Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
0002 
0003 /* Authors: Bernard Metzler <bmt@zurich.ibm.com> */
0004 /*          Fredy Neeser */
0005 /*          Greg Joyce <greg@opengridcomputing.com> */
0006 /* Copyright (c) 2008-2019, IBM Corporation */
0007 /* Copyright (c) 2017, Open Grid Computing, Inc. */
0008 
0009 #include <linux/errno.h>
0010 #include <linux/types.h>
0011 #include <linux/net.h>
0012 #include <linux/inetdevice.h>
0013 #include <net/addrconf.h>
0014 #include <linux/workqueue.h>
0015 #include <net/sock.h>
0016 #include <net/tcp.h>
0017 #include <linux/inet.h>
0018 #include <linux/tcp.h>
0019 
0020 #include <rdma/iw_cm.h>
0021 #include <rdma/ib_verbs.h>
0022 #include <rdma/ib_user_verbs.h>
0023 
0024 #include "siw.h"
0025 #include "siw_cm.h"
0026 
0027 /*
0028  * Set to any combination of
0029  * MPA_V2_RDMA_NO_RTR, MPA_V2_RDMA_READ_RTR, MPA_V2_RDMA_WRITE_RTR
0030  */
0031 static __be16 rtr_type = MPA_V2_RDMA_READ_RTR | MPA_V2_RDMA_WRITE_RTR;
0032 static const bool relaxed_ird_negotiation = true;
0033 
0034 static void siw_cm_llp_state_change(struct sock *s);
0035 static void siw_cm_llp_data_ready(struct sock *s);
0036 static void siw_cm_llp_write_space(struct sock *s);
0037 static void siw_cm_llp_error_report(struct sock *s);
0038 static int siw_cm_upcall(struct siw_cep *cep, enum iw_cm_event_type reason,
0039              int status);
0040 
0041 static void siw_sk_assign_cm_upcalls(struct sock *sk)
0042 {
0043     write_lock_bh(&sk->sk_callback_lock);
0044     sk->sk_state_change = siw_cm_llp_state_change;
0045     sk->sk_data_ready = siw_cm_llp_data_ready;
0046     sk->sk_write_space = siw_cm_llp_write_space;
0047     sk->sk_error_report = siw_cm_llp_error_report;
0048     write_unlock_bh(&sk->sk_callback_lock);
0049 }
0050 
0051 static void siw_sk_save_upcalls(struct sock *sk)
0052 {
0053     struct siw_cep *cep = sk_to_cep(sk);
0054 
0055     write_lock_bh(&sk->sk_callback_lock);
0056     cep->sk_state_change = sk->sk_state_change;
0057     cep->sk_data_ready = sk->sk_data_ready;
0058     cep->sk_write_space = sk->sk_write_space;
0059     cep->sk_error_report = sk->sk_error_report;
0060     write_unlock_bh(&sk->sk_callback_lock);
0061 }
0062 
0063 static void siw_sk_restore_upcalls(struct sock *sk, struct siw_cep *cep)
0064 {
0065     sk->sk_state_change = cep->sk_state_change;
0066     sk->sk_data_ready = cep->sk_data_ready;
0067     sk->sk_write_space = cep->sk_write_space;
0068     sk->sk_error_report = cep->sk_error_report;
0069     sk->sk_user_data = NULL;
0070 }
0071 
0072 static void siw_qp_socket_assoc(struct siw_cep *cep, struct siw_qp *qp)
0073 {
0074     struct socket *s = cep->sock;
0075     struct sock *sk = s->sk;
0076 
0077     write_lock_bh(&sk->sk_callback_lock);
0078 
0079     qp->attrs.sk = s;
0080     sk->sk_data_ready = siw_qp_llp_data_ready;
0081     sk->sk_write_space = siw_qp_llp_write_space;
0082 
0083     write_unlock_bh(&sk->sk_callback_lock);
0084 }
0085 
0086 static void siw_socket_disassoc(struct socket *s)
0087 {
0088     struct sock *sk = s->sk;
0089     struct siw_cep *cep;
0090 
0091     if (sk) {
0092         write_lock_bh(&sk->sk_callback_lock);
0093         cep = sk_to_cep(sk);
0094         if (cep) {
0095             siw_sk_restore_upcalls(sk, cep);
0096             siw_cep_put(cep);
0097         } else {
0098             pr_warn("siw: cannot restore sk callbacks: no ep\n");
0099         }
0100         write_unlock_bh(&sk->sk_callback_lock);
0101     } else {
0102         pr_warn("siw: cannot restore sk callbacks: no sk\n");
0103     }
0104 }
0105 
0106 static void siw_rtr_data_ready(struct sock *sk)
0107 {
0108     struct siw_cep *cep;
0109     struct siw_qp *qp = NULL;
0110     read_descriptor_t rd_desc;
0111 
0112     read_lock(&sk->sk_callback_lock);
0113 
0114     cep = sk_to_cep(sk);
0115     if (!cep) {
0116         WARN(1, "No connection endpoint\n");
0117         goto out;
0118     }
0119     qp = sk_to_qp(sk);
0120 
0121     memset(&rd_desc, 0, sizeof(rd_desc));
0122     rd_desc.arg.data = qp;
0123     rd_desc.count = 1;
0124 
0125     tcp_read_sock(sk, &rd_desc, siw_tcp_rx_data);
0126     /*
0127      * Check if first frame was successfully processed.
0128      * Signal connection full establishment if yes.
0129      * Failed data processing would have already scheduled
0130      * connection drop.
0131      */
0132     if (!qp->rx_stream.rx_suspend)
0133         siw_cm_upcall(cep, IW_CM_EVENT_ESTABLISHED, 0);
0134 out:
0135     read_unlock(&sk->sk_callback_lock);
0136     if (qp)
0137         siw_qp_socket_assoc(cep, qp);
0138 }
0139 
0140 static void siw_sk_assign_rtr_upcalls(struct siw_cep *cep)
0141 {
0142     struct sock *sk = cep->sock->sk;
0143 
0144     write_lock_bh(&sk->sk_callback_lock);
0145     sk->sk_data_ready = siw_rtr_data_ready;
0146     sk->sk_write_space = siw_qp_llp_write_space;
0147     write_unlock_bh(&sk->sk_callback_lock);
0148 }
0149 
0150 static void siw_cep_socket_assoc(struct siw_cep *cep, struct socket *s)
0151 {
0152     cep->sock = s;
0153     siw_cep_get(cep);
0154     s->sk->sk_user_data = cep;
0155 
0156     siw_sk_save_upcalls(s->sk);
0157     siw_sk_assign_cm_upcalls(s->sk);
0158 }
0159 
0160 static struct siw_cep *siw_cep_alloc(struct siw_device *sdev)
0161 {
0162     struct siw_cep *cep = kzalloc(sizeof(*cep), GFP_KERNEL);
0163     unsigned long flags;
0164 
0165     if (!cep)
0166         return NULL;
0167 
0168     INIT_LIST_HEAD(&cep->listenq);
0169     INIT_LIST_HEAD(&cep->devq);
0170     INIT_LIST_HEAD(&cep->work_freelist);
0171 
0172     kref_init(&cep->ref);
0173     cep->state = SIW_EPSTATE_IDLE;
0174     init_waitqueue_head(&cep->waitq);
0175     spin_lock_init(&cep->lock);
0176     cep->sdev = sdev;
0177     cep->enhanced_rdma_conn_est = false;
0178 
0179     spin_lock_irqsave(&sdev->lock, flags);
0180     list_add_tail(&cep->devq, &sdev->cep_list);
0181     spin_unlock_irqrestore(&sdev->lock, flags);
0182 
0183     siw_dbg_cep(cep, "new endpoint\n");
0184     return cep;
0185 }
0186 
0187 static void siw_cm_free_work(struct siw_cep *cep)
0188 {
0189     struct list_head *w, *tmp;
0190     struct siw_cm_work *work;
0191 
0192     list_for_each_safe(w, tmp, &cep->work_freelist) {
0193         work = list_entry(w, struct siw_cm_work, list);
0194         list_del(&work->list);
0195         kfree(work);
0196     }
0197 }
0198 
0199 static void siw_cancel_mpatimer(struct siw_cep *cep)
0200 {
0201     spin_lock_bh(&cep->lock);
0202     if (cep->mpa_timer) {
0203         if (cancel_delayed_work(&cep->mpa_timer->work)) {
0204             siw_cep_put(cep);
0205             kfree(cep->mpa_timer); /* not needed again */
0206         }
0207         cep->mpa_timer = NULL;
0208     }
0209     spin_unlock_bh(&cep->lock);
0210 }
0211 
0212 static void siw_put_work(struct siw_cm_work *work)
0213 {
0214     INIT_LIST_HEAD(&work->list);
0215     spin_lock_bh(&work->cep->lock);
0216     list_add(&work->list, &work->cep->work_freelist);
0217     spin_unlock_bh(&work->cep->lock);
0218 }
0219 
0220 static void siw_cep_set_inuse(struct siw_cep *cep)
0221 {
0222     unsigned long flags;
0223 retry:
0224     spin_lock_irqsave(&cep->lock, flags);
0225 
0226     if (cep->in_use) {
0227         spin_unlock_irqrestore(&cep->lock, flags);
0228         wait_event_interruptible(cep->waitq, !cep->in_use);
0229         if (signal_pending(current))
0230             flush_signals(current);
0231         goto retry;
0232     } else {
0233         cep->in_use = 1;
0234         spin_unlock_irqrestore(&cep->lock, flags);
0235     }
0236 }
0237 
0238 static void siw_cep_set_free(struct siw_cep *cep)
0239 {
0240     unsigned long flags;
0241 
0242     spin_lock_irqsave(&cep->lock, flags);
0243     cep->in_use = 0;
0244     spin_unlock_irqrestore(&cep->lock, flags);
0245 
