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, *t;
 506         struct llist_node *llnode;
 507
 508         mutex_lock(&vq->mutex);
 509         llnode = llist_del_all(&vs->vs_event_list);
 510         llist_for_each_entry_safe(evt, t, llnode, list) {
 511                 vhost_scsi_do_evt_work(vs, evt);
 512                 vhost_scsi_free_evt(vs, evt);
 513         }
 514         mutex_unlock(&vq->mutex);
 515 }
 516
 517 /* Fill in status and signal that we are done processing this command
 518  *
 519  * This is scheduled in the vhost work queue so we are called with the owner
 520  * process mm and can access the vring.
 521  */
 522 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 523 {
 524         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 525                                         vs_completion_work);
 526         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
 527         struct virtio_scsi_cmd_resp v_rsp;
 528         struct vhost_scsi_cmd *cmd;
 529         struct llist_node *llnode;
 530         struct se_cmd *se_cmd;
 531         struct iov_iter iov_iter;
 532         int ret, vq;
 533
 534         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
 535         llnode = llist_del_all(&vs->vs_completion_list);
 536         llist_for_each_entry(cmd, llnode, tvc_completion_list) {
 537                 se_cmd = &cmd->tvc_se_cmd;
 538
 539                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
 540                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
 541
 542                 memset(&v_rsp, 0, sizeof(v_rsp));
 543                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
 544                 /* TODO is status_qualifier field needed? */
 545                 v_rsp.status = se_cmd->scsi_status;
 546                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
 547                                                  se_cmd->scsi_sense_length);
 548                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
 549                        se_cmd->scsi_sense_length);
 550
 551                 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
 552                               cmd->tvc_in_iovs, sizeof(v_rsp));
 553                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
 554                 if (likely(ret == sizeof(v_rsp))) {
 555                         struct vhost_scsi_virtqueue *q;
 556                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
 557                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
 558                         vq = q - vs->vqs;
 559                         __set_bit(vq, signal);
 560                 } else
 561                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
 562
 563                 vhost_scsi_free_cmd(cmd);
 564         }
 565
 566         vq = -1;
 567         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
 568                 < VHOST_SCSI_MAX_VQ)
 569                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
 570 }
 571
 572 static struct vhost_scsi_cmd *
 573 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
 574                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
 575                    u32 exp_data_len, int data_direction)
 576 {
 577         struct vhost_scsi_cmd *cmd;
 578         struct vhost_scsi_nexus *tv_nexus;
 579         struct se_session *se_sess;
 580         struct scatterlist *sg, *prot_sg;
 581         struct page **pages;
 582         int tag;
 583
 584         tv_nexus = tpg->tpg_nexus;
 585         if (!tv_nexus) {
 586                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
 587                 return ERR_PTR(-EIO);
 588         }
 589         se_sess = tv_nexus->tvn_se_sess;
 590
 591         tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
 592         if (tag < 0) {
 593                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
 594                 return ERR_PTR(-ENOMEM);
 595         }
 596
 597         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
 598         sg = cmd->tvc_sgl;
 599         prot_sg = cmd->tvc_prot_sgl;
 600         pages = cmd->tvc_upages;
 601         memset(cmd, 0, sizeof(struct vhost_scsi_cmd));
 602
 603         cmd->tvc_sgl = sg;
 604         cmd->tvc_prot_sgl = prot_sg;
 605         cmd->tvc_upages = pages;
 606         cmd->tvc_se_cmd.map_tag = tag;
 607         cmd->tvc_tag = scsi_tag;
 608         cmd->tvc_lun = lun;
 609         cmd->tvc_task_attr = task_attr;
 610         cmd->tvc_exp_data_len = exp_data_len;
 611         cmd->tvc_data_direction = data_direction;
 612         cmd->tvc_nexus = tv_nexus;
 613         cmd->inflight = vhost_scsi_get_inflight(vq);
 614
 615         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
 616
 617         return cmd;
 618 }
 619
 620 /*
 621  * Map a user memory range into a scatterlist
 622  *
 623  * Returns the number of scatterlist entries used or -errno on error.
 624  */
 625 static int
 626 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
 627                       void __user *ptr,
 628                       size_t len,
 629                       struct scatterlist *sgl,
 630                       bool write)
 631 {
 632         unsigned int npages = 0, offset, nbytes;
 633         unsigned int pages_nr = iov_num_pages(ptr, len);
 634         struct scatterlist *sg = sgl;
 635         struct page **pages = cmd->tvc_upages;
 636         int ret, i;
 637
 638         if (pages_nr > VHOST_SCSI_PREALLOC_UPAGES) {
 639                 pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
 640                        " preallocated VHOST_SCSI_PREALLOC_UPAGES: %u\n",
 641                         pages_nr, VHOST_SCSI_PREALLOC_UPAGES);
 642                 return -ENOBUFS;
 643         }
 644
 645         ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
 646         /* No pages were pinned */
 647         if (ret < 0)
 648                 goto out;
 649         /* Less pages pinned than wanted */
 650         if (ret != pages_nr) {
 651                 for (i = 0; i < ret; i++)
 652                         put_page(pages[i]);
 653                 ret = -EFAULT;
 654                 goto out;
 655         }
 656
 657         while (len > 0) {
 658                 offset = (uintptr_t)ptr & ~PAGE_MASK;
 659                 nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
 660                 sg_set_page(sg, pages[npages], nbytes, offset);
 661                 ptr += nbytes;
 662                 len -= nbytes;
 663                 sg++;
 664                 npages++;
 665         }
 666
 667 out:
 668         return ret;
 669 }
 670
 671 static int
 672 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
 673 {
 674         int sgl_count = 0;
 675
 676         if (!iter || !iter->iov) {
 677                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
 678                        " present\n", __func__, bytes);
 679                 return -EINVAL;
 680         }
 681
 682         sgl_count = iov_iter_npages(iter, 0xffff);
 683         if (sgl_count > max_sgls) {
 684                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
 685                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
 686                 return -EINVAL;
 687         }
 688         return sgl_count;
 689 }
 690
 691 static int
 692 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
 693                       struct iov_iter *iter,
 694                       struct scatterlist *sg, int sg_count)
 695 {
 696         size_t off = iter->iov_offset;
 697         struct scatterlist *p = sg;
 698         int i, ret;
 699
 700         for (i = 0; i < iter->nr_segs; i++) {
 701                 void __user *base = iter->iov[i].iov_base + off;
 702                 size_t len = iter->iov[i].iov_len - off;
 703
 704                 ret = vhost_scsi_map_to_sgl(cmd, base, len, sg, write);
 705                 if (ret < 0) {
 706                         while (p < sg) {
 707                                 struct page *page = sg_page(p++);
 708                                 if (page)
 709                                         put_page(page);
 710                         }
 711                         return ret;
 712                 }
 713                 sg += ret;
 714                 off = 0;
 715         }
 716         return 0;
 717 }
 718
 719 static int
 720 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
 721                  size_t prot_bytes, struct iov_iter *prot_iter,
 722                  size_t data_bytes, struct iov_iter *data_iter)
 723 {
 724         int sgl_count, ret;
 725         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
 726
 727         if (prot_bytes) {
 728                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
 729                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
 730                 if (sgl_count < 0)
 731                         return sgl_count;
 732
 733                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
 734                 cmd->tvc_prot_sgl_count = sgl_count;
 735                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
 736                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
 737
 738                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
 739                                             cmd->tvc_prot_sgl,
 740                                             cmd->tvc_prot_sgl_count);
 741                 if (ret < 0) {
 742                         cmd->tvc_prot_sgl_count = 0;
 743                         return ret;
 744                 }
 745         }
 746         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
 747                                          VHOST_SCSI_PREALLOC_SGLS);
 748         if (sgl_count < 0)
 749                 return sgl_count;
 750
 751         sg_init_table(cmd->tvc_sgl, sgl_count);
 752         cmd->tvc_sgl_count = sgl_count;
 753         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
 754                   cmd->tvc_sgl, cmd->tvc_sgl_count);
 755
 756         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
 757                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
 758         if (ret < 0) {
 759                 cmd->tvc_sgl_count = 0;
 760                 return ret;
 761         }
 762         return 0;
 763 }
 764
 765 static int vhost_scsi_to_tcm_attr(int attr)
 766 {
 767         switch (attr) {
 768         case VIRTIO_SCSI_S_SIMPLE:
 769                 return TCM_SIMPLE_TAG;
 770         case VIRTIO_SCSI_S_ORDERED:
 771                 return TCM_ORDERED_TAG;
 772         case VIRTIO_SCSI_S_HEAD:
 773                 return TCM_HEAD_TAG;
 774         case VIRTIO_SCSI_S_ACA:
 775                 return TCM_ACA_TAG;
 776         default:
 777                 break;
 778         }
 779         return TCM_SIMPLE_TAG;
 780 }
 781
 782 static void vhost_scsi_submission_work(struct work_struct *work)
 783 {
 784         struct vhost_scsi_cmd *cmd =
 785                 container_of(work, struct vhost_scsi_cmd, work);
 786         struct vhost_scsi_nexus *tv_nexus;
 787         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 788         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
 789         int rc;
 790
 791         /* FIXME: BIDI operation */
 792         if (cmd->tvc_sgl_count) {
 793                 sg_ptr = cmd->tvc_sgl;
 794
 795                 if (cmd->tvc_prot_sgl_count)
 796                         sg_prot_ptr = cmd->tvc_prot_sgl;
 797                 else
 798                         se_cmd->prot_pto = true;
 799         } else {
 800                 sg_ptr = NULL;
 801         }
 802         tv_nexus = cmd->tvc_nexus;
 803
 804         se_cmd->tag = 0;
 805         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
 806                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
 807                         cmd->tvc_lun, cmd->tvc_exp_data_len,
 808                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
 809                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
 810                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
 811                         cmd->tvc_prot_sgl_count);
 812         if (rc < 0) {
 813                 transport_send_check_condition_and_sense(se_cmd,
 814                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
 815                 transport_generic_free_cmd(se_cmd, 0);
 816         }
 817 }
 818
 819 static void
 820 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
 821                            struct vhost_virtqueue *vq,
 822                            int head, unsigned out)
 823 {
 824         struct virtio_scsi_cmd_resp __user *resp;
 825         struct virtio_scsi_cmd_resp rsp;
 826         int ret;
 827
 828         memset(&rsp, 0, sizeof(rsp));
 829         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
 830         resp = vq->iov[out].iov_base;
 831         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
 832         if (!ret)
 833                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 834         else
 835                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
 836 }
 837
 838 static void
 839 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
 840 {
 841         struct vhost_scsi_tpg **vs_tpg, *tpg;
 842         struct virtio_scsi_cmd_req v_req;
 843         struct virtio_scsi_cmd_req_pi v_req_pi;
 844         struct vhost_scsi_cmd *cmd;
 845         struct iov_iter out_iter, in_iter, prot_iter, data_iter;
 846         u64 tag;
 847         u32 exp_data_len, data_direction;
 848         unsigned int out = 0, in = 0;
 849         int head, ret, prot_bytes, c = 0;
 850         size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
 851         size_t out_size, in_size;
 852         u16 lun;
 853         u8 *target, *lunp, task_attr;
 854         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
 855         void *req, *cdb;
 856
 857         mutex_lock(&vq->mutex);
 858         /*
 859          * We can handle the vq only after the endpoint is setup by calling the
 860          * VHOST_SCSI_SET_ENDPOINT ioctl.
 861          */
 862         vs_tpg = vq->private_data;
 863         if (!vs_tpg)
 864                 goto out;
 865
 866         vhost_disable_notify(&vs->dev, vq);
 867
 868         do {
 869                 head = vhost_get_vq_desc(vq, vq->iov,
 870                                          ARRAY_SIZE(vq->iov), &out, &in,
 871                                          NULL, NULL);
 872                 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
 873                          head, out, in);
 874                 /* On error, stop handling until the next kick. */
 875                 if (unlikely(head < 0))
 876                         break;
 877                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 878                 if (head == vq->num) {
 879                         if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
 880                                 vhost_disable_notify(&vs->dev, vq);
 881                                 continue;
 882                         }
 883                         break;
 884                 }
 885                 /*
 886                  * Check for a sane response buffer so we can report early
 887                  * errors back to the guest.
 888                  */
 889                 if (unlikely(vq->iov[out].iov_len < rsp_size)) {
 890                         vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
 891                                 " size, got %zu bytes\n", vq->iov[out].iov_len);
 892                         break;
 893                 }
 894                 /*
 895                  * Setup pointers and values based upon different virtio-scsi
 896                  * request header if T10_PI is enabled in KVM guest.
 897                  */
 898                 if (t10_pi) {
 899                         req = &v_req_pi;
 900                         req_size = sizeof(v_req_pi);
 901                         lunp = &v_req_pi.lun[0];
 902                         target = &v_req_pi.lun[1];
 903                 } else {
 904                         req = &v_req;
 905                         req_size = sizeof(v_req);
 906                         lunp = &v_req.lun[0];
 907                         target = &v_req.lun[1];
 908                 }
 909                 /*
 910                  * FIXME: Not correct for BIDI operation
 911                  */
 912                 out_size = iov_length(vq->iov, out);
 913                 in_size = iov_length(&vq->iov[out], in);
 914
 915                 /*
 916                  * Copy over the virtio-scsi request header, which for a
 917                  * ANY_LAYOUT enabled guest may span multiple iovecs, or a
 918                  * single iovec may contain both the header + outgoing
 919                  * WRITE payloads.
 920                  *
 921                  * copy_from_iter() will advance out_iter, so that it will
 922                  * point at the start of the outgoing WRITE payload, if
 923                  * DMA_TO_DEVICE is set.
 924                  */
 925                 iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
 926
 927                 if (unlikely(!copy_from_iter_full(req, req_size, &out_iter))) {
 928                         vq_err(vq, "Faulted on copy_from_iter\n");
 929                         vhost_scsi_send_bad_target(vs, vq, head, out);
 930                         continue;
 931                 }
 932                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
 933                 if (unlikely(*lunp != 1)) {
 934                         vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
 935                         vhost_scsi_send_bad_target(vs, vq, head, out);
 936                         continue;
 937                 }
 938
 939                 tpg = ACCESS_ONCE(vs_tpg[*target]);
 940                 if (unlikely(!tpg)) {
 941                         /* Target does not exist, fail the request */
 942                         vhost_scsi_send_bad_target(vs, vq, head, out);
 943                         continue;
 944                 }
 945                 /*
 946                  * Determine data_direction by calculating the total outgoing
 947                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
 948                  * response headers respectively.
 949                  *
 950                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
 951                  * to the right place.
 952                  *
 953                  * For DMA_FROM_DEVICE, the iovec will be just past the end
 954                  * of the virtio-scsi response header in either the same
 955                  * or immediately following iovec.
 956                  *
 957                  * Any associated T10_PI bytes for the outgoing / incoming
 958                  * payloads are included in calculation of exp_data_len here.
 959                  */
 960                 prot_bytes = 0;
 961
 962                 if (out_size > req_size) {
 963                         data_direction = DMA_TO_DEVICE;
 964                         exp_data_len = out_size - req_size;
 965                         data_iter = out_iter;
 966                 } else if (in_size > rsp_size) {
 967                         data_direction = DMA_FROM_DEVICE;
 968                         exp_data_len = in_size - rsp_size;
 969
 970                         iov_iter_init(&in_iter, READ, &vq->iov[out], in,
 971                                       rsp_size + exp_data_len);
 972                         iov_iter_advance(&in_iter, rsp_size);
 973                         data_iter = in_iter;
 974                 } else {
 975                         data_direction = DMA_NONE;
 976                         exp_data_len = 0;
 977                 }
 978                 /*
 979                  * If T10_PI header + payload is present, setup prot_iter values
 980                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
 981                  * host scatterlists via get_user_pages_fast().
 982                  */
 983                 if (t10_pi) {
 984                         if (v_req_pi.pi_bytesout) {
 985                                 if (data_direction != DMA_TO_DEVICE) {
 986                                         vq_err(vq, "Received non zero pi_bytesout,"
 987                                                 " but wrong data_direction\n");
 988                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 989                                         continue;
 990                                 }
 991                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
 992                         } else if (v_req_pi.pi_bytesin) {
 993                                 if (data_direction != DMA_FROM_DEVICE) {
 994                                         vq_err(vq, "Received non zero pi_bytesin,"
 995                                                 " but wrong data_direction\n");
 996                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 997                                         continue;
 998                                 }
 999                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1000                         }
1001                         /*
1002                          * Set prot_iter to data_iter and truncate it to
1003                          * prot_bytes, and advance data_iter past any
1004                          * preceeding prot_bytes that may be present.
1005                          *
1006                          * Also fix up the exp_data_len to reflect only the
1007                          * actual data payload length.
1008                          */
1009                         if (prot_bytes) {
1010                                 exp_data_len -= prot_bytes;
1011                                 prot_iter = data_iter;
1012                                 iov_iter_truncate(&prot_iter, prot_bytes);
1013                                 iov_iter_advance(&data_iter, prot_bytes);
1014                         }
1015                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
1016                         task_attr = v_req_pi.task_attr;
1017                         cdb = &v_req_pi.cdb[0];
1018                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1019                 } else {
1020                         tag = vhost64_to_cpu(vq, v_req.tag);
1021                         task_attr = v_req.task_attr;
1022                         cdb = &v_req.cdb[0];
1023                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1024                 }
1025                 /*
1026                  * Check that the received CDB size does not exceeded our
1027                  * hardcoded max for vhost-scsi, then get a pre-allocated
1028                  * cmd descriptor for the new virtio-scsi tag.
1029                  *
1030                  * TODO what if cdb was too small for varlen cdb header?
1031                  */
1032                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1033                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
1034                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1035                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1036                         vhost_scsi_send_bad_target(vs, vq, head, out);
1037                         continue;
1038                 }
1039                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1040                                          exp_data_len + prot_bytes,
1041                                          data_direction);
1042                 if (IS_ERR(cmd)) {
1043                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1044                                PTR_ERR(cmd));
1045                         vhost_scsi_send_bad_target(vs, vq, head, out);
1046                         continue;
1047                 }
1048                 cmd->tvc_vhost = vs;
1049                 cmd->tvc_vq = vq;
1050                 cmd->tvc_resp_iov = vq->iov[out];
1051                 cmd->tvc_in_iovs = in;
1052
1053                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1054                          cmd->tvc_cdb[0], cmd->tvc_lun);
1055                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1056                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1057
1058                 if (data_direction != DMA_NONE) {
1059                         ret = vhost_scsi_mapal(cmd,
1060                                                prot_bytes, &prot_iter,
1061                                                exp_data_len, &data_iter);
1062                         if (unlikely(ret)) {
1063                                 vq_err(vq, "Failed to map iov to sgl\n");
1064                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1065                                 vhost_scsi_send_bad_target(vs, vq, head, out);
1066                                 continue;
1067                         }
1068                 }
1069                 /*
1070                  * Save the descriptor from vhost_get_vq_desc() to be used to
1071                  * complete the virtio-scsi request in TCM callback context via
1072                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1073                  */
1074                 cmd->tvc_vq_desc = head;
1075                 /*
1076                  * Dispatch cmd descriptor for cmwq execution in process
1077                  * context provided by vhost_scsi_workqueue.  This also ensures
1078                  * cmd is executed on the same kworker CPU as this vhost
1079                  * thread to gain positive L2 cache locality effects.
