return pl;
0586         }
0587         /* prog to replace not found for cgroup */
0588         return ERR_PTR(-ENOENT);
0589     }
0590 
0591     return NULL;
0592 }
0593 
0594 /**
0595  * __cgroup_bpf_attach() - Attach the program or the link to a cgroup, and
0596  *                         propagate the change to descendants
0597  * @cgrp: The cgroup which descendants to traverse
0598  * @prog: A program to attach
0599  * @link: A link to attach
0600  * @replace_prog: Previously attached program to replace if BPF_F_REPLACE is set
0601  * @type: Type of attach operation
0602  * @flags: Option flags
0603  *
0604  * Exactly one of @prog or @link can be non-null.
0605  * Must be called with cgroup_mutex held.
0606  */
0607 static int __cgroup_bpf_attach(struct cgroup *cgrp,
0608                    struct bpf_prog *prog, struct bpf_prog *replace_prog,
0609                    struct bpf_cgroup_link *link,
0610                    enum bpf_attach_type type, u32 flags)
0611 {
0612     u32 saved_flags = (flags & (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI));
0613     struct bpf_prog *old_prog = NULL;
0614     struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {};
0615     struct bpf_cgroup_storage *new_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {};
0616     struct bpf_prog *new_prog = prog ? : link->link.prog;
0617     enum cgroup_bpf_attach_type atype;
0618     struct bpf_prog_list *pl;
0619     struct hlist_head *progs;
0620     int err;
0621 
0622     if (((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI)) ||
0623         ((flags & BPF_F_REPLACE) && !(flags & BPF_F_ALLOW_MULTI)))
0624         /* invalid combination */
0625         return -EINVAL;
0626     if (link && (prog || replace_prog))
0627         /* only either link or prog/replace_prog can be specified */
0628         return -EINVAL;
0629     if (!!replace_prog != !!(flags & BPF_F_REPLACE))
0630         /* replace_prog implies BPF_F_REPLACE, and vice versa */
0631         return -EINVAL;
0632 
0633     atype = bpf_cgroup_atype_find(type, new_prog->aux->attach_btf_id);
0634     if (atype < 0)
0635         return -EINVAL;
0636 
0637     progs = &cgrp->bpf.progs[atype];
0638 
0639     if (!hierarchy_allows_attach(cgrp, atype))
0640         return -EPERM;
0641 
0642     if (!hlist_empty(progs) && cgrp->bpf.flags[atype] != saved_flags)
0643         /* Disallow attaching non-overridable on top
0644          * of existing overridable in this cgroup.
0645          * Disallow attaching multi-prog if overridable or none
0646          */
0647         return -EPERM;
0648 
0649     if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS)
0650         return -E2BIG;
0651 
0652     pl = find_attach_entry(progs, prog, link, replace_prog,
0653                    flags & BPF_F_ALLOW_MULTI);
0654     if (IS_ERR(pl))
0655         return PTR_ERR(pl);
0656 
0657     if (bpf_cgroup_storages_alloc(storage, new_storage, type,
0658                       prog ? : link->link.prog, cgrp))
0659         return -ENOMEM;
0660 
0661     if (pl) {
0662         old_prog = pl->prog;
0663     } else {
0664         struct hlist_node *last = NULL;
0665 
0666         pl = kmalloc(sizeof(*pl), GFP_KERNEL);
0667         if (!pl) {
0668             bpf_cgroup_storages_free(new_storage);
0669             return -ENOMEM;
0670         }
0671         if (hlist_empty(progs))
0672             hlist_add_head(&pl->node, progs);
0673         else
0674             hlist_for_each(last, progs) {
0675                 if (last->next)
0676                     continue;
0677                 hlist_add_behind(&pl->node, last);
0678                 break;
0679             }
0680     }
0681 
0682     pl->prog = prog;
0683     pl->link = link;
0684     bpf_cgroup_storages_assign(pl->storage, storage);
0685     cgrp->bpf.flags[atype] = saved_flags;
0686 
0687     if (type == BPF_LSM_CGROUP) {
0688         err = bpf_trampoline_link_cgroup_shim(new_prog, atype);
0689         if (err)
0690             goto cleanup;
0691     }
0692 
0693     err = update_effective_progs(cgrp, atype);
0694     if (err)
0695         goto cleanup_trampoline;
0696 
0697     if (old_prog) {
0698         if (type == BPF_LSM_CGROUP)
0699             bpf_trampoline_unlink_cgroup_shim(old_prog);
0700         bpf_prog_put(old_prog);
0701     } else {
0702         static_branch_inc(&cgroup_bpf_enabled_key[atype]);
0703     }
0704     bpf_cgroup_storages_link(new_storage, cgrp, type);
0705     return 0;
0706 
0707 cleanup_trampoline:
0708     if (type == BPF_LSM_CGROUP)
0709         bpf_trampoline_unlink_cgroup_shim(new_prog);
0710 
0711 cleanup:
0712     if (old_prog) {
0713         pl->prog = old_prog;
0714         pl->link = NULL;
0715     }
0716     bpf_cgroup_storages_free(new_storage);
0717     if (!old_prog) {
0718         hlist_del(&pl->node);
0719         kfree(pl);
0720     }
0721     return err;
0722 }
0723 
0724 static int cgroup_bpf_attach(struct cgroup *cgrp,
0725                  struct bpf_prog *prog, struct bpf_prog *replace_prog,
0726                  struct bpf_cgroup_link *link,
0727                  enum bpf_attach_type type,
0728                  u32 flags)
0729 {
0730     int ret;
0731 
0732     mutex_lock(&cgroup_mutex);
0733     ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
0734     mutex_unlock(&cgroup_mutex);
0735     return ret;
0736 }
0737 
0738 /* Swap updated BPF program for given link in effective program arrays across
0739  * all descendant cgroups. This function is guaranteed to succeed.
0740  */
0741 static void replace_effective_prog(struct cgroup *cgrp,
0742                    enum cgroup_bpf_attach_type atype,
0743                    struct bpf_cgroup_link *link)
0744 {
0745     struct bpf_prog_array_item *item;
0746     struct cgroup_subsys_state *css;
0747     struct bpf_prog_array *progs;
0748     struct bpf_prog_list *pl;
0749     struct hlist_head *head;
0750     struct cgroup *cg;
0751     int pos;
0752 
0753     css_for_each_descendant_pre(css, &cgrp->self) {
0754         struct cgroup *desc = container_of(css, struct cgroup, self);
0755 
0756         if (percpu_ref_is_zero(&desc->bpf.refcnt))
0757             continue;
0758 
0759         /* find position of link in effective progs array */
0760         for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) {
0761             if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
0762                 continue;
0763 
0764             head = &cg->bpf.progs[atype];
0765             hlist_for_each_entry(pl, head, node) {
0766                 if (!prog_list_prog(pl))
0767                     continue;
0768                 if (pl->link == link)
0769                     goto found;
0770                 pos++;
0771             }
0772         }
0773 found:
0774         BUG_ON(!cg);
0775         progs = rcu_dereference_protected(
0776                 desc->bpf.effective[atype],
0777                 lockdep_is_held(&cgroup_mutex));
0778         item = &progs->items[pos];
0779         WRITE_ONCE(item->prog, link->link.prog);
0780     }
0781 }
0782 
0783 /**
0784  * __cgroup_bpf_replace() - Replace link's program and propagate the change
0785  *                          to descendants
0786  * @cgrp: The cgroup which descendants to traverse
0787  * @link: A link for which to replace BPF program
0788  * @type: Type of attach operation
0789  *
0790  * Must be called with cgroup_mutex held.
0791  */
0792 static int __cgroup_bpf_replace(struct cgroup *cgrp,
0793                 struct bpf_cgroup_link *link,
0794                 struct bpf_prog *new_prog)
0795 {
0796     enum cgroup_bpf_attach_type atype;
0797     struct bpf_prog *old_prog;
0798     struct bpf_prog_list *pl;
0799     struct hlist_head *progs;
0800     bool found = false;
0801 
0802     atype = bpf_cgroup_atype_find(link->type, new_prog->aux->attach_btf_id);
0803     if (atype < 0)
0804         return -EINVAL;
0805 
0806     progs = &cgrp->bpf.progs[atype];
0807 
0808     if (link->link.prog->type != new_prog->type)
0809         return -EINVAL;
0810 
0811     hlist_for_each_entry(pl, progs, node) {
0812         if (pl->link == link) {
0813             found = true;
0814             break;
0815         }
0816     }
0817     if (!found)
0818         return -ENOENT;
0819 
0820     old_prog = xchg(&link->link.prog, new_prog);
0821     replace_effective_prog(cgrp, atype, link);
0822     bpf_prog_put(old_prog);
0823     return 0;
0824 }
0825 
0826 static int cgroup_bpf_replace(struct bpf_link *link, struct bpf_prog *new_prog,
0827                   struct bpf_prog *old_prog)
0828 {
0829     struct bpf_cgroup_link *cg_link;
0830     int ret;
0831 
0832     cg_link = container_of(link, struct bpf_cgroup_link, link);
0833 
0834     mutex_lock(&cgroup_mutex);
0835     /* link might have been auto-released by dying cgroup, so fail */
0836     if (!cg_link->cgroup) {
0837         ret = -ENOLINK;
0838         goto out_unlock;
0839     }
0840     if (old_prog && link->prog != old_prog) {
0841         ret = -EPERM;
0842         goto out_unlock;
0843     }
0844     ret = __cgroup_bpf_replace(cg_link->cgroup, cg_link, new_prog);
0845 out_unlock:
0846     mutex_unlock(&cgroup_mutex);
0847     return ret;
0848 }
0849 
0850 static struct bpf_prog_list *find_detach_entry(struct hlist_head *progs,
0851                            struct bpf_prog *prog,
0852                            struct bpf_cgroup_link *link,
0853                            bool allow_multi)
0854 {
0855     struct bpf_prog_list *pl;
0856 
0857     if (!allow_multi) {
0858         if (hlist_empty(progs))
0859             /* report error when trying to detach and nothing is attached */
0860             return ERR_PTR(-ENOENT);
0861 
0862         /* to maintain backward compatibility NONE and OVERRIDE cgroups
0863          * allow detaching with invalid FD (prog==NULL) in legacy mode
0864          */
0865         return hlist_entry(progs->first, typeof(*pl), node);
0866     }
0867 
0868     if (!prog && !link)
0869         /* to detach MULTI prog the user has to specify valid FD
0870          * of the program or link to be detached
0871          */
0872         return ERR_PTR(-EINVAL);
0873 
0874     /* find the prog or link and detach it */
0875     hlist_for_each_entry(pl, progs, node) {
0876         if (pl->prog == prog && pl->link == link)
0877             return pl;
0878     }
0879     return ERR_PTR(-ENOENT);
0880 }
0881 
0882 /**
0883  * purge_effective_progs() - After compute_effective_progs fails to alloc new
0884  *                           cgrp->bpf.inactive table we can recover by
0885  *                           recomputing the array in place.
0886  *
0887  * @cgrp: The cgroup which descendants to travers
0888  * @prog: A program to detach or NULL
0889  * @link: A link to detach or NULL
0890  * @atype: Type of detach operation
0891  */
0892 static void purge_effective_progs(struct cgroup *cgrp, struct bpf_prog *prog,
0893                   struct bpf_cgroup_link *link,
0894                   enum cgroup_bpf_attach_type atype)
0895 {
0896     struct cgroup_subsys_state *css;
0897     struct bpf_prog_array *progs;
0898     struct bpf_prog_list *pl;
0899     struct hlist_head *head;
0900     struct cgroup *cg;
0901     int pos;
0902 
0903     /* recompute effective prog array in place */
0904     css_for_each_descendant_pre(css, &cgrp->self) {
0905         struct cgroup *desc = container_of(css, struct cgroup, self);
0906 
0907         if (percpu_ref_is_zero(&desc->bpf.refcnt))
0908             continue;
0909 
0910         /* find position of link or prog in effective progs array */
0911         for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) {
0912             if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
0913                 continue;
0914 
0915             head = &cg->bpf.progs[atype];
0916             hlist_for_each_entry(pl, head, node) {
0917                 if (!prog_list_prog(pl))
0918                     continue;
0919                 if (pl->prog == prog && pl->link == link)
0920                     goto found;
0921                 pos++;
0922             }
0923         }
0924 
0925         /* no link or prog match, skip the cgroup of this layer */
0926         continue;
0927 found:
0928         progs = rcu_dereference_protected(
0929                 desc->bpf.effective[atype],
0930                 lockdep_is_held(&cgroup_mutex));
0931 
0932         /* Remove the program from the array */
0933         WARN_ONCE(bpf_prog_array_delete_safe_at(progs, pos),
0934               "Failed to purge a prog from array at index %d", pos);
0935     }
0936 }
0937 
0938 /**
0939  * __cgroup_bpf_detach() - Detach the program or link from a cgroup, and
0940  *                         propagate the change to descendants
0941  * @cgrp: The cgroup which descendants to traverse
0942  * @prog: A program to detach or NULL
0943  * @link: A link to detach or NULL
0944  * @type: Type of detach operation
0945  *
0946  * At most one of @prog or @link can be non-NULL.
0947  * Must be called with cgroup_mutex held.
0948  */
0949 static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
0950                    struct bpf_cgroup_link *link, enum bpf_attach_type type)
0951 {
0952     enum cgroup_bpf_attach_type atype;
0953     struct bpf_prog *old_prog;
0954     struct bpf_prog_list *pl;
0955     struct hlist_head *progs;
0956     u32 attach_btf_id = 0;
0957     u32 flags;
0958 
0959     if (prog)
0960         attach_btf_id = prog->aux->attach_btf_id;
0961     if (link)
0962         attach_btf_id = link->link.prog->aux->attach_btf_id;
0963 
0964     atype = bpf_cgroup_atype_find(type, attach_btf_id);
0965     if (atype < 0)
0966         return -EINVAL;
0967 
0968     progs = &cgrp->bpf.progs[atype];
0969     flags = cgrp->bpf.flags[atype];
0970 
0971     if (prog && link)
0972         /* only one of prog or link can be specified */
0973         return -EINVAL;
0974 
0975     pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI);
0976     if (IS_ERR(pl))
0977         return PTR_ERR(pl);
0978 
0979     /* mark it deleted, so it's ignored while recomputing effective */
0980     old_prog = pl->prog;
0981     pl->prog = NULL;
0982     pl->link = NULL;
0983 
0984     if (update_effective_progs(cgrp, atype)) {
0985         /* if update effective array failed replace the prog with a dummy prog*/
0986         pl->prog = old_prog;
0987         pl->link = link;
0988         purge_effective_progs(cgrp, old_prog, link, atype);
0989     }
0990 
0991     /* now can actually delete it from this cgroup list */
0992     hlist_del(&pl->node);
0993 
0994     kfree(pl);
0995     if (hlist_empty(progs))
0996         /* last program was detached, reset flags to zero */
0997         cgrp->bpf.flags[atype] = 0;
0998     if (old_prog) {
0999         if (type == BPF_LSM_CGROUP)
1000             bpf_trampoline_unlink_cgroup_shim(old_prog);
1001         bpf_prog_put(old_prog);
1002     }
1003     static_branch_dec(&cgroup_bpf_enabled_key[atype]);
1004     return 0;
1005 }
1006 
1007 static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
1008                  enum bpf_attach_type type)
1009 {
1010     int ret;
1011 
1012     mutex_lock(&cgroup_mutex);
1013     ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
1014     mutex_unlock(&cgroup_mutex);
1015     return ret;
1016 }
1017 
1018 /* Must be called with cgroup_mutex held to avoid races. */
1019 static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1020                   union bpf_attr __user *uattr)
1021 {
1022     __u32 __user *prog_attach_flags = u64_to_user_ptr(attr->query.prog_attach_flags);
1023     __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1024     enum bpf_attach_type type = attr->query.attach_type;
1025     enum cgroup_bpf_attach_type from_atype, to_atype;
1026     enum cgroup_bpf_attach_type atype;
1027     struct bpf_prog_array *effective;
1028     int cnt, ret = 0, i;
1029     int total_cnt = 0;
1030     u32 flags;
1031 
1032     if (type == BPF_LSM_CGROUP) {
1033         if (attr->query.prog_cnt && prog_ids && !prog_attach_flags)
1034             return -EINVAL;
1035 
1036         from_atype = CGROUP_LSM_START;
1037         to_atype = CGROUP_LSM_END;
1038         flags = 0;
1039     } else {
1040         from_atype = to_cgroup_bpf_attach_type(type);
1041         if (from_atype < 0)
1042             return -EINVAL;
1043         to_atype = from_atype;
1044         flags = cgrp->bpf.flags[from_atype];
1045     }
1046 
1047     for (atype = from_atype; atype <= to_atype; atype++) {
1048         if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) {
1049             effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
1050                                   lockdep_is_held(&cgroup_mutex));
1051             total_cnt += bpf_prog_array_length(effective);
1052         } else {
1053             total_cnt += prog_list_length(&cgrp->bpf.progs[atype]);
1054         }
1055     }
1056 
1057     if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
1058         return -EFAULT;
1059     if (copy_to_user(&uattr->query.prog_cnt, &total_cnt, sizeof(total_cnt)))
1060         return -EFAULT;
1061     if (attr->query.prog_cnt == 0 || !prog_ids || !total_cnt)
1062         /* return early if user requested only program count + flags */
1063         return 0;
1064 
1065     if (attr->query.prog_cnt < total_cnt) {
1066         total_cnt = attr->query.prog_cnt;
1067         ret = -ENOSPC;
1068     }
1069 
1070     for (atype = from_atype; atype <= to_atype && total_cnt; atype++) {
1071         if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) {
1072             effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
1073                                   lockdep_is_held(&cgroup_mutex));
1074             cnt = min_t(int, bpf_prog_array_length(effective), total_cnt);
1075             ret = bpf_prog_array_copy_to_user(effective, prog_ids, cnt);
1076         } else {
1077             struct hlist_head *progs;
1078             struct bpf_prog_list *pl;
1079             struct bpf_prog *prog;
1080             u32 id;
1081 
1082             progs = &cgrp->bpf.progs[atype];
1083             cnt = min_t(int, prog_list_length(progs), total_cnt);
1084             i = 0;
1085             hlist_for_each_entry(pl, progs, node) {
1086                 prog = prog_list_prog(pl);
1087                 id = prog->aux->id;
1088                 if (copy_to_user(prog_ids + i, &id, sizeof(id)))
1089                     return -EFAULT;
1090                 if (++i == cnt)
1091                     break;
1092             }
1093         }
1094 
1095         if (prog_attach_flags) {
1096             flags = cgrp->bpf.flags[atype];
1097 
1098             for (i = 0; i < cnt; i++)
1099                 if (copy_to_user(prog_attach_flags + i, &flags, sizeof(flags)))
1100                     return -EFAULT;
1101             prog_attach_flags += cnt;
1102         }
1103 
1104         prog_ids += cnt;
1105         total_cnt -= cnt;
1106     }
1107     return ret;
1108 }
1109 
1110 static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1111                 union bpf_attr __user *uattr)
1112 {
1113     int ret;
1114 
1115     mutex_lock(&cgroup_mutex);
1116     ret = __cgroup_bpf_query(cgrp, attr, uattr);
1117     mutex_unlock(&cgroup_mutex);
1118     return ret;
1119 }
1120 
1121 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
1122                enum bpf_prog_type ptype, struct bpf_prog *prog)
1123 {
1124     struct bpf_prog *replace_prog = NULL;
1125     struct cgroup *cgrp;
1126     int ret;
1127 
1128     cgrp = cgroup_get_from_fd(attr->target_fd);
1129     if (IS_ERR(cgrp))
1130         return PTR_ERR(cgrp);
1131 
1132     if ((attr->attach_flags & BPF_F_ALLOW_MULTI) &&
1133         (attr->attach_flags & BPF_F_REPLACE)) {
1134         replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype);
1135         if (IS_ERR(replace_prog)) {
1136             cgroup_put(cgrp);
1137             return PTR_ERR(replace_prog);
1138         }
1139     }
1140 
1141     ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL,
1142                 attr->attach_type, attr->attach_flags);
1143 
1144     if (replace_prog)
1145         bpf_prog_put(replace_prog);
1146     cgroup_put(cgrp);
1147     return ret;
1148 }
1149 
1150 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
1151 {
1152     struct bpf_prog *prog;
1153     struct cgroup *cgrp;
1154     int ret;
1155 
1156     cgrp = cgroup_get_from_fd(attr->target_fd);
1157     if (IS_ERR(cgrp))
1158         return PTR_ERR(cgrp);
1159 
1160     prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1161     if (IS_ERR(prog))
1162         prog = NULL;
1163 
1164     ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type);
1165     if (prog)
1166         bpf_prog_put(prog);
1167 
1168     cgroup_put(cgrp);
1169     return ret;
1170 }
1171 
1172 static void bpf_cgroup_link_release(struct bpf_link *link)
1173 {
1174     struct bpf_cgroup_link *cg_link =
1175         container_of(link, struct bpf_cgroup_link, link);
1176     struct cgroup *cg;
1177 
1178     /* link might have been auto-detached by dying cgroup already,
1179      * in that case our work is done here
1180      */
1181     if (!cg_link->cgroup)
1182         return;
1183 
1184     mutex_lock(&cgroup_mutex);
1185 
1186     /* re-check cgroup under lock again */
1187     if (!cg_link->cgroup) {
1188         mutex_unlock(&cgroup_mutex);
1189         return;
1190     }
1191 
1192     WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link,
1193                     cg_link->type));
1194     if (cg_link->type == BPF_LSM_CGROUP)
1195         bpf_trampoline_unlink_cgroup_shim(cg_link->link.prog);
1196 
1197     cg = cg_link->cgroup;
1198     cg_link->cgroup = NULL;
1199 
1200     mutex_unlock(&cgroup_mutex);
1201 
1202     cgroup_put(cg);
1203 }
1204 
1205 static void bpf_cgroup_link_dealloc(struct bpf_link *link)
1206 {
1207     struct bpf_cgroup_link *cg_link =
1208         container_of(link, struct bpf_cgroup_link, link);
1209 
1210     kfree(cg_link);
1211 }
1212 
1213 static int bpf_cgroup_link_detach(struct bpf_link *link)
1214 {
1215     bpf_cgroup_link_release(link);
1216 
1217     return 0;
1218 }
1219 
1220 static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link,
1221                     struct seq_file *seq)
1222 {
1223     struct bpf_cgroup_link *cg_link =
1224         container_of(link, struct bpf_cgroup_link, link);
1225     u64 cg_id = 0;
1226 
1227     mutex_lock(&cgroup_mutex);
1228     if (cg_link->cgroup)
1229         cg_id = cgroup_id(cg_link->cgroup);
1230     mutex_unlock(&cgroup_mutex);
1231 
1232     seq_printf(seq,
1233            "cgroup_id:\t%llu\n"
1234            "attach_type:\t%d\n",
1235            cg_id,
1236            cg_link->type);
1237 }
1238 
1239 static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link,
1240                       struct bpf_link_info *info)
1241 {
1242     struct bpf_cgroup_link *cg_link =
1243         container_of(link, struct bpf_cgroup_link, link);
1244     u64 cg_id = 0;
1245 
1246     mutex_lock(&cgroup_mutex);
1247     if (cg_link->cgroup)
1248         cg_id = cgroup_id(cg_link->cgroup);
1249     mutex_unlock(&cgroup_mutex);
1250 
1251     info->cgroup.cgroup_id = cg_id;
1252     info->cgroup.attach_type = cg_link->type;
1253     return 0;
1254 }
1255 
1256 static const struct bpf_link_ops bpf_cgroup_link_lops = {
1257     .release = bpf_cgroup_link_release,
1258     .dealloc = bpf_cgroup_link_dealloc,
1259     .detach = bpf_cgroup_link_detach,
1260     .update_prog = cgroup_bpf_replace,
1261     .show_fdinfo = bpf_cgroup_link_show_fdinfo,
1262     .fill_link_info = bpf_cgroup_link_fill_link_info,
1263 };
1264 
1265 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
1266 {
1267     struct bpf_link_primer link_primer;
1268     struct bpf_cgroup_link *link;
1269     struct cgroup *cgrp;
1270     int err;
1271 
1272     if (attr->link_create.flags)
1273         return -EINVAL;
1274 
1275     cgrp = cgroup_get_from_fd(attr->link_create.target_fd);
1276     if (IS_ERR(cgrp))
1277         return PTR_ERR(cgrp);
1278 
1279     link = kzalloc(sizeof(*link), GFP_USER);
1280     if (!link) {
1281         err = -ENOMEM;
1282         goto out_put_cgroup;
1283     }
1284     bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops,
1285               prog);
1286     link->cgroup = cgrp;
1287     link->type = attr->link_create.attach_type;
1288 
1289     err = bpf_link_prime(&link->link, &link_primer);
1290     if (err) {
1291         kfree(link);
1292         goto out_put_cgroup;
1293     }
1294 
1295     err = cgroup_bpf_attach(cgrp, NULL, NULL, link,
1296                 link->type, BPF_F_ALLOW_MULTI);
1297     if (err) {
1298         bpf_link_cleanup(&link_primer);
1299         goto out_put_cgroup;
1300     }
1301 
1302     return bpf_link_settle(&link_primer);
1303 
1304 out_put_cgroup:
1305     cgroup_put(cgrp);
1306     return err;
1307 }
1308 
1309 int cgroup_bpf_prog_query(const union bpf_attr *attr,
1310               union bpf_attr __user *uattr)
1311 {
1312     struct cgroup *cgrp;
1313     int ret;
1314 
1315     cgrp = cgroup_get_from_fd(attr->query.target_fd);
1316     if (IS_ERR(cgrp))
1317         return PTR_ERR(cgrp);
1318 
1319     ret = cgroup_bpf_query(cgrp, attr, uattr);
1320 
1321     cgroup_put(cgrp);
1322     return ret;
1323 }
1324 
1325 /**
1326  * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering
1327  * @sk: The socket sending or receiving traffic
1328  * @skb: The skb that is being sent or received
1329  * @type: The type of program to be executed
1330  *
1331  * If no socket is passed, or the socket is not of type INET or INET6,
1332  * this function does nothing and returns 0.
1333  *
1334  * The program type passed in via @type must be suitable for network
1335  * filtering. No further check is performed to assert that.
1336  *
1337  * For egress packets, this function can return:
1338  *   NET_XMIT_SUCCESS    (0)    - continue with packet output
1339  *   NET_XMIT_DROP       (1)    - drop packet and notify TCP to call cwr
1340  *   NET_XMIT_CN         (2)    - continue with packet output and notify TCP
1341  *                to call cwr
1342  *   -err           - drop packet
1343  *
1344  * For ingress packets, this function will return -EPERM if any
1345  * attached program was found and if it returned != 1 during execution.
1346  * Otherwise 0 is returned.
1347  */
1348 int __cgroup_bpf_run_filter_skb(struct sock *sk,
1349                 struct sk_buff *skb,
1350                 enum cgroup_bpf_attach_type atype)
1351 {
1352     unsigned int offset = skb->data - skb_network_header(skb);
1353     struct sock *save_sk;
1354     void *saved_data_end;
1355     struct cgroup *cgrp;
1356     int ret;
1357 
1358     if (!sk || !sk_fullsock(sk))
1359         return 0;
1360 
1361     if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1362         return 0;
1363 
1364     cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1365     save_sk = skb->sk;
1366     skb->sk = sk;
1367     __skb_push(skb, offset);
1368 
1369     /* compute pointers for the bpf prog */
1370     bpf_compute_and_save_data_end(skb, &saved_data_end);
1371 
1372     if (atype == CGROUP_INET_EGRESS) {
1373         u32 flags = 0;
1374         bool cn;
1375 
1376         ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, skb,
1377                         __bpf_prog_run_save_cb, 0, &flags);
1378 
1379         /* Return values of CGROUP EGRESS BPF programs are:
1380          *   0: drop packet
1381          *   1: keep packet
1382          *   2: drop packet and cn
1383          *   3: keep packet and cn
1384          *
1385          * The returned value is then converted to one of the NET_XMIT
1386          * or an error code that is then interpreted as drop packet
1387          * (and no cn):
1388          *   0: NET_XMIT_SUCCESS  skb should be transmitted
1389          *   1: NET_XMIT_DROP     skb should be dropped and cn
1390          *   2: NET_XMIT_CN       skb should be transmitted and cn
1391          *   3: -err              skb should be dropped
1392          */
1393 
1394         cn = flags & BPF_RET_SET_CN;
1395         if (ret && !IS_ERR_VALUE((long)ret))
1396             ret = -EFAULT;
1397         if (!ret)
1398             ret = (cn ? NET_XMIT_CN : NET_XMIT_SUCCESS);
1399         else
1400             ret = (cn ? NET_XMIT_DROP : ret);
1401     } else {
1402         ret = bpf_prog_run_array_cg(&cgrp->bpf, atype,
1403                         skb, __bpf_prog_run_save_cb, 0,
1404                         NULL);
1405         if (ret && !IS_ERR_VALUE((long)ret))
1406             ret = -EFAULT;
1407     }
1408     bpf_restore_data_end(skb, saved_data_end);
1409     __skb_pull(skb, offset);
1410     skb->sk = save_sk;
1411 
1412     return ret;
1413 }
1414 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
1415 
1416 /**
1417  * __cgroup_bpf_run_filter_sk() - Run a program on a sock
1418  * @sk: sock structure to manipulate
1419  * @type: The type of program to be executed
1420  *
1421  * socket is passed is expected to be of type INET or INET6.
1422  *
1423  * The program type passed in via @type must be suitable for sock
1424  * filtering. No further check is performed to assert that.
1425  *
1426  * This function will return %-EPERM if any if an attached program was found
1427  * and if it returned != 1 during execution. In all other cases, 0 is returned.
1428  */
1429 int __cgroup_bpf_run_filter_sk(struct sock *sk,
1430                    enum cgroup_bpf_attach_type atype)
1431 {
1432     struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1433 
1434     return bpf_prog_run_array_cg(&cgrp->bpf, atype, sk, bpf_prog_run, 0,
1435                      NULL);
1436 }
1437 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
1438 
1439 /**
1440  * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and
1441  *                                       provided by user sockaddr
1442  * @sk: sock struct that will use sockaddr
1443  * @uaddr: sockaddr struct provided by user
1444  * @type: The type of program to be executed
1445  * @t_ctx: Pointer to attach type specific context
1446  * @flags: Pointer to u32 which contains higher bits of BPF program
1447  *         return value (OR'ed together).
1448  *
1449  * socket is expected to be of type INET or INET6.
1450  *
1451  * This function will return %-EPERM if an attached program is found and
1452  * returned value != 1 during execution. In all other cases, 0 is returned.
1453  */
1454 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
1455                       struct sockaddr *uaddr,
1456                       enum cgroup_bpf_attach_type atype,
1457                       void *t_ctx,
1458                       u32 *flags)
1459 {
1460     struct bpf_sock_addr_kern ctx = {
1461         .sk = sk,
1462         .uaddr = uaddr,
1463         .t_ctx = t_ctx,
1464     };
1465     struct sockaddr_storage unspec;
1466     struct cgroup *cgrp;
1467 
1468     /* Check socket family since not all sockets represent network
1469      * endpoint (e.g. AF_UNIX).