0246     wake_up(&cep->waitq);
0247 }
0248 
0249 static void __siw_cep_dealloc(struct kref *ref)
0250 {
0251     struct siw_cep *cep = container_of(ref, struct siw_cep, ref);
0252     struct siw_device *sdev = cep->sdev;
0253     unsigned long flags;
0254 
0255     WARN_ON(cep->listen_cep);
0256 
0257     /* kfree(NULL) is safe */
0258     kfree(cep->mpa.pdata);
0259     spin_lock_bh(&cep->lock);
0260     if (!list_empty(&cep->work_freelist))
0261         siw_cm_free_work(cep);
0262     spin_unlock_bh(&cep->lock);
0263 
0264     spin_lock_irqsave(&sdev->lock, flags);
0265     list_del(&cep->devq);
0266     spin_unlock_irqrestore(&sdev->lock, flags);
0267 
0268     siw_dbg_cep(cep, "free endpoint\n");
0269     kfree(cep);
0270 }
0271 
0272 static struct siw_cm_work *siw_get_work(struct siw_cep *cep)
0273 {
0274     struct siw_cm_work *work = NULL;
0275 
0276     spin_lock_bh(&cep->lock);
0277     if (!list_empty(&cep->work_freelist)) {
0278         work = list_entry(cep->work_freelist.next, struct siw_cm_work,
0279                   list);
0280         list_del_init(&work->list);
0281     }
0282     spin_unlock_bh(&cep->lock);
0283     return work;
0284 }
0285 
0286 static int siw_cm_alloc_work(struct siw_cep *cep, int num)
0287 {
0288     struct siw_cm_work *work;
0289 
0290     while (num--) {
0291         work = kmalloc(sizeof(*work), GFP_KERNEL);
0292         if (!work) {
0293             if (!(list_empty(&cep->work_freelist)))
0294                 siw_cm_free_work(cep);
0295             return -ENOMEM;
0296         }
0297         work->cep = cep;
0298         INIT_LIST_HEAD(&work->list);
0299         list_add(&work->list, &cep->work_freelist);
0300     }
0301     return 0;
0302 }
0303 
0304 /*
0305  * siw_cm_upcall()
0306  *
0307  * Upcall to IWCM to inform about async connection events
0308  */
0309 static int siw_cm_upcall(struct siw_cep *cep, enum iw_cm_event_type reason,
0310              int status)
0311 {
0312     struct iw_cm_event event;
0313     struct iw_cm_id *id;
0314 
0315     memset(&event, 0, sizeof(event));
0316     event.status = status;
0317     event.event = reason;
0318 
0319     if (reason == IW_CM_EVENT_CONNECT_REQUEST) {
0320         event.provider_data = cep;
0321         id = cep->listen_cep->cm_id;
0322     } else {
0323         id = cep->cm_id;
0324     }
0325     /* Signal IRD and ORD */
0326     if (reason == IW_CM_EVENT_ESTABLISHED ||
0327         reason == IW_CM_EVENT_CONNECT_REPLY) {
0328         /* Signal negotiated IRD/ORD values we will use */
0329         event.ird = cep->ird;
0330         event.ord = cep->ord;
0331     } else if (reason == IW_CM_EVENT_CONNECT_REQUEST) {
0332         event.ird = cep->ord;
0333         event.ord = cep->ird;
0334     }
0335     /* Signal private data and address information */
0336     if (reason == IW_CM_EVENT_CONNECT_REQUEST ||
0337         reason == IW_CM_EVENT_CONNECT_REPLY) {
0338         u16 pd_len = be16_to_cpu(cep->mpa.hdr.params.pd_len);
0339 
0340         if (pd_len) {
0341             /*
0342              * hand over MPA private data
0343              */
0344             event.private_data_len = pd_len;
0345             event.private_data = cep->mpa.pdata;
0346 
0347             /* Hide MPA V2 IRD/ORD control */
0348             if (cep->enhanced_rdma_conn_est) {
0349                 event.private_data_len -=
0350                     sizeof(struct mpa_v2_data);
0351                 event.private_data +=
0352                     sizeof(struct mpa_v2_data);
0353             }
0354         }
0355         getname_local(cep->sock, &event.local_addr);
0356         getname_peer(cep->sock, &event.remote_addr);
0357     }
0358     siw_dbg_cep(cep, "[QP %u]: reason=%d, status=%d\n",
0359             cep->qp ? qp_id(cep->qp) : UINT_MAX, reason, status);
0360 
0361     return id->event_handler(id, &event);
0362 }
0363 
0364 /*
0365  * siw_qp_cm_drop()
0366  *
0367  * Drops established LLP connection if present and not already
0368  * scheduled for dropping. Called from user context, SQ workqueue
0369  * or receive IRQ. Caller signals if socket can be immediately
0370  * closed (basically, if not in IRQ).
0371  */
0372 void siw_qp_cm_drop(struct siw_qp *qp, int schedule)
0373 {
0374     struct siw_cep *cep = qp->cep;
0375 
0376     qp->rx_stream.rx_suspend = 1;
0377     qp->tx_ctx.tx_suspend = 1;
0378 
0379     if (!qp->cep)
0380         return;
0381 
0382     if (schedule) {
0383         siw_cm_queue_work(cep, SIW_CM_WORK_CLOSE_LLP);
0384     } else {
0385         siw_cep_set_inuse(cep);
0386 
0387         if (cep->state == SIW_EPSTATE_CLOSED) {
0388             siw_dbg_cep(cep, "already closed\n");
0389             goto out;
0390         }
0391         siw_dbg_cep(cep, "immediate close, state %d\n", cep->state);
0392 
0393         if (qp->term_info.valid)
0394             siw_send_terminate(qp);
0395 
0396         if (cep->cm_id) {
0397             switch (cep->state) {
0398             case SIW_EPSTATE_AWAIT_MPAREP:
0399                 siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
0400                           -EINVAL);
0401                 break;
0402 
0403             case SIW_EPSTATE_RDMA_MODE:
0404                 siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
0405                 break;
0406 
0407             case SIW_EPSTATE_IDLE:
0408             case SIW_EPSTATE_LISTENING:
0409             case SIW_EPSTATE_CONNECTING:
0410             case SIW_EPSTATE_AWAIT_MPAREQ:
0411             case SIW_EPSTATE_RECVD_MPAREQ:
0412             case SIW_EPSTATE_CLOSED:
0413             default:
0414                 break;
0415             }
0416             cep->cm_id->rem_ref(cep->cm_id);
0417             cep->cm_id = NULL;
0418             siw_cep_put(cep);
0419         }
0420         cep->state = SIW_EPSTATE_CLOSED;
0421 
0422         if (cep->sock) {
0423             siw_socket_disassoc(cep->sock);
0424             /*
0425              * Immediately close socket
0426              */
0427             sock_release(cep->sock);
0428             cep->sock = NULL;
0429         }
0430         if (cep->qp) {
0431             cep->qp = NULL;
0432             siw_qp_put(qp);
0433         }
0434 out:
0435         siw_cep_set_free(cep);
0436     }
0437 }
0438 
0439 void siw_cep_put(struct siw_cep *cep)
0440 {
0441     WARN_ON(kref_read(&cep->ref) < 1);
0442     kref_put(&cep->ref, __siw_cep_dealloc);
0443 }
0444 
0445 void siw_cep_get(struct siw_cep *cep)
0446 {
0447     kref_get(&cep->ref);
0448 }
0449 
0450 /*
0451  * Expects params->pd_len in host byte order
0452  */
0453 static int siw_send_mpareqrep(struct siw_cep *cep, const void *pdata, u8 pd_len)
0454 {
0455     struct socket *s = cep->sock;
0456     struct mpa_rr *rr = &cep->mpa.hdr;
0457     struct kvec iov[3];
0458     struct msghdr msg;
0459     int rv;
0460     int iovec_num = 0;
0461     int mpa_len;
0462 
0463     memset(&msg, 0, sizeof(msg));
0464 
0465     iov[iovec_num].iov_base = rr;
0466     iov[iovec_num].iov_len = sizeof(*rr);
0467     mpa_len = sizeof(*rr);
0468 
0469     if (cep->enhanced_rdma_conn_est) {
0470         iovec_num++;
0471         iov[iovec_num].iov_base = &cep->mpa.v2_ctrl;
0472         iov[iovec_num].iov_len = sizeof(cep->mpa.v2_ctrl);
0473         mpa_len += sizeof(cep->mpa.v2_ctrl);
0474     }
0475     if (pd_len) {
0476         iovec_num++;
0477         iov[iovec_num].iov_base = (char *)pdata;
0478         iov[iovec_num].iov_len = pd_len;
0479         mpa_len += pd_len;
0480     }
0481     if (cep->enhanced_rdma_conn_est)
0482         pd_len += sizeof(cep->mpa.v2_ctrl);
0483 
0484     rr->params.pd_len = cpu_to_be16(pd_len);
0485 
0486     rv = kernel_sendmsg(s, &msg, iov, iovec_num + 1, mpa_len);
0487 
0488     return rv < 0 ? rv : 0;
0489 }
0490 
0491 /*
0492  * Receive MPA Request/Reply header.