1080                  */
1081                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1082                 queue_work(vhost_scsi_workqueue, &cmd->work);
1083         } while (likely(!vhost_exceeds_weight(vq, ++c, 0)));
1084 out:
1085         mutex_unlock(&vq->mutex);
1086 }
1087
1088 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1089 {
1090         pr_debug("%s: The handling func for control queue.\n", __func__);
1091 }
1092
1093 static void
1094 vhost_scsi_send_evt(struct vhost_scsi *vs,
1095                    struct vhost_scsi_tpg *tpg,
1096                    struct se_lun *lun,
1097                    u32 event,
1098                    u32 reason)
1099 {
1100         struct vhost_scsi_evt *evt;
1101
1102         evt = vhost_scsi_allocate_evt(vs, event, reason);
1103         if (!evt)
1104                 return;
1105
1106         if (tpg && lun) {
1107                 /* TODO: share lun setup code with virtio-scsi.ko */
1108                 /*
1109                  * Note: evt->event is zeroed when we allocate it and
1110                  * lun[4-7] need to be zero according to virtio-scsi spec.
1111                  */
1112                 evt->event.lun[0] = 0x01;
1113                 evt->event.lun[1] = tpg->tport_tpgt;
1114                 if (lun->unpacked_lun >= 256)
1115                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1116                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1117         }
1118
1119         llist_add(&evt->list, &vs->vs_event_list);
1120         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1121 }
1122
1123 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1124 {
1125         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1126                                                 poll.work);
1127         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1128
1129         mutex_lock(&vq->mutex);
1130         if (!vq->private_data)
1131                 goto out;
1132
1133         if (vs->vs_events_missed)
1134                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1135 out:
1136         mutex_unlock(&vq->mutex);
1137 }
1138
1139 static void vhost_scsi_handle_kick(struct vhost_work *work)
1140 {
1141         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1142                                                 poll.work);
1143         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1144
1145         vhost_scsi_handle_vq(vs, vq);
1146 }
1147
1148 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1149 {
1150         vhost_poll_flush(&vs->vqs[index].vq.poll);
1151 }
1152
1153 /* Callers must hold dev mutex */
1154 static void vhost_scsi_flush(struct vhost_scsi *vs)
1155 {
1156         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1157         int i;
1158
1159         /* Init new inflight and remember the old inflight */
1160         vhost_scsi_init_inflight(vs, old_inflight);
1161
1162         /*
1163          * The inflight->kref was initialized to 1. We decrement it here to
1164          * indicate the start of the flush operation so that it will reach 0
1165          * when all the reqs are finished.
1166          */
1167         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1168                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1169
1170         /* Flush both the vhost poll and vhost work */
1171         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1172                 vhost_scsi_flush_vq(vs, i);
1173         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1174         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1175
1176         /* Wait for all reqs issued before the flush to be finished */
1177         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1178                 wait_for_completion(&old_inflight[i]->comp);
1179 }
1180
1181 /*
1182  * Called from vhost_scsi_ioctl() context to walk the list of available
1183  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1184  *
1185  *  The lock nesting rule is:
1186  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1187  */
1188 static int
1189 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1190                         struct vhost_scsi_target *t)
1191 {
1192         struct se_portal_group *se_tpg;
1193         struct vhost_scsi_tport *tv_tport;
1194         struct vhost_scsi_tpg *tpg;
1195         struct vhost_scsi_tpg **vs_tpg;
1196         struct vhost_virtqueue *vq;
1197         int index, ret, i, len;
1198         bool match = false;
1199
1200         mutex_lock(&vhost_scsi_mutex);
1201         mutex_lock(&vs->dev.mutex);
1202
1203         /* Verify that ring has been setup correctly. */
1204         for (index = 0; index < vs->dev.nvqs; ++index) {
1205                 /* Verify that ring has been setup correctly. */
1206                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1207                         ret = -EFAULT;
1208                         goto out;
1209                 }
1210         }
1211
1212         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1213         vs_tpg = kzalloc(len, GFP_KERNEL);
1214         if (!vs_tpg) {
1215                 ret = -ENOMEM;
1216                 goto out;
1217         }
1218         if (vs->vs_tpg)
1219                 memcpy(vs_tpg, vs->vs_tpg, len);
1220
1221         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1222                 mutex_lock(&tpg->tv_tpg_mutex);
1223                 if (!tpg->tpg_nexus) {
1224                         mutex_unlock(&tpg->tv_tpg_mutex);
1225                         continue;
1226                 }
1227                 if (tpg->tv_tpg_vhost_count != 0) {
1228                         mutex_unlock(&tpg->tv_tpg_mutex);
1229                         continue;
1230                 }
1231                 tv_tport = tpg->tport;
1232
1233                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1234                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1235                                 kfree(vs_tpg);
1236                                 mutex_unlock(&tpg->tv_tpg_mutex);
1237                                 ret = -EEXIST;
1238                                 goto out;
1239                         }
1240                         /*
1241                          * In order to ensure individual vhost-scsi configfs
1242                          * groups cannot be removed while in use by vhost ioctl,
1243                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1244                          * dependency now.
1245                          */
1246                         se_tpg = &tpg->se_tpg;
1247                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1248                         if (ret) {
1249                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1250                                 kfree(vs_tpg);
1251                                 mutex_unlock(&tpg->tv_tpg_mutex);
1252                                 goto out;
1253                         }
1254                         tpg->tv_tpg_vhost_count++;
1255                         tpg->vhost_scsi = vs;
1256                         vs_tpg[tpg->tport_tpgt] = tpg;
1257                         smp_mb__after_atomic();
1258                         match = true;
1259                 }
1260                 mutex_unlock(&tpg->tv_tpg_mutex);
1261         }
1262
1263         if (match) {
1264                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1265                        sizeof(vs->vs_vhost_wwpn));
1266                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1267                         vq = &vs->vqs[i].vq;
1268                         mutex_lock(&vq->mutex);
1269                         vq->private_data = vs_tpg;
1270                         vhost_vq_init_access(vq);
1271                         mutex_unlock(&vq->mutex);
1272                 }
1273                 ret = 0;
1274         } else {
1275                 ret = -EEXIST;
1276         }
1277
1278         /*
1279          * Act as synchronize_rcu to make sure access to
1280          * old vs->vs_tpg is finished.