1470      */
1471     if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1472         return 0;
1473 
1474     if (!ctx.uaddr) {
1475         memset(&unspec, 0, sizeof(unspec));
1476         ctx.uaddr = (struct sockaddr *)&unspec;
1477     }
1478 
1479     cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1480     return bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run,
1481                      0, flags);
1482 }
1483 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
1484 
1485 /**
1486  * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock
1487  * @sk: socket to get cgroup from
1488  * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains
1489  * sk with connection information (IP addresses, etc.) May not contain
1490  * cgroup info if it is a req sock.
1491  * @type: The type of program to be executed
1492  *
1493  * socket passed is expected to be of type INET or INET6.
1494  *
1495  * The program type passed in via @type must be suitable for sock_ops
1496  * filtering. No further check is performed to assert that.
1497  *
1498  * This function will return %-EPERM if any if an attached program was found
1499  * and if it returned != 1 during execution. In all other cases, 0 is returned.
1500  */
1501 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
1502                      struct bpf_sock_ops_kern *sock_ops,
1503                      enum cgroup_bpf_attach_type atype)
1504 {
1505     struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1506 
1507     return bpf_prog_run_array_cg(&cgrp->bpf, atype, sock_ops, bpf_prog_run,
1508                      0, NULL);
1509 }
1510 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
1511 
1512 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
1513                       short access, enum cgroup_bpf_attach_type atype)
1514 {
1515     struct cgroup *cgrp;
1516     struct bpf_cgroup_dev_ctx ctx = {
1517         .access_type = (access << 16) | dev_type,
1518         .major = major,
1519         .minor = minor,
1520     };
1521     int ret;
1522 
1523     rcu_read_lock();
1524     cgrp = task_dfl_cgroup(current);
1525     ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1526                     NULL);
1527     rcu_read_unlock();
1528 
1529     return ret;
1530 }
1531 
1532 BPF_CALL_0(bpf_get_retval)
1533 {
1534     struct bpf_cg_run_ctx *ctx =
1535         container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1536 
1537     return ctx->retval;
1538 }
1539 
1540 const struct bpf_func_proto bpf_get_retval_proto = {
1541     .func       = bpf_get_retval,
1542     .gpl_only   = false,
1543     .ret_type   = RET_INTEGER,
1544 };
1545 
1546 BPF_CALL_1(bpf_set_retval, int, retval)
1547 {
1548     struct bpf_cg_run_ctx *ctx =
1549         container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1550 
1551     ctx->retval = retval;
1552     return 0;
1553 }
1554 
1555 const struct bpf_func_proto bpf_set_retval_proto = {
1556     .func       = bpf_set_retval,
1557     .gpl_only   = false,
1558     .ret_type   = RET_INTEGER,
1559     .arg1_type  = ARG_ANYTHING,
1560 };
1561 
1562 static const struct bpf_func_proto *
1563 cgroup_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1564 {
1565     switch (func_id) {
1566     case BPF_FUNC_get_current_uid_gid:
1567         return &bpf_get_current_uid_gid_proto;
1568     case BPF_FUNC_get_local_storage:
1569         return &bpf_get_local_storage_proto;
1570     case BPF_FUNC_get_current_cgroup_id:
1571         return &bpf_get_current_cgroup_id_proto;
1572     case BPF_FUNC_perf_event_output:
1573         return &bpf_event_output_data_proto;
1574     case BPF_FUNC_get_retval:
1575         return &bpf_get_retval_proto;
1576     case BPF_FUNC_set_retval:
1577         return &bpf_set_retval_proto;
1578     default:
1579         return bpf_base_func_proto(func_id);
1580     }
1581 }
1582 
1583 static const struct bpf_func_proto *
1584 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1585 {
1586     return cgroup_base_func_proto(func_id, prog);
1587 }
1588 
1589 static bool cgroup_dev_is_valid_access(int off, int size,
1590                        enum bpf_access_type type,
1591                        const struct bpf_prog *prog,
1592                        struct bpf_insn_access_aux *info)
1593 {
1594     const int size_default = sizeof(__u32);
1595 
1596     if (type == BPF_WRITE)
1597         return false;
1598 
1599     if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
1600         return false;
1601     /* The verifier guarantees that size > 0. */
1602     if (off % size != 0)
1603         return false;
1604 
1605     switch (off) {
1606     case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type):
1607         bpf_ctx_record_field_size(info, size_default);
1608         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
1609             return false;
1610         break;
1611     default:
1612         if (size != size_default)
1613             return false;
1614     }
1615 
1616     return true;
1617 }
1618 
1619 const struct bpf_prog_ops cg_dev_prog_ops = {
1620 };
1621 
1622 const struct bpf_verifier_ops cg_dev_verifier_ops = {
1623     .get_func_proto     = cgroup_dev_func_proto,
1624     .is_valid_access    = cgroup_dev_is_valid_access,
1625 };
1626 
1627 /**
1628  * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl
1629  *
1630  * @head: sysctl table header
1631  * @table: sysctl table
1632  * @write: sysctl is being read (= 0) or written (= 1)
1633  * @buf: pointer to buffer (in and out)
1634  * @pcount: value-result argument: value is size of buffer pointed to by @buf,
1635  *  result is size of @new_buf if program set new value, initial value
1636  *  otherwise
1637  * @ppos: value-result argument: value is position at which read from or write
1638  *  to sysctl is happening, result is new position if program overrode it,
1639  *  initial value otherwise
1640  * @type: type of program to be executed
1641  *
1642  * Program is run when sysctl is being accessed, either read or written, and
1643  * can allow or deny such access.
1644  *
1645  * This function will return %-EPERM if an attached program is found and
1646  * returned value != 1 during execution. In all other cases 0 is returned.
1647  */
1648 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
1649                    struct ctl_table *table, int write,
1650                    char **buf, size_t *pcount, loff_t *ppos,
1651                    enum cgroup_bpf_attach_type atype)
1652 {
1653     struct bpf_sysctl_kern ctx = {
1654         .head = head,
1655         .table = table,
1656         .write = write,
1657         .ppos = ppos,
1658         .cur_val = NULL,
1659         .cur_len = PAGE_SIZE,
1660         .new_val = NULL,
1661         .new_len = 0,
1662         .new_updated = 0,
1663     };
1664     struct cgroup *cgrp;
1665     loff_t pos = 0;
1666     int ret;
1667 
1668     ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL);
1669     if (!ctx.cur_val ||
1670         table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) {
1671         /* Let BPF program decide how to proceed. */
1672         ctx.cur_len = 0;
1673     }
1674 
1675     if (write && *buf && *pcount) {
1676         /* BPF program should be able to override new value with a
1677          * buffer bigger than provided by user.
1678          */
1679         ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL);
1680         ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount);
1681         if (ctx.new_val) {
1682             memcpy(ctx.new_val, *buf, ctx.new_len);
1683         } else {
1684             /* Let BPF program decide how to proceed. */
1685             ctx.new_len = 0;
1686         }
1687     }
1688 
1689     rcu_read_lock();
1690     cgrp = task_dfl_cgroup(current);
1691     ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1692                     NULL);
1693     rcu_read_unlock();
1694 
1695     kfree(ctx.cur_val);
1696 
1697     if (ret == 1 && ctx.new_updated) {
1698         kfree(*buf);
1699         *buf = ctx.new_val;
1700         *pcount = ctx.new_len;
1701     } else {
1702         kfree(ctx.new_val);
1703     }
1704 
1705     return ret;
1706 }
1707 
1708 #ifdef CONFIG_NET
1709 static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen,
1710                  struct bpf_sockopt_buf *buf)
1711 {
1712     if (unlikely(max_optlen < 0))
1713         return -EINVAL;
1714 
1715     if (unlikely(max_optlen > PAGE_SIZE)) {
1716         /* We don't expose optvals that are greater than PAGE_SIZE
1717          * to the BPF program.
1718          */
1719         max_optlen = PAGE_SIZE;
1720     }
1721 
1722     if (max_optlen <= sizeof(buf->data)) {
1723         /* When the optval fits into BPF_SOCKOPT_KERN_BUF_SIZE
1724          * bytes avoid the cost of kzalloc.