0493  *
0494  * Returns 0 if complete MPA Request/Reply header including
0495  * eventual private data was received. Returns -EAGAIN if
0496  * header was partially received or negative error code otherwise.
0497  *
0498  * Context: May be called in process context only
0499  */
0500 static int siw_recv_mpa_rr(struct siw_cep *cep)
0501 {
0502     struct mpa_rr *hdr = &cep->mpa.hdr;
0503     struct socket *s = cep->sock;
0504     u16 pd_len;
0505     int rcvd, to_rcv;
0506 
0507     if (cep->mpa.bytes_rcvd < sizeof(struct mpa_rr)) {
0508         rcvd = ksock_recv(s, (char *)hdr + cep->mpa.bytes_rcvd,
0509                   sizeof(struct mpa_rr) - cep->mpa.bytes_rcvd,
0510                   0);
0511         if (rcvd <= 0)
0512             return -ECONNABORTED;
0513 
0514         cep->mpa.bytes_rcvd += rcvd;
0515 
0516         if (cep->mpa.bytes_rcvd < sizeof(struct mpa_rr))
0517             return -EAGAIN;
0518 
0519         if (be16_to_cpu(hdr->params.pd_len) > MPA_MAX_PRIVDATA)
0520             return -EPROTO;
0521     }
0522     pd_len = be16_to_cpu(hdr->params.pd_len);
0523 
0524     /*
0525      * At least the MPA Request/Reply header (frame not including
0526      * private data) has been received.
0527      * Receive (or continue receiving) any private data.
0528      */
0529     to_rcv = pd_len - (cep->mpa.bytes_rcvd - sizeof(struct mpa_rr));
0530 
0531     if (!to_rcv) {
0532         /*
0533          * We must have hdr->params.pd_len == 0 and thus received a
0534          * complete MPA Request/Reply frame.
0535          * Check against peer protocol violation.
0536          */
0537         u32 word;
0538 
0539         rcvd = ksock_recv(s, (char *)&word, sizeof(word), MSG_DONTWAIT);
0540         if (rcvd == -EAGAIN)
0541             return 0;
0542 
0543         if (rcvd == 0) {
0544             siw_dbg_cep(cep, "peer EOF\n");
0545             return -EPIPE;
0546         }
0547         if (rcvd < 0) {
0548             siw_dbg_cep(cep, "error: %d\n", rcvd);
0549             return rcvd;
0550         }
0551         siw_dbg_cep(cep, "peer sent extra data: %d\n", rcvd);
0552 
0553         return -EPROTO;
0554     }
0555 
0556     /*
0557      * At this point, we must have hdr->params.pd_len != 0.
0558      * A private data buffer gets allocated if hdr->params.pd_len != 0.
0559      */
0560     if (!cep->mpa.pdata) {
0561         cep->mpa.pdata = kmalloc(pd_len + 4, GFP_KERNEL);
0562         if (!cep->mpa.pdata)
0563             return -ENOMEM;
0564     }
0565     rcvd = ksock_recv(
0566         s, cep->mpa.pdata + cep->mpa.bytes_rcvd - sizeof(struct mpa_rr),
0567         to_rcv + 4, MSG_DONTWAIT);
0568 
0569     if (rcvd < 0)
0570         return rcvd;
0571 
0572     if (rcvd > to_rcv)
0573         return -EPROTO;
0574 
0575     cep->mpa.bytes_rcvd += rcvd;
0576 
0577     if (to_rcv == rcvd) {
0578         siw_dbg_cep(cep, "%d bytes private data received\n", pd_len);
0579         return 0;
0580     }
0581     return -EAGAIN;
0582 }
0583 
0584 /*
0585  * siw_proc_mpareq()
0586  *
0587  * Read MPA Request from socket and signal new connection to IWCM
0588  * if success. Caller must hold lock on corresponding listening CEP.
0589  */
0590 static int siw_proc_mpareq(struct siw_cep *cep)
0591 {
0592     struct mpa_rr *req;
0593     int version, rv;
0594     u16 pd_len;
0595 
0596     rv = siw_recv_mpa_rr(cep);
0597     if (rv)
0598         return rv;
0599 
0600     req = &cep->mpa.hdr;
0601 
0602     version = __mpa_rr_revision(req->params.bits);
0603     pd_len = be16_to_cpu(req->params.pd_len);
0604 
0605     if (version > MPA_REVISION_2)
0606         /* allow for 0, 1, and 2 only */
0607         return -EPROTO;
0608 
0609     if (memcmp(req->key, MPA_KEY_REQ, 16))
0610         return -EPROTO;
0611 
0612     /* Prepare for sending MPA reply */
0613     memcpy(req->key, MPA_KEY_REP, 16);
0614 
0615     if (version == MPA_REVISION_2 &&
0616         (req->params.bits & MPA_RR_FLAG_ENHANCED)) {
0617         /*
0618          * MPA version 2 must signal IRD/ORD values and P2P mode
0619          * in private data if header flag MPA_RR_FLAG_ENHANCED
0620          * is set.
0621          */
0622         if (pd_len < sizeof(struct mpa_v2_data))
0623             goto reject_conn;
0624 
0625         cep->enhanced_rdma_conn_est = true;
0626     }
0627 
0628     /* MPA Markers: currently not supported. Marker TX to be added. */
0629     if (req->params.bits & MPA_RR_FLAG_MARKERS)
0630         goto reject_conn;
0631 
0632     if (req->params.bits & MPA_RR_FLAG_CRC) {
0633         /*
0634          * RFC 5044, page 27: CRC MUST be used if peer requests it.
0635          * siw specific: 'mpa_crc_strict' parameter to reject
0636          * connection with CRC if local CRC off enforced by
0637          * 'mpa_crc_strict' module parameter.
0638          */
0639         if (!mpa_crc_required && mpa_crc_strict)
0640             goto reject_conn;
0641 
0642         /* Enable CRC if requested by module parameter */
0643         if (mpa_crc_required)
0644             req->params.bits |= MPA_RR_FLAG_CRC;
0645     }
0646     if (cep->enhanced_rdma_conn_est) {
0647         struct mpa_v2_data *v2 = (struct mpa_v2_data *)cep->mpa.pdata;
0648 
0649         /*
0650          * Peer requested ORD becomes requested local IRD,
0651          * peer requested IRD becomes requested local ORD.
0652          * IRD and ORD get limited by global maximum values.
0653          */
0654         cep->ord = ntohs(v2->ird) & MPA_IRD_ORD_MASK;
0655         cep->ord = min(cep->ord, SIW_MAX_ORD_QP);
0656         cep->ird = ntohs(v2->ord) & MPA_IRD_ORD_MASK;
0657         cep->ird = min(cep->ird, SIW_MAX_IRD_QP);
0658 
0659         /* May get overwritten by locally negotiated values */
0660         cep->mpa.v2_ctrl.ird = htons(cep->ird);
0661         cep->mpa.v2_ctrl.ord = htons(cep->ord);
0662 
0663         /*
0664          * Support for peer sent zero length Write or Read to
0665          * let local side enter RTS. Writes are preferred.
0666          * Sends would require pre-posting a Receive and are
0667          * not supported.
0668          * Propose zero length Write if none of Read and Write
0669          * is indicated.