1281          */
1282         vhost_scsi_flush(vs);
1283         kfree(vs->vs_tpg);
1284         vs->vs_tpg = vs_tpg;
1285
1286 out:
1287         mutex_unlock(&vs->dev.mutex);
1288         mutex_unlock(&vhost_scsi_mutex);
1289         return ret;
1290 }
1291
1292 static int
1293 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1294                           struct vhost_scsi_target *t)
1295 {
1296         struct se_portal_group *se_tpg;
1297         struct vhost_scsi_tport *tv_tport;
1298         struct vhost_scsi_tpg *tpg;
1299         struct vhost_virtqueue *vq;
1300         bool match = false;
1301         int index, ret, i;
1302         u8 target;
1303
1304         mutex_lock(&vhost_scsi_mutex);
1305         mutex_lock(&vs->dev.mutex);
1306         /* Verify that ring has been setup correctly. */
1307         for (index = 0; index < vs->dev.nvqs; ++index) {
1308                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1309                         ret = -EFAULT;
1310                         goto err_dev;
1311                 }
1312         }
1313
1314         if (!vs->vs_tpg) {
1315                 ret = 0;
1316                 goto err_dev;
1317         }
1318
1319         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1320                 target = i;
1321                 tpg = vs->vs_tpg[target];
1322                 if (!tpg)
1323                         continue;
1324
1325                 mutex_lock(&tpg->tv_tpg_mutex);
1326                 tv_tport = tpg->tport;
1327                 if (!tv_tport) {
1328                         ret = -ENODEV;
1329                         goto err_tpg;
1330                 }
1331
1332                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1333                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1334                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1335                                 tv_tport->tport_name, tpg->tport_tpgt,
1336                                 t->vhost_wwpn, t->vhost_tpgt);
1337                         ret = -EINVAL;
1338                         goto err_tpg;
1339                 }
1340                 tpg->tv_tpg_vhost_count--;
1341                 tpg->vhost_scsi = NULL;
1342                 vs->vs_tpg[target] = NULL;
1343                 match = true;
1344                 mutex_unlock(&tpg->tv_tpg_mutex);
1345                 /*
1346                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1347                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1348                  */
1349                 se_tpg = &tpg->se_tpg;
1350                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1351         }
1352         if (match) {
1353                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1354                         vq = &vs->vqs[i].vq;
1355                         mutex_lock(&vq->mutex);
1356                         vq->private_data = NULL;
1357                         mutex_unlock(&vq->mutex);
1358                 }
1359         }
1360         /*
1361          * Act as synchronize_rcu to make sure access to
1362          * old vs->vs_tpg is finished.
1363          */
1364         vhost_scsi_flush(vs);
1365         kfree(vs->vs_tpg);
1366         vs->vs_tpg = NULL;
1367         WARN_ON(vs->vs_events_nr);
1368         mutex_unlock(&vs->dev.mutex);
1369         mutex_unlock(&vhost_scsi_mutex);
1370         return 0;
1371
1372 err_tpg:
1373         mutex_unlock(&tpg->tv_tpg_mutex);
1374 err_dev:
1375         mutex_unlock(&vs->dev.mutex);
1376         mutex_unlock(&vhost_scsi_mutex);
1377         return ret;
1378 }
1379
1380 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1381 {
1382         struct vhost_virtqueue *vq;
1383         int i;
1384
1385         if (features & ~VHOST_SCSI_FEATURES)
1386                 return -EOPNOTSUPP;
1387
1388         mutex_lock(&vs->dev.mutex);
1389         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1390             !vhost_log_access_ok(&vs->dev)) {
1391                 mutex_unlock(&vs->dev.mutex);
1392                 return -EFAULT;
1393         }
1394
1395         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1396                 vq = &vs->vqs[i].vq;
1397                 mutex_lock(&vq->mutex);
1398                 vq->acked_features = features;
1399                 mutex_unlock(&vq->mutex);
1400         }
1401         mutex_unlock(&vs->dev.mutex);
1402         return 0;
1403 }
1404
1405 static int vhost_scsi_open(struct inode *inode, struct file *f)
1406 {
1407         struct vhost_scsi *vs;
1408         struct vhost_virtqueue **vqs;
1409         int r = -ENOMEM, i;
1410
1411         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_RETRY_MAYFAIL);
1412         if (!vs) {
1413                 vs = vzalloc(sizeof(*vs));
1414                 if (!vs)
1415                         goto err_vs;
1416         }
1417
1418         vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1419         if (!vqs)
1420                 goto err_vqs;
1421
1422         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1423         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1424
1425         vs->vs_events_nr = 0;
1426         vs->vs_events_missed = false;
1427
1428         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1429         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1430         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1431         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1432         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1433                 vqs[i] = &vs->vqs[i].vq;
1434                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1435         }
1436         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ, VHOST_SCSI_WEIGHT, 0);
1437
1438         vhost_scsi_init_inflight(vs, NULL);
1439
1440         f->private_data = vs;
1441         return 0;
1442
1443 err_vqs:
1444         kvfree(vs);
1445 err_vs:
1446         return r;
1447 }
1448
1449 static int vhost_scsi_release(struct inode *inode, struct file *f)
1450 {
1451         struct vhost_scsi *vs = f->private_data;
1452         struct vhost_scsi_target t;
1453
1454         mutex_lock(&vs->dev.mutex);
1455         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1456         mutex_unlock(&vs->dev.mutex);
1457         vhost_scsi_clear_endpoint(vs, &t);
1458         vhost_dev_stop(&vs->dev);
1459         vhost_dev_cleanup(&vs->dev, false);
1460         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1461         vhost_scsi_flush(vs);
1462         kfree(vs->dev.vqs);
1463         kvfree(vs);
1464         return 0;
1465 }
1466
1467 static long
1468 vhost_scsi_ioctl(struct file *f,
1469                  unsigned int ioctl,
1470                  unsigned long arg)
1471 {
1472         struct vhost_scsi *vs = f->private_data;
1473         struct vhost_scsi_target backend;
1474         void __user *argp = (void __user *)arg;
1475         u64 __user *featurep = argp;
1476         u32 __user *eventsp = argp;
1477         u32 events_missed;
1478         u64 features;
1479         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1480         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1481
1482         switch (ioctl) {
1483         case VHOST_SCSI_SET_ENDPOINT:
1484                 if (copy_from_user(&backend, argp, sizeof backend))
1485                         return -EFAULT;
1486                 if (backend.reserved != 0)
1487                         return -EOPNOTSUPP;
1488
1489                 return vhost_scsi_set_endpoint(vs, &backend);
1490         case VHOST_SCSI_CLEAR_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_clear_endpoint(vs, &backend);
1497         case VHOST_SCSI_GET_ABI_VERSION:
1498                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1499                         return -EFAULT;
1500                 return 0;
1501         case VHOST_SCSI_SET_EVENTS_MISSED:
1502                 if (get_user(events_missed, eventsp))
1503                         return -EFAULT;
1504                 mutex_lock(&vq->mutex);
1505                 vs->vs_events_missed = events_missed;
1506                 mutex_unlock(&vq->mutex);
1507                 return 0;
1508         case VHOST_SCSI_GET_EVENTS_MISSED:
1509                 mutex_lock(&vq->mutex);
1510                 events_missed = vs->vs_events_missed;
1511                 mutex_unlock(&vq->mutex);
1512                 if (put_user(events_missed, eventsp))
1513                         return -EFAULT;
1514                 return 0;
1515         case VHOST_GET_FEATURES:
1516                 features = VHOST_SCSI_FEATURES;
1517                 if (copy_to_user(featurep, &features, sizeof features))
1518                         return -EFAULT;
1519                 return 0;
1520         case VHOST_SET_FEATURES:
1521                 if (copy_from_user(&features, featurep, sizeof features))
1522                         return -EFAULT;
1523                 return vhost_scsi_set_features(vs, features);
1524         default:
1525                 mutex_lock(&vs->dev.mutex);
1526                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1527                 /* TODO: flush backend after dev ioctl. */
1528                 if (r == -ENOIOCTLCMD)
1529                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1530                 mutex_unlock(&vs->dev.mutex);
1531                 return r;
1532         }
1533 }
1534
1535 #ifdef CONFIG_COMPAT
1536 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1537                                 unsigned long arg)
1538 {
1539         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1540 }
1541 #endif
1542
1543 static const struct file_operations vhost_scsi_fops = {
1544         .owner          = THIS_MODULE,
1545         .release        = vhost_scsi_release,
1546         .unlocked_ioctl = vhost_scsi_ioctl,
1547 #ifdef CONFIG_COMPAT
1548         .compat_ioctl   = vhost_scsi_compat_ioctl,
1549 #endif
1550         .open           = vhost_scsi_open,
1551         .llseek         = noop_llseek,
1552 };
1553
1554 static struct miscdevice vhost_scsi_misc = {
1555         MISC_DYNAMIC_MINOR,
1556         "vhost-scsi",
1557         &vhost_scsi_fops,
1558 };
1559
1560 static int __init vhost_scsi_register(void)
1561 {
1562         return misc_register(&vhost_scsi_misc);
1563 }
1564
1565 static void vhost_scsi_deregister(void)
1566 {
1567         misc_deregister(&vhost_scsi_misc);
1568 }
1569
1570 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1571 {
1572         switch (tport->tport_proto_id) {
1573         case SCSI_PROTOCOL_SAS:
1574                 return "SAS";
1575         case SCSI_PROTOCOL_FCP:
1576                 return "FCP";
1577         case SCSI_PROTOCOL_ISCSI:
1578                 return "iSCSI";
1579         default:
1580                 break;
1581         }
1582
1583         return "Unknown";
1584 }
1585
1586 static void
1587 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1588                   struct se_lun *lun, bool plug)
1589 {
1590
1591         struct vhost_scsi *vs = tpg->vhost_scsi;
1592         struct vhost_virtqueue *vq;
1593         u32 reason;
1594
1595         if (!vs)
1596                 return;
1597
1598         mutex_lock(&vs->dev.mutex);
1599
1600         if (plug)
1601                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1602         else
1603                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1604
1605         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1606         mutex_lock(&vq->mutex);
1607         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1608                 vhost_scsi_send_evt(vs, tpg, lun,
1609                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1610         mutex_unlock(&vq->mutex);
1611         mutex_unlock(&vs->dev.mutex);
1612 }
1613
1614 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1615 {
1616         vhost_scsi_do_plug(tpg, lun, true);
1617 }
1618
1619 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1620 {
1621         vhost_scsi_do_plug(tpg, lun, false);
1622 }
1623
1624 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1625                                struct se_lun *lun)
1626 {
1627         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1628                                 struct vhost_scsi_tpg, se_tpg);
1629
1630         mutex_lock(&vhost_scsi_mutex);
1631
1632         mutex_lock(&tpg->tv_tpg_mutex);
1633         tpg->tv_tpg_port_count++;
1634         mutex_unlock(&tpg->tv_tpg_mutex);
1635
1636         vhost_scsi_hotplug(tpg, lun);
1637
1638         mutex_unlock(&vhost_scsi_mutex);
1639
1640         return 0;
1641 }
1642
1643 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1644                                   struct se_lun *lun)
1645 {
1646         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1647                                 struct vhost_scsi_tpg, se_tpg);
1648
1649         mutex_lock(&vhost_scsi_mutex);
1650
1651         mutex_lock(&tpg->tv_tpg_mutex);
1652         tpg->tv_tpg_port_count--;
1653         mutex_unlock(&tpg->tv_tpg_mutex);
1654
1655         vhost_scsi_hotunplug(tpg, lun);
1656
1657         mutex_unlock(&vhost_scsi_mutex);
1658 }
1659
1660 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1661 {
1662         struct vhost_scsi_cmd *tv_cmd;
1663         unsigned int i;
1664
1665         if (!se_sess->sess_cmd_map)
1666                 return;
1667
1668         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1669                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1670
1671                 kfree(tv_cmd->tvc_sgl);
1672                 kfree(tv_cmd->tvc_prot_sgl);
1673                 kfree(tv_cmd->tvc_upages);
1674         }
1675 }
1676
1677 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1678                 struct config_item *item, const char *page, size_t count)
1679 {
1680         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1681         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1682                                 struct vhost_scsi_tpg, se_tpg);
1683         unsigned long val;
1684         int ret = kstrtoul(page, 0, &val);
1685
1686         if (ret) {
1687                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1688                 return ret;
1689         }
1690         if (val != 0 && val != 1 && val != 3) {
1691                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1692                 return -EINVAL;
1693         }
1694         tpg->tv_fabric_prot_type = val;
1695
1696         return count;
1697 }
1698
1699 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1700                 struct config_item *item, char *page)
1701 {
1702         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1703         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1704                                 struct vhost_scsi_tpg, se_tpg);
1705
1706         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1707 }
1708
1709 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1710
1711 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1712         &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1713         NULL,
1714 };
1715
1716 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1717                                struct se_session *se_sess, void *p)
1718 {
1719         struct vhost_scsi_cmd *tv_cmd;
1720         unsigned int i;
1721
1722         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1723                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1724
1725                 tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1726                                         VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1727                 if (!tv_cmd->tvc_sgl) {
1728                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1729                         goto out;
1730                 }
1731
1732                 tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1733                                 VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1734                 if (!tv_cmd->tvc_upages) {
1735                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1736                         goto out;
1737                 }
1738
1739                 tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1740                                 VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1741                 if (!tv_cmd->tvc_prot_sgl) {
1742                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1743                         goto out;
1744                 }
1745         }
1746         return 0;
1747 out:
1748         vhost_scsi_free_cmd_map_res(se_sess);
1749         return -ENOMEM;
1750 }
1751
1752 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1753                                 const char *name)
1754 {
1755         struct vhost_scsi_nexus *tv_nexus;
1756
1757         mutex_lock(&tpg->tv_tpg_mutex);
1758         if (tpg->tpg_nexus) {
1759                 mutex_unlock(&tpg->tv_tpg_mutex);
1760                 pr_debug("tpg->tpg_nexus already exists\n");
1761                 return -EEXIST;
1762         }
1763
1764         tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1765         if (!tv_nexus) {
1766                 mutex_unlock(&tpg->tv_tpg_mutex);
1767                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1768                 return -ENOMEM;
1769         }
1770         /*
1771          * Since we are running in 'demo mode' this call with generate a
1772          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1773          * the SCSI Initiator port name of the passed configfs group 'name'.
1774          */
1775         tv_nexus->tvn_se_sess = target_alloc_session(&tpg->se_tpg,
1776                                         VHOST_SCSI_DEFAULT_TAGS,
1777                                         sizeof(struct vhost_scsi_cmd),
1778                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1779                                         (unsigned char *)name, tv_nexus,
1780                                         vhost_scsi_nexus_cb);
1781         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1782                 mutex_unlock(&tpg->tv_tpg_mutex);
1783                 kfree(tv_nexus);
1784                 return -ENOMEM;
1785         }
1786         tpg->tpg_nexus = tv_nexus;
1787
1788         mutex_unlock(&tpg->tv_tpg_mutex);
1789         return 0;
1790 }
1791
1792 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1793 {
1794         struct se_session *se_sess;
1795         struct vhost_scsi_nexus *tv_nexus;
1796
1797         mutex_lock(&tpg->tv_tpg_mutex);
1798         tv_nexus = tpg->tpg_nexus;
1799         if (!tv_nexus) {
1800                 mutex_unlock(&tpg->tv_tpg_mutex);
1801                 return -ENODEV;
1802         }
1803
1804         se_sess = tv_nexus->tvn_se_sess;
1805         if (!se_sess) {
1806                 mutex_unlock(&tpg->tv_tpg_mutex);
1807                 return -ENODEV;
1808         }
1809
1810         if (tpg->tv_tpg_port_count != 0) {
1811                 mutex_unlock(&tpg->tv_tpg_mutex);
1812                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1813                         " active TPG port count: %d\n",
1814                         tpg->tv_tpg_port_count);
1815                 return -EBUSY;
1816         }
1817
1818         if (tpg->tv_tpg_vhost_count != 0) {
1819                 mutex_unlock(&tpg->tv_tpg_mutex);
1820                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1821                         " active TPG vhost count: %d\n",
1822                         tpg->tv_tpg_vhost_count);
1823                 return -EBUSY;
1824         }
1825
1826         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1827                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1828                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1829
1830         vhost_scsi_free_cmd_map_res(se_sess);
1831         /*
1832          * Release the SCSI I_T Nexus to the emulated vhost Target Port
1833          */
1834         transport_deregister_session(tv_nexus->tvn_se_sess);
1835         tpg->tpg_nexus = NULL;
1836         mutex_unlock(&tpg->tv_tpg_mutex);
1837
1838         kfree(tv_nexus);
1839         return 0;
1840 }
1841
1842 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1843 {
1844         struct se_portal_group *se_tpg = to_tpg(item);
1845         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1846                                 struct vhost_scsi_tpg, se_tpg);
1847         struct vhost_scsi_nexus *tv_nexus;
1848         ssize_t ret;
1849
1850         mutex_lock(&tpg->tv_tpg_mutex);
1851         tv_nexus = tpg->tpg_nexus;
1852         if (!tv_nexus) {
1853                 mutex_unlock(&tpg->tv_tpg_mutex);
1854                 return -ENODEV;
1855         }
1856         ret = snprintf(page, PAGE_SIZE, "%s\n",
1857                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1858         mutex_unlock(&tpg->tv_tpg_mutex);
1859
1860         return ret;
1861 }
1862
1863 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1864                 const char *page, size_t count)
1865 {
1866         struct se_portal_group *se_tpg = to_tpg(item);
1867         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1868                                 struct vhost_scsi_tpg, se_tpg);
1869         struct vhost_scsi_tport *tport_wwn = tpg->tport;
1870         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1871         int ret;
1872         /*
1873          * Shutdown the active I_T nexus if 'NULL' is passed..
1874          */
1875         if (!strncmp(page, "NULL", 4)) {
1876                 ret = vhost_scsi_drop_nexus(tpg);
1877                 return (!ret) ? count : ret;
1878         }
1879         /*
1880          * Otherwise make sure the passed virtual Initiator port WWN matches
1881          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1882          * vhost_scsi_make_nexus().
1883          */
1884         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1885                 pr_err("Emulated NAA Sas Address: %s, exceeds"
1886                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
1887                 return -EINVAL;
1888         }
1889         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1890
1891         ptr = strstr(i_port, "naa.");
1892         if (ptr) {
1893                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1894                         pr_err("Passed SAS Initiator Port %s does not"
1895                                 " match target port protoid: %s\n", i_port,
1896                                 vhost_scsi_dump_proto_id(tport_wwn));
1897                         return -EINVAL;
1898                 }
1899                 port_ptr = &i_port[0];
1900                 goto check_newline;
1901         }
1902         ptr = strstr(i_port, "fc.");
1903         if (ptr) {
1904                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1905                         pr_err("Passed FCP Initiator Port %s does not"
1906                                 " match target port protoid: %s\n", i_port,
1907                                 vhost_scsi_dump_proto_id(tport_wwn));
1908                         return -EINVAL;
1909                 }
1910                 port_ptr = &i_port[3]; /* Skip over "fc." */
1911                 goto check_newline;
1912         }
1913         ptr = strstr(i_port, "iqn.");
1914         if (ptr) {
1915                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1916                         pr_err("Passed iSCSI Initiator Port %s does not"
1917                                 " match target port protoid: %s\n", i_port,
1918                                 vhost_scsi_dump_proto_id(tport_wwn));
1919                         return -EINVAL;
1920                 }
1921                 port_ptr = &i_port[0];
1922                 goto check_newline;
1923         }
1924         pr_err("Unable to locate prefix for emulated Initiator Port:"
1925                         " %s\n", i_port);
1926         return -EINVAL;
1927         /*
1928          * Clear any trailing newline for the NAA WWN
1929          */
1930 check_newline:
1931         if (i_port[strlen(i_port)-1] == '\n')
1932                 i_port[strlen(i_port)-1] = '\0';
1933
1934         ret = vhost_scsi_make_nexus(tpg, port_ptr);
1935         if (ret < 0)
1936                 return ret;
1937
1938         return count;
1939 }
1940
1941 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1942
1943 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1944         &vhost_scsi_tpg_attr_nexus,
1945         NULL,
1946 };
1947
1948 static struct se_portal_group *
1949 vhost_scsi_make_tpg(struct se_wwn *wwn,
1950                    struct config_group *group,
1951                    const char *name)
1952 {
1953         struct vhost_scsi_tport *tport = container_of(wwn,
1954                         struct vhost_scsi_tport, tport_wwn);
1955
1956         struct vhost_scsi_tpg *tpg;
1957         u16 tpgt;
1958         int ret;
1959
1960         if (strstr(name, "tpgt_") != name)
1961                 return ERR_PTR(-EINVAL);
1962         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1963                 return ERR_PTR(-EINVAL);
1964
1965         tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1966         if (!tpg) {
1967                 pr_err("Unable to allocate struct vhost_scsi_tpg");
1968                 return ERR_PTR(-ENOMEM);
1969         }
1970         mutex_init(&tpg->tv_tpg_mutex);
1971         INIT_LIST_HEAD(&tpg->tv_tpg_list);
1972         tpg->tport = tport;
1973         tpg->tport_tpgt = tpgt;
1974
1975         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1976         if (ret < 0) {
1977                 kfree(tpg);
1978                 return NULL;
1979         }
1980         mutex_lock(&vhost_scsi_mutex);
1981         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1982         mutex_unlock(&vhost_scsi_mutex);
1983
1984         return &tpg->se_tpg;
1985 }
1986
1987 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1988 {
1989         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1990                                 struct vhost_scsi_tpg, se_tpg);
1991
1992         mutex_lock(&vhost_scsi_mutex);
1993         list_del(&tpg->tv_tpg_list);
1994         mutex_unlock(&vhost_scsi_mutex);
1995         /*
1996          * Release the virtual I_T Nexus for this vhost TPG
1997          */
1998         vhost_scsi_drop_nexus(tpg);
1999         /*
2000          * Deregister the se_tpg from TCM..
