drivers/vhost/scsi.c

   1 /*******************************************************************************
   2  * Vhost kernel TCM fabric driver for virtio SCSI initiators
   3  *
   4  * (C) Copyright 2010-2013 Datera, Inc.
   5  * (C) Copyright 2010-2012 IBM Corp.
   6  *
   7  * Licensed to the Linux Foundation under the General Public License (GPL) version 2.
   8  *
   9  * Authors: Nicholas A. Bellinger <nab@daterainc.com>
  10  *          Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2 of the License, or
  15  * (at your option) any later version.
  16  *
  17  * This program is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  * GNU General Public License for more details.
  21  *
  22  ****************************************************************************/
  23
  24 #include <linux/module.h>
  25 #include <linux/moduleparam.h>
  26 #include <generated/utsrelease.h>
  27 #include <linux/utsname.h>
  28 #include <linux/init.h>
  29 #include <linux/slab.h>
  30 #include <linux/kthread.h>
  31 #include <linux/types.h>
  32 #include <linux/string.h>
  33 #include <linux/configfs.h>
  34 #include <linux/ctype.h>
  35 #include <linux/compat.h>
  36 #include <linux/eventfd.h>
  37 #include <linux/fs.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/miscdevice.h>
  40 #include <asm/unaligned.h>
  41 #include <scsi/scsi_common.h>
  42 #include <scsi/scsi_proto.h>
  43 #include <target/target_core_base.h>
  44 #include <target/target_core_fabric.h>
  45 #include <linux/vhost.h>
  46 #include <linux/virtio_scsi.h>
  47 #include <linux/llist.h>
  48 #include <linux/bitmap.h>
  49 #include <linux/percpu_ida.h>
  50
  51 #include "vhost.h"
  52
  53 #define VHOST_SCSI_VERSION  "v0.1"
  54 #define VHOST_SCSI_NAMELEN 256
  55 #define VHOST_SCSI_MAX_CDB_SIZE 32
  56 #define VHOST_SCSI_DEFAULT_TAGS 256
  57 #define VHOST_SCSI_PREALLOC_SGLS 2048
  58 #define VHOST_SCSI_PREALLOC_UPAGES 2048
  59 #define VHOST_SCSI_PREALLOC_PROT_SGLS 512
  60
  61 /* Max number of requests before requeueing the job.
  62  * Using this limit prevents one virtqueue from starving others with
  63  * request.
  64  */
  65 #define VHOST_SCSI_WEIGHT 256
  66
  67 struct vhost_scsi_inflight {
  68         /* Wait for the flush operation to finish */
  69         struct completion comp;
  70         /* Refcount for the inflight reqs */
  71         struct kref kref;
  72 };
  73
  74 struct vhost_scsi_cmd {
  75         /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
  76         int tvc_vq_desc;
  77         /* virtio-scsi initiator task attribute */
  78         int tvc_task_attr;
  79         /* virtio-scsi response incoming iovecs */
  80         int tvc_in_iovs;
  81         /* virtio-scsi initiator data direction */
  82         enum dma_data_direction tvc_data_direction;
  83         /* Expected data transfer length from virtio-scsi header */
  84         u32 tvc_exp_data_len;
  85         /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
  86         u64 tvc_tag;
  87         /* The number of scatterlists associated with this cmd */
  88         u32 tvc_sgl_count;
  89         u32 tvc_prot_sgl_count;
  90         /* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
  91         u32 tvc_lun;
  92         /* Pointer to the SGL formatted memory from virtio-scsi */
  93         struct scatterlist *tvc_sgl;
  94         struct scatterlist *tvc_prot_sgl;
  95         struct page **tvc_upages;
  96         /* Pointer to response header iovec */
  97         struct iovec tvc_resp_iov;
  98         /* Pointer to vhost_scsi for our device */
  99         struct vhost_scsi *tvc_vhost;
 100         /* Pointer to vhost_virtqueue for the cmd */
 101         struct vhost_virtqueue *tvc_vq;
 102         /* Pointer to vhost nexus memory */
 103         struct vhost_scsi_nexus *tvc_nexus;
 104         /* The TCM I/O descriptor that is accessed via container_of() */
 105         struct se_cmd tvc_se_cmd;
 106         /* work item used for cmwq dispatch to vhost_scsi_submission_work() */
 107         struct work_struct work;
 108         /* Copy of the incoming SCSI command descriptor block (CDB) */
 109         unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
 110         /* Sense buffer that will be mapped into outgoing status */
 111         unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
 112         /* Completed commands list, serviced from vhost worker thread */
 113         struct llist_node tvc_completion_list;
 114         /* Used to track inflight cmd */
 115         struct vhost_scsi_inflight *inflight;
 116 };
 117
 118 struct vhost_scsi_nexus {
 119         /* Pointer to TCM session for I_T Nexus */
 120         struct se_session *tvn_se_sess;
 121 };
 122
 123 struct vhost_scsi_tpg {
 124         /* Vhost port target portal group tag for TCM */
 125         u16 tport_tpgt;
 126         /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
 127         int tv_tpg_port_count;
 128         /* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
 129         int tv_tpg_vhost_count;
 130         /* Used for enabling T10-PI with legacy devices */
 131         int tv_fabric_prot_type;
 132         /* list for vhost_scsi_list */
 133         struct list_head tv_tpg_list;
 134         /* Used to protect access for tpg_nexus */
 135         struct mutex tv_tpg_mutex;
 136         /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
 137         struct vhost_scsi_nexus *tpg_nexus;
 138         /* Pointer back to vhost_scsi_tport */
 139         struct vhost_scsi_tport *tport;
 140         /* Returned by vhost_scsi_make_tpg() */
 141         struct se_portal_group se_tpg;
 142         /* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
 143         struct vhost_scsi *vhost_scsi;
 144 };
 145
 146 struct vhost_scsi_tport {
 147         /* SCSI protocol the tport is providing */
 148         u8 tport_proto_id;
 149         /* Binary World Wide unique Port Name for Vhost Target port */
 150         u64 tport_wwpn;
 151         /* ASCII formatted WWPN for Vhost Target port */
 152         char tport_name[VHOST_SCSI_NAMELEN];
 153         /* Returned by vhost_scsi_make_tport() */
 154         struct se_wwn tport_wwn;
 155 };
 156
 157 struct vhost_scsi_evt {
 158         /* event to be sent to guest */
 159         struct virtio_scsi_event event;
 160         /* event list, serviced from vhost worker thread */
 161         struct llist_node list;
 162 };
 163
 164 enum {
 165         VHOST_SCSI_VQ_CTL = 0,
 166         VHOST_SCSI_VQ_EVT = 1,
 167         VHOST_SCSI_VQ_IO = 2,
 168 };
 169
 170 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
 171 enum {
 172         VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
 173                                                (1ULL << VIRTIO_SCSI_F_T10_PI)
 174 };
 175
 176 #define VHOST_SCSI_MAX_TARGET   256
 177 #define VHOST_SCSI_MAX_VQ       128
 178 #define VHOST_SCSI_MAX_EVENT    128
 179
 180 struct vhost_scsi_virtqueue {
 181         struct vhost_virtqueue vq;
 182         /*
 183          * Reference counting for inflight reqs, used for flush operation. At
 184          * each time, one reference tracks new commands submitted, while we
 185          * wait for another one to reach 0.
 186          */
 187         struct vhost_scsi_inflight inflights[2];
 188         /*
 189          * Indicate current inflight in use, protected by vq->mutex.
 190          * Writers must also take dev mutex and flush under it.