0670          */
0671         if (v2->ird & MPA_V2_PEER_TO_PEER) {
0672             cep->mpa.v2_ctrl.ird |= MPA_V2_PEER_TO_PEER;
0673 
0674             if (v2->ord & MPA_V2_RDMA_WRITE_RTR)
0675                 cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_WRITE_RTR;
0676             else if (v2->ord & MPA_V2_RDMA_READ_RTR)
0677                 cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_READ_RTR;
0678             else
0679                 cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_WRITE_RTR;
0680         }
0681     }
0682 
0683     cep->state = SIW_EPSTATE_RECVD_MPAREQ;
0684 
0685     /* Keep reference until IWCM accepts/rejects */
0686     siw_cep_get(cep);
0687     rv = siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REQUEST, 0);
0688     if (rv)
0689         siw_cep_put(cep);
0690 
0691     return rv;
0692 
0693 reject_conn:
0694     siw_dbg_cep(cep, "reject: crc %d:%d:%d, m %d:%d\n",
0695             req->params.bits & MPA_RR_FLAG_CRC ? 1 : 0,
0696             mpa_crc_required, mpa_crc_strict,
0697             req->params.bits & MPA_RR_FLAG_MARKERS ? 1 : 0, 0);
0698 
0699     req->params.bits &= ~MPA_RR_FLAG_MARKERS;
0700     req->params.bits |= MPA_RR_FLAG_REJECT;
0701 
0702     if (!mpa_crc_required && mpa_crc_strict)
0703         req->params.bits &= ~MPA_RR_FLAG_CRC;
0704 
0705     if (pd_len)
0706         kfree(cep->mpa.pdata);
0707 
0708     cep->mpa.pdata = NULL;
0709 
0710     siw_send_mpareqrep(cep, NULL, 0);
0711 
0712     return -EOPNOTSUPP;
0713 }
0714 
0715 static int siw_proc_mpareply(struct siw_cep *cep)
0716 {
0717     struct siw_qp_attrs qp_attrs;
0718     enum siw_qp_attr_mask qp_attr_mask;
0719     struct siw_qp *qp = cep->qp;
0720     struct mpa_rr *rep;
0721     int rv;
0722     u16 rep_ord;
0723     u16 rep_ird;
0724     bool ird_insufficient = false;
0725     enum mpa_v2_ctrl mpa_p2p_mode = MPA_V2_RDMA_NO_RTR;
0726 
0727     rv = siw_recv_mpa_rr(cep);
0728     if (rv)
0729         goto out_err;
0730 
0731     siw_cancel_mpatimer(cep);
0732 
0733     rep = &cep->mpa.hdr;
0734 
0735     if (__mpa_rr_revision(rep->params.bits) > MPA_REVISION_2) {
0736         /* allow for 0, 1,  and 2 only */
0737         rv = -EPROTO;
0738         goto out_err;
0739     }
0740     if (memcmp(rep->key, MPA_KEY_REP, 16)) {
0741         siw_init_terminate(qp, TERM_ERROR_LAYER_LLP, LLP_ETYPE_MPA,
0742                    LLP_ECODE_INVALID_REQ_RESP, 0);
0743         siw_send_terminate(qp);
0744         rv = -EPROTO;
0745         goto out_err;
0746     }
0747     if (rep->params.bits & MPA_RR_FLAG_REJECT) {
0748         siw_dbg_cep(cep, "got mpa reject\n");
0749         siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -ECONNRESET);
0750 
0751         return -ECONNRESET;
0752     }
0753     if (try_gso && rep->params.bits & MPA_RR_FLAG_GSO_EXP) {
0754         siw_dbg_cep(cep, "peer allows GSO on TX\n");
0755         qp->tx_ctx.gso_seg_limit = 0;
0756     }
0757     if ((rep->params.bits & MPA_RR_FLAG_MARKERS) ||
0758         (mpa_crc_required && !(rep->params.bits & MPA_RR_FLAG_CRC)) ||
0759         (mpa_crc_strict && !mpa_crc_required &&
0760          (rep->params.bits & MPA_RR_FLAG_CRC))) {
0761         siw_dbg_cep(cep, "reply unsupp: crc %d:%d:%d, m %d:%d\n",
0762                 rep->params.bits & MPA_RR_FLAG_CRC ? 1 : 0,
0763                 mpa_crc_required, mpa_crc_strict,
0764                 rep->params.bits & MPA_RR_FLAG_MARKERS ? 1 : 0, 0);
0765 
0766         siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED);
0767 
0768         return -EINVAL;
0769     }
0770     if (cep->enhanced_rdma_conn_est) {
0771         struct mpa_v2_data *v2;
0772 
0773         if (__mpa_rr_revision(rep->params.bits) < MPA_REVISION_2 ||
0774             !(rep->params.bits & MPA_RR_FLAG_ENHANCED)) {
0775             /*
0776              * Protocol failure: The responder MUST reply with
0777              * MPA version 2 and MUST set MPA_RR_FLAG_ENHANCED.
0778              */
0779             siw_dbg_cep(cep, "mpa reply error: vers %d, enhcd %d\n",
0780                     __mpa_rr_revision(rep->params.bits),
0781                     rep->params.bits & MPA_RR_FLAG_ENHANCED ?
0782                         1 :
0783                         0);
0784 
0785             siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
0786                       -ECONNRESET);
0787             return -EINVAL;
0788         }
0789         v2 = (struct mpa_v2_data *)cep->mpa.pdata;
0790         rep_ird = ntohs(v2->ird) & MPA_IRD_ORD_MASK;
0791         rep_ord = ntohs(v2->ord) & MPA_IRD_ORD_MASK;
0792 
0793         if (cep->ird < rep_ord &&
0794             (relaxed_ird_negotiation == false ||
0795              rep_ord > cep->sdev->attrs.max_ird)) {
0796             siw_dbg_cep(cep, "ird %d, rep_ord %d, max_ord %d\n",
0797                     cep->ird, rep_ord,
0798                     cep->sdev->attrs.max_ord);
0799             ird_insufficient = true;
0800         }
0801         if (cep->ord > rep_ird && relaxed_ird_negotiation == false) {
0802             siw_dbg_cep(cep, "ord %d, rep_ird %d\n", cep->ord,
0803                     rep_ird);
0804             ird_insufficient = true;
0805         }
0806         /*
0807          * Always report negotiated peer values to user,
0808          * even if IRD/ORD negotiation failed
0809          */
0810         cep->ird = rep_ord;
0811         cep->ord = rep_ird;
0812 
0813         if (ird_insufficient) {
0814             /*
0815              * If the initiator IRD is insuffient for the
0816              * responder ORD, send a TERM.
0817              */
0818             siw_init_terminate(qp, TERM_ERROR_LAYER_LLP,
0819                        LLP_ETYPE_MPA,
0820                        LLP_ECODE_INSUFFICIENT_IRD, 0);
0821             siw_send_terminate(qp);
0822             rv = -ENOMEM;
0823             goto out_err;
0824         }
0825         if (cep->mpa.v2_ctrl_req.ird & MPA_V2_PEER_TO_PEER)
0826             mpa_p2p_mode =
0827                 cep->mpa.v2_ctrl_req.ord &
0828                 (MPA_V2_RDMA_WRITE_RTR | MPA_V2_RDMA_READ_RTR);
0829 
0830         /*
0831          * Check if we requested P2P mode, and if peer agrees
0832          */
0833         if (mpa_p2p_mode != MPA_V2_RDMA_NO_RTR) {
0834             if ((mpa_p2p_mode & v2->ord) == 0) {
0835                 /*
0836                  * We requested RTR mode(s), but the peer
0837                  * did not pick any mode we support.
0838                  */
0839                 siw_dbg_cep(cep,
0840                         "rtr mode:  req %2x, got %2x\n",
0841                         mpa_p2p_mode,
0842                         v2->ord & (MPA_V2_RDMA_WRITE_RTR |
0843                                MPA_V2_RDMA_READ_RTR));
0844 
0845                 siw_init_terminate(qp, TERM_ERROR_LAYER_LLP,
0846                            LLP_ETYPE_MPA,
0847                            LLP_ECODE_NO_MATCHING_RTR,
0848                            0);
0849                 siw_send_terminate(qp);
0850                 rv = -EPROTO;
0851                 goto out_err;
0852             }
0853             mpa_p2p_mode = v2->ord & (MPA_V2_RDMA_WRITE_RTR |
0854                           MPA_V2_RDMA_READ_RTR);
0855         }
0856     }
0857     memset(&qp_attrs, 0, sizeof(qp_attrs));
0858 
0859     if (rep->params.bits & MPA_RR_FLAG_CRC)
0860         qp_attrs.flags = SIW_MPA_CRC;
0861 
0862     qp_attrs.irq_size = cep->ird;
0863     qp_attrs.orq_size = cep->ord;
0864     qp_attrs.sk = cep->sock;
0865     qp_attrs.state = SIW_QP_STATE_RTS;
0866 
0867     qp_attr_mask = SIW_QP_ATTR_STATE | SIW_QP_ATTR_LLP_HANDLE |
0868                SIW_QP_ATTR_ORD | SIW_QP_ATTR_IRD | SIW_QP_ATTR_MPA;
0869 
0870     /* Move socket RX/TX under QP control */
0871     down_write(&qp->state_lock);
0872     if (qp->attrs.state > SIW_QP_STATE_RTR) {
0873         rv = -EINVAL;
0874         up_write(&qp->state_lock);
0875         goto out_err;
0876     }
0877     rv = siw_qp_modify(qp, &qp_attrs, qp_attr_mask);
0878 
0879     siw_qp_socket_assoc(cep, qp);
0880 
0881     up_write(&qp->state_lock);
0882 
0883     /* Send extra RDMA frame to trigger peer RTS if negotiated */
0884     if (mpa_p2p_mode != MPA_V2_RDMA_NO_RTR) {
0885         rv = siw_qp_mpa_rts(qp, mpa_p2p_mode);
0886         if (rv)
0887             goto out_err;
0888     }
0889     if (!rv) {
0890         rv = siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, 0);
0891         if (!rv)
0892             cep->state = SIW_EPSTATE_RDMA_MODE;
0893 
0894         return 0;
0895     }
0896 
0897 out_err:
0898     if (rv != -EAGAIN)
0899         siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -EINVAL);
0900 
0901     return rv;
0902 }
0903 
0904 /*
0905  * siw_accept_newconn - accept an incoming pending connection
0906  *
0907  */
0908 static void siw_accept_newconn(struct siw_cep *cep)
0909 {
0910     struct socket *s = cep->sock;
0911     struct socket *new_s = NULL;
0912     struct siw_cep *new_cep = NULL;
0913     int rv = 0; /* debug only. should disappear */
0914 
0915     if (cep->state != SIW_EPSTATE_LISTENING)
0916         goto error;
0917 
0918     new_cep = siw_cep_alloc(cep->sdev);
0919     if (!new_cep)
0920         goto error;
0921 
0922     /*
0923      * 4: Allocate a sufficient number of work elements
0924      * to allow concurrent handling of local + peer close
0925      * events, MPA header processing + MPA timeout.
0926      */
0927     if (siw_cm_alloc_work(new_cep, 4) != 0)
0928         goto error;
0929 
0930     /*
0931      * Copy saved socket callbacks from listening CEP
0932      * and assign new socket with new CEP
0933      */
0934     new_cep->sk_state_change = cep->sk_state_change;
0935     new_cep->sk_data_ready = cep->sk_data_ready;
0936     new_cep->sk_write_space = cep->sk_write_space;
0937     new_cep->sk_error_report = cep->sk_error_report;
0938 
0939     rv = kernel_accept(s, &new_s, O_NONBLOCK);
0940     if (rv != 0) {
0941         /*
0942          * Connection already aborted by peer..?
0943          */
0944         siw_dbg_cep(cep, "kernel_accept() error: %d\n", rv);
0945         goto error;
0946     }
0947     new_cep->sock = new_s;
0948     siw_cep_get(new_cep);
0949     new_s->sk->sk_user_data = new_cep;
0950 
0951     if (siw_tcp_nagle == false)
0952         tcp_sock_set_nodelay(new_s->sk);
0953     new_cep->state = SIW_EPSTATE_AWAIT_MPAREQ;
0954 
0955     rv = siw_cm_queue_work(new_cep, SIW_CM_WORK_MPATIMEOUT);
0956     if (rv)
0957         goto error;
0958     /*
0959      * See siw_proc_mpareq() etc. for the use of new_cep->listen_cep.
0960      */
0961     new_cep->listen_cep = cep;
0962     siw_cep_get(cep);
0963 
0964     if (atomic_read(&new_s->sk->sk_rmem_alloc)) {
0965         /*
0966          * MPA REQ already queued
0967          */
0968         siw_dbg_cep(cep, "immediate mpa request\n");
0969 
0970         siw_cep_set_inuse(new_cep);
0971         rv = siw_proc_mpareq(new_cep);
0972         if (rv != -EAGAIN) {
0973             siw_cep_put(cep);
0974             new_cep->listen_cep = NULL;
0975             if (rv) {
0976                 siw_cep_set_free(new_cep);
0977                 goto error;
0978             }
0979         }
0980         siw_cep_set_free(new_cep);
0981     }
0982     return;
0983 
0984 error:
0985     if (new_cep)
0986         siw_cep_put(new_cep);
0987 
0988     if (new_s) {
0989         siw_socket_disassoc(new_s);
0990         sock_release(new_s);
0991         new_cep->sock = NULL;
0992     }
0993     siw_dbg_cep(cep, "error %d\n", rv);
0994 }
0995 
0996 static void siw_cm_work_handler(struct work_struct *w)
0997 {
0998     struct siw_cm_work *work;
0999     struct siw_cep *cep;
1000     int release_cep = 0, rv = 0;
1001 
1002     work = container_of(w, struct siw_cm_work, work.work);
1003     cep = work->cep;
1004 
1005     siw_dbg_cep(cep, "[QP %u]: work type: %d, state %d\n",
1006             cep->qp ? qp_id(cep->qp) : UINT_MAX,
1007             work->type, cep->state);
1008 
1009     siw_cep_set_inuse(cep);
1010 
1011     switch (work->type) {
1012     case SIW_CM_WORK_ACCEPT:
1013         siw_accept_newconn(cep);
1014         break;
1015 
1016     case SIW_CM_WORK_READ_MPAHDR:
1017         if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1018             if (cep->listen_cep) {
1019                 siw_cep_set_inuse(cep->listen_cep);
1020 
1021                 if (cep->listen_cep->state ==
1022                     SIW_EPSTATE_LISTENING)
1023                     rv = siw_proc_mpareq(cep);
1024                 else
1025                     rv = -EFAULT;
1026 
1027                 siw_cep_set_free(cep->listen_cep);
1028 
1029                 if (rv != -EAGAIN) {
1030                     siw_cep_put(cep->listen_cep);
1031                     cep->listen_cep = NULL;
1032                     if (rv)
1033                         siw_cep_put(cep);
1034                 }
1035             }
1036         } else if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1037             rv = siw_proc_mpareply(cep);
1038         } else {
1039             /*
1040              * CEP already moved out of MPA handshake.
1041              * any connection management already done.
1042              * silently ignore the mpa packet.
1043              */
1044             if (cep->state == SIW_EPSTATE_RDMA_MODE) {
1045                 cep->sock->sk->sk_data_ready(cep->sock->sk);
1046                 siw_dbg_cep(cep, "already in RDMA mode");
1047             } else {
1048                 siw_dbg_cep(cep, "out of state: %d\n",
1049                         cep->state);
1050             }
1051         }
1052         if (rv && rv != -EAGAIN)
1053             release_cep = 1;
1054         break;
1055 
1056     case SIW_CM_WORK_CLOSE_LLP:
1057         /*
1058          * QP scheduled LLP close
1059          */
1060         if (cep->qp && cep->qp->term_info.valid)
1061             siw_send_terminate(cep->qp);
1062 
1063         if (cep->cm_id)
1064             siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
1065 
1066         release_cep = 1;
1067         break;
1068 
1069     case SIW_CM_WORK_PEER_CLOSE:
1070         if (cep->cm_id) {
1071             if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1072                 /*
1073                  * MPA reply not received, but connection drop
1074                  */
1075                 siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
1076                           -ECONNRESET);
1077             } else if (cep->state == SIW_EPSTATE_RDMA_MODE) {
1078                 /*
1079                  * NOTE: IW_CM_EVENT_DISCONNECT is given just
1080                  *       to transition IWCM into CLOSING.
1081                  */
1082                 siw_cm_upcall(cep, IW_CM_EVENT_DISCONNECT, 0);
1083                 siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
1084             }
1085             /*
1086              * for other states there is no connection
1087              * known to the IWCM.
1088              */
1089         } else {
1090             if (cep->state == SIW_EPSTATE_RECVD_MPAREQ) {
1091                 /*
1092                  * Wait for the ulp/CM to call accept/reject
1093                  */
1094                 siw_dbg_cep(cep,
1095                         "mpa req recvd, wait for ULP\n");
1096             } else if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1097                 /*
1098                  * Socket close before MPA request received.
1099                  */
1100                 siw_dbg_cep(cep, "no mpareq: drop listener\n");
1101                 siw_cep_put(cep->listen_cep);
1102                 cep->listen_cep = NULL;
1103             }
1104         }
1105         release_cep = 1;
1106         break;
1107 
1108     case SIW_CM_WORK_MPATIMEOUT:
1109         cep->mpa_timer = NULL;
1110 
1111         if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1112             /*
1113              * MPA request timed out:
1114              * Hide any partially received private data and signal
1115              * timeout
1116              */
1117             cep->mpa.hdr.params.pd_len = 0;
1118 
1119             if (cep->cm_id)
1120                 siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
1121                           -ETIMEDOUT);
1122             release_cep = 1;
1123 
1124         } else if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1125             /*
1126              * No MPA request received after peer TCP stream setup.
1127              */
1128             if (cep->listen_cep) {
1129                 siw_cep_put(cep->listen_cep);
1130                 cep->listen_cep = NULL;
1131             }
1132             release_cep = 1;
1133         }
1134         break;
1135 
1136     default:
1137         WARN(1, "Undefined CM work type: %d\n", work->type);
1138     }
1139     if (release_cep) {
1140         siw_dbg_cep(cep,
1141                 "release: timer=%s, QP[%u]\n",
1142                 cep->mpa_timer ? "y" : "n",
1143                 cep->qp ? qp_id(cep->qp) : UINT_MAX);
1144 
1145         siw_cancel_mpatimer(cep);
1146 
1147         cep->state = SIW_EPSTATE_CLOSED;
1148 
1149         if (cep->qp) {
1150             struct siw_qp *qp = cep->qp;
1151             /*
1152              * Serialize a potential race with application
1153              * closing the QP and calling siw_qp_cm_drop()
1154              */
1155             siw_qp_get(qp);
1156             siw_cep_set_free(cep);
1157 
1158             siw_qp_llp_close(qp);
1159             siw_qp_put(qp);
1160 
1161             siw_cep_set_inuse(cep);
1162             cep->qp = NULL;
1163             siw_qp_put(qp);
1164         }
1165         if (cep->sock) {
1166             siw_socket_disassoc(cep->sock);
1167             sock_release(cep->sock);
1168             cep->sock = NULL;
1169         }
1170         if (cep->cm_id) {
1171             cep->cm_id->rem_ref(cep->cm_id);
1172             cep->cm_id = NULL;
1173             siw_cep_put(cep);
1174         }
1175     }
1176     siw_cep_set_free(cep);
1177     siw_put_work(work);
1178     siw_cep_put(cep);
1179 }
1180 
1181 static struct workqueue_struct *siw_cm_wq;
1182 
1183 int siw_cm_queue_work(struct siw_cep *cep, enum siw_work_type type)
1184 {