1725          */
1726         ctx->optval = buf->data;
1727         ctx->optval_end = ctx->optval + max_optlen;
1728         return max_optlen;
1729     }
1730 
1731     ctx->optval = kzalloc(max_optlen, GFP_USER);
1732     if (!ctx->optval)
1733         return -ENOMEM;
1734 
1735     ctx->optval_end = ctx->optval + max_optlen;
1736 
1737     return max_optlen;
1738 }
1739 
1740 static void sockopt_free_buf(struct bpf_sockopt_kern *ctx,
1741                  struct bpf_sockopt_buf *buf)
1742 {
1743     if (ctx->optval == buf->data)
1744         return;
1745     kfree(ctx->optval);
1746 }
1747 
1748 static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx,
1749                   struct bpf_sockopt_buf *buf)
1750 {
1751     return ctx->optval != buf->data;
1752 }
1753 
1754 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
1755                        int *optname, char __user *optval,
1756                        int *optlen, char **kernel_optval)
1757 {
1758     struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1759     struct bpf_sockopt_buf buf = {};
1760     struct bpf_sockopt_kern ctx = {
1761         .sk = sk,
1762         .level = *level,
1763         .optname = *optname,
1764     };
1765     int ret, max_optlen;
1766 
1767     /* Allocate a bit more than the initial user buffer for
1768      * BPF program. The canonical use case is overriding
1769      * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic).
1770      */
1771     max_optlen = max_t(int, 16, *optlen);
1772     max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1773     if (max_optlen < 0)
1774         return max_optlen;
1775 
1776     ctx.optlen = *optlen;
1777 
1778     if (copy_from_user(ctx.optval, optval, min(*optlen, max_optlen)) != 0) {
1779         ret = -EFAULT;
1780         goto out;
1781     }
1782 
1783     lock_sock(sk);
1784     ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_SETSOCKOPT,
1785                     &ctx, bpf_prog_run, 0, NULL);
1786     release_sock(sk);
1787 
1788     if (ret)
1789         goto out;
1790 
1791     if (ctx.optlen == -1) {
1792         /* optlen set to -1, bypass kernel */
1793         ret = 1;
1794     } else if (ctx.optlen > max_optlen || ctx.optlen < -1) {
1795         /* optlen is out of bounds */
1796         ret = -EFAULT;
1797     } else {
1798         /* optlen within bounds, run kernel handler */
1799         ret = 0;
1800 
1801         /* export any potential modifications */
1802         *level = ctx.level;
1803         *optname = ctx.optname;
1804 
1805         /* optlen == 0 from BPF indicates that we should
1806          * use original userspace data.
1807          */
1808         if (ctx.optlen != 0) {
1809             *optlen = ctx.optlen;
1810             /* We've used bpf_sockopt_kern->buf as an intermediary
1811              * storage, but the BPF program indicates that we need
1812              * to pass this data to the kernel setsockopt handler.
1813              * No way to export on-stack buf, have to allocate a
1814              * new buffer.
1815              */
1816             if (!sockopt_buf_allocated(&ctx, &buf)) {
1817                 void *p = kmalloc(ctx.optlen, GFP_USER);
1818 
1819                 if (!p) {
1820                     ret = -ENOMEM;
1821                     goto out;
1822                 }
1823                 memcpy(p, ctx.optval, ctx.optlen);
1824                 *kernel_optval = p;
1825             } else {
1826                 *kernel_optval = ctx.optval;
1827             }
1828             /* export and don't free sockopt buf */
1829             return 0;
1830         }
1831     }
1832 
1833 out:
1834     sockopt_free_buf(&ctx, &buf);
1835     return ret;
1836 }
1837 
1838 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
1839                        int optname, char __user *optval,
1840                        int __user *optlen, int max_optlen,
1841                        int retval)
1842 {
1843     struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1844     struct bpf_sockopt_buf buf = {};
1845     struct bpf_sockopt_kern ctx = {
1846         .sk = sk,
1847         .level = level,
1848         .optname = optname,
1849         .current_task = current,
1850     };
1851     int ret;
1852 
1853     ctx.optlen = max_optlen;
1854     max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1855     if (max_optlen < 0)
1856         return max_optlen;
1857 
1858     if (!retval) {
1859         /* If kernel getsockopt finished successfully,
1860          * copy whatever was returned to the user back
1861          * into our temporary buffer. Set optlen to the
1862          * one that kernel returned as well to let
1863          * BPF programs inspect the value.
1864          */
1865 
1866         if (get_user(ctx.optlen, optlen)) {
1867             ret = -EFAULT;
1868             goto out;
1869         }
1870 
1871         if (ctx.optlen < 0) {
1872             ret = -EFAULT;
1873             goto out;
1874         }
1875 
1876         if (copy_from_user(ctx.optval, optval,
1877                    min(ctx.optlen, max_optlen)) != 0) {
1878             ret = -EFAULT;
1879             goto out;
1880         }
1881     }
1882 
1883     lock_sock(sk);
1884     ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1885                     &ctx, bpf_prog_run, retval, NULL);
1886     release_sock(sk);
1887 
1888     if (ret < 0)
1889         goto out;
1890 
1891     if (ctx.optlen > max_optlen || ctx.optlen < 0) {
1892         ret = -EFAULT;
1893         goto out;
1894     }
1895 
1896     if (ctx.optlen != 0) {
1897         if (copy_to_user(optval, ctx.optval, ctx.optlen) ||
1898             put_user(ctx.optlen, optlen)) {
1899             ret = -EFAULT;
1900             goto out;
1901         }
1902     }
1903 
1904 out:
1905     sockopt_free_buf(&ctx, &buf);
1906     return ret;
1907 }
1908 
1909 int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
1910                         int optname, void *optval,
1911                         int *optlen, int retval)
1912 {
1913     struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1914     struct bpf_sockopt_kern ctx = {
1915         .sk = sk,
1916         .level = level,
1917         .optname = optname,
1918         .optlen = *optlen,
1919         .optval = optval,
1920         .optval_end = optval + *optlen,
1921         .current_task = current,
1922     };
1923     int ret;
1924 
1925     /* Note that __cgroup_bpf_run_filter_getsockopt doesn't copy
1926      * user data back into BPF buffer when reval != 0. This is
1927      * done as an optimization to avoid extra copy, assuming
1928      * kernel won't populate the data in case of an error.
1929      * Here we always pass the data and memset() should
1930      * be called if that data shouldn't be "exported".
1931      */
1932 
1933     ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1934                     &ctx, bpf_prog_run, retval, NULL);
1935     if (ret < 0)
1936         return ret;
1937 
1938     if (ctx.optlen > *optlen)
1939         return -EFAULT;
1940 
1941     /* BPF programs can shrink the buffer, export the modifications.
1942      */
1943     if (ctx.optlen != 0)
1944         *optlen = ctx.optlen;
1945 
1946     return ret;
1947 }
1948 #endif
1949 
1950 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp,
1951                   size_t *lenp)
1952 {
1953     ssize_t tmp_ret = 0, ret;
1954 
1955     if (dir->header.parent) {
1956         tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp);
1957         if (tmp_ret < 0)
1958             return tmp_ret;
1959     }
1960 
1961     ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp);
1962     if (ret < 0)
1963         return ret;
1964     *bufp += ret;
1965     *lenp -= ret;
1966     ret += tmp_ret;
1967 
1968     /* Avoid leading slash. */
1969     if (!ret)
1970         return ret;
1971 
1972     tmp_ret = strscpy(*bufp, "/", *lenp);
1973     if (tmp_ret < 0)
1974         return tmp_ret;
1975     *bufp += tmp_ret;
1976     *lenp -= tmp_ret;
1977 
1978     return ret + tmp_ret;
1979 }
1980 
1981 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf,
1982        size_t, buf_len, u64, flags)
1983 {
1984     ssize_t tmp_ret = 0, ret;
1985 
1986     if (!buf)
1987         return -EINVAL;
1988 
1989     if (!(flags & BPF_F_SYSCTL_BASE_NAME)) {
1990         if (!ctx->head)
1991             return -EINVAL;
1992         tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len);
1993         if (tmp_ret < 0)
1994             return tmp_ret;
1995     }
1996 
1997     ret = strscpy(buf, ctx->table->procname, buf_len);
1998 
1999     return ret < 0 ? ret : tmp_ret + ret;
2000 }
2001 
2002 static const struct bpf_func_proto bpf_sysctl_get_name_proto = {
2003     .func       = bpf_sysctl_get_name,
2004     .gpl_only   = false,
2005     .ret_type   = RET_INTEGER,
2006     .arg1_type  = ARG_PTR_TO_CTX,
2007     .arg2_type  = ARG_PTR_TO_MEM,
2008     .arg3_type  = ARG_CONST_SIZE,
2009     .arg4_type  = ARG_ANYTHING,
2010 };
2011 
2012 static int copy_sysctl_value(char *dst, size_t dst_len, char *src,
2013                  size_t src_len)
2014 {
2015     if (!dst)
2016         return -EINVAL;
2017 
2018     if (!dst_len)
2019         return -E2BIG;
2020 
2021     if (!src || !src_len) {
2022         memset(dst, 0, dst_len);
2023         return -EINVAL;
2024     }
2025 
2026     memcpy(dst, src, min(dst_len, src_len));
2027 
2028     if (dst_len > src_len) {
2029         memset(dst + src_len, '\0', dst_len - src_len);
2030         return src_len;
2031     }
2032 
2033     dst[dst_len - 1] = '\0';
2034 
2035     return -E2BIG;
2036 }
2037 
2038 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx,
2039        char *, buf, size_t, buf_len)
2040 {
2041     return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len);
2042 }
2043 
2044 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = {
2045     .func       = bpf_sysctl_get_current_value,
2046     .gpl_only   = false,
2047     .ret_type   = RET_INTEGER,
2048     .arg1_type  = ARG_PTR_TO_CTX,
2049     .arg2_type  = ARG_PTR_TO_UNINIT_MEM,
2050     .arg3_type  = ARG_CONST_SIZE,
2051 };
2052 
2053 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf,
2054        size_t, buf_len)
2055 {
2056     if (!ctx->write) {
2057         if (buf && buf_len)
2058             memset(buf, '\0', buf_len);
2059         return -EINVAL;
2060     }
2061     return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len);
2062 }
2063 
2064 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = {
2065     .func       = bpf_sysctl_get_new_value,
2066     .gpl_only   = false,
2067     .ret_type   = RET_INTEGER,
2068     .arg1_type  = ARG_PTR_TO_CTX,
2069     .arg2_type  = ARG_PTR_TO_UNINIT_MEM,
2070     .arg3_type  = ARG_CONST_SIZE,
2071 };
2072 
2073 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx,
2074        const char *, buf, size_t, buf_len)
2075 {
2076     if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len)
2077         return -EINVAL;
2078 
2079     if (buf_len > PAGE_SIZE - 1)
2080         return -E2BIG;
2081 
2082     memcpy(ctx->new_val, buf, buf_len);
2083     ctx->new_len = buf_len;
2084     ctx->new_updated = 1;
2085 
2086     return 0;
2087 }
2088 
2089 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = {
2090     .func       = bpf_sysctl_set_new_value,
2091     .gpl_only   = false,
2092     .ret_type   = RET_INTEGER,
2093     .arg1_type  = ARG_PTR_TO_CTX,
2094     .arg2_type  = ARG_PTR_TO_MEM | MEM_RDONLY,
2095     .arg3_type  = ARG_CONST_SIZE,
2096 };
2097 
2098 static const struct bpf_func_proto *
2099 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2100 {
2101     switch (func_id) {
2102     case BPF_FUNC_strtol:
2103         return &bpf_strtol_proto;
2104     case BPF_FUNC_strtoul:
2105         return &bpf_strtoul_proto;
2106     case BPF_FUNC_sysctl_get_name:
2107         return &bpf_sysctl_get_name_proto;
2108     case BPF_FUNC_sysctl_get_current_value:
2109         return &bpf_sysctl_get_current_value_proto;
2110     case BPF_FUNC_sysctl_get_new_value:
2111         return &bpf_sysctl_get_new_value_proto;
2112     case BPF_FUNC_sysctl_set_new_value:
2113         return &bpf_sysctl_set_new_value_proto;
2114     case BPF_FUNC_ktime_get_coarse_ns:
2115         return &bpf_ktime_get_coarse_ns_proto;
2116     default:
2117         return cgroup_base_func_proto(func_id, prog);
2118     }
2119 }
2120 
2121 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type,
2122                    const struct bpf_prog *prog,
2123                    struct bpf_insn_access_aux *info)
2124 {
2125     const int size_default = sizeof(__u32);
2126 
2127     if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size)
2128         return false;
2129 
2130     switch (off) {
2131     case bpf_ctx_range(struct bpf_sysctl, write):
2132         if (type != BPF_READ)
2133             return false;
2134         bpf_ctx_record_field_size(info, size_default);
2135         return bpf_ctx_narrow_access_ok(off, size, size_default);
2136     case bpf_ctx_range(struct bpf_sysctl, file_pos):
2137         if (type == BPF_READ) {
2138             bpf_ctx_record_field_size(info, size_default);
2139             return bpf_ctx_narrow_access_ok(off, size, size_default);
2140         } else {
2141             return size == size_default;
2142         }
2143     default:
2144         return false;
2145     }
2146 }
2147 
2148 static u32 sysctl_convert_ctx_access(enum bpf_access_type type,
2149                      const struct bpf_insn *si,
2150                      struct bpf_insn *insn_buf,
2151                      struct bpf_prog *prog, u32 *target_size)
2152 {
2153     struct bpf_insn *insn = insn_buf;
2154     u32 read_size;
2155 
2156     switch (si->off) {
2157     case offsetof(struct bpf_sysctl, write):
2158         *insn++ = BPF_LDX_MEM(
2159             BPF_SIZE(si->code), si->dst_reg, si->src_reg,
2160             bpf_target_off(struct bpf_sysctl_kern, write,
2161                        sizeof_field(struct bpf_sysctl_kern,
2162                             write),
2163                        target_size));
2164         break;
2165     case offsetof(struct bpf_sysctl, file_pos):
2166         /* ppos is a pointer so it should be accessed via indirect
2167          * loads and stores. Also for stores additional temporary
2168          * register is used since neither src_reg nor dst_reg can be
2169          * overridden.