2001          */
2002         core_tpg_deregister(se_tpg);
2003         kfree(tpg);
2004 }
2005
2006 static struct se_wwn *
2007 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2008                      struct config_group *group,
2009                      const char *name)
2010 {
2011         struct vhost_scsi_tport *tport;
2012         char *ptr;
2013         u64 wwpn = 0;
2014         int off = 0;
2015
2016         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2017                 return ERR_PTR(-EINVAL); */
2018
2019         tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2020         if (!tport) {
2021                 pr_err("Unable to allocate struct vhost_scsi_tport");
2022                 return ERR_PTR(-ENOMEM);
2023         }
2024         tport->tport_wwpn = wwpn;
2025         /*
2026          * Determine the emulated Protocol Identifier and Target Port Name
2027          * based on the incoming configfs directory name.
2028          */
2029         ptr = strstr(name, "naa.");
2030         if (ptr) {
2031                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2032                 goto check_len;
2033         }
2034         ptr = strstr(name, "fc.");
2035         if (ptr) {
2036                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2037                 off = 3; /* Skip over "fc." */
2038                 goto check_len;
2039         }
2040         ptr = strstr(name, "iqn.");
2041         if (ptr) {
2042                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2043                 goto check_len;
2044         }
2045
2046         pr_err("Unable to locate prefix for emulated Target Port:"
2047                         " %s\n", name);
2048         kfree(tport);
2049         return ERR_PTR(-EINVAL);
2050
2051 check_len:
2052         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2053                 pr_err("Emulated %s Address: %s, exceeds"
2054                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2055                         VHOST_SCSI_NAMELEN);
2056                 kfree(tport);
2057                 return ERR_PTR(-EINVAL);
2058         }
2059         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2060
2061         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2062                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2063
2064         return &tport->tport_wwn;
2065 }
2066
2067 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2068 {
2069         struct vhost_scsi_tport *tport = container_of(wwn,
2070                                 struct vhost_scsi_tport, tport_wwn);
2071
2072         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2073                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2074                 tport->tport_name);
2075
2076         kfree(tport);
2077 }
2078
2079 static ssize_t
2080 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2081 {
2082         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2083                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2084                 utsname()->machine);
2085 }
2086
2087 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2088
2089 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2090         &vhost_scsi_wwn_attr_version,
2091         NULL,
2092 };
2093
2094 static const struct target_core_fabric_ops vhost_scsi_ops = {
2095         .module                         = THIS_MODULE,
2096         .name                           = "vhost",
2097         .get_fabric_name                = vhost_scsi_get_fabric_name,
2098         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2099         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2100         .tpg_check_demo_mode            = vhost_scsi_check_true,
2101         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2102         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2103         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2104         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2105         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2106         .release_cmd                    = vhost_scsi_release_cmd,
2107         .check_stop_free                = vhost_scsi_check_stop_free,
2108         .sess_get_index                 = vhost_scsi_sess_get_index,
2109         .sess_get_initiator_sid         = NULL,
2110         .write_pending                  = vhost_scsi_write_pending,
2111         .write_pending_status           = vhost_scsi_write_pending_status,
2112         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2113         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2114         .queue_data_in                  = vhost_scsi_queue_data_in,
2115         .queue_status                   = vhost_scsi_queue_status,
2116         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2117         .aborted_task                   = vhost_scsi_aborted_task,
2118         /*
2119          * Setup callers for generic logic in target_core_fabric_configfs.c
2120          */
2121         .fabric_make_wwn                = vhost_scsi_make_tport,
2122         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2123         .fabric_make_tpg                = vhost_scsi_make_tpg,
2124         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2125         .fabric_post_link               = vhost_scsi_port_link,
2126         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2127
2128         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2129         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2130         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2131 };
2132
2133 static int __init vhost_scsi_init(void)
2134 {
2135         int ret = -ENOMEM;
2136
2137         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2138                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2139                 utsname()->machine);
2140
2141         /*
2142          * Use our own dedicated workqueue for submitting I/O into
2143          * target core to avoid contention within system_wq.
2144          */
2145         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2146         if (!vhost_scsi_workqueue)
2147                 goto out;
2148
2149         ret = vhost_scsi_register();
2150         if (ret < 0)
2151                 goto out_destroy_workqueue;
2152
2153         ret = target_register_template(&vhost_scsi_ops);
2154         if (ret < 0)
2155                 goto out_vhost_scsi_deregister;
2156
2157         return 0;
2158
2159 out_vhost_scsi_deregister:
2160         vhost_scsi_deregister();
2161 out_destroy_workqueue:
2162         destroy_workqueue(vhost_scsi_workqueue);
2163 out:
2164         return ret;
2165 };
2166
2167 static void vhost_scsi_exit(void)
2168 {
2169         target_unregister_template(&vhost_scsi_ops);
2170         vhost_scsi_deregister();
2171         destroy_workqueue(vhost_scsi_workqueue);
2172 };
2173
2174 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2175 MODULE_ALIAS("tcm_vhost");
2176 MODULE_LICENSE("GPL");
2177 module_init(vhost_scsi_init);
2178 module_exit(vhost_scsi_exit);