1185     struct siw_cm_work *work = siw_get_work(cep);
1186     unsigned long delay = 0;
1187 
1188     if (!work) {
1189         siw_dbg_cep(cep, "failed with no work available\n");
1190         return -ENOMEM;
1191     }
1192     work->type = type;
1193     work->cep = cep;
1194 
1195     siw_cep_get(cep);
1196 
1197     INIT_DELAYED_WORK(&work->work, siw_cm_work_handler);
1198 
1199     if (type == SIW_CM_WORK_MPATIMEOUT) {
1200         cep->mpa_timer = work;
1201 
1202         if (cep->state == SIW_EPSTATE_AWAIT_MPAREP)
1203             delay = MPAREQ_TIMEOUT;
1204         else
1205             delay = MPAREP_TIMEOUT;
1206     }
1207     siw_dbg_cep(cep, "[QP %u]: work type: %d, timeout %lu\n",
1208             cep->qp ? qp_id(cep->qp) : -1, type, delay);
1209 
1210     queue_delayed_work(siw_cm_wq, &work->work, delay);
1211 
1212     return 0;
1213 }
1214 
1215 static void siw_cm_llp_data_ready(struct sock *sk)
1216 {
1217     struct siw_cep *cep;
1218 
1219     read_lock(&sk->sk_callback_lock);
1220 
1221     cep = sk_to_cep(sk);
1222     if (!cep)
1223         goto out;
1224 
1225     siw_dbg_cep(cep, "state: %d\n", cep->state);
1226 
1227     switch (cep->state) {
1228     case SIW_EPSTATE_RDMA_MODE:
1229     case SIW_EPSTATE_LISTENING:
1230         break;
1231 
1232     case SIW_EPSTATE_AWAIT_MPAREQ:
1233     case SIW_EPSTATE_AWAIT_MPAREP:
1234         siw_cm_queue_work(cep, SIW_CM_WORK_READ_MPAHDR);
1235         break;
1236 
1237     default:
1238         siw_dbg_cep(cep, "unexpected data, state %d\n", cep->state);
1239         break;
1240     }
1241 out:
1242     read_unlock(&sk->sk_callback_lock);
1243 }
1244 
1245 static void siw_cm_llp_write_space(struct sock *sk)
1246 {
1247     struct siw_cep *cep = sk_to_cep(sk);
1248 
1249     if (cep)
1250         siw_dbg_cep(cep, "state: %d\n", cep->state);
1251 }
1252 
1253 static void siw_cm_llp_error_report(struct sock *sk)
1254 {
1255     struct siw_cep *cep = sk_to_cep(sk);
1256 
1257     if (cep) {
1258         siw_dbg_cep(cep, "error %d, socket state: %d, cep state: %d\n",
1259                 sk->sk_err, sk->sk_state, cep->state);
1260         cep->sk_error_report(sk);
1261     }
1262 }
1263 
1264 static void siw_cm_llp_state_change(struct sock *sk)
1265 {
1266     struct siw_cep *cep;
1267     void (*orig_state_change)(struct sock *s);
1268 
1269     read_lock(&sk->sk_callback_lock);
1270 
1271     cep = sk_to_cep(sk);
1272     if (!cep) {
1273         /* endpoint already disassociated */
1274         read_unlock(&sk->sk_callback_lock);
1275         return;
1276     }
1277     orig_state_change = cep->sk_state_change;
1278 
1279     siw_dbg_cep(cep, "state: %d\n", cep->state);
1280 
1281     switch (sk->sk_state) {
1282     case TCP_ESTABLISHED:
1283         /*
1284          * handle accepting socket as special case where only
1285          * new connection is possible
1286          */
1287         siw_cm_queue_work(cep, SIW_CM_WORK_ACCEPT);
1288         break;
1289 
1290     case TCP_CLOSE:
1291     case TCP_CLOSE_WAIT:
1292         if (cep->qp)
1293             cep->qp->tx_ctx.tx_suspend = 1;
1294         siw_cm_queue_work(cep, SIW_CM_WORK_PEER_CLOSE);
1295         break;
1296 
1297     default:
1298         siw_dbg_cep(cep, "unexpected socket state %d\n", sk->sk_state);
1299     }
1300     read_unlock(&sk->sk_callback_lock);
1301     orig_state_change(sk);
1302 }
1303 
1304 static int kernel_bindconnect(struct socket *s, struct sockaddr *laddr,
1305                   struct sockaddr *raddr, bool afonly)
1306 {
1307     int rv, flags = 0;
1308     size_t size = laddr->sa_family == AF_INET ?
1309         sizeof(struct sockaddr_in) : sizeof(struct sockaddr_in6);
1310 
1311     /*
1312      * Make address available again asap.
1313      */
1314     sock_set_reuseaddr(s->sk);
1315 
1316     if (afonly) {
1317         rv = ip6_sock_set_v6only(s->sk);
1318         if (rv)
1319             return rv;
1320     }
1321 
1322     rv = s->ops->bind(s, laddr, size);
1323     if (rv < 0)
1324         return rv;
1325 
1326     rv = s->ops->connect(s, raddr, size, flags);
1327 
1328     return rv < 0 ? rv : 0;
1329 }
1330 
1331 int siw_connect(struct iw_cm_id *id, struct iw_cm_conn_param *params)
1332 {
1333     struct siw_device *sdev = to_siw_dev(id->device);
1334     struct siw_qp *qp;
1335     struct siw_cep *cep = NULL;
1336     struct socket *s = NULL;
1337     struct sockaddr *laddr = (struct sockaddr *)&id->local_addr,
1338             *raddr = (struct sockaddr *)&id->remote_addr;
1339     bool p2p_mode = peer_to_peer, v4 = true;
1340     u16 pd_len = params->private_data_len;
1341     int version = mpa_version, rv;
1342 
1343     if (pd_len > MPA_MAX_PRIVDATA)
1344         return -EINVAL;
1345 
1346     if (params->ird > sdev->attrs.max_ird ||
1347         params->ord > sdev->attrs.max_ord)
1348         return -ENOMEM;
1349 
1350     if (laddr->sa_family == AF_INET6)
1351         v4 = false;
1352     else if (laddr->sa_family != AF_INET)
1353         return -EAFNOSUPPORT;
1354 
1355     /*
1356      * Respect any iwarp port mapping: Use mapped remote address
1357      * if valid. Local address must not be mapped, since siw
1358      * uses kernel TCP stack.
1359      */
1360     if ((v4 && to_sockaddr_in(id->remote_addr).sin_port != 0) ||
1361          to_sockaddr_in6(id->remote_addr).sin6_port != 0)
1362         raddr = (struct sockaddr *)&id->m_remote_addr;
1363 
1364     qp = siw_qp_id2obj(sdev, params->qpn);
1365     if (!qp) {
1366         WARN(1, "[QP %u] does not exist\n", params->qpn);
1367         rv = -EINVAL;
1368         goto error;
1369     }
1370     siw_dbg_qp(qp, "pd_len %d, laddr %pISp, raddr %pISp\n", pd_len, laddr,
1371            raddr);
1372 
1373     rv = sock_create(v4 ? AF_INET : AF_INET6, SOCK_STREAM, IPPROTO_TCP, &s);
1374     if (rv < 0)
1375         goto error;
1376 
1377     /*
1378      * NOTE: For simplification, connect() is called in blocking
1379      * mode. Might be reconsidered for async connection setup at
1380      * TCP level.
1381      */
1382     rv = kernel_bindconnect(s, laddr, raddr, id->afonly);
1383     if (rv != 0) {
1384         siw_dbg_qp(qp, "kernel_bindconnect: error %d\n", rv);
1385         goto error;
1386     }
1387     if (siw_tcp_nagle == false)
1388         tcp_sock_set_nodelay(s->sk);
1389     cep = siw_cep_alloc(sdev);
1390     if (!cep) {
1391         rv = -ENOMEM;
1392         goto error;
1393     }
1394     siw_cep_set_inuse(cep);
1395 
1396     /* Associate QP with CEP */
1397     siw_cep_get(cep);
1398     qp->cep = cep;
1399 
1400     /* siw_qp_get(qp) already done by QP lookup */
1401     cep->qp = qp;
1402 
1403     id->add_ref(id);
1404     cep->cm_id = id;
1405 
1406     /*
1407      * 4: Allocate a sufficient number of work elements
1408      * to allow concurrent handling of local + peer close
1409      * events, MPA header processing + MPA timeout.
1410      */
1411     rv = siw_cm_alloc_work(cep, 4);
1412     if (rv != 0) {
1413         rv = -ENOMEM;
1414         goto error;
1415     }
1416     cep->ird = params->ird;
1417     cep->ord = params->ord;
1418 
1419     if (p2p_mode && cep->ord == 0)
1420         cep->ord = 1;
1421 
1422     cep->state = SIW_EPSTATE_CONNECTING;
1423 
1424     /*
1425      * Associate CEP with socket
1426      */
1427     siw_cep_socket_assoc(cep, s);
1428 
1429     cep->state = SIW_EPSTATE_AWAIT_MPAREP;
1430 
1431     /*
1432      * Set MPA Request bits: CRC if required, no MPA Markers,
1433      * MPA Rev. according to module parameter 'mpa_version', Key 'Request'.
1434      */
1435     cep->mpa.hdr.params.bits = 0;
1436     if (version > MPA_REVISION_2) {
1437         pr_warn("Setting MPA version to %u\n", MPA_REVISION_2);
1438         version = MPA_REVISION_2;
1439         /* Adjust also module parameter */
1440         mpa_version = MPA_REVISION_2;
1441     }
1442     __mpa_rr_set_revision(&cep->mpa.hdr.params.bits, version);
1443 
1444     if (try_gso)
1445         cep->mpa.hdr.params.bits |= MPA_RR_FLAG_GSO_EXP;
1446 
1447     if (mpa_crc_required)
1448         cep->mpa.hdr.params.bits |= MPA_RR_FLAG_CRC;
1449 
1450     /*
1451      * If MPA version == 2:
1452      * o Include ORD and IRD.
1453      * o Indicate peer-to-peer mode, if required by module
1454      *   parameter 'peer_to_peer'.