 191          */
 192         int inflight_idx;
 193 };
 194
 195 struct vhost_scsi {
 196         /* Protected by vhost_scsi->dev.mutex */
 197         struct vhost_scsi_tpg **vs_tpg;
 198         char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
 199
 200         struct vhost_dev dev;
 201         struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
 202
 203         struct vhost_work vs_completion_work; /* cmd completion work item */
 204         struct llist_head vs_completion_list; /* cmd completion queue */
 205
 206         struct vhost_work vs_event_work; /* evt injection work item */
 207         struct llist_head vs_event_list; /* evt injection queue */
 208
 209         bool vs_events_missed; /* any missed events, protected by vq->mutex */
 210         int vs_events_nr; /* num of pending events, protected by vq->mutex */
 211 };
 212
 213 static struct workqueue_struct *vhost_scsi_workqueue;
 214
 215 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
 216 static DEFINE_MUTEX(vhost_scsi_mutex);
 217 static LIST_HEAD(vhost_scsi_list);
 218
 219 static int iov_num_pages(void __user *iov_base, size_t iov_len)
 220 {
 221         return (PAGE_ALIGN((unsigned long)iov_base + iov_len) -
 222                ((unsigned long)iov_base & PAGE_MASK)) >> PAGE_SHIFT;
 223 }
 224
 225 static void vhost_scsi_done_inflight(struct kref *kref)
 226 {
 227         struct vhost_scsi_inflight *inflight;
 228
 229         inflight = container_of(kref, struct vhost_scsi_inflight, kref);
 230         complete(&inflight->comp);
 231 }
 232
 233 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
 234                                     struct vhost_scsi_inflight *old_inflight[])
 235 {
 236         struct vhost_scsi_inflight *new_inflight;
 237         struct vhost_virtqueue *vq;
 238         int idx, i;
 239
 240         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
 241                 vq = &vs->vqs[i].vq;
 242
 243                 mutex_lock(&vq->mutex);
 244
 245                 /* store old infight */
 246                 idx = vs->vqs[i].inflight_idx;
 247                 if (old_inflight)
 248                         old_inflight[i] = &vs->vqs[i].inflights[idx];
 249
 250                 /* setup new infight */
 251                 vs->vqs[i].inflight_idx = idx ^ 1;
 252                 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
 253                 kref_init(&new_inflight->kref);
 254                 init_completion(&new_inflight->comp);
 255
 256                 mutex_unlock(&vq->mutex);
 257         }
 258 }
 259
 260 static struct vhost_scsi_inflight *
 261 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
 262 {
 263         struct vhost_scsi_inflight *inflight;
 264         struct vhost_scsi_virtqueue *svq;
 265
 266         svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
 267         inflight = &svq->inflights[svq->inflight_idx];
 268         kref_get(&inflight->kref);
 269
 270         return inflight;
 271 }
 272
 273 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
 274 {
 275         kref_put(&inflight->kref, vhost_scsi_done_inflight);
 276 }
 277
 278 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
 279 {
 280         return 1;
 281 }
 282
 283 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
 284 {
 285         return 0;
 286 }
 287
 288 static char *vhost_scsi_get_fabric_name(void)
 289 {
 290         return "vhost";
 291 }
 292
 293 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
 294 {
 295         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 296                                 struct vhost_scsi_tpg, se_tpg);
 297         struct vhost_scsi_tport *tport = tpg->tport;
 298
 299         return &tport->tport_name[0];
 300 }
 301
 302 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
 303 {
 304         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 305                                 struct vhost_scsi_tpg, se_tpg);
 306         return tpg->tport_tpgt;
 307 }
 308
 309 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
 310 {
 311         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 312                                 struct vhost_scsi_tpg, se_tpg);
 313
 314         return tpg->tv_fabric_prot_type;
 315 }
 316
 317 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
 318 {
 319         return 1;
 320 }
 321
 322 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
 323 {
 324         struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
 325                                 struct vhost_scsi_cmd, tvc_se_cmd);
 326         struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
 327         int i;
 328
 329         if (tv_cmd->tvc_sgl_count) {
 330                 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
 331                         put_page(sg_page(&tv_cmd->tvc_sgl[i]));
 332         }
 333         if (tv_cmd->tvc_prot_sgl_count) {
 334                 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
 335                         put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
 336         }
 337
 338         vhost_scsi_put_inflight(tv_cmd->inflight);
 339         percpu_ida_free(&se_sess->sess_tag_pool, se_cmd->map_tag);
 340 }
 341
 342 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
 343 {
 344         return 0;
 345 }
 346
 347 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
 348 {
 349         /* Go ahead and process the write immediately */
 350         target_execute_cmd(se_cmd);
 351         return 0;
 352 }
 353
 354 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
 355 {
 356         return 0;
 357 }
 358
 359 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
 360 {
 361         return;
 362 }
 363
 364 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
 365 {
 366         return 0;
 367 }
 368
 369 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
 370 {
 371         struct vhost_scsi *vs = cmd->tvc_vhost;
 372
 373         llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
 374
 375         vhost_work_queue(&vs->dev, &vs->vs_completion_work);
 376 }
 377
 378 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
 379 {
 380         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 381                                 struct vhost_scsi_cmd, tvc_se_cmd);
 382         vhost_scsi_complete_cmd(cmd);
 383         return 0;
 384 }
 385
 386 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
 387 {
 388         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 389                                 struct vhost_scsi_cmd, tvc_se_cmd);
 390         vhost_scsi_complete_cmd(cmd);
 391         return 0;
 392 }
 393
 394 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
 395 {
 396         return;
 397 }
 398
 399 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
 400 {
 401         return;
 402 }
 403
 404 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 405 {
 406         vs->vs_events_nr--;
 407         kfree(evt);
 408 }
 409
 410 static struct vhost_scsi_evt *
 411 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
 412                        u32 event, u32 reason)
 413 {
 414         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 415         struct vhost_scsi_evt *evt;
 416
 417         if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
 418                 vs->vs_events_missed = true;
 419                 return NULL;
 420         }
 421
 422         evt = kzalloc(sizeof(*evt), GFP_KERNEL);
 423         if (!evt) {
 424                 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
 425                 vs->vs_events_missed = true;
 426                 return NULL;
 427         }
 428
 429         evt->event.event = cpu_to_vhost32(vq, event);
 430         evt->event.reason = cpu_to_vhost32(vq, reason);
 431         vs->vs_events_nr++;
 432
 433         return evt;
 434 }
 435
 436 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
 437 {
 438         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 439
 440         /* TODO locking against target/backend threads? */
 441         transport_generic_free_cmd(se_cmd, 0);
 442
 443 }
 444
 445 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
 446 {
 447         return target_put_sess_cmd(se_cmd);
 448 }
 449
 450 static void
 451 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 452 {
 453         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 454         struct virtio_scsi_event *event = &evt->event;
 455         struct virtio_scsi_event __user *eventp;
 456         unsigned out, in;
 457         int head, ret;
 458
 459         if (!vq->private_data) {
 460                 vs->vs_events_missed = true;
 461                 return;
 462         }
 463
 464 again:
 465         vhost_disable_notify(&vs->dev, vq);
 466         head = vhost_get_vq_desc(vq, vq->iov,
 467                         ARRAY_SIZE(vq->iov), &out, &in,
 468                         NULL, NULL);
 469         if (head < 0) {
 470                 vs->vs_events_missed = true;
 471                 return;
 472         }
 473         if (head == vq->num) {
 474                 if (vhost_enable_notify(&vs->dev, vq))
 475                         goto again;
 476                 vs->vs_events_missed = true;
 477                 return;
 478         }
 479
 480         if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
 481                 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
 482                                 vq->iov[out].iov_len);
 483                 vs->vs_events_missed = true;
 484                 return;
 485         }
 486
 487         if (vs->vs_events_missed) {
 488                 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
 489                 vs->vs_events_missed = false;
 490         }
 491
 492         eventp = vq->iov[out].iov_base;
 493         ret = __copy_to_user(eventp, event, sizeof(*event));
 494         if (!ret)
 495                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 496         else
 497                 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
 498 }
 499
 500 static void vhost_scsi_evt_work(struct vhost_work *work)
 501 {
 502         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 503                                         vs_event_work);
 504         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 505         struct vhost_scsi_evt *evt;
 506         struct llist_node *llnode;
 507
 508         mutex_lock(&vq->mutex);
 509         llnode = llist_del_all(&vs->vs_event_list);
 510         while (llnode) {
 511                 evt = llist_entry(llnode, struct vhost_scsi_evt, list);
 512                 llnode = llist_next(llnode);
 513                 vhost_scsi_do_evt_work(vs, evt);
 514                 vhost_scsi_free_evt(vs, evt);
 515         }
 516         mutex_unlock(&vq->mutex);
 517 }
 518
 519 /* Fill in status and signal that we are done processing this command
 520  *
 521  * This is scheduled in the vhost work queue so we are called with the owner
 522  * process mm and can access the vring.
