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 <target/target_core_fabric_configfs.h>
  46 #include <target/configfs_macros.h>
  47 #include <linux/vhost.h>
  48 #include <linux/virtio_scsi.h>
  49 #include <linux/llist.h>
  50 #include <linux/bitmap.h>
  51 #include <linux/percpu_ida.h>
  52
  53 #include "vhost.h"
  54
  55 #define VHOST_SCSI_VERSION  "v0.1"
  56 #define VHOST_SCSI_NAMELEN 256
  57 #define VHOST_SCSI_MAX_CDB_SIZE 32
  58 #define VHOST_SCSI_DEFAULT_TAGS 256
  59 #define VHOST_SCSI_PREALLOC_SGLS 2048
  60 #define VHOST_SCSI_PREALLOC_UPAGES 2048
  61 #define VHOST_SCSI_PREALLOC_PROT_SGLS 512
  62
  63 struct vhost_scsi_inflight {
  64         /* Wait for the flush operation to finish */
  65         struct completion comp;
  66         /* Refcount for the inflight reqs */
  67         struct kref kref;
  68 };
  69
  70 struct vhost_scsi_cmd {
  71         /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
  72         int tvc_vq_desc;
  73         /* virtio-scsi initiator task attribute */
  74         int tvc_task_attr;
  75         /* virtio-scsi response incoming iovecs */
  76         int tvc_in_iovs;
  77         /* virtio-scsi initiator data direction */
  78         enum dma_data_direction tvc_data_direction;
  79         /* Expected data transfer length from virtio-scsi header */
  80         u32 tvc_exp_data_len;
  81         /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
  82         u64 tvc_tag;
  83         /* The number of scatterlists associated with this cmd */
  84         u32 tvc_sgl_count;
  85         u32 tvc_prot_sgl_count;
  86         /* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
  87         u32 tvc_lun;
  88         /* Pointer to the SGL formatted memory from virtio-scsi */
  89         struct scatterlist *tvc_sgl;
  90         struct scatterlist *tvc_prot_sgl;
  91         struct page **tvc_upages;
  92         /* Pointer to response header iovec */
  93         struct iovec *tvc_resp_iov;
  94         /* Pointer to vhost_scsi for our device */
  95         struct vhost_scsi *tvc_vhost;
  96         /* Pointer to vhost_virtqueue for the cmd */
  97         struct vhost_virtqueue *tvc_vq;
  98         /* Pointer to vhost nexus memory */
  99         struct vhost_scsi_nexus *tvc_nexus;
 100         /* The TCM I/O descriptor that is accessed via container_of() */
 101         struct se_cmd tvc_se_cmd;
 102         /* work item used for cmwq dispatch to vhost_scsi_submission_work() */
 103         struct work_struct work;
 104         /* Copy of the incoming SCSI command descriptor block (CDB) */
 105         unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
 106         /* Sense buffer that will be mapped into outgoing status */
 107         unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
 108         /* Completed commands list, serviced from vhost worker thread */
 109         struct llist_node tvc_completion_list;
 110         /* Used to track inflight cmd */
 111         struct vhost_scsi_inflight *inflight;
 112 };
 113
 114 struct vhost_scsi_nexus {
 115         /* Pointer to TCM session for I_T Nexus */
 116         struct se_session *tvn_se_sess;
 117 };
 118
 119 struct vhost_scsi_tpg {
 120         /* Vhost port target portal group tag for TCM */
 121         u16 tport_tpgt;
 122         /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
 123         int tv_tpg_port_count;
 124         /* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
 125         int tv_tpg_vhost_count;
 126         /* Used for enabling T10-PI with legacy devices */
 127         int tv_fabric_prot_type;
 128         /* list for vhost_scsi_list */
 129         struct list_head tv_tpg_list;
 130         /* Used to protect access for tpg_nexus */
 131         struct mutex tv_tpg_mutex;
 132         /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
 133         struct vhost_scsi_nexus *tpg_nexus;
 134         /* Pointer back to vhost_scsi_tport */
 135         struct vhost_scsi_tport *tport;
 136         /* Returned by vhost_scsi_make_tpg() */
 137         struct se_portal_group se_tpg;
 138         /* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
 139         struct vhost_scsi *vhost_scsi;
 140 };
 141
 142 struct vhost_scsi_tport {
 143         /* SCSI protocol the tport is providing */
 144         u8 tport_proto_id;
 145         /* Binary World Wide unique Port Name for Vhost Target port */
 146         u64 tport_wwpn;
 147         /* ASCII formatted WWPN for Vhost Target port */
 148         char tport_name[VHOST_SCSI_NAMELEN];
 149         /* Returned by vhost_scsi_make_tport() */
 150         struct se_wwn tport_wwn;
 151 };
 152
 153 struct vhost_scsi_evt {
 154         /* event to be sent to guest */
 155         struct virtio_scsi_event event;
 156         /* event list, serviced from vhost worker thread */
 157         struct llist_node list;
 158 };
 159
 160 enum {
 161         VHOST_SCSI_VQ_CTL = 0,
 162         VHOST_SCSI_VQ_EVT = 1,
 163         VHOST_SCSI_VQ_IO = 2,
 164 };
 165
 166 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
 167 enum {
 168         VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
 169                                                (1ULL << VIRTIO_SCSI_F_T10_PI) |
 170                                                (1ULL << VIRTIO_F_ANY_LAYOUT) |
 171                                                (1ULL << VIRTIO_F_VERSION_1)
 172 };
 173
 174 #define VHOST_SCