Back to home page

OSCL-LXR
 
 

     

0001 /* Copyright (c) 2018, Mellanox Technologies All rights reserved.
0002  *
0003  * This software is available to you under a choice of one of two
0004  * licenses.  You may choose to be licensed under the terms of the GNU
0005  * General Public License (GPL) Version 2, available from the file
0006  * COPYING in the main directory of this source tree, or the
0007  * OpenIB.org BSD license below:
0008  *
0009  *     Redistribution and use in source and binary forms, with or
0010  *     without modification, are permitted provided that the following
0011  *     conditions are met:
0012  *
0013  *      - Redistributions of source code must retain the above
0014  *        copyright notice, this list of conditions and the following
0015  *        disclaimer.
0016  *
0017  *      - Redistributions in binary form must reproduce the above
0018  *        copyright notice, this list of conditions and the following
0019  *        disclaimer in the documentation and/or other materials
0020  *        provided with the distribution.
0021  *
0022  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
0023  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
0024  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
0025  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
0026  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
0027  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
0028  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
0029  * SOFTWARE.
0030  */
0031 
0032 #include <crypto/aead.h>
0033 #include <linux/highmem.h>
0034 #include <linux/module.h>
0035 #include <linux/netdevice.h>
0036 #include <net/dst.h>
0037 #include <net/inet_connection_sock.h>
0038 #include <net/tcp.h>
0039 #include <net/tls.h>
0040 
0041 #include "tls.h"
0042 #include "trace.h"
0043 
0044 /* device_offload_lock is used to synchronize tls_dev_add
0045  * against NETDEV_DOWN notifications.
0046  */
0047 static DECLARE_RWSEM(device_offload_lock);
0048 
0049 static struct workqueue_struct *destruct_wq __read_mostly;
0050 
0051 static LIST_HEAD(tls_device_list);
0052 static LIST_HEAD(tls_device_down_list);
0053 static DEFINE_SPINLOCK(tls_device_lock);
0054 
0055 static void tls_device_free_ctx(struct tls_context *ctx)
0056 {
0057     if (ctx->tx_conf == TLS_HW) {
0058         kfree(tls_offload_ctx_tx(ctx));
0059         kfree(ctx->tx.rec_seq);
0060         kfree(ctx->tx.iv);
0061     }
0062 
0063     if (ctx->rx_conf == TLS_HW)
0064         kfree(tls_offload_ctx_rx(ctx));
0065 
0066     tls_ctx_free(NULL, ctx);
0067 }
0068 
0069 static void tls_device_tx_del_task(struct work_struct *work)
0070 {
0071     struct tls_offload_context_tx *offload_ctx =
0072         container_of(work, struct tls_offload_context_tx, destruct_work);
0073     struct tls_context *ctx = offload_ctx->ctx;
0074     struct net_device *netdev;
0075 
0076     /* Safe, because this is the destroy flow, refcount is 0, so
0077      * tls_device_down can't store this field in parallel.
0078      */
0079     netdev = rcu_dereference_protected(ctx->netdev,
0080                        !refcount_read(&ctx->refcount));
0081 
0082     netdev->tlsdev_ops->tls_dev_del(netdev, ctx, TLS_OFFLOAD_CTX_DIR_TX);
0083     dev_put(netdev);
0084     ctx->netdev = NULL;
0085     tls_device_free_ctx(ctx);
0086 }
0087 
0088 static void tls_device_queue_ctx_destruction(struct tls_context *ctx)
0089 {
0090     struct net_device *netdev;
0091     unsigned long flags;
0092     bool async_cleanup;
0093 
0094     spin_lock_irqsave(&tls_device_lock, flags);
0095     if (unlikely(!refcount_dec_and_test(&ctx->refcount))) {
0096         spin_unlock_irqrestore(&tls_device_lock, flags);
0097         return;
0098     }
0099 
0100     list_del(&ctx->list); /* Remove from tls_device_list / tls_device_down_list */
0101 
0102     /* Safe, because this is the destroy flow, refcount is 0, so
0103      * tls_device_down can't store this field in parallel.
0104      */
0105     netdev = rcu_dereference_protected(ctx->netdev,
0106                        !refcount_read(&ctx->refcount));
0107 
0108     async_cleanup = netdev && ctx->tx_conf == TLS_HW;
0109     if (async_cleanup) {
0110         struct tls_offload_context_tx *offload_ctx = tls_offload_ctx_tx(ctx);
0111 
0112         /* queue_work inside the spinlock
0113          * to make sure tls_device_down waits for that work.
0114          */
0115         queue_work(destruct_wq, &offload_ctx->destruct_work);
0116     }
0117     spin_unlock_irqrestore(&tls_device_lock, flags);
0118 
0119     if (!async_cleanup)
0120         tls_device_free_ctx(ctx);
0121 }
0122 
0123 /* We assume that the socket is already connected */
0124 static struct net_device *get_netdev_for_sock(struct sock *sk)
0125 {
0126     struct dst_entry *dst = sk_dst_get(sk);
0127     struct net_device *netdev = NULL;
0128 
0129     if (likely(dst)) {
0130         netdev = netdev_sk_get_lowest_dev(dst->dev, sk);
0131         dev_hold(netdev);
0132     }
0133 
0134     dst_release(dst);
0135 
0136     return netdev;
0137 }
0138 
0139 static void destroy_record(struct tls_record_info *record)
0140 {
0141     int i;
0142 
0143     for (i = 0; i < record->num_frags; i++)
0144         __skb_frag_unref(&record->frags[i], false);
0145     kfree(record);
0146 }
0147 
0148 static void delete_all_records(struct tls_offload_context_tx *offload_ctx)
0149 {
0150     struct tls_record_info *info, *temp;
0151 
0152     list_for_each_entry_safe(info, temp, &offload_ctx->records_list, list) {
0153         list_del(&info->list);
0154         destroy_record(info);
0155     }
0156 
0157     offload_ctx->retransmit_hint = NULL;
0158 }
0159 
0160 static void tls_icsk_clean_acked(struct sock *sk, u32 acked_seq)
0161 {
0162     struct tls_context *tls_ctx = tls_get_ctx(sk);
0163     struct tls_record_info *info, *temp;
0164     struct tls_offload_context_tx *ctx;
0165     u64 deleted_records = 0;
0166     unsigned long flags;
0167 
0168     if (!tls_ctx)
0169         return;
0170 
0171     ctx = tls_offload_ctx_tx(tls_ctx);
0172 
0173     spin_lock_irqsave(&ctx->lock, flags);
0174     info = ctx->retransmit_hint;
0175     if (info && !before(acked_seq, info->end_seq))
0176         ctx->retransmit_hint = NULL;
0177 
0178     list_for_each_entry_safe(info, temp, &ctx->records_list, list) {
0179         if (before(acked_seq, info->end_seq))
0180             break;
0181         list_del(&info->list);
0182 
0183         destroy_record(info);
0184         deleted_records++;
0185     }
0186 
0187     ctx->unacked_record_sn += deleted_records;
0188     spin_unlock_irqrestore(&ctx->lock, flags);
0189 }
0190 
0191 /* At this point, there should be no references on this
0192  * socket and no in-flight SKBs associated with this
0193  * socket, so it is safe to free all the resources.
0194  */
0195 void tls_device_sk_destruct(struct sock *sk)
0196 {
0197     struct tls_context *tls_ctx = tls_get_ctx(sk);
0198     struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
0199 
0200     tls_ctx->sk_destruct(sk);
0201 
0202     if (tls_ctx->tx_conf == TLS_HW) {
0203         if (ctx->open_record)
0204             destroy_record(ctx->open_record);
0205         delete_all_records(ctx);
0206         crypto_free_aead(ctx->aead_send);
0207         clean_acked_data_disable(inet_csk(sk));
0208     }
0209 
0210     tls_device_queue_ctx_destruction(tls_ctx);
0211 }
0212 EXPORT_SYMBOL_GPL(tls_device_sk_destruct);
0213 
0214 void tls_device_free_resources_tx(struct sock *sk)
0215 {
0216     struct tls_context *tls_ctx = tls_get_ctx(sk);
0217 
0218     tls_free_partial_record(sk, tls_ctx);
0219 }
0220 
0221 void tls_offload_tx_resync_request(struct sock *sk, u32 got_seq, u32 exp_seq)
0222 {
0223     struct tls_context *tls_ctx = tls_get_ctx(sk);
0224 
0225     trace_tls_device_tx_resync_req(sk, got_seq, exp_seq);
0226     WARN_ON(test_and_set_bit(TLS_TX_SYNC_SCHED, &tls_ctx->flags));
0227 }
0228 EXPORT_SYMBOL_GPL(tls_offload_tx_resync_request);
0229 
0230 static void tls_device_resync_tx(struct sock *sk, struct tls_context *tls_ctx,
0231                  u32 seq)
0232 {
0233     struct net_device *netdev;
0234     struct sk_buff *skb;
0235     int err = 0;
0236     u8 *rcd_sn;
0237 
0238     skb = tcp_write_queue_tail(sk);
0239     if (skb)
0240         TCP_SKB_CB(skb)->eor = 1;
0241 
0242     rcd_sn = tls_ctx->tx.rec_seq;
0243 
0244     trace_tls_device_tx_resync_send(sk, seq, rcd_sn);
0245     down_read(&device_offload_lock);
0246     netdev = rcu_dereference_protected(tls_ctx->netdev,
0247                        lockdep_is_held(&device_offload_lock));
0248     if (netdev)
0249         err = netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq,
0250                              rcd_sn,
0251                              TLS_OFFLOAD_CTX_DIR_TX);
0252     up_read(&device_offload_lock);
0253     if (err)
0254         return;
0255 
0256     clear_bit_unlock(TLS_TX_SYNC_SCHED, &tls_ctx->flags);
0257 }
0258 
0259 static void tls_append_frag(struct tls_record_info *record,
0260                 struct page_frag *pfrag,
0261                 int size)
0262 {
0263     skb_frag_t *frag;
0264 
0265     frag = &record->frags[record->num_frags - 1];
0266     if (skb_frag_page(frag) == pfrag->page &&
0267         skb_frag_off(frag) + skb_frag_size(frag) == pfrag->offset) {
0268         skb_frag_size_add(frag, size);
0269     } else {
0270         ++frag;
0271         __skb_frag_set_page(frag, pfrag->page);
0272         skb_frag_off_set(frag, pfrag->offset);
0273         skb_frag_size_set(frag, size);
0274         ++record->num_frags;
0275         get_page(pfrag->page);
0276     }
0277 
0278     pfrag->offset += size;
0279     record->len += size;
0280 }
0281 
0282 static int tls_push_record(struct sock *sk,
0283                struct tls_context *ctx,
0284                struct tls_offload_context_tx *offload_ctx,
0285                struct tls_record_info *record,
0286                int flags)
0287 {
0288     struct tls_prot_info *prot = &ctx->prot_info;
0289     struct tcp_sock *tp = tcp_sk(sk);
0290     skb_frag_t *frag;
0291     int i;
0292 
0293     record->end_seq = tp->write_seq + record->len;
0294     list_add_tail_rcu(&record->list, &offload_ctx->records_list);
0295     offload_ctx->open_record = NULL;
0296 
0297     if (test_bit(TLS_TX_SYNC_SCHED, &ctx->flags))
0298         tls_device_resync_tx(sk, ctx, tp->write_seq);
0299 
0300     tls_advance_record_sn(sk, prot, &ctx->tx);
0301 
0302     for (i = 0; i < record->num_frags; i++) {
0303         frag = &record->frags[i];
0304         sg_unmark_end(&offload_ctx->sg_tx_data[i]);
0305         sg_set_page(&offload_ctx->sg_tx_data[i], skb_frag_page(frag),
0306                 skb_frag_size(frag), skb_frag_off(frag));
0307         sk_mem_charge(sk, skb_frag_size(frag));
0308         get_page(skb_frag_page(frag));
0309     }
0310     sg_mark_end(&offload_ctx->sg_tx_data[record->num_frags - 1]);
0311 
0312     /* all ready, send */
0313     return tls_push_sg(sk, ctx, offload_ctx->sg_tx_data, 0, flags);
0314 }
0315 
0316 static int tls_device_record_close(struct sock *sk,
0317                    struct tls_context *ctx,
0318                    struct tls_record_info *record,
0319                    struct page_frag *pfrag,
0320                    unsigned char record_type)
0321 {
0322     struct tls_prot_info *prot = &ctx->prot_info;
0323     int ret;
0324 
0325     /* append tag
0326      * device will fill in the tag, we just need to append a placeholder
0327      * use socket memory to improve coalescing (re-using a single buffer
0328      * increases frag count)
0329      * if we can't allocate memory now, steal some back from data
0330      */
0331     if (likely(skb_page_frag_refill(prot->tag_size, pfrag,
0332                     sk->sk_allocation))) {
0333         ret = 0;
0334         tls_append_frag(record, pfrag, prot->tag_size);
0335     } else {
0336         ret = prot->tag_size;
0337         if (record->len <= prot->overhead_size)
0338             return -ENOMEM;
0339     }
0340 
0341     /* fill prepend */
0342     tls_fill_prepend(ctx, skb_frag_address(&record->frags[0]),
0343              record->len - prot->overhead_size,
0344              record_type);
0345     return ret;
0346 }
0347 
0348 static int tls_create_new_record(struct tls_offload_context_tx *offload_ctx,
0349                  struct page_frag *pfrag,
0350                  size_t prepend_size)
0351 {
0352     struct tls_record_info *record;
0353     skb_frag_t *frag;
0354 
0355     record = kmalloc(sizeof(*record), GFP_KERNEL);
0356     if (!record)
0357         return -ENOMEM;
0358 
0359     frag = &record->frags[0];
0360     __skb_frag_set_page(frag, pfrag->page);
0361     skb_frag_off_set(frag, pfrag->offset);
0362     skb_frag_size_set(frag, prepend_size);
0363 
0364     get_page(pfrag->page);
0365     pfrag->offset += prepend_size;
0366 
0367     record->num_frags = 1;
0368     record->len = prepend_size;
0369     offload_ctx->open_record = record;
0370     return 0;
0371 }
0372 
0373 static int tls_do_allocation(struct sock *sk,
0374                  struct tls_offload_context_tx *offload_ctx,
0375                  struct page_frag *pfrag,
0376                  size_t prepend_size)
0377 {
0378     int ret;
0379 
0380     if (!offload_ctx->open_record) {
0381         if (unlikely(!skb_page_frag_refill(prepend_size, pfrag,
0382                            sk->sk_allocation))) {
0383             READ_ONCE(sk->sk_prot)->enter_memory_pressure(sk);
0384             sk_stream_moderate_sndbuf(sk);
0385             return -ENOMEM;
0386         }
0387 
0388         ret = tls_create_new_record(offload_ctx, pfrag, prepend_size);
0389         if (ret)
0390             return ret;
0391 
0392         if (pfrag->size > pfrag->offset)
0393             return 0;
0394     }
0395 
0396     if (!sk_page_frag_refill(sk, pfrag))
0397         return -ENOMEM;
0398 
0399     return 0;
0400 }
0401 
0402 static int tls_device_copy_data(void *addr, size_t bytes, struct iov_iter *i)
0403 {
0404     size_t pre_copy, nocache;
0405 
0406     pre_copy = ~((unsigned long)addr - 1) & (SMP_CACHE_BYTES - 1);
0407     if (pre_copy) {
0408         pre_copy = min(pre_copy, bytes);
0409         if (copy_from_iter(addr, pre_copy, i) != pre_copy)
0410             return -EFAULT;
0411         bytes -= pre_copy;
0412         addr += pre_copy;
0413     }
0414 
0415     nocache = round_down(bytes, SMP_CACHE_BYTES);
0416     if (copy_from_iter_nocache(addr, nocache, i) != nocache)
0417         return -EFAULT;
0418     bytes -= nocache;
0419     addr += nocache;
0420 
0421     if (bytes && copy_from_iter(addr, bytes, i) != bytes)
0422         return -EFAULT;
0423 
0424     return 0;
0425 }
0426 
0427 union tls_iter_offset {
0428     struct iov_iter *msg_iter;
0429     int offset;
0430 };
0431 
0432 static int tls_push_data(struct sock *sk,
0433              union tls_iter_offset iter_offset,
0434              size_t size, int flags,
0435              unsigned char record_type,
0436              struct page *zc_page)
0437 {
0438     struct tls_context *tls_ctx = tls_get_ctx(sk);
0439     struct tls_prot_info *prot = &tls_ctx->prot_info;
0440     struct tls_offload_context_tx *ctx = tls_offload_ctx_tx(tls_ctx);
0441     struct tls_record_info *record;
0442     int tls_push_record_flags;
0443     struct page_frag *pfrag;
0444     size_t orig_size = size;
0445     u32 max_open_record_len;
0446     bool more = false;
0447     bool done = false;
0448     int copy, rc = 0;
0449     long timeo;
0450 
0451     if (flags &
0452         ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_SENDPAGE_NOTLAST))
0453         return -EOPNOTSUPP;
0454 
0455     if (unlikely(sk->sk_err))
0456         return -sk->sk_err;
0457 
0458     flags |= MSG_SENDPAGE_DECRYPTED;
0459     tls_push_record_flags = flags | MSG_SENDPAGE_NOTLAST;
0460 
0461     timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
0462     if (tls_is_partially_sent_record(tls_ctx)) {
0463         rc = tls_push_partial_record(sk, tls_ctx, flags);
0464         if (rc < 0)
0465             return rc;
0466     }
0467 
0468     pfrag = sk_page_frag(sk);
0469 
0470     /* TLS_HEADER_SIZE is not counted as part of the TLS record, and
0471      * we need to leave room for an authentication tag.
0472      */
0473     max_open_record_len = TLS_MAX_PAYLOAD_SIZE +
0474                   prot->prepend_size;
0475     do {
0476         rc = tls_do_allocation(sk, ctx, pfrag, prot->prepend_size);
0477         if (unlikely(rc)) {
0478             rc = sk_stream_wait_memory(sk, &timeo);
0479             if (!rc)
0480                 continue;
0481 
0482             record = ctx->open_record;
0483             if (!record)
0484                 break;
0485 handle_error:
0486             if (record_type != TLS_RECORD_TYPE_DATA) {
0487                 /* avoid sending partial
0488                  * record with type !=
0489                  * application_data
0490                  */
0491                 size = orig_size;
0492                 destroy_record(record);
0493                 ctx->open_record = NULL;
0494             } else if (record->len > prot->prepend_size) {
0495                 goto last_record;
0496             }
0497 
0498             break;
0499         }
0500 
0501         record = ctx->open_record;
0502 
0503         copy = min_t(size_t, size, max_open_record_len - record->len);
0504         if (copy && zc_page) {
0505             struct page_frag zc_pfrag;
0506 
0507             zc_pfrag.page = zc_page;
0508             zc_pfrag.offset = iter_offset.offset;
0509             zc_pfrag.size = copy;
0510             tls_append_frag(record, &zc_pfrag, copy);
0511         } else if (copy) {
0512             copy = min_t(size_t, copy, pfrag->size - pfrag->offset);
0513 
0514             rc = tls_device_copy_data(page_address(pfrag->page) +
0515                           pfrag->offset, copy,
0516                           iter_offset.msg_iter);
0517             if (rc)
0518                 goto handle_error;
0519             tls_append_frag(record, pfrag, copy);
0520         }
0521 
0522         size -= copy;
0523         if (!size) {
0524 last_record:
0525             tls_push_record_flags = flags;
0526             if (flags & (MSG_SENDPAGE_NOTLAST | MSG_MORE)) {
0527                 more = true;
0528                 break;
0529             }
0530 
0531             done = true;
0532         }
0533 
0534         if (done || record->len >= max_open_record_len ||
0535             (record->num_frags >= MAX_SKB_FRAGS - 1)) {
0536             rc = tls_device_record_close(sk, tls_ctx, record,
0537                              pfrag, record_type);
0538             if (rc) {
0539                 if (rc > 0) {
0540                     size += rc;
0541                 } else {
0542                     size = orig_size;
0543                     destroy_record(record);
0544                     ctx->open_record = NULL;
0545                     break;
0546                 }
0547             }
0548 
0549             rc = tls_push_record(sk,
0550                          tls_ctx,
0551                          ctx,
0552                          record,
0553                          tls_push_record_flags);
0554             if (rc < 0)
0555                 break;
0556         }
0557     } while (!done);
0558 
0559     tls_ctx->pending_open_record_frags = more;
0560 
0561     if (orig_size - size > 0)
0562         rc = orig_size - size;
0563 
0564     return rc;
0565 }
0566 
0567 int tls_device_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
0568 {
0569     unsigned char record_type = TLS_RECORD_TYPE_DATA;
0570     struct tls_context *tls_ctx = tls_get_ctx(sk);
0571     union tls_iter_offset iter;
0572     int rc;
0573 
0574     mutex_lock(&tls_ctx->tx_lock);
0575     lock_sock(sk);
0576 
0577     if (unlikely(msg->msg_controllen)) {
0578         rc = tls_process_cmsg(sk, msg, &record_type);
0579         if (rc)
0580             goto out;
0581     }
0582 
0583     iter.msg_iter = &msg->msg_iter;
0584     rc = tls_push_data(sk, iter, size, msg->msg_flags, record_type, NULL);
0585 
0586 out:
0587     release_sock(sk);
0588     mutex_unlock(&tls_ctx->tx_lock);
0589     return rc;
0590 }
0591 
0592 int tls_device_sendpage(struct sock *sk, struct page *page,
0593             int offset, size_t size, int flags)
0594 {
0595     struct tls_context *tls_ctx = tls_get_ctx(sk);
0596     union tls_iter_offset iter_offset;
0597     struct iov_iter msg_iter;
0598     char *kaddr;
0599     struct kvec iov;
0600     int rc;
0601 
0602     if (flags & MSG_SENDPAGE_NOTLAST)
0603         flags |= MSG_MORE;
0604 
0605     mutex_lock(&tls_ctx->tx_lock);
0606     lock_sock(sk);
0607 
0608     if (flags & MSG_OOB) {
0609         rc = -EOPNOTSUPP;
0610         goto out;
0611     }
0612 
0613     if (tls_ctx->zerocopy_sendfile) {
0614         iter_offset.offset = offset;
0615         rc = tls_push_data(sk, iter_offset, size,
0616                    flags, TLS_RECORD_TYPE_DATA, page);
0617         goto out;
0618     }
0619 
0620     kaddr = kmap(page);
0621     iov.iov_base = kaddr + offset;
0622     iov.iov_len = size;
0623     iov_iter_kvec(&msg_iter, WRITE, &iov, 1, size);
0624     iter_offset.msg_iter = &msg_iter;
0625     rc = tls_push_data(sk, iter_offset, size, flags, TLS_RECORD_TYPE_DATA,
0626                NULL);
0627     kunmap(page);
0628 
0629 out:
0630     release_sock(sk);
0631     mutex_unlock(&tls_ctx->tx_lock);
0632     return rc;
0633 }
0634 
0635 struct tls_record_info *tls_get_record(struct tls_offload_context_tx *context,
0636                        u32 seq, u64 *p_record_sn)
0637 {
0638     u64 record_sn = context->hint_record_sn;
0639     struct tls_record_info *info, *last;
0640 
0641     info = context->retransmit_hint;
0642     if (!info ||
0643         before(seq, info->end_seq - info->len)) {
0644         /* if retransmit_hint is irrelevant start
0645          * from the beginning of the list
0646          */
0647         info = list_first_entry_or_null(&context->records_list,
0648                         struct tls_record_info, list);
0649         if (!info)
0650             return NULL;
0651         /* send the start_marker record if seq number is before the
0652          * tls offload start marker sequence number. This record is
0653          * required to handle TCP packets which are before TLS offload
0654          * started.
0655          *  And if it's not start marker, look if this seq number
0656          * belongs to the list.
0657          */
0658         if (likely(!tls_record_is_start_marker(info))) {
0659             /* we have the first record, get the last record to see
0660              * if this seq number belongs to the list.
0661              */
0662             last = list_last_entry(&context->records_list,
0663                            struct tls_record_info, list);
0664 
0665             if (!between(seq, tls_record_start_seq(info),
0666                      last->end_seq))
0667                 return NULL;
0668         }
0669         record_sn = context->unacked_record_sn;
0670     }
0671 
0672     /* We just need the _rcu for the READ_ONCE() */
0673     rcu_read_lock();
0674     list_for_each_entry_from_rcu(info, &context->records_list, list) {
0675         if (before(seq, info->end_seq)) {
0676             if (!context->retransmit_hint ||
0677                 after(info->end_seq,
0678                   context->retransmit_hint->end_seq)) {
0679                 context->hint_record_sn = record_sn;
0680                 context->retransmit_hint = info;
0681             }
0682             *p_record_sn = record_sn;
0683             goto exit_rcu_unlock;
0684         }
0685         record_sn++;
0686     }
0687     info = NULL;
0688 
0689 exit_rcu_unlock:
0690     rcu_read_unlock();
0691     return info;
0692 }
0693 EXPORT_SYMBOL(tls_get_record);
0694 
0695 static int tls_device_push_pending_record(struct sock *sk, int flags)
0696 {
0697     union tls_iter_offset iter;
0698     struct iov_iter msg_iter;
0699 
0700     iov_iter_kvec(&msg_iter, WRITE, NULL, 0, 0);
0701     iter.msg_iter = &msg_iter;
0702     return tls_push_data(sk, iter, 0, flags, TLS_RECORD_TYPE_DATA, NULL);
0703 }
0704 
0705 void tls_device_write_space(struct sock *sk, struct tls_context *ctx)
0706 {
0707     if (tls_is_partially_sent_record(ctx)) {
0708         gfp_t sk_allocation = sk->sk_allocation;
0709 
0710         WARN_ON_ONCE(sk->sk_write_pending);
0711 
0712         sk->sk_allocation = GFP_ATOMIC;
0713         tls_push_partial_record(sk, ctx,
0714                     MSG_DONTWAIT | MSG_NOSIGNAL |
0715                     MSG_SENDPAGE_DECRYPTED);
0716         sk->sk_allocation = sk_allocation;
0717     }
0718 }
0719 
0720 static void tls_device_resync_rx(struct tls_context *tls_ctx,
0721                  struct sock *sk, u32 seq, u8 *rcd_sn)
0722 {
0723     struct tls_offload_context_rx *rx_ctx = tls_offload_ctx_rx(tls_ctx);
0724     struct net_device *netdev;
0725 
0726     trace_tls_device_rx_resync_send(sk, seq, rcd_sn, rx_ctx->resync_type);
0727     rcu_read_lock();
0728     netdev = rcu_dereference(tls_ctx->netdev);
0729     if (netdev)
0730         netdev->tlsdev_ops->tls_dev_resync(netdev, sk, seq, rcd_sn,
0731                            TLS_OFFLOAD_CTX_DIR_RX);
0732     rcu_read_unlock();
0733     TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSRXDEVICERESYNC);
0734 }
0735 
0736 static bool
0737 tls_device_rx_resync_async(struct tls_offload_resync_async *resync_async,
0738                s64 resync_req, u32 *seq, u16 *rcd_delta)
0739 {
0740     u32 is_async = resync_req & RESYNC_REQ_ASYNC;
0741     u32 req_seq = resync_req >> 32;
0742     u32 req_end = req_seq + ((resync_req >> 16) & 0xffff);
0743     u16 i;
0744 
0745     *rcd_delta = 0;
0746 
0747     if (is_async) {
0748         /* shouldn't get to wraparound:
0749          * too long in async stage, something bad happened
0750          */
0751         if (WARN_ON_ONCE(resync_async->rcd_delta == USHRT_MAX))
0752             return false;
0753 
0754         /* asynchronous stage: log all headers seq such that
0755          * req_seq <= seq <= end_seq, and wait for real resync request
0756          */
0757         if (before(*seq, req_seq))
0758             return false;
0759         if (!after(*seq, req_end) &&
0760             resync_async->loglen < TLS_DEVICE_RESYNC_ASYNC_LOGMAX)
0761             resync_async->log[resync_async->loglen++] = *seq;
0762 
0763         resync_async->rcd_delta++;
0764 
0765         return false;
0766     }
0767 
0768     /* synchronous stage: check against the logged entries and
0769      * proceed to check the next entries if no match was found
0770      */
0771     for (i = 0; i < resync_async->loglen; i++)
0772         if (req_seq == resync_async->log[i] &&
0773             atomic64_try_cmpxchg(&resync_async->req, &resync_req, 0)) {
0774             *rcd_delta = resync_async->rcd_delta - i;
0775             *seq = req_seq;
0776             resync_async->loglen = 0;
0777             resync_async->rcd_delta = 0;
0778             return true;
0779         }
0780 
0781     resync_async->loglen = 0;
0782     resync_async->rcd_delta = 0;
0783 
0784     if (req_seq == *seq &&
0785         atomic64_try_cmpxchg(&resync_async->req,
0786                  &resync_req, 0))
0787         return true;
0788 
0789     return false;
0790 }
0791 
0792 void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq)
0793 {
0794     struct tls_context *tls_ctx = tls_get_ctx(sk);
0795     struct tls_offload_context_rx *rx_ctx;
0796     u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
0797     u32 sock_data, is_req_pending;
0798     struct tls_prot_info *prot;
0799     s64 resync_req;
0800     u16 rcd_delta;
0801     u32 req_seq;
0802 
0803     if (tls_ctx->rx_conf != TLS_HW)
0804         return;
0805     if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags)))
0806         return;
0807 
0808     prot = &tls_ctx->prot_info;
0809     rx_ctx = tls_offload_ctx_rx(tls_ctx);
0810     memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
0811 
0812     switch (rx_ctx->resync_type) {
0813     case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ:
0814         resync_req = atomic64_read(&rx_ctx->resync_req);
0815         req_seq = resync_req >> 32;
0816         seq += TLS_HEADER_SIZE - 1;
0817         is_req_pending = resync_req;
0818 
0819         if (likely(!is_req_pending) || req_seq != seq ||
0820             !atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))
0821             return;
0822         break;
0823     case TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT:
0824         if (likely(!rx_ctx->resync_nh_do_now))
0825             return;
0826 
0827         /* head of next rec is already in, note that the sock_inq will
0828          * include the currently parsed message when called from parser
0829          */
0830         sock_data = tcp_inq(sk);
0831         if (sock_data > rcd_len) {
0832             trace_tls_device_rx_resync_nh_delay(sk, sock_data,
0833                                 rcd_len);
0834             return;
0835         }
0836 
0837         rx_ctx->resync_nh_do_now = 0;
0838         seq += rcd_len;
0839         tls_bigint_increment(rcd_sn, prot->rec_seq_size);
0840         break;
0841     case TLS_OFFLOAD_SYNC_TYPE_DRIVER_REQ_ASYNC:
0842         resync_req = atomic64_read(&rx_ctx->resync_async->req);
0843         is_req_pending = resync_req;
0844         if (likely(!is_req_pending))
0845             return;
0846 
0847         if (!tls_device_rx_resync_async(rx_ctx->resync_async,
0848                         resync_req, &seq, &rcd_delta))
0849             return;
0850         tls_bigint_subtract(rcd_sn, rcd_delta);
0851         break;
0852     }
0853 
0854     tls_device_resync_rx(tls_ctx, sk, seq, rcd_sn);
0855 }
0856 
0857 static void tls_device_core_ctrl_rx_resync(struct tls_context *tls_ctx,
0858                        struct tls_offload_context_rx *ctx,
0859                        struct sock *sk, struct sk_buff *skb)
0860 {
0861     struct strp_msg *rxm;
0862 
0863     /* device will request resyncs by itself based on stream scan */
0864     if (ctx->resync_type != TLS_OFFLOAD_SYNC_TYPE_CORE_NEXT_HINT)
0865         return;
0866     /* already scheduled */
0867     if (ctx->resync_nh_do_now)
0868         return;
0869     /* seen decrypted fragments since last fully-failed record */
0870     if (ctx->resync_nh_reset) {
0871         ctx->resync_nh_reset = 0;
0872         ctx->resync_nh.decrypted_failed = 1;
0873         ctx->resync_nh.decrypted_tgt = TLS_DEVICE_RESYNC_NH_START_IVAL;
0874         return;
0875     }
0876 
0877     if (++ctx->resync_nh.decrypted_failed <= ctx->resync_nh.decrypted_tgt)
0878         return;
0879 
0880     /* doing resync, bump the next target in case it fails */
0881     if (ctx->resync_nh.decrypted_tgt < TLS_DEVICE_RESYNC_NH_MAX_IVAL)
0882         ctx->resync_nh.decrypted_tgt *= 2;
0883     else
0884         ctx->resync_nh.decrypted_tgt += TLS_DEVICE_RESYNC_NH_MAX_IVAL;
0885 
0886     rxm = strp_msg(skb);
0887 
0888     /* head of next rec is already in, parser will sync for us */
0889     if (tcp_inq(sk) > rxm->full_len) {
0890         trace_tls_device_rx_resync_nh_schedule(sk);
0891         ctx->resync_nh_do_now = 1;
0892     } else {
0893         struct tls_prot_info *prot = &tls_ctx->prot_info;
0894         u8 rcd_sn[TLS_MAX_REC_SEQ_SIZE];
0895 
0896         memcpy(rcd_sn, tls_ctx->rx.rec_seq, prot->rec_seq_size);
0897         tls_bigint_increment(rcd_sn, prot->rec_seq_size);
0898 
0899         tls_device_resync_rx(tls_ctx, sk, tcp_sk(sk)->copied_seq,
0900                      rcd_sn);
0901     }
0902 }
0903 
0904 static int
0905 tls_device_reencrypt(struct sock *sk, struct tls_sw_context_rx *sw_ctx)
0906 {
0907     int err, offset, copy, data_len, pos;
0908     struct sk_buff *skb, *skb_iter;
0909     struct scatterlist sg[1];
0910     struct strp_msg *rxm;
0911     char *orig_buf, *buf;
0912 
0913     rxm = strp_msg(tls_strp_msg(sw_ctx));
0914     orig_buf = kmalloc(rxm->full_len + TLS_HEADER_SIZE +
0915                TLS_CIPHER_AES_GCM_128_IV_SIZE, sk->sk_allocation);
0916     if (!orig_buf)
0917         return -ENOMEM;
0918     buf = orig_buf;
0919 
0920     err = tls_strp_msg_cow(sw_ctx);
0921     if (unlikely(err))
0922         goto free_buf;
0923 
0924     skb = tls_strp_msg(sw_ctx);
0925     rxm = strp_msg(skb);
0926     offset = rxm->offset;
0927 
0928     sg_init_table(sg, 1);
0929     sg_set_buf(&sg[0], buf,
0930            rxm->full_len + TLS_HEADER_SIZE +
0931            TLS_CIPHER_AES_GCM_128_IV_SIZE);
0932     err = skb_copy_bits(skb, offset, buf,
0933                 TLS_HEADER_SIZE + TLS_CIPHER_AES_GCM_128_IV_SIZE);
0934     if (err)
0935         goto free_buf;
0936 
0937     /* We are interested only in the decrypted data not the auth */
0938     err = decrypt_skb(sk, sg);
0939     if (err != -EBADMSG)
0940         goto free_buf;
0941     else
0942         err = 0;
0943 
0944     data_len = rxm->full_len - TLS_CIPHER_AES_GCM_128_TAG_SIZE;
0945 
0946     if (skb_pagelen(skb) > offset) {
0947         copy = min_t(int, skb_pagelen(skb) - offset, data_len);
0948 
0949         if (skb->decrypted) {
0950             err = skb_store_bits(skb, offset, buf, copy);
0951             if (err)
0952                 goto free_buf;
0953         }
0954 
0955         offset += copy;
0956         buf += copy;
0957     }
0958 
0959     pos = skb_pagelen(skb);
0960     skb_walk_frags(skb, skb_iter) {
0961         int frag_pos;
0962 
0963         /* Practically all frags must belong to msg if reencrypt
0964          * is needed with current strparser and coalescing logic,
0965          * but strparser may "get optimized", so let's be safe.
0966          */
0967         if (pos + skb_iter->len <= offset)
0968             goto done_with_frag;
0969         if (pos >= data_len + rxm->offset)
0970             break;
0971 
0972         frag_pos = offset - pos;
0973         copy = min_t(int, skb_iter->len - frag_pos,
0974                  data_len + rxm->offset - offset);
0975 
0976         if (skb_iter->decrypted) {
0977             err = skb_store_bits(skb_iter, frag_pos, buf, copy);
0978             if (err)
0979                 goto free_buf;
0980         }
0981 
0982         offset += copy;
0983         buf += copy;
0984 done_with_frag:
0985         pos += skb_iter->len;
0986     }
0987 
0988 free_buf:
0989     kfree(orig_buf);
0990     return err;
0991 }
0992 
0993 int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx)
0994 {
0995     struct tls_offload_context_rx *ctx = tls_offload_ctx_rx(tls_ctx);
0996     struct tls_sw_context_rx *sw_ctx = tls_sw_ctx_rx(tls_ctx);
0997     struct sk_buff *skb = tls_strp_msg(sw_ctx);
0998     struct strp_msg *rxm = strp_msg(skb);
0999     int is_decrypted = skb->decrypted;
1000     int is_encrypted = !is_decrypted;
1001     struct sk_buff *skb_iter;
1002     int left;
1003 
1004     left = rxm->full_len - skb->len;
1005     /* Check if all the data is decrypted already */
1006     skb_iter = skb_shinfo(skb)->frag_list;
1007     while (skb_iter && left > 0) {
1008         is_decrypted &= skb_iter->decrypted;
1009         is_encrypted &= !skb_iter->decrypted;
1010 
1011         left -= skb_iter->len;
1012         skb_iter = skb_iter->next;
1013     }
1014 
1015     trace_tls_device_decrypted(sk, tcp_sk(sk)->copied_seq - rxm->full_len,
1016                    tls_ctx->rx.rec_seq, rxm->full_len,
1017                    is_encrypted, is_decrypted);
1018 
1019     if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) {
1020         if (likely(is_encrypted || is_decrypted))
1021             return is_decrypted;
1022 
1023         /* After tls_device_down disables the offload, the next SKB will
1024          * likely have initial fragments decrypted, and final ones not
1025          * decrypted. We need to reencrypt that single SKB.
1026          */
1027         return tls_device_reencrypt(sk, sw_ctx);
1028     }
1029 
1030     /* Return immediately if the record is either entirely plaintext or
1031      * entirely ciphertext. Otherwise handle reencrypt partially decrypted
1032      * record.
1033      */
1034     if (is_decrypted) {
1035         ctx->resync_nh_reset = 1;
1036         return is_decrypted;
1037     }
1038     if (is_encrypted) {
1039         tls_device_core_ctrl_rx_resync(tls_ctx, ctx, sk, skb);
1040         return 0;
1041     }
1042 
1043     ctx->resync_nh_reset = 1;
1044     return tls_device_reencrypt(sk, sw_ctx);
1045 }
1046 
1047 static void tls_device_attach(struct tls_context *ctx, struct sock *sk,
1048                   struct net_device *netdev)
1049 {
1050     if (sk->sk_destruct != tls_device_sk_destruct) {
1051         refcount_set(&ctx->refcount, 1);
1052         dev_hold(netdev);
1053         RCU_INIT_POINTER(ctx->netdev, netdev);
1054         spin_lock_irq(&tls_device_lock);
1055         list_add_tail(&ctx->list, &tls_device_list);
1056         spin_unlock_irq(&tls_device_lock);
1057 
1058         ctx->sk_destruct = sk->sk_destruct;
1059         smp_store_release(&sk->sk_destruct, tls_device_sk_destruct);
1060     }
1061 }
1062 
1063 int tls_set_device_offload(struct sock *sk, struct tls_context *ctx)
1064 {
1065     u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
1066     struct tls_context *tls_ctx = tls_get_ctx(sk);
1067     struct tls_prot_info *prot = &tls_ctx->prot_info;
1068     struct tls_record_info *start_marker_record;
1069     struct tls_offload_context_tx *offload_ctx;
1070     struct tls_crypto_info *crypto_info;
1071     struct net_device *netdev;
1072     char *iv, *rec_seq;
1073     struct sk_buff *skb;
1074     __be64 rcd_sn;
1075     int rc;
1076 
1077     if (!ctx)
1078         return -EINVAL;
1079 
1080     if (ctx->priv_ctx_tx)
1081         return -EEXIST;
1082 
1083     netdev = get_netdev_for_sock(sk);
1084     if (!netdev) {
1085         pr_err_ratelimited("%s: netdev not found\n", __func__);
1086         return -EINVAL;
1087     }
1088 
1089     if (!(netdev->features & NETIF_F_HW_TLS_TX)) {
1090         rc = -EOPNOTSUPP;
1091         goto release_netdev;
1092     }
1093 
1094     crypto_info = &ctx->crypto_send.info;
1095     if (crypto_info->version != TLS_1_2_VERSION) {
1096         rc = -EOPNOTSUPP;
1097         goto release_netdev;
1098     }
1099 
1100     switch (crypto_info->cipher_type) {
1101     case TLS_CIPHER_AES_GCM_128:
1102         nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1103         tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
1104         iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
1105         iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
1106         rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
1107         salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
1108         rec_seq =
1109          ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
1110         break;
1111     default:
1112         rc = -EINVAL;
1113         goto release_netdev;
1114     }
1115 
1116     /* Sanity-check the rec_seq_size for stack allocations */
1117     if (rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
1118         rc = -EINVAL;
1119         goto release_netdev;
1120     }
1121 
1122     prot->version = crypto_info->version;
1123     prot->cipher_type = crypto_info->cipher_type;
1124     prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
1125     prot->tag_size = tag_size;
1126     prot->overhead_size = prot->prepend_size + prot->tag_size;
1127     prot->iv_size = iv_size;
1128     prot->salt_size = salt_size;
1129     ctx->tx.iv = kmalloc(iv_size + TLS_CIPHER_AES_GCM_128_SALT_SIZE,
1130                  GFP_KERNEL);
1131     if (!ctx->tx.iv) {
1132         rc = -ENOMEM;
1133         goto release_netdev;
1134     }
1135 
1136     memcpy(ctx->tx.iv + TLS_CIPHER_AES_GCM_128_SALT_SIZE, iv, iv_size);
1137 
1138     prot->rec_seq_size = rec_seq_size;
1139     ctx->tx.rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
1140     if (!ctx->tx.rec_seq) {
1141         rc = -ENOMEM;
1142         goto free_iv;
1143     }
1144 
1145     start_marker_record = kmalloc(sizeof(*start_marker_record), GFP_KERNEL);
1146     if (!start_marker_record) {
1147         rc = -ENOMEM;
1148         goto free_rec_seq;
1149     }
1150 
1151     offload_ctx = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_TX, GFP_KERNEL);
1152     if (!offload_ctx) {
1153         rc = -ENOMEM;
1154         goto free_marker_record;
1155     }
1156 
1157     rc = tls_sw_fallback_init(sk, offload_ctx, crypto_info);
1158     if (rc)
1159         goto free_offload_ctx;
1160 
1161     /* start at rec_seq - 1 to account for the start marker record */
1162     memcpy(&rcd_sn, ctx->tx.rec_seq, sizeof(rcd_sn));
1163     offload_ctx->unacked_record_sn = be64_to_cpu(rcd_sn) - 1;
1164 
1165     start_marker_record->end_seq = tcp_sk(sk)->write_seq;
1166     start_marker_record->len = 0;
1167     start_marker_record->num_frags = 0;
1168 
1169     INIT_WORK(&offload_ctx->destruct_work, tls_device_tx_del_task);
1170     offload_ctx->ctx = ctx;
1171 
1172     INIT_LIST_HEAD(&offload_ctx->records_list);
1173     list_add_tail(&start_marker_record->list, &offload_ctx->records_list);
1174     spin_lock_init(&offload_ctx->lock);
1175     sg_init_table(offload_ctx->sg_tx_data,
1176               ARRAY_SIZE(offload_ctx->sg_tx_data));
1177 
1178     clean_acked_data_enable(inet_csk(sk), &tls_icsk_clean_acked);
1179     ctx->push_pending_record = tls_device_push_pending_record;
1180 
1181     /* TLS offload is greatly simplified if we don't send
1182      * SKBs where only part of the payload needs to be encrypted.
1183      * So mark the last skb in the write queue as end of record.
1184      */
1185     skb = tcp_write_queue_tail(sk);
1186     if (skb)
1187         TCP_SKB_CB(skb)->eor = 1;
1188 
1189     /* Avoid offloading if the device is down
1190      * We don't want to offload new flows after
1191      * the NETDEV_DOWN event
1192      *
1193      * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1194      * handler thus protecting from the device going down before
1195      * ctx was added to tls_device_list.
1196      */
1197     down_read(&device_offload_lock);
1198     if (!(netdev->flags & IFF_UP)) {
1199         rc = -EINVAL;
1200         goto release_lock;
1201     }
1202 
1203     ctx->priv_ctx_tx = offload_ctx;
1204     rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_TX,
1205                          &ctx->crypto_send.info,
1206                          tcp_sk(sk)->write_seq);
1207     trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_TX,
1208                      tcp_sk(sk)->write_seq, rec_seq, rc);
1209     if (rc)
1210         goto release_lock;
1211 
1212     tls_device_attach(ctx, sk, netdev);
1213     up_read(&device_offload_lock);
1214 
1215     /* following this assignment tls_is_sk_tx_device_offloaded
1216      * will return true and the context might be accessed
1217      * by the netdev's xmit function.
1218      */
1219     smp_store_release(&sk->sk_validate_xmit_skb, tls_validate_xmit_skb);
1220     dev_put(netdev);
1221 
1222     return 0;
1223 
1224 release_lock:
1225     up_read(&device_offload_lock);
1226     clean_acked_data_disable(inet_csk(sk));
1227     crypto_free_aead(offload_ctx->aead_send);
1228 free_offload_ctx:
1229     kfree(offload_ctx);
1230     ctx->priv_ctx_tx = NULL;
1231 free_marker_record:
1232     kfree(start_marker_record);
1233 free_rec_seq:
1234     kfree(ctx->tx.rec_seq);
1235 free_iv:
1236     kfree(ctx->tx.iv);
1237 release_netdev:
1238     dev_put(netdev);
1239     return rc;
1240 }
1241 
1242 int tls_set_device_offload_rx(struct sock *sk, struct tls_context *ctx)
1243 {
1244     struct tls12_crypto_info_aes_gcm_128 *info;
1245     struct tls_offload_context_rx *context;
1246     struct net_device *netdev;
1247     int rc = 0;
1248 
1249     if (ctx->crypto_recv.info.version != TLS_1_2_VERSION)
1250         return -EOPNOTSUPP;
1251 
1252     netdev = get_netdev_for_sock(sk);
1253     if (!netdev) {
1254         pr_err_ratelimited("%s: netdev not found\n", __func__);
1255         return -EINVAL;
1256     }
1257 
1258     if (!(netdev->features & NETIF_F_HW_TLS_RX)) {
1259         rc = -EOPNOTSUPP;
1260         goto release_netdev;
1261     }
1262 
1263     /* Avoid offloading if the device is down
1264      * We don't want to offload new flows after
1265      * the NETDEV_DOWN event
1266      *
1267      * device_offload_lock is taken in tls_devices's NETDEV_DOWN
1268      * handler thus protecting from the device going down before
1269      * ctx was added to tls_device_list.
1270      */
1271     down_read(&device_offload_lock);
1272     if (!(netdev->flags & IFF_UP)) {
1273         rc = -EINVAL;
1274         goto release_lock;
1275     }
1276 
1277     context = kzalloc(TLS_OFFLOAD_CONTEXT_SIZE_RX, GFP_KERNEL);
1278     if (!context) {
1279         rc = -ENOMEM;
1280         goto release_lock;
1281     }
1282     context->resync_nh_reset = 1;
1283 
1284     ctx->priv_ctx_rx = context;
1285     rc = tls_set_sw_offload(sk, ctx, 0);
1286     if (rc)
1287         goto release_ctx;
1288 
1289     rc = netdev->tlsdev_ops->tls_dev_add(netdev, sk, TLS_OFFLOAD_CTX_DIR_RX,
1290                          &ctx->crypto_recv.info,
1291                          tcp_sk(sk)->copied_seq);
1292     info = (void *)&ctx->crypto_recv.info;
1293     trace_tls_device_offload_set(sk, TLS_OFFLOAD_CTX_DIR_RX,
1294                      tcp_sk(sk)->copied_seq, info->rec_seq, rc);
1295     if (rc)
1296         goto free_sw_resources;
1297 
1298     tls_device_attach(ctx, sk, netdev);
1299     up_read(&device_offload_lock);
1300 
1301     dev_put(netdev);
1302 
1303     return 0;
1304 
1305 free_sw_resources:
1306     up_read(&device_offload_lock);
1307     tls_sw_free_resources_rx(sk);
1308     down_read(&device_offload_lock);
1309 release_ctx:
1310     ctx->priv_ctx_rx = NULL;
1311 release_lock:
1312     up_read(&device_offload_lock);
1313 release_netdev:
1314     dev_put(netdev);
1315     return rc;
1316 }
1317 
1318 void tls_device_offload_cleanup_rx(struct sock *sk)
1319 {
1320     struct tls_context *tls_ctx = tls_get_ctx(sk);
1321     struct net_device *netdev;
1322 
1323     down_read(&device_offload_lock);
1324     netdev = rcu_dereference_protected(tls_ctx->netdev,
1325                        lockdep_is_held(&device_offload_lock));
1326     if (!netdev)
1327         goto out;
1328 
1329     netdev->tlsdev_ops->tls_dev_del(netdev, tls_ctx,
1330                     TLS_OFFLOAD_CTX_DIR_RX);
1331 
1332     if (tls_ctx->tx_conf != TLS_HW) {
1333         dev_put(netdev);
1334         rcu_assign_pointer(tls_ctx->netdev, NULL);
1335     } else {
1336         set_bit(TLS_RX_DEV_CLOSED, &tls_ctx->flags);
1337     }
1338 out:
1339     up_read(&device_offload_lock);
1340     tls_sw_release_resources_rx(sk);
1341 }
1342 
1343 static int tls_device_down(struct net_device *netdev)
1344 {
1345     struct tls_context *ctx, *tmp;
1346     unsigned long flags;
1347     LIST_HEAD(list);
1348 
1349     /* Request a write lock to block new offload attempts */
1350     down_write(&device_offload_lock);
1351 
1352     spin_lock_irqsave(&tls_device_lock, flags);
1353     list_for_each_entry_safe(ctx, tmp, &tls_device_list, list) {
1354         struct net_device *ctx_netdev =
1355             rcu_dereference_protected(ctx->netdev,
1356                           lockdep_is_held(&device_offload_lock));
1357 
1358         if (ctx_netdev != netdev ||
1359             !refcount_inc_not_zero(&ctx->refcount))
1360             continue;
1361 
1362         list_move(&ctx->list, &list);
1363     }
1364     spin_unlock_irqrestore(&tls_device_lock, flags);
1365 
1366     list_for_each_entry_safe(ctx, tmp, &list, list) {
1367         /* Stop offloaded TX and switch to the fallback.
1368          * tls_is_sk_tx_device_offloaded will return false.
1369          */
1370         WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw);
1371 
1372         /* Stop the RX and TX resync.
1373          * tls_dev_resync must not be called after tls_dev_del.
1374          */
1375         rcu_assign_pointer(ctx->netdev, NULL);
1376 
1377         /* Start skipping the RX resync logic completely. */
1378         set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags);
1379 
1380         /* Sync with inflight packets. After this point:
1381          * TX: no non-encrypted packets will be passed to the driver.
1382          * RX: resync requests from the driver will be ignored.
1383          */
1384         synchronize_net();
1385 
1386         /* Release the offload context on the driver side. */
1387         if (ctx->tx_conf == TLS_HW)
1388             netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1389                             TLS_OFFLOAD_CTX_DIR_TX);
1390         if (ctx->rx_conf == TLS_HW &&
1391             !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags))
1392             netdev->tlsdev_ops->tls_dev_del(netdev, ctx,
1393                             TLS_OFFLOAD_CTX_DIR_RX);
1394 
1395         dev_put(netdev);
1396 
1397         /* Move the context to a separate list for two reasons:
1398          * 1. When the context is deallocated, list_del is called.
1399          * 2. It's no longer an offloaded context, so we don't want to
1400          *    run offload-specific code on this context.
1401          */
1402         spin_lock_irqsave(&tls_device_lock, flags);
1403         list_move_tail(&ctx->list, &tls_device_down_list);
1404         spin_unlock_irqrestore(&tls_device_lock, flags);
1405 
1406         /* Device contexts for RX and TX will be freed in on sk_destruct
1407          * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW.
1408          * Now release the ref taken above.
1409          */
1410         if (refcount_dec_and_test(&ctx->refcount)) {
1411             /* sk_destruct ran after tls_device_down took a ref, and
1412              * it returned early. Complete the destruction here.
1413              */
1414             list_del(&ctx->list);
1415             tls_device_free_ctx(ctx);
1416         }
1417     }
1418 
1419     up_write(&device_offload_lock);
1420 
1421     flush_workqueue(destruct_wq);
1422 
1423     return NOTIFY_DONE;
1424 }
1425 
1426 static int tls_dev_event(struct notifier_block *this, unsigned long event,
1427              void *ptr)
1428 {
1429     struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1430 
1431     if (!dev->tlsdev_ops &&
1432         !(dev->features & (NETIF_F_HW_TLS_RX | NETIF_F_HW_TLS_TX)))
1433         return NOTIFY_DONE;
1434 
1435     switch (event) {
1436     case NETDEV_REGISTER:
1437     case NETDEV_FEAT_CHANGE:
1438         if (netif_is_bond_master(dev))
1439             return NOTIFY_DONE;
1440         if ((dev->features & NETIF_F_HW_TLS_RX) &&
1441             !dev->tlsdev_ops->tls_dev_resync)
1442             return NOTIFY_BAD;
1443 
1444         if  (dev->tlsdev_ops &&
1445              dev->tlsdev_ops->tls_dev_add &&
1446              dev->tlsdev_ops->tls_dev_del)
1447             return NOTIFY_DONE;
1448         else
1449             return NOTIFY_BAD;
1450     case NETDEV_DOWN:
1451         return tls_device_down(dev);
1452     }
1453     return NOTIFY_DONE;
1454 }
1455 
1456 static struct notifier_block tls_dev_notifier = {
1457     .notifier_call  = tls_dev_event,
1458 };
1459 
1460 int __init tls_device_init(void)
1461 {
1462     int err;
1463 
1464     destruct_wq = alloc_workqueue("ktls_device_destruct", 0, 0);
1465     if (!destruct_wq)
1466         return -ENOMEM;
1467 
1468     err = register_netdevice_notifier(&tls_dev_notifier);
1469     if (err)
1470         destroy_workqueue(destruct_wq);
1471 
1472     return err;
1473 }
1474 
1475 void __exit tls_device_cleanup(void)
1476 {
1477     unregister_netdevice_notifier(&tls_dev_notifier);
1478     destroy_workqueue(destruct_wq);
1479     clean_acked_data_flush();
1480 }
[ 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!