 523  */
 524 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 525 {
 526         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 527                                         vs_completion_work);
 528         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
 529         struct virtio_scsi_cmd_resp v_rsp;
 530         struct vhost_scsi_cmd *cmd;
 531         struct llist_node *llnode;
 532         struct se_cmd *se_cmd;
 533         struct iov_iter iov_iter;
 534         int ret, vq;
 535
 536         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
 537         llnode = llist_del_all(&vs->vs_completion_list);
 538         while (llnode) {
 539                 cmd = llist_entry(llnode, struct vhost_scsi_cmd,
 540                                      tvc_completion_list);
 541                 llnode = llist_next(llnode);
 542                 se_cmd = &cmd->tvc_se_cmd;
 543
 544                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
 545                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
 546
 547                 memset(&v_rsp, 0, sizeof(v_rsp));
 548                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
 549                 /* TODO is status_qualifier field needed? */
 550                 v_rsp.status = se_cmd->scsi_status;
 551                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
 552                                                  se_cmd->scsi_sense_length);
 553                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
 554                        se_cmd->scsi_sense_length);
 555
 556                 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
 557                               cmd->tvc_in_iovs, sizeof(v_rsp));
 558                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
 559                 if (likely(ret == sizeof(v_rsp))) {
 560                         struct vhost_scsi_virtqueue *q;
 561                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
 562                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
 563                         vq = q - vs->vqs;
 564                         __set_bit(vq, signal);
 565                 } else
 566                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
 567
 568                 vhost_scsi_free_cmd(cmd);
 569         }
 570
 571         vq = -1;
 572         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
 573                 < VHOST_SCSI_MAX_VQ)
 574                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
 575 }
 576
 577 static struct vhost_scsi_cmd *
 578 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
 579                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
 580                    u32 exp_data_len, int data_direction)
 581 {
 582         struct vhost_scsi_cmd *cmd;
 583         struct vhost_scsi_nexus *tv_nexus;
 584         struct se_session *se_sess;
 585         struct scatterlist *sg, *prot_sg;
 586         struct page **pages;
 587         int tag;
 588
 589         tv_nexus = tpg->tpg_nexus;
 590         if (!tv_nexus) {
 591                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
 592                 return ERR_PTR(-EIO);
 593         }
 594         se_sess = tv_nexus->tvn_se_sess;
 595
 596         tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
 597         if (tag < 0) {
 598                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
 599                 return ERR_PTR(-ENOMEM);
 600         }
 601
 602         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
 603         sg = cmd->tvc_sgl;
 604         prot_sg = cmd->tvc_prot_sgl;
 605         pages = cmd->tvc_upages;
 606         memset(cmd, 0, sizeof(struct vhost_scsi_cmd));
 607
 608         cmd->tvc_sgl = sg;
 609         cmd->tvc_prot_sgl = prot_sg;
 610         cmd->tvc_upages = pages;
 611         cmd->tvc_se_cmd.map_tag = tag;
 612         cmd->tvc_tag = scsi_tag;
 613         cmd->tvc_lun = lun;
 614         cmd->tvc_task_attr = task_attr;
 615         cmd->tvc_exp_data_len = exp_data_len;
 616         cmd->tvc_data_direction = data_direction;
 617         cmd->tvc_nexus = tv_nexus;
 618         cmd->inflight = vhost_scsi_get_inflight(vq);
 619
 620         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
 621
 622         return cmd;
 623 }
 624
 625 /*
 626  * Map a user memory range into a scatterlist
 627  *
 628  * Returns the number of scatterlist entries used or -errno on error.
 629  */
 630 static int
 631 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
 632                       void __user *ptr,
 633                       size_t len,
 634                       struct scatterlist *sgl,
 635                       bool write)
 636 {
 637         unsigned int npages = 0, offset, nbytes;
 638         unsigned int pages_nr = iov_num_pages(ptr, len);
 639         struct scatterlist *sg = sgl;
 640         struct page **pages = cmd->tvc_upages;
 641         int ret, i;
 642
 643         if (pages_nr > VHOST_SCSI_PREALLOC_UPAGES) {
 644                 pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
 645                        " preallocated VHOST_SCSI_PREALLOC_UPAGES: %u\n",
 646                         pages_nr, VHOST_SCSI_PREALLOC_UPAGES);
 647                 return -ENOBUFS;
 648         }
 649
 650         ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
 651         /* No pages were pinned */
 652         if (ret < 0)
 653                 goto out;
 654         /* Less pages pinned than wanted */
 655         if (ret != pages_nr) {
 656                 for (i = 0; i < ret; i++)
 657                         put_page(pages[i]);
 658                 ret = -EFAULT;
 659                 goto out;
 660         }
 661
 662         while (len > 0) {
 663                 offset = (uintptr_t)ptr & ~PAGE_MASK;
 664                 nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
 665                 sg_set_page(sg, pages[npages], nbytes, offset);
 666                 ptr += nbytes;
 667                 len -= nbytes;
 668                 sg++;
 669                 npages++;
 670         }
 671
 672 out:
 673         return ret;
 674 }
 675
 676 static int
 677 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
 678 {
 679         int sgl_count = 0;
 680
 681         if (!iter || !iter->iov) {
 682                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
 683                        " present\n", __func__, bytes);
 684                 return -EINVAL;
 685         }
 686
 687         sgl_count = iov_iter_npages(iter, 0xffff);
 688         if (sgl_count > max_sgls) {
 689                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
 690                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
 691                 return -EINVAL;
 692         }
 693         return sgl_count;
 694 }
 695
 696 static int
 697 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
 698                       struct iov_iter *iter,
 699                       struct scatterlist *sg, int sg_count)
 700 {
 701         size_t off = iter->iov_offset;
 702         struct scatterlist *p = sg;
 703         int i, ret;
 704
 705         for (i = 0; i < iter->nr_segs; i++) {
 706                 void __user *base = iter->iov[i].iov_base + off;
 707                 size_t len = iter->iov[i].iov_len - off;
 708
 709                 ret = vhost_scsi_map_to_sgl(cmd, base, len, sg, write);
 710                 if (ret < 0) {
 711                         while (p < sg) {
 712                                 struct page *page = sg_page(p++);
 713                                 if (page)
 714                                         put_page(page);
 715                         }
 716                         return ret;
 717                 }
 718                 sg += ret;
 719                 off = 0;
 720         }
 721         return 0;
 722 }
 723
 724 static int
 725 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
 726                  size_t prot_bytes, struct iov_iter *prot_iter,
 727                  size_t data_bytes, struct iov_iter *data_iter)
 728 {
 729         int sgl_count, ret;
 730         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
 731
 732         if (prot_bytes) {
 733                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
 734                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
 735                 if (sgl_count < 0)
 736                         return sgl_count;
 737
 738                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
 739                 cmd->tvc_prot_sgl_count = sgl_count;
 740                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
 741                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
 742
 743                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
 744                                             cmd->tvc_prot_sgl,
 745                                             cmd->tvc_prot_sgl_count);
 746                 if (ret < 0) {
 747                         cmd->tvc_prot_sgl_count = 0;
 748                         return ret;
 749                 }
 750         }
 751         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
 752                                          VHOST_SCSI_PREALLOC_SGLS);
 753         if (sgl_count < 0)
 754                 return sgl_count;
 755
 756         sg_init_table(cmd->tvc_sgl, sgl_count);
 757         cmd->tvc_sgl_count = sgl_count;
 758         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
 759                   cmd->tvc_sgl, cmd->tvc_sgl_count);
 760
 761         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
 762                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
 763         if (ret < 0) {
 764                 cmd->tvc_sgl_count = 0;
 765                 return ret;
 766         }
 767         return 0;
 768 }
 769
 770 static int vhost_scsi_to_tcm_attr(int attr)
 771 {
 772         switch (attr) {
 773         case VIRTIO_SCSI_S_SIMPLE:
 774                 return TCM_SIMPLE_TAG;
 775         case VIRTIO_SCSI_S_ORDERED:
 776                 return TCM_ORDERED_TAG;
 777         case VIRTIO_SCSI_S_HEAD:
 778                 return TCM_HEAD_TAG;
 779         case VIRTIO_SCSI_S_ACA:
 780                 return TCM_ACA_TAG;
 781         default:
 782                 break;
 783         }
 784         return TCM_SIMPLE_TAG;
 785 }
 786
 787 static void vhost_scsi_submission_work(struct work_struct *work)
 788 {
 789         struct vhost_scsi_cmd *cmd =
 790                 container_of(work, struct vhost_scsi_cmd, work);
 791         struct vhost_scsi_nexus *tv_nexus;
 792         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 793         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
 794         int rc;
 795
 796         /* FIXME: BIDI operation */
 797         if (cmd->tvc_sgl_count) {
 798                 sg_ptr = cmd->tvc_sgl;
 799
 800                 if (cmd->tvc_prot_sgl_count)
 801                         sg_prot_ptr = cmd->tvc_prot_sgl;
 802                 else
 803                         se_cmd->prot_pto = true;
 804         } else {
 805                 sg_ptr = NULL;
 806         }
 807         tv_nexus = cmd->tvc_nexus;
 808
 809         se_cmd->tag = 0;
 810         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
 811                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
 812                         cmd->tvc_lun, cmd->tvc_exp_data_len,
 813                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
 814                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
 815                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
 816                         cmd->tvc_prot_sgl_count);
 817         if (rc < 0) {
 818                 transport_send_check_condition_and_sense(se_cmd,
 819                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
 820                 transport_generic_free_cmd(se_cmd, 0);
 821         }
 822 }
 823
 824 static void
 825 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
 826                            struct vhost_virtqueue *vq,
 827                            int head, unsigned out)
 828 {
 829         struct virtio_scsi_cmd_resp __user *resp;
 830         struct virtio_scsi_cmd_resp rsp;
 831         int ret;
 832
 833         memset(&rsp, 0, sizeof(rsp));
 834         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
 835         resp = vq->iov[out].iov_base;
 836         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
 837         if (!ret)
 838                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 839         else
 840                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
 841 }
 842
 843 static void
 844 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
 845 {
 846         struct vhost_scsi_tpg **vs_tpg, *tpg;
 847         struct virtio_scsi_cmd_req v_req;
 848         struct virtio_scsi_cmd_req_pi v_req_pi;
 849         struct vhost_scsi_cmd *cmd;
 850         struct iov_iter out_iter, in_iter, prot_iter, data_iter;
 851         u64 tag;
 852         u32 exp_data_len, data_direction;
 853         unsigned out, in;
 854         int head, ret, prot_bytes, c = 0;
 855         size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
 856         size_t out_size, in_size;
 857         u16 lun;
 858         u8 *target, *lunp, task_attr;
 859         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
 860         void *req, *cdb;
 861
 862         mutex_lock(&vq->mutex);
 863         /*
 864          * We can handle the vq only after the endpoint is setup by calling the
 865          * VHOST_SCSI_SET_ENDPOINT ioctl.
 866          */
 867         vs_tpg = vq->private_data;
 868         if (!vs_tpg)
 869                 goto out;
 870
 871         vhost_disable_notify(&vs->dev, vq);
 872
 873         do {
 874                 head = vhost_get_vq_desc(vq, vq->iov,
 875                                          ARRAY_SIZE(vq->iov), &out, &in,
 876                                          NULL, NULL);
 877                 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
 878                          head, out, in);
 879                 /* On error, stop handling until the next kick. */
 880                 if (unlikely(head < 0))
 881                         break;
 882                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 883                 if (head == vq->num) {
 884                         if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
 885                                 vhost_disable_notify(&vs->dev, vq);
 886                                 continue;
 887                         }
 888                         break;
 889                 }
 890                 /*
 891                  * Check for a sane response buffer so we can report early
 892                  * errors back to the guest.
 893                  */
 894                 if (unlikely(vq->iov[out].iov_len < rsp_size)) {
 895                         vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
 896                                 " size, got %zu bytes\n", vq->iov[out].iov_len);
 897                         break;
 898                 }
 899                 /*
 900                  * Setup pointers and values based upon different virtio-scsi
 901                  * request header if T10_PI is enabled in KVM guest.
 902                  */
 903                 if (t10_pi) {
 904                         req = &v_req_pi;
 905                         req_size = sizeof(v_req_pi);
 906                         lunp = &v_req_pi.lun[0];
 907                         target = &v_req_pi.lun[1];
 908                 } else {
 909                         req = &v_req;
 910                         req_size = sizeof(v_req);
 911                         lunp = &v_req.lun[0];
 912                         target = &v_req.lun[1];
 913                 }
 914                 /*
 915                  * FIXME: Not correct for BIDI operation
 916                  */
 917                 out_size = iov_length(vq->iov, out);
 918                 in_size = iov_length(&vq->iov[out], in);
 919
 920                 /*
 921                  * Copy over the virtio-scsi request header, which for a
 922                  * ANY_LAYOUT enabled guest may span multiple iovecs, or a
 923                  * single iovec may contain both the header + outgoing
 924                  * WRITE payloads.
 925                  *
 926                  * copy_from_iter() will advance out_iter, so that it will
 927                  * point at the start of the outgoing WRITE payload, if
 928                  * DMA_TO_DEVICE is set.
 929                  */
 930                 iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
 931
 932                 ret = copy_from_iter(req, req_size, &out_iter);
 933                 if (unlikely(ret != req_size)) {
 934                         vq_err(vq, "Faulted on copy_from_iter\n");
 935                         vhost_scsi_send_bad_target(vs, vq, head, out);
 936                         continue;
 937                 }
 938                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
 939                 if (unlikely(*lunp != 1)) {
 940                         vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
 941                         vhost_scsi_send_bad_target(vs, vq, head, out);
 942                         continue;
 943                 }
 944
 945                 tpg = ACCESS_ONCE(vs_tpg[*target]);
 946                 if (unlikely(!tpg)) {
 947                         /* Target does not exist, fail the request */
 948                         vhost_scsi_send_bad_target(vs, vq, head, out);
 949                         continue;
 950                 }
 951                 /*
 952                  * Determine data_direction by calculating the total outgoing
 953                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
 954                  * response headers respectively.
 955                  *
 956                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
 957                  * to the right place.
 958                  *
 959                  * For DMA_FROM_DEVICE, the iovec will be just past the end
 960                  * of the virtio-scsi response header in either the same
 961                  * or immediately following iovec.
 962                  *
 963                  * Any associated T10_PI bytes for the outgoing / incoming
 964                  * payloads are included in calculation of exp_data_len here.
 965                  */
 966                 prot_bytes = 0;
 967
 968                 if (out_size > req_size) {
 969                         data_direction = DMA_TO_DEVICE;
 970                         exp_data_len = out_size - req_size;
 971                         data_iter = out_iter;
 972                 } else if (in_size > rsp_size) {
 973                         data_direction = DMA_FROM_DEVICE;
 974                         exp_data_len = in_size - rsp_size;
 975
 976                         iov_iter_init(&in_iter, READ, &vq->iov[out], in,
 977                                       rsp_size + exp_data_len);
 978                         iov_iter_advance(&in_iter, rsp_size);
 979                         data_iter = in_iter;
 980                 } else {
 981                         data_direction = DMA_NONE;
 982                         exp_data_len = 0;
 983                 }
 984                 /*
 985                  * If T10_PI header + payload is present, setup prot_iter values
 986                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
 987                  * host scatterlists via get_user_pages_fast().
 988                  */
 989                 if (t10_pi) {
 990                         if (v_req_pi.pi_bytesout) {
 991                                 if (data_direction != DMA_TO_DEVICE) {
 992                                         vq_err(vq, "Received non zero pi_bytesout,"
 993                                                 " but wrong data_direction\n");
 994                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 995                                         continue;
 996                                 }
 997                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
 998                         } else if (v_req_pi.pi_bytesin) {
 999                                 if (data_direction != DMA_FROM_DEVICE) {
1000                                         vq_err(vq, "Received non zero pi_bytesin,"
1001                                                 " but wrong data_direction\n");
1002                                         vhost_scsi_send_bad_target(vs, vq, head, out);
1003                                         continue;
1004                                 }
1005                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1006                         }
1007                         /*
1008                          * Set prot_iter to data_iter and truncate it to
1009                          * prot_bytes, and advance data_iter past any
1010                          * preceeding prot_bytes that may be present.
1011                          *
1012                          * Also fix up the exp_data_len to reflect only the
1013                          * actual data payload length.
1014                          */
1015                         if (prot_bytes) {
1016                                 exp_data_len -= prot_bytes;
1017                                 prot_iter = data_iter;
1018                                 iov_iter_truncate(&prot_iter, prot_bytes);
1019                                 iov_iter_advance(&data_iter, prot_bytes);
1020                         }
1021                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
1022                         task_attr = v_req_pi.task_attr;
1023                         cdb = &v_req_pi.cdb[0];
1024                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1025                 } else {
1026                         tag = vhost64_to_cpu(vq, v_req.tag);
1027                         task_attr = v_req.task_attr;
1028                         cdb = &v_req.cdb[0];
1029                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1030                 }
1031                 /*
1032                  * Check that the received CDB size does not exceeded our
1033                  * hardcoded max for vhost-scsi, then get a pre-allocated
1034                  * cmd descriptor for the new virtio-scsi tag.
1035                  *
1036                  * TODO what if cdb was too small for varlen cdb header?
1037                  */
1038                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1039                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
1040                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1041                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1042                         vhost_scsi_send_bad_target(vs, vq, head, out);
1043                         continue;
1044                 }
1045                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1046                                          exp_data_len + prot_bytes,
1047                                          data_direction);
1048                 if (IS_ERR(cmd)) {
1049                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1050                                PTR_ERR(cmd));
1051                         vhost_scsi_send_bad_target(vs, vq, head, out);
1052                         continue;
1053                 }
1054                 cmd->tvc_vhost = vs;
1055                 cmd->tvc_vq = vq;
1056                 cmd->tvc_resp_iov = vq->iov[out];
1057                 cmd->tvc_in_iovs = in;
1058
1059                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1060                          cmd->tvc_cdb[0], cmd->tvc_lun);
1061                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1062                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1063
1064                 if (data_direction != DMA_NONE) {
1065                         ret = vhost_scsi_mapal(cmd,
1066                                                prot_bytes, &prot_iter,
1067                                                exp_data_len, &data_iter);
1068                         if (unlikely(ret)) {
1069                                 vq_err(vq, "Failed to map iov to sgl\n");
1070                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1071                                 vhost_scsi_send_bad_target(vs, vq, head, out);
1072                                 continue;
1073                         }
1074                 }
1075                 /*
1076                  * Save the descriptor from vhost_get_vq_desc() to be used to
1077                  * complete the virtio-scsi request in TCM callback context via
1078                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1079                  */
1080                 cmd->tvc_vq_desc = head;
1081                 /*
1082                  * Dispatch cmd descriptor for cmwq execution in process
1083                  * context provided by vhost_scsi_workqueue.  This also ensures
1084                  * cmd is executed on the same kworker CPU as this vhost
1085                  * thread to gain positive L2 cache locality effects.
1086                  */
1087                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1088                 queue_work(vhost_scsi_workqueue, &cmd->work);
1089         } while (likely(!vhost_exceeds_weight(vq, ++c, 0)));
1090 out:
1091         mutex_unlock(&vq->mutex);
1092 }
1093
1094 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1095 {
1096         pr_debug("%s: The handling func for control queue.\n", __func__);
1097 }
1098
1099 static void
1100 vhost_scsi_send_evt(struct vhost_scsi *vs,
1101                    struct vhost_scsi_tpg *tpg,
1102                    struct se_lun *lun,
1103                    u32 event,
1104                    u32 reason)
1105 {
1106         struct vhost_scsi_evt *evt;
1107
1108         evt = vhost_scsi_allocate_evt(vs, event, reason);
1109         if (!evt)
1110                 return;
1111
1112         if (tpg && lun) {
1113                 /* TODO: share lun setup code with virtio-scsi.ko */
1114                 /*
1115                  * Note: evt->event is zeroed when we allocate it and
1116                  * lun[4-7] need to be zero according to virtio-scsi spec.
1117                  */
1118                 evt->event.lun[0] = 0x01;
1119                 evt->event.lun[1] = tpg->tport_tpgt;
1120                 if (lun->unpacked_lun >= 256)
1121                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1122                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1123         }
1124
1125         llist_add(&evt->list, &vs->vs_event_list);
1126         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1127 }
1128
1129 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1130 {
1131         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1132                                                 poll.work);
1133         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1134
1135         mutex_lock(&vq->mutex);
1136         if (!vq->private_data)
1137                 goto out;
1138
1139         if (vs->vs_events_missed)
1140                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1141 out:
1142         mutex_unlock(&vq->mutex);
1143 }
1144
1145 static void vhost_scsi_handle_kick(struct vhost_work *work)
1146 {
1147         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1148                                                 poll.work);
1149         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1150
1151         vhost_scsi_handle_vq(vs, vq);
1152 }
1153
1154 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1155 {
1156         vhost_poll_flush(&vs->vqs[index].vq.poll);
1157 }
1158
1159 /* Callers must hold dev mutex */
1160 static void vhost_scsi_flush(struct vhost_scsi *vs)
1161 {
1162         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1163         int i;
1164
1165         /* Init new inflight and remember the old inflight */
1166         vhost_scsi_init_inflight(vs, old_inflight);
1167
1168         /*
1169          * The inflight->kref was initialized to 1. We decrement it here to
1170          * indicate the start of the flush operation so that it will reach 0
1171          * when all the reqs are finished.
1172          */
1173         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1174                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1175
1176         /* Flush both the vhost poll and vhost work */
1177         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1178                 vhost_scsi_flush_vq(vs, i);
1179         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1180         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1181
1182         /* Wait for all reqs issued before the flush to be finished */
1183         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1184                 wait_for_completion(&old_inflight[i]->comp);
1185 }
1186
1187 /*
1188  * Called from vhost_scsi_ioctl() context to walk the list of available
1189  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1190  *
1191  *  The lock nesting rule is:
1192  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1193  */
1194 static int
1195 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1196                         struct vhost_scsi_target *t)
1197 {
1198         struct se_portal_group *se_tpg;
1199         struct vhost_scsi_tport *tv_tport;
1200         struct vhost_scsi_tpg *tpg;
1201         struct vhost_scsi_tpg **vs_tpg;
1202         struct vhost_virtqueue *vq;
1203         int index, ret, i, len;
1204         bool match = false;
1205
1206         mutex_lock(&vhost_scsi_mutex);
1207         mutex_lock(&vs->dev.mutex);
1208
1209         /* Verify that ring has been setup correctly. */
1210         for (index = 0; index < vs->dev.nvqs; ++index) {
1211                 /* Verify that ring has been setup correctly. */
1212                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1213                         ret = -EFAULT;
1214                         goto out;
1215                 }
1216         }
1217
1218         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1219         vs_tpg = kzalloc(len, GFP_KERNEL);
1220         if (!vs_tpg) {
1221                 ret = -ENOMEM;
1222                 goto out;
1223         }
1224         if (vs->vs_tpg)
1225                 memcpy(vs_tpg, vs->vs_tpg, len);
1226
1227         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1228                 mutex_lock(&tpg->tv_tpg_mutex);
1229                 if (!tpg->tpg_nexus) {
1230                         mutex_unlock(&tpg->tv_tpg_mutex);
1231                         continue;
1232                 }
1233                 if (tpg->tv_tpg_vhost_count != 0) {
1234                         mutex_unlock(&tpg->tv_tpg_mutex);
1235                         continue;
1236                 }
1237                 tv_tport = tpg->tport;
1238
1239                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1240                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1241                                 kfree(vs_tpg);
1242                                 mutex_unlock(&tpg->tv_tpg_mutex);
1243                                 ret = -EEXIST;
1244                                 goto out;
1245                         }
1246                         /*
1247                          * In order to ensure individual vhost-scsi configfs
1248                          * groups cannot be removed while in use by vhost ioctl,
1249                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1250                          * dependency now.
1251                          */
1252                         se_tpg = &tpg->se_tpg;
1253                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1254                         if (ret) {
1255                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1256                                 kfree(vs_tpg);
1257                                 mutex_unlock(&tpg->tv_tpg_mutex);
1258                                 goto out;
1259                         }
1260                         tpg->tv_tpg_vhost_count++;
1261                         tpg->vhost_scsi = vs;
1262                         vs_tpg[tpg->tport_tpgt] = tpg;
1263                         smp_mb__after_atomic();
1264                         match = true;
1265                 }
1266                 mutex_unlock(&tpg->tv_tpg_mutex);
1267         }
1268
1269         if (match) {
1270                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1271                        sizeof(vs->vs_vhost_wwpn));
1272                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1273                         vq = &vs->vqs[i].vq;
1274                         mutex_lock(&vq->mutex);
1275                         vq->private_data = vs_tpg;
1276                         vhost_vq_init_access(vq);
1277                         mutex_unlock(&vq->mutex);
1278                 }
1279                 ret = 0;
1280         } else {
1281                 ret = -EEXIST;
1282         }
1283
1284         /*
1285          * Act as synchronize_rcu to make sure access to
1286          * old vs->vs_tpg is finished.
1287          */
1288         vhost_scsi_flush(vs);
1289         kfree(vs->vs_tpg);
1290         vs->vs_tpg = vs_tpg;
1291
1292 out:
1293         mutex_unlock(&vs->dev.mutex);
1294         mutex_unlock(&vhost_scsi_mutex);
1295         return ret;
1296 }
1297
1298 static int
1299 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1300                           struct vhost_scsi_target *t)
1301 {
1302         struct se_portal_group *se_tpg;
1303         struct vhost_scsi_tport *tv_tport;
1304         struct vhost_scsi_tpg *tpg;
1305         struct vhost_virtqueue *vq;
1306         bool match = false;
1307         int index, ret, i;
1308         u8 target;
1309
1310         mutex_lock(&vhost_scsi_mutex);
1311         mutex_lock(&vs->dev.mutex);
1312         /* Verify that ring has been setup correctly. */
1313         for (index = 0; index < vs->dev.nvqs; ++index) {
1314                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1315                         ret = -EFAULT;
1316                         goto err_dev;
1317                 }
1318         }
1319
1320         if (!vs->vs_tpg) {
1321                 ret = 0;
1322                 goto err_dev;
1323         }
1324
1325         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1326                 target = i;
1327                 tpg = vs->vs_tpg[target];
1328                 if (!tpg)
1329                         continue;
1330
1331                 mutex_lock(&tpg->tv_tpg_mutex);
1332                 tv_tport = tpg->tport;
1333                 if (!tv_tport) {
1334                         ret = -ENODEV;
1335                         goto err_tpg;
1336                 }
1337
1338                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1339                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1340                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1341                                 tv_tport->tport_name, tpg->tport_tpgt,
1342                                 t->vhost_wwpn, t->vhost_tpgt);
1343                         ret = -EINVAL;
1344                         goto err_tpg;
1345                 }
1346                 tpg->tv_tpg_vhost_count--;
1347                 tpg->vhost_scsi = NULL;
1348                 vs->vs_tpg[target] = NULL;
1349                 match = true;
1350                 mutex_unlock(&tpg->tv_tpg_mutex);
1351                 /*
1352                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1353                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1354                  */
1355                 se_tpg = &tpg->se_tpg;
1356                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1357         }
1358         if (match) {
1359                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1360                         vq = &vs->vqs[i].vq;
1361                         mutex_lock(&vq->mutex);
1362                         vq->private_data = NULL;
1363                         mutex_unlock(&vq->mutex);
1364                 }
1365         }
1366         /*
1367          * Act as synchronize_rcu to make sure access to
1368          * old vs->vs_tpg is finished.
1369          */
1370         vhost_scsi_flush(vs);
1371         kfree(vs->vs_tpg);
1372         vs->vs_tpg = NULL;
1373         WARN_ON(vs->vs_events_nr);
1374         mutex_unlock(&vs->dev.mutex);
1375         mutex_unlock(&vhost_scsi_mutex);
1376         return 0;
1377
1378 err_tpg:
1379         mutex_unlock(&tpg->tv_tpg_mutex);
1380 err_dev:
1381         mutex_unlock(&vs->dev.mutex);
1382         mutex_unlock(&vhost_scsi_mutex);
1383         return ret;
1384 }
1385
1386 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1387 {
1388         struct vhost_virtqueue *vq;
1389         int i;
1390
1391         if (features & ~VHOST_SCSI_FEATURES)
1392                 return -EOPNOTSUPP;
1393
1394         mutex_lock(&vs->dev.mutex);
1395         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1396             !vhost_log_access_ok(&vs->dev)) {
1397                 mutex_unlock(&vs->dev.mutex);
1398                 return -EFAULT;
1399         }
1400
1401         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1402                 vq = &vs->vqs[i].vq;
1403                 mutex_lock(&vq->mutex);
1404                 vq->acked_features = features;
1405                 mutex_unlock(&vq->mutex);
1406         }
1407         mutex_unlock(&vs->dev.mutex);
1408         return 0;
1409 }
1410
1411 static int vhost_scsi_open(struct inode *inode, struct file *f)
1412 {
1413         struct vhost_scsi *vs;
1414         struct vhost_virtqueue **vqs;
1415         int r = -ENOMEM, i;
1416
1417         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
1418         if (!vs) {
1419                 vs = vzalloc(sizeof(*vs));
1420                 if (!vs)
1421                         goto err_vs;
1422         }
1423
1424         vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1425         if (!vqs)
1426                 goto err_vqs;
1427
1428         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1429         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1430
1431         vs->vs_events_nr = 0;
1432         vs->vs_events_missed = false;
1433
1434         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1435         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1436         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1437         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1438         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1439                 vqs[i] = &vs->vqs[i].vq;
1440                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1441         }
1442         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ,
1443                        VHOST_SCSI_WEIGHT, 0);
1444
1445         vhost_scsi_init_inflight(vs, NULL);
1446
1447         f->private_data = vs;
1448         return 0;
1449
1450 err_vqs:
1451         kvfree(vs);
1452 err_vs:
1453         return r;
1454 }
1455
1456 static int vhost_scsi_release(struct inode *inode, struct file *f)
1457 {
1458         struct vhost_scsi *vs = f->private_data;
1459         struct vhost_scsi_target t;
1460
1461         mutex_lock(&vs->dev.mutex);
1462         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1463         mutex_unlock(&vs->dev.mutex);
1464         vhost_scsi_clear_endpoint(vs, &t);
1465         vhost_dev_stop(&vs->dev);
1466         vhost_dev_cleanup(&vs->dev, false);
1467         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1468         vhost_scsi_flush(vs);
1469         kfree(vs->dev.vqs);
1470         kvfree(vs);
1471         return 0;
1472 }
1473
1474 static long
1475 vhost_scsi_ioctl(struct file *f,
1476                  unsigned int ioctl,
1477                  unsigned long arg)
1478 {
1479         struct vhost_scsi *vs = f->private_data;
1480         struct vhost_scsi_target backend;
1481         void __user *argp = (void __user *)arg;
1482         u64 __user *featurep = argp;
1483         u32 __user *eventsp = argp;
1484         u32 events_missed;
1485         u64 features;
1486         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1487         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1488
1489         switch (ioctl) {
1490         case VHOST_SCSI_SET_ENDPOINT:
1491                 if (copy_from_user(&backend, argp, sizeof backend))
1492                         return -EFAULT;
1493                 if (backend.reserved != 0)
1494                         return -EOPNOTSUPP;
1495
1496                 return vhost_scsi_set_endpoint(vs, &backend);
1497         case VHOST_SCSI_CLEAR_ENDPOINT:
1498                 if (copy_from_user(&backend, argp, sizeof backend))
1499                         return -EFAULT;
1500                 if (backend.reserved != 0)
1501                         return -EOPNOTSUPP;
1502
1503                 return vhost_scsi_clear_endpoint(vs, &backend);
1504         case VHOST_SCSI_GET_ABI_VERSION:
1505                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1506                         return -EFAULT;
1507                 return 0;
1508         case VHOST_SCSI_SET_EVENTS_MISSED:
1509                 if (get_user(events_missed, eventsp))
1510                         return -EFAULT;
1511                 mutex_lock(&vq->mutex);
1512                 vs->vs_events_missed = events_missed;
1513                 mutex_unlock(&vq->mutex);
1514                 return 0;
1515         case VHOST_SCSI_GET_EVENTS_MISSED:
1516                 mutex_lock(&vq->mutex);
1517                 events_missed = vs->vs_events_missed;
1518                 mutex_unlock(&vq->mutex);
1519                 if (put_user(events_missed, eventsp))
1520                         return -EFAULT;
1521                 return 0;
1522         case VHOST_GET_FEATURES:
1523                 features = VHOST_SCSI_FEATURES;
1524                 if (copy_to_user(featurep, &features, sizeof features))
1525                         return -EFAULT;
1526                 return 0;
1527         case VHOST_SET_FEATURES:
1528                 if (copy_from_user(&features, featurep, sizeof features))
1529                         return -EFAULT;
1530                 return vhost_scsi_set_features(vs, features);
1531         default:
1532                 mutex_lock(&vs->dev.mutex);
1533                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1534                 /* TODO: flush backend after dev ioctl. */
1535                 if (r == -ENOIOCTLCMD)
1536                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1537                 mutex_unlock(&vs->dev.mutex);
1538                 return r;
1539         }
1540 }
1541
1542 #ifdef CONFIG_COMPAT
1543 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1544                                 unsigned long arg)
1545 {
1546         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1547 }
1548 #endif
1549
1550 static const struct file_operations vhost_scsi_fops = {
1551         .owner          = THIS_MODULE,
1552         .release        = vhost_scsi_release,
1553         .unlocked_ioctl = vhost_scsi_ioctl,
1554 #ifdef CONFIG_COMPAT
1555         .compat_ioctl   = vhost_scsi_compat_ioctl,
1556 #endif
1557         .open           = vhost_scsi_open,
1558         .llseek         = noop_llseek,
1559 };
1560
1561 static struct miscdevice vhost_scsi_misc = {
1562         MISC_DYNAMIC_MINOR,
1563         "vhost-scsi",
1564         &vhost_scsi_fops,
1565 };
1566
1567 static int __init vhost_scsi_register(void)
1568 {
1569         return misc_register(&vhost_scsi_misc);
1570 }
1571
1572 static void vhost_scsi_deregister(void)
1573 {
1574         misc_deregister(&vhost_scsi_misc);
1575 }
1576
1577 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1578 {
1579         switch (tport->tport_proto_id) {
1580         case SCSI_PROTOCOL_SAS:
1581                 return "SAS";
1582         case SCSI_PROTOCOL_FCP:
1583                 return "FCP";
1584         case SCSI_PROTOCOL_ISCSI:
1585                 return "iSCSI";
1586         default:
1587                 break;
1588         }
1589
1590         return "Unknown";
1591 }
1592
1593 static void
1594 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1595                   struct se_lun *lun, bool plug)
1596 {
1597
1598         struct vhost_scsi *vs = tpg->vhost_scsi;
1599         struct vhost_virtqueue *vq;
1600         u32 reason;
1601
1602         if (!vs)
1603                 return;
1604
1605         mutex_lock(&vs->dev.mutex);
1606
1607         if (plug)
1608                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1609         else
1610                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1611
1612         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1613         mutex_lock(&vq->mutex);
1614         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1615                 vhost_scsi_send_evt(vs, tpg, lun,
1616                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1617         mutex_unlock(&vq->mutex);
1618         mutex_unlock(&vs->dev.mutex);
1619 }
1620
1621 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1622 {
1623         vhost_scsi_do_plug(tpg, lun, true);
1624 }
1625
1626 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1627 {
1628         vhost_scsi_do_plug(tpg, lun, false);
1629 }
1630
1631 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1632                                struct se_lun *lun)
1633 {
1634         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1635                                 struct vhost_scsi_tpg, se_tpg);
1636
1637         mutex_lock(&vhost_scsi_mutex);
1638
1639         mutex_lock(&tpg->tv_tpg_mutex);
1640         tpg->tv_tpg_port_count++;
1641         mutex_unlock(&tpg->tv_tpg_mutex);
1642
1643         vhost_scsi_hotplug(tpg, lun);
1644
1645         mutex_unlock(&vhost_scsi_mutex);
1646
1647         return 0;
1648 }
1649
1650 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1651                                   struct se_lun *lun)
1652 {
1653         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1654                                 struct vhost_scsi_tpg, se_tpg);
1655
1656         mutex_lock(&vhost_scsi_mutex);
1657
1658         mutex_lock(&tpg->tv_tpg_mutex);
1659         tpg->tv_tpg_port_count--;
1660         mutex_unlock(&tpg->tv_tpg_mutex);
1661
1662         vhost_scsi_hotunplug(tpg, lun);
1663
1664         mutex_unlock(&vhost_scsi_mutex);
1665 }
1666
1667 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1668 {
1669         struct vhost_scsi_cmd *tv_cmd;
1670         unsigned int i;
1671
1672         if (!se_sess->sess_cmd_map)
1673                 return;
1674
1675         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1676                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1677
1678                 kfree(tv_cmd->tvc_sgl);
1679                 kfree(tv_cmd->tvc_prot_sgl);
1680                 kfree(tv_cmd->tvc_upages);
1681         }
1682 }
1683
1684 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1685                 struct config_item *item, const char *page, size_t count)
1686 {
1687         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1688         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1689                                 struct vhost_scsi_tpg, se_tpg);
1690         unsigned long val;
1691         int ret = kstrtoul(page, 0, &val);
1692
1693         if (ret) {
1694                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1695                 return ret;
1696         }
1697         if (val != 0 && val != 1 && val != 3) {
1698                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1699                 return -EINVAL;
1700         }
1701         tpg->tv_fabric_prot_type = val;
1702
1703         return count;
1704 }
1705
1706 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1707                 struct config_item *item, char *page)
1708 {
1709         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1710         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1711                                 struct vhost_scsi_tpg, se_tpg);
1712
1713         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1714 }
1715
1716 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1717
1718 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1719         &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1720         NULL,
1721 };
1722
1723 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1724                                struct se_session *se_sess, void *p)
1725 {
1726         struct vhost_scsi_cmd *tv_cmd;
1727         unsigned int i;
1728
1729         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1730                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1731
1732                 tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1733                                         VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1734                 if (!tv_cmd->tvc_sgl) {
1735                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1736                         goto out;
1737                 }
1738
1739                 tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1740                                 VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1741                 if (!tv_cmd->tvc_upages) {
1742                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1743                         goto out;
1744                 }
1745
1746                 tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1747                                 VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1748                 if (!tv_cmd->tvc_prot_sgl) {
1749                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1750                         goto out;
1751                 }
1752         }
1753         return 0;
1754 out:
1755         vhost_scsi_free_cmd_map_res(se_sess);
1756         return -ENOMEM;
1757 }
1758
1759 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1760                                 const char *name)
1761 {
1762         struct se_portal_group *se_tpg;
1763         struct vhost_scsi_nexus *tv_nexus;
1764
1765         mutex_lock(&tpg->tv_tpg_mutex);
1766         if (tpg->tpg_nexus) {
1767                 mutex_unlock(&tpg->tv_tpg_mutex);
1768                 pr_debug("tpg->tpg_nexus already exists\n");
1769                 return -EEXIST;
1770         }
1771         se_tpg = &tpg->se_tpg;
1772
1773         tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1774         if (!tv_nexus) {
1775                 mutex_unlock(&tpg->tv_tpg_mutex);
1776                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1777                 return -ENOMEM;
1778         }
1779         /*
1780          * Since we are running in 'demo mode' this call with generate a
1781          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1782          * the SCSI Initiator port name of the passed configfs group 'name'.
1783          */
1784         tv_nexus->tvn_se_sess = target_alloc_session(&tpg->se_tpg,
1785                                         VHOST_SCSI_DEFAULT_TAGS,
1786                                         sizeof(struct vhost_scsi_cmd),
1787                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1788                                         (unsigned char *)name, tv_nexus,
1789                                         vhost_scsi_nexus_cb);
1790         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1791                 mutex_unlock(&tpg->tv_tpg_mutex);
1792                 kfree(tv_nexus);
1793                 return -ENOMEM;
1794         }
1795         tpg->tpg_nexus = tv_nexus;
1796
1797         mutex_unlock(&tpg->tv_tpg_mutex);
1798         return 0;
1799 }
1800
1801 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1802 {
1803         struct se_session *se_sess;
1804         struct vhost_scsi_nexus *tv_nexus;
1805
1806         mutex_lock(&tpg->tv_tpg_mutex);
1807         tv_nexus = tpg->tpg_nexus;
1808         if (!tv_nexus) {
1809                 mutex_unlock(&tpg->tv_tpg_mutex);
1810                 return -ENODEV;
1811         }
1812
1813         se_sess = tv_nexus->tvn_se_sess;
1814         if (!se_sess) {
1815                 mutex_unlock(&tpg->tv_tpg_mutex);
1816                 return -ENODEV;
1817         }
1818
1819         if (tpg->tv_tpg_port_count != 0) {
1820                 mutex_unlock(&tpg->tv_tpg_mutex);
1821                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1822                         " active TPG port count: %d\n",
1823                         tpg->tv_tpg_port_count);
1824                 return -EBUSY;
1825         }
1826
1827         if (tpg->tv_tpg_vhost_count != 0) {
1828                 mutex_unlock(&tpg->tv_tpg_mutex);
1829                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1830                         " active TPG vhost count: %d\n",
1831                         tpg->tv_tpg_vhost_count);
1832                 return -EBUSY;
1833         }
1834
1835         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1836                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1837                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1838
1839         vhost_scsi_free_cmd_map_res(se_sess);
1840         /*
1841          * Release the SCSI I_T Nexus to the emulated vhost Target Port
1842          */
1843         transport_deregister_session(tv_nexus->tvn_se_sess);
1844         tpg->tpg_nexus = NULL;
1845         mutex_unlock(&tpg->tv_tpg_mutex);
1846
1847         kfree(tv_nexus);
1848         return 0;
1849 }
1850
1851 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1852 {
1853         struct se_portal_group *se_tpg = to_tpg(item);
1854         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1855                                 struct vhost_scsi_tpg, se_tpg);
1856         struct vhost_scsi_nexus *tv_nexus;
1857         ssize_t ret;
1858
1859         mutex_lock(&tpg->tv_tpg_mutex);
1860         tv_nexus = tpg->tpg_nexus;
1861         if (!tv_nexus) {
1862                 mutex_unlock(&tpg->tv_tpg_mutex);
1863                 return -ENODEV;
1864         }
1865         ret = snprintf(page, PAGE_SIZE, "%s\n",
1866                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1867         mutex_unlock(&tpg->tv_tpg_mutex);
1868
1869         return ret;
1870 }
1871
1872 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1873                 const char *page, size_t count)
1874 {
1875         struct se_portal_group *se_tpg = to_tpg(item);
1876         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1877                                 struct vhost_scsi_tpg, se_tpg);
1878         struct vhost_scsi_tport *tport_wwn = tpg->tport;
1879         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1880         int ret;
1881         /*
1882          * Shutdown the active I_T nexus if 'NULL' is passed..
1883          */
1884         if (!strncmp(page, "NULL", 4)) {
1885                 ret = vhost_scsi_drop_nexus(tpg);
1886                 return (!ret) ? count : ret;
1887         }
1888         /*
1889          * Otherwise make sure the passed virtual Initiator port WWN matches
1890          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1891          * vhost_scsi_make_nexus().
1892          */
1893         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1894                 pr_err("Emulated NAA Sas Address: %s, exceeds"
1895                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
1896                 return -EINVAL;
1897         }
1898         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1899
1900         ptr = strstr(i_port, "naa.");
1901         if (ptr) {
1902                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1903                         pr_err("Passed SAS Initiator Port %s does not"
1904                                 " match target port protoid: %s\n", i_port,
1905                                 vhost_scsi_dump_proto_id(tport_wwn));
1906                         return -EINVAL;
1907                 }
1908                 port_ptr = &i_port[0];
1909                 goto check_newline;
1910         }
1911         ptr = strstr(i_port, "fc.");
1912         if (ptr) {
1913                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1914                         pr_err("Passed FCP Initiator Port %s does not"
1915                                 " match target port protoid: %s\n", i_port,
1916                                 vhost_scsi_dump_proto_id(tport_wwn));
1917                         return -EINVAL;
1918                 }
1919                 port_ptr = &i_port[3]; /* Skip over "fc." */
1920                 goto check_newline;
1921         }
1922         ptr = strstr(i_port, "iqn.");
1923         if (ptr) {
1924                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1925                         pr_err("Passed iSCSI Initiator Port %s does not"
1926                                 " match target port protoid: %s\n", i_port,
1927                                 vhost_scsi_dump_proto_id(tport_wwn));
1928                         return -EINVAL;
1929                 }
1930                 port_ptr = &i_port[0];
1931                 goto check_newline;
1932         }
1933         pr_err("Unable to locate prefix for emulated Initiator Port:"
1934                         " %s\n", i_port);
1935         return -EINVAL;
1936         /*
1937          * Clear any trailing newline for the NAA WWN
1938          */
1939 check_newline:
1940         if (i_port[strlen(i_port)-1] == '\n')
1941                 i_port[strlen(i_port)-1] = '\0';
1942
1943         ret = vhost_scsi_make_nexus(tpg, port_ptr);
1944         if (ret < 0)
1945                 return ret;
1946
1947         return count;
1948 }
1949
1950 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1951
1952 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1953         &vhost_scsi_tpg_attr_nexus,
1954         NULL,
1955 };
1956
1957 static struct se_portal_group *
1958 vhost_scsi_make_tpg(struct se_wwn *wwn,
1959                    struct config_group *group,
1960                    const char *name)
1961 {
1962         struct vhost_scsi_tport *tport = container_of(wwn,
1963                         struct vhost_scsi_tport, tport_wwn);
1964
1965         struct vhost_scsi_tpg *tpg;
1966         u16 tpgt;
1967         int ret;
1968
1969         if (strstr(name, "tpgt_") != name)
1970                 return ERR_PTR(-EINVAL);
1971         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1972                 return ERR_PTR(-EINVAL);
1973
1974         tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1975         if (!tpg) {
1976                 pr_err("Unable to allocate struct vhost_scsi_tpg");
1977                 return ERR_PTR(-ENOMEM);
1978         }
1979         mutex_init(&tpg->tv_tpg_mutex);
1980         INIT_LIST_HEAD(&tpg->tv_tpg_list);
1981         tpg->tport = tport;
1982         tpg->tport_tpgt = tpgt;
1983
1984         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1985         if (ret < 0) {
1986                 kfree(tpg);
1987                 return NULL;
1988         }
1989         mutex_lock(&vhost_scsi_mutex);
1990         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1991         mutex_unlock(&vhost_scsi_mutex);
1992
1993         return &tpg->se_tpg;
1994 }
1995
1996 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1997 {
1998         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1999                                 struct vhost_scsi_tpg, se_tpg);
2000
2001         mutex_lock(&vhost_scsi_mutex);
2002         list_del(&tpg->tv_tpg_list);
2003         mutex_unlock(&vhost_scsi_mutex);
2004         /*
2005          * Release the virtual I_T Nexus for this vhost TPG
2006          */
2007         vhost_scsi_drop_nexus(tpg);
2008         /*
2009          * Deregister the se_tpg from TCM..
2010          */
2011         core_tpg_deregister(se_tpg);
2012         kfree(tpg);
2013 }
2014
2015 static struct se_wwn *
2016 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2017                      struct config_group *group,
2018                      const char *name)
2019 {
2020         struct vhost_scsi_tport *tport;
2021         char *ptr;
2022         u64 wwpn = 0;
2023         int off = 0;
2024
2025         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2026                 return ERR_PTR(-EINVAL); */
2027
2028         tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2029         if (!tport) {
2030                 pr_err("Unable to allocate struct vhost_scsi_tport");
2031                 return ERR_PTR(-ENOMEM);
2032         }
2033         tport->tport_wwpn = wwpn;
2034         /*
2035          * Determine the emulated Protocol Identifier and Target Port Name
2036          * based on the incoming configfs directory name.
2037          */
2038         ptr = strstr(name, "naa.");
2039         if (ptr) {
2040                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2041                 goto check_len;
2042         }
2043         ptr = strstr(name, "fc.");
2044         if (ptr) {
2045                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2046                 off = 3; /* Skip over "fc." */
2047                 goto check_len;
2048         }
2049         ptr = strstr(name, "iqn.");
2050         if (ptr) {
2051                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2052                 goto check_len;
2053         }
2054
2055         pr_err("Unable to locate prefix for emulated Target Port:"
2056                         " %s\n", name);
2057         kfree(tport);
2058         return ERR_PTR(-EINVAL);
2059
2060 check_len:
2061         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2062                 pr_err("Emulated %s Address: %s, exceeds"
2063                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2064                         VHOST_SCSI_NAMELEN);
2065                 kfree(tport);
2066                 return ERR_PTR(-EINVAL);
2067         }
2068         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2069
2070         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2071                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2072
2073         return &tport->tport_wwn;
2074 }
2075
2076 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2077 {
2078         struct vhost_scsi_tport *tport = container_of(wwn,
2079                                 struct vhost_scsi_tport, tport_wwn);
2080
2081         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2082                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2083                 tport->tport_name);
2084
2085         kfree(tport);
2086 }
2087
2088 static ssize_t
2089 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2090 {
2091         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2092                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2093                 utsname()->machine);
2094 }
2095
2096 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2097
2098 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2099         &vhost_scsi_wwn_attr_version,
2100         NULL,
2101 };
2102
2103 static struct target_core_fabric_ops vhost_scsi_ops = {
2104         .module                         = THIS_MODULE,
2105         .name                           = "vhost",
2106         .get_fabric_name                = vhost_scsi_get_fabric_name,
2107         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2108         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2109         .tpg_check_demo_mode            = vhost_scsi_check_true,
2110         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2111         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2112         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2113         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2114         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2115         .release_cmd                    = vhost_scsi_release_cmd,
2116         .check_stop_free                = vhost_scsi_check_stop_free,
2117         .sess_get_index                 = vhost_scsi_sess_get_index,
2118         .sess_get_initiator_sid         = NULL,
2119         .write_pending                  = vhost_scsi_write_pending,
2120         .write_pending_status           = vhost_scsi_write_pending_status,
2121         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2122         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2123         .queue_data_in                  = vhost_scsi_queue_data_in,
2124         .queue_status                   = vhost_scsi_queue_status,
2125         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2126         .aborted_task                   = vhost_scsi_aborted_task,
2127         /*
2128          * Setup callers for generic logic in target_core_fabric_configfs.c
2129          */
2130         .fabric_make_wwn                = vhost_scsi_make_tport,
2131         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2132         .fabric_make_tpg                = vhost_scsi_make_tpg,
2133         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2134         .fabric_post_link               = vhost_scsi_port_link,
2135         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2136
2137         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2138         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2139         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2140 };
2141
2142 static int __init vhost_scsi_init(void)
2143 {
2144         int ret = -ENOMEM;
2145
2146         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2147                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2148                 utsname()->machine);
2149
2150         /*
2151          * Use our own dedicated workqueue for submitting I/O into
2152          * target core to avoid contention within system_wq.
2153          */
2154         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2155         if (!vhost_scsi_workqueue)
2156                 goto out;
2157
2158         ret = vhost_scsi_register();
2159         if (ret < 0)
2160                 goto out_destroy_workqueue;
2161
2162         ret = target_register_template(&vhost_scsi_ops);
2163         if (ret < 0)
2164                 goto out_vhost_scsi_deregister;
2165
2166         return 0;
2167
2168 out_vhost_scsi_deregister:
2169         vhost_scsi_deregister();
2170 out_destroy_workqueue:
2171         destroy_workqueue(vhost_scsi_workqueue);
2172 out:
2173         return ret;
2174 };
2175
2176 static void vhost_scsi_exit(void)
2177 {
2178         target_unregister_template(&vhost_scsi_ops);
2179         vhost_scsi_deregister();
2180         destroy_workqueue(vhost_scsi_workqueue);
2181 };
2182
2183 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2184 MODULE_ALIAS("tcm_vhost");
2185 MODULE_LICENSE("GPL");
2186 module_init(vhost_scsi_init);
2187 module_exit(vhost_scsi_exit);