1455      */
1456     if (version == MPA_REVISION_2) {
1457         cep->enhanced_rdma_conn_est = true;
1458         cep->mpa.hdr.params.bits |= MPA_RR_FLAG_ENHANCED;
1459 
1460         cep->mpa.v2_ctrl.ird = htons(cep->ird);
1461         cep->mpa.v2_ctrl.ord = htons(cep->ord);
1462 
1463         if (p2p_mode) {
1464             cep->mpa.v2_ctrl.ird |= MPA_V2_PEER_TO_PEER;
1465             cep->mpa.v2_ctrl.ord |= rtr_type;
1466         }
1467         /* Remember own P2P mode requested */
1468         cep->mpa.v2_ctrl_req.ird = cep->mpa.v2_ctrl.ird;
1469         cep->mpa.v2_ctrl_req.ord = cep->mpa.v2_ctrl.ord;
1470     }
1471     memcpy(cep->mpa.hdr.key, MPA_KEY_REQ, 16);
1472 
1473     rv = siw_send_mpareqrep(cep, params->private_data, pd_len);
1474     /*
1475      * Reset private data.
1476      */
1477     cep->mpa.hdr.params.pd_len = 0;
1478 
1479     if (rv >= 0) {
1480         rv = siw_cm_queue_work(cep, SIW_CM_WORK_MPATIMEOUT);
1481         if (!rv) {
1482             siw_dbg_cep(cep, "[QP %u]: exit\n", qp_id(qp));
1483             siw_cep_set_free(cep);
1484             return 0;
1485         }
1486     }
1487 error:
1488     siw_dbg(id->device, "failed: %d\n", rv);
1489 
1490     if (cep) {
1491         siw_socket_disassoc(s);
1492         sock_release(s);
1493         cep->sock = NULL;
1494 
1495         cep->qp = NULL;
1496 
1497         cep->cm_id = NULL;
1498         id->rem_ref(id);
1499         siw_cep_put(cep);
1500 
1501         qp->cep = NULL;
1502         siw_cep_put(cep);
1503 
1504         cep->state = SIW_EPSTATE_CLOSED;
1505 
1506         siw_cep_set_free(cep);
1507 
1508         siw_cep_put(cep);
1509 
1510     } else if (s) {
1511         sock_release(s);
1512     }
1513     if (qp)
1514         siw_qp_put(qp);
1515 
1516     return rv;
1517 }
1518 
1519 /*
1520  * siw_accept - Let SoftiWARP accept an RDMA connection request
1521  *
1522  * @id:     New connection management id to be used for accepted
1523  *      connection request
1524  * @params: Connection parameters provided by ULP for accepting connection
1525  *
1526  * Transition QP to RTS state, associate new CM id @id with accepted CEP
1527  * and get prepared for TCP input by installing socket callbacks.
1528  * Then send MPA Reply and generate the "connection established" event.
1529  * Socket callbacks must be installed before sending MPA Reply, because
1530  * the latter may cause a first RDMA message to arrive from the RDMA Initiator
1531  * side very quickly, at which time the socket callbacks must be ready.
1532  */
1533 int siw_accept(struct iw_cm_id *id, struct iw_cm_conn_param *params)
1534 {
1535     struct siw_device *sdev = to_siw_dev(id->device);
1536     struct siw_cep *cep = (struct siw_cep *)id->provider_data;
1537     struct siw_qp *qp;
1538     struct siw_qp_attrs qp_attrs;
1539     int rv, max_priv_data = MPA_MAX_PRIVDATA;
1540     bool wait_for_peer_rts = false;
1541 
1542     siw_cep_set_inuse(cep);
1543     siw_cep_put(cep);
1544 
1545     /* Free lingering inbound private data */
1546     if (cep->mpa.hdr.params.pd_len) {
1547         cep->mpa.hdr.params.pd_len = 0;
1548         kfree(cep->mpa.pdata);
1549         cep->mpa.pdata = NULL;
1550     }
1551     siw_cancel_mpatimer(cep);
1552 
1553     if (cep->state != SIW_EPSTATE_RECVD_MPAREQ) {
1554         siw_dbg_cep(cep, "out of state\n");
1555 
1556         siw_cep_set_free(cep);
1557         siw_cep_put(cep);
1558 
1559         return -ECONNRESET;
1560     }
1561     qp = siw_qp_id2obj(sdev, params->qpn);
1562     if (!qp) {
1563         WARN(1, "[QP %d] does not exist\n", params->qpn);
1564         siw_cep_set_free(cep);
1565         siw_cep_put(cep);
1566 
1567         return -EINVAL;
1568     }
1569     down_write(&qp->state_lock);
1570     if (qp->attrs.state > SIW_QP_STATE_RTR) {
1571         rv = -EINVAL;
1572         up_write(&qp->state_lock);
1573         goto error;
1574     }
1575     siw_dbg_cep(cep, "[QP %d]\n", params->qpn);
1576 
1577     if (try_gso && cep->mpa.hdr.params.bits & MPA_RR_FLAG_GSO_EXP) {
1578         siw_dbg_cep(cep, "peer allows GSO on TX\n");
1579         qp->tx_ctx.gso_seg_limit = 0;
1580     }
1581     if (params->ord > sdev->attrs.max_ord ||
1582         params->ird > sdev->attrs.max_ird) {
1583         siw_dbg_cep(
1584             cep,
1585             "[QP %u]: ord %d (max %d), ird %d (max %d)\n",
1586             qp_id(qp), params->ord, sdev->attrs.max_ord,
1587             params->ird, sdev->attrs.max_ird);
1588         rv = -EINVAL;
1589         up_write(&qp->state_lock);
1590         goto error;
1591     }
1592     if (cep->enhanced_rdma_conn_est)
1593         max_priv_data -= sizeof(struct mpa_v2_data);
1594 
1595     if (params->private_data_len > max_priv_data) {
1596         siw_dbg_cep(
1597             cep,
1598             "[QP %u]: private data length: %d (max %d)\n",
1599             qp_id(qp), params->private_data_len, max_priv_data);
1600         rv = -EINVAL;
1601         up_write(&qp->state_lock);
1602         goto error;
1603     }
1604     if (cep->enhanced_rdma_conn_est) {
1605         if (params->ord > cep->ord) {
1606             if (relaxed_ird_negotiation) {
1607                 params->ord = cep->ord;
1608             } else {
1609                 cep->ird = params->ird;
1610                 cep->ord = params->ord;
1611                 rv = -EINVAL;
1612                 up_write(&qp->state_lock);
1613                 goto error;
1614             }
1615         }
1616         if (params->ird < cep->ird) {
1617             if (relaxed_ird_negotiation &&
1618                 cep->ird <= sdev->attrs.max_ird)
1619                 params->ird = cep->ird;
1620             else {
1621                 rv = -ENOMEM;
1622                 up_write(&qp->state_lock);
1623                 goto error;
1624             }
1625         }
1626         if (cep->mpa.v2_ctrl.ord &
1627             (MPA_V2_RDMA_WRITE_RTR | MPA_V2_RDMA_READ_RTR))
1628             wait_for_peer_rts = true;
1629         /*
1630          * Signal back negotiated IRD and ORD values
1631          */
1632         cep->mpa.v2_ctrl.ord =
1633             htons(params->ord & MPA_IRD_ORD_MASK) |
1634             (cep->mpa.v2_ctrl.ord & ~MPA_V2_MASK_IRD_ORD);
1635         cep->mpa.v2_ctrl.ird =
1636             htons(params->ird & MPA_IRD_ORD_MASK) |
1637             (cep->mpa.v2_ctrl.ird & ~MPA_V2_MASK_IRD_ORD);
1638     }
1639     cep->ird = params->ird;
1640     cep->ord = params->ord;
1641 
1642     cep->cm_id = id;
1643     id->add_ref(id);
1644 
1645     memset(&qp_attrs, 0, sizeof(qp_attrs));
1646     qp_attrs.orq_size = cep->ord;
1647     qp_attrs.irq_size = cep->ird;
1648     qp_attrs.sk = cep->sock;
1649     if (cep->mpa.hdr.params.bits & MPA_RR_FLAG_CRC)
1650         qp_attrs.flags = SIW_MPA_CRC;
1651     qp_attrs.state = SIW_QP_STATE_RTS;
1652 
1653     siw_dbg_cep(cep, "[QP%u]: moving to rts\n", qp_id(qp));
1654 
1655     /* Associate QP with CEP */
1656     siw_cep_get(cep);
1657     qp->cep = cep;
1658 
1659     /* siw_qp_get(qp) already done by QP lookup */
1660     cep->qp = qp;
1661 
1662     cep->state = SIW_EPSTATE_RDMA_MODE;
1663 
1664     /* Move socket RX/TX under QP control */
1665     rv = siw_qp_modify(qp, &qp_attrs,
1666                SIW_QP_ATTR_STATE | SIW_QP_ATTR_LLP_HANDLE |
1667                    SIW_QP_ATTR_ORD | SIW_QP_ATTR_IRD |
1668                    SIW_QP_ATTR_MPA);
1669     up_write(&qp->state_lock);
1670 
1671     if (rv)
1672         goto error;
1673 
1674     siw_dbg_cep(cep, "[QP %u]: send mpa reply, %d byte pdata\n",
1675             qp_id(qp), params->private_data_len);
1676 
1677     rv = siw_send_mpareqrep(cep, params->private_data,
1678                 params->private_data_len);
1679     if (rv != 0)
1680         goto error;
1681 
1682     if (wait_for_peer_rts) {
1683         siw_sk_assign_rtr_upcalls(cep);
1684     } else {
1685         siw_qp_socket_assoc(cep, qp);
1686         rv = siw_cm_upcall(cep, IW_CM_EVENT_ESTABLISHED, 0);
1687         if (rv)
1688             goto error;
1689     }
1690     siw_cep_set_free(cep);
1691 
1692     return 0;
1693 error:
1694     siw_socket_disassoc(cep->sock);
1695     sock_release(cep->sock);
1696     cep->sock = NULL;
1697 
1698     cep->state = SIW_EPSTATE_CLOSED;
1699 
1700     if (cep->cm_id) {
1701         cep->cm_id->rem_ref(id);
1702         cep->cm_id = NULL;
1703     }
1704     if (qp->cep) {
1705         siw_cep_put(cep);
1706         qp->cep = NULL;
1707     }
1708     cep->qp = NULL;
1709     siw_qp_put(qp);
1710 
1711     siw_cep_set_free(cep);
1712     siw_cep_put(cep);
1713 
1714     return rv;
1715 }
1716 
1717 /*
1718  * siw_reject()
1719  *
1720  * Local connection reject case. Send private data back to peer,
1721  * close connection and dereference connection id.
1722  */
1723 int siw_reject(struct iw_cm_id *id, const void *pdata, u8 pd_len)
1724 {
1725     struct siw_cep *cep = (struct siw_cep *)id->provider_data;
1726 
1727     siw_cep_set_inuse(cep);
1728     siw_cep_put(cep);
1729 
1730     siw_cancel_mpatimer(cep);
1731 
1732     if (cep->state != SIW_EPSTATE_RECVD_MPAREQ) {
1733         siw_dbg_cep(cep, "out of state\n");
1734 
1735         siw_cep_set_free(cep);
1736         siw_cep_put(cep); /* put last reference */
1737 
1738         return -ECONNRESET;
1739     }
1740     siw_dbg_cep(cep, "cep->state %d, pd_len %d\n", cep->state,
1741             pd_len);
1742 
1743     if (__mpa_rr_revision(cep->mpa.hdr.params.bits) >= MPA_REVISION_1) {
1744         cep->mpa.hdr.params.bits |= MPA_RR_FLAG_REJECT; /* reject */
1745         siw_send_mpareqrep(cep, pdata, pd_len);
1746     }
1747     siw_socket_disassoc(cep->sock);
1748     sock_release(cep->sock);
1749     cep->sock = NULL;
1750 
1751     cep->state = SIW_EPSTATE_CLOSED;
1752 
1753     siw_cep_set_free(cep);
1754     siw_cep_put(cep);
1755 
1756     return 0;
1757 }
1758 
1759 /*
1760  * siw_create_listen - Create resources for a listener's IWCM ID @id
1761  *
1762  * Starts listen on the socket address id->local_addr.
1763  *
1764  */
1765 int siw_create_listen(struct iw_cm_id *id, int backlog)
1766 {
1767     struct socket *s;
1768     struct siw_cep *cep = NULL;
1769     struct siw_device *sdev = to_siw_dev(id->device);
1770     int addr_family = id->local_addr.ss_family;
1771     int rv = 0;
1772 
1773     if (addr_family != AF_INET && addr_family != AF_INET6)
1774         return -EAFNOSUPPORT;
1775 
1776     rv = sock_create(addr_family, SOCK_STREAM, IPPROTO_TCP, &s);
1777     if (rv < 0)
1778         return rv;
1779 
1780     /*
1781      * Allow binding local port when still in TIME_WAIT from last close.
1782      */
1783     sock_set_reuseaddr(s->sk);
1784 
1785     if (addr_family == AF_INET) {
1786         struct sockaddr_in *laddr = &to_sockaddr_in(id->local_addr);
1787 
1788         /* For wildcard addr, limit binding to current device only */
1789         if (ipv4_is_zeronet(laddr->sin_addr.s_addr))
1790             s->sk->sk_bound_dev_if = sdev->netdev->ifindex;
1791 
1792         rv = s->ops->bind(s, (struct sockaddr *)laddr,
1793                   sizeof(struct sockaddr_in));
1794     } else {
1795         struct sockaddr_in6 *laddr = &to_sockaddr_in6(id->local_addr);
1796 
1797         if (id->afonly) {
1798             rv = ip6_sock_set_v6only(s->sk);
1799             if (rv) {
1800                 siw_dbg(id->device,
1801                     "ip6_sock_set_v6only erro: %d\n", rv);
1802                 goto error;
1803             }
1804         }
1805 
1806         /* For wildcard addr, limit binding to current device only */
1807         if (ipv6_addr_any(&laddr->sin6_addr))
1808             s->sk->sk_bound_dev_if = sdev->netdev->ifindex;
1809 
1810         rv = s->ops->bind(s, (struct sockaddr *)laddr,
1811                   sizeof(struct sockaddr_in6));
1812     }
1813     if (rv) {
1814         siw_dbg(id->device, "socket bind error: %d\n", rv);
1815         goto error;
1816     }
1817     cep = siw_cep_alloc(sdev);
1818     if (!cep) {
1819         rv = -ENOMEM;
1820         goto error;
1821     }
1822     siw_cep_socket_assoc(cep, s);
1823 
1824     rv = siw_cm_alloc_work(cep, backlog);
1825     if (rv) {
1826         siw_dbg(id->device,
1827             "alloc_work error %d, backlog %d\n",
1828             rv, backlog);
1829         goto error;
1830     }
1831     rv = s->ops->listen(s, backlog);
1832     if (rv) {
1833         siw_dbg(id->device, "listen error %d\n", rv);
1834         goto error;
1835     }
1836     cep->cm_id = id;
1837     id->add_ref(id);
1838 
1839     /*
1840      * In case of a wildcard rdma_listen on a multi-homed device,
1841      * a listener's IWCM id is associated with more than one listening CEP.
1842      *
1843      * We currently use id->provider_data in three different ways:
1844      *
1845      * o For a listener's IWCM id, id->provider_data points to
1846      *   the list_head of the list of listening CEPs.
1847      *   Uses: siw_create_listen(), siw_destroy_listen()
1848      *
1849      * o For each accepted passive-side IWCM id, id->provider_data
1850      *   points to the CEP itself. This is a consequence of
1851      *   - siw_cm_upcall() setting event.provider_data = cep and
1852      *   - the IWCM's cm_conn_req_handler() setting provider_data of the
1853      *     new passive-side IWCM id equal to event.provider_data
1854      *   Uses: siw_accept(), siw_reject()
1855      *
1856      * o For an active-side IWCM id, id->provider_data is not used at all.
1857      *
1858      */
1859     if (!id->provider_data) {
1860         id->provider_data =
1861             kmalloc(sizeof(struct list_head), GFP_KERNEL);
1862         if (!id->provider_data) {
1863             rv = -ENOMEM;
1864             goto error;
1865         }
1866         INIT_LIST_HEAD((struct list_head *)id->provider_data);
1867     }
1868     list_add_tail(&cep->listenq, (struct list_head *)id->provider_data);
1869     cep->state = SIW_EPSTATE_LISTENING;
1870 
1871     siw_dbg(id->device, "Listen at laddr %pISp\n", &id->local_addr);
1872 
1873     return 0;
1874 
1875 error:
1876     siw_dbg(id->device, "failed: %d\n", rv);
1877 
1878     if (cep) {
1879         siw_cep_set_inuse(cep);
1880 
1881         if (cep->cm_id) {
1882             cep->cm_id->rem_ref(cep->cm_id);
1883             cep->cm_id = NULL;
1884         }
1885         cep->sock = NULL;
1886         siw_socket_disassoc(s);
1887         cep->state = SIW_EPSTATE_CLOSED;
1888 
1889         siw_cep_set_free(cep);
1890         siw_cep_put(cep);
1891     }
1892     sock_release(s);
1893 
1894     return rv;
1895 }
1896 
1897 static void siw_drop_listeners(struct iw_cm_id *id)
1898 {
1899     struct list_head *p, *tmp;
1900 
1901     /*
1902      * In case of a wildcard rdma_listen on a multi-homed device,
1903      * a listener's IWCM id is associated with more than one listening CEP.
1904      */
1905     list_for_each_safe(p, tmp, (struct list_head *)id->provider_data) {
1906         struct siw_cep *cep = list_entry(p, struct siw_cep, listenq);
1907 
1908         list_del(p);
1909 
1910         siw_dbg_cep(cep, "drop cep, state %d\n", cep->state);
1911 
1912         siw_cep_set_inuse(cep);
1913 
1914         if (cep->cm_id) {
1915             cep->cm_id->rem_ref(cep->cm_id);
1916             cep->cm_id = NULL;
1917         }
1918         if (cep->sock) {
1919             siw_socket_disassoc(cep->sock);
1920             sock_release(cep->sock);
1921             cep->sock = NULL;
1922         }
1923         cep->state = SIW_EPSTATE_CLOSED;
1924         siw_cep_set_free(cep);
1925         siw_cep_put(cep);
1926     }
1927 }
1928 
1929 int siw_destroy_listen(struct iw_cm_id *id)
1930 {
1931     if (!id->provider_data) {
1932         siw_dbg(id->device, "no cep(s)\n");
1933         return 0;
1934     }
1935     siw_drop_listeners(id);
1936     kfree(id->provider_data);
1937     id->provider_data = NULL;
1938 
1939     return 0;
1940 }
1941 
1942 int siw_cm_init(void)
1943 {
1944     /*
1945      * create_single_workqueue for strict ordering
1946      */
1947     siw_cm_wq = create_singlethread_workqueue("siw_cm_wq");
1948     if (!siw_cm_wq)
1949         return -ENOMEM;
1950 
1951     return 0;
1952 }
1953 
1954 void siw_cm_exit(void)
1955 {
1956     if (siw_cm_wq)
1957         destroy_workqueue(siw_cm_wq);
1958 }
[ 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!