2170          */
2171         if (type == BPF_WRITE) {
2172             int treg = BPF_REG_9;
2173 
2174             if (si->src_reg == treg || si->dst_reg == treg)
2175                 --treg;
2176             if (si->src_reg == treg || si->dst_reg == treg)
2177                 --treg;
2178             *insn++ = BPF_STX_MEM(
2179                 BPF_DW, si->dst_reg, treg,
2180                 offsetof(struct bpf_sysctl_kern, tmp_reg));
2181             *insn++ = BPF_LDX_MEM(
2182                 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2183                 treg, si->dst_reg,
2184                 offsetof(struct bpf_sysctl_kern, ppos));
2185             *insn++ = BPF_STX_MEM(
2186                 BPF_SIZEOF(u32), treg, si->src_reg,
2187                 bpf_ctx_narrow_access_offset(
2188                     0, sizeof(u32), sizeof(loff_t)));
2189             *insn++ = BPF_LDX_MEM(
2190                 BPF_DW, treg, si->dst_reg,
2191                 offsetof(struct bpf_sysctl_kern, tmp_reg));
2192         } else {
2193             *insn++ = BPF_LDX_MEM(
2194                 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2195                 si->dst_reg, si->src_reg,
2196                 offsetof(struct bpf_sysctl_kern, ppos));
2197             read_size = bpf_size_to_bytes(BPF_SIZE(si->code));
2198             *insn++ = BPF_LDX_MEM(
2199                 BPF_SIZE(si->code), si->dst_reg, si->dst_reg,
2200                 bpf_ctx_narrow_access_offset(
2201                     0, read_size, sizeof(loff_t)));
2202         }
2203         *target_size = sizeof(u32);
2204         break;
2205     }
2206 
2207     return insn - insn_buf;
2208 }
2209 
2210 const struct bpf_verifier_ops cg_sysctl_verifier_ops = {
2211     .get_func_proto     = sysctl_func_proto,
2212     .is_valid_access    = sysctl_is_valid_access,
2213     .convert_ctx_access = sysctl_convert_ctx_access,
2214 };
2215 
2216 const struct bpf_prog_ops cg_sysctl_prog_ops = {
2217 };
2218 
2219 #ifdef CONFIG_NET
2220 BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx)
2221 {
2222     const struct net *net = ctx ? sock_net(ctx->sk) : &init_net;
2223 
2224     return net->net_cookie;
2225 }
2226 
2227 static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = {
2228     .func       = bpf_get_netns_cookie_sockopt,
2229     .gpl_only   = false,
2230     .ret_type   = RET_INTEGER,
2231     .arg1_type  = ARG_PTR_TO_CTX_OR_NULL,
2232 };
2233 #endif
2234 
2235 static const struct bpf_func_proto *
2236 cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2237 {
2238     switch (func_id) {
2239 #ifdef CONFIG_NET
2240     case BPF_FUNC_get_netns_cookie:
2241         return &bpf_get_netns_cookie_sockopt_proto;
2242     case BPF_FUNC_sk_storage_get:
2243         return &bpf_sk_storage_get_proto;
2244     case BPF_FUNC_sk_storage_delete:
2245         return &bpf_sk_storage_delete_proto;
2246     case BPF_FUNC_setsockopt:
2247         if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2248             return &bpf_sk_setsockopt_proto;
2249         return NULL;
2250     case BPF_FUNC_getsockopt:
2251         if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2252             return &bpf_sk_getsockopt_proto;
2253         return NULL;
2254 #endif
2255 #ifdef CONFIG_INET
2256     case BPF_FUNC_tcp_sock:
2257         return &bpf_tcp_sock_proto;
2258 #endif
2259     default:
2260         return cgroup_base_func_proto(func_id, prog);
2261     }
2262 }
2263 
2264 static bool cg_sockopt_is_valid_access(int off, int size,
2265                        enum bpf_access_type type,
2266                        const struct bpf_prog *prog,
2267                        struct bpf_insn_access_aux *info)
2268 {
2269     const int size_default = sizeof(__u32);
2270 
2271     if (off < 0 || off >= sizeof(struct bpf_sockopt))
2272         return false;
2273 
2274     if (off % size != 0)
2275         return false;
2276 
2277     if (type == BPF_WRITE) {
2278         switch (off) {
2279         case offsetof(struct bpf_sockopt, retval):
2280             if (size != size_default)
2281                 return false;
2282             return prog->expected_attach_type ==
2283                 BPF_CGROUP_GETSOCKOPT;
2284         case offsetof(struct bpf_sockopt, optname):
2285             fallthrough;
2286         case offsetof(struct bpf_sockopt, level):
2287             if (size != size_default)
2288                 return false;
2289             return prog->expected_attach_type ==
2290                 BPF_CGROUP_SETSOCKOPT;
2291         case offsetof(struct bpf_sockopt, optlen):
2292             return size == size_default;
2293         default:
2294             return false;
2295         }
2296     }
2297 
2298     switch (off) {
2299     case offsetof(struct bpf_sockopt, sk):
2300         if (size != sizeof(__u64))
2301             return false;
2302         info->reg_type = PTR_TO_SOCKET;
2303         break;
2304     case offsetof(struct bpf_sockopt, optval):
2305         if (size != sizeof(__u64))
2306             return false;
2307         info->reg_type = PTR_TO_PACKET;
2308         break;
2309     case offsetof(struct bpf_sockopt, optval_end):
2310         if (size != sizeof(__u64))
2311             return false;
2312         info->reg_type = PTR_TO_PACKET_END;
2313         break;
2314     case offsetof(struct bpf_sockopt, retval):
2315         if (size != size_default)
2316             return false;
2317         return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT;
2318     default:
2319         if (size != size_default)
2320             return false;
2321         break;
2322     }
2323     return true;
2324 }
2325 
2326 #define CG_SOCKOPT_ACCESS_FIELD(T, F)                   \
2327     T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F),         \
2328       si->dst_reg, si->src_reg,                 \
2329       offsetof(struct bpf_sockopt_kern, F))
2330 
2331 static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type,
2332                      const struct bpf_insn *si,
2333                      struct bpf_insn *insn_buf,
2334                      struct bpf_prog *prog,
2335                      u32 *target_size)
2336 {
2337     struct bpf_insn *insn = insn_buf;
2338 
2339     switch (si->off) {
2340     case offsetof(struct bpf_sockopt, sk):
2341         *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk);
2342         break;
2343     case offsetof(struct bpf_sockopt, level):
2344         if (type == BPF_WRITE)
2345             *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level);
2346         else
2347             *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level);
2348         break;
2349     case offsetof(struct bpf_sockopt, optname):
2350         if (type == BPF_WRITE)
2351             *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname);
2352         else
2353             *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname);
2354         break;
2355     case offsetof(struct bpf_sockopt, optlen):
2356         if (type == BPF_WRITE)
2357             *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen);
2358         else
2359             *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen);
2360         break;
2361     case offsetof(struct bpf_sockopt, retval):
2362         BUILD_BUG_ON(offsetof(struct bpf_cg_run_ctx, run_ctx) != 0);
2363 
2364         if (type == BPF_WRITE) {
2365             int treg = BPF_REG_9;
2366 
2367             if (si->src_reg == treg || si->dst_reg == treg)
2368                 --treg;
2369             if (si->src_reg == treg || si->dst_reg == treg)
2370                 --treg;
2371             *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, treg,
2372                           offsetof(struct bpf_sockopt_kern, tmp_reg));
2373             *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2374                           treg, si->dst_reg,
2375                           offsetof(struct bpf_sockopt_kern, current_task));
2376             *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2377                           treg, treg,
2378                           offsetof(struct task_struct, bpf_ctx));
2379             *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2380                           treg, si->src_reg,
2381                           offsetof(struct bpf_cg_run_ctx, retval));
2382             *insn++ = BPF_LDX_MEM(BPF_DW, treg, si->dst_reg,
2383                           offsetof(struct bpf_sockopt_kern, tmp_reg));
2384         } else {
2385             *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2386                           si->dst_reg, si->src_reg,
2387                           offsetof(struct bpf_sockopt_kern, current_task));
2388             *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2389                           si->dst_reg, si->dst_reg,
2390                           offsetof(struct task_struct, bpf_ctx));
2391             *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2392                           si->dst_reg, si->dst_reg,
2393                           offsetof(struct bpf_cg_run_ctx, retval));
2394         }
2395         break;
2396     case offsetof(struct bpf_sockopt, optval):
2397         *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval);
2398         break;
2399     case offsetof(struct bpf_sockopt, optval_end):
2400         *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end);
2401         break;
2402     }
2403 
2404     return insn - insn_buf;
2405 }
2406 
2407 static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf,
2408                    bool direct_write,
2409                    const struct bpf_prog *prog)
2410 {
2411     /* Nothing to do for sockopt argument. The data is kzalloc'ated.
2412      */
2413     return 0;
2414 }
2415 
2416 const struct bpf_verifier_ops cg_sockopt_verifier_ops = {
2417     .get_func_proto     = cg_sockopt_func_proto,
2418     .is_valid_access    = cg_sockopt_is_valid_access,
2419     .convert_ctx_access = cg_sockopt_convert_ctx_access,
2420     .gen_prologue       = cg_sockopt_get_prologue,
2421 };
2422 
2423 const struct bpf_prog_ops cg_sockopt_prog_ops = {
2424 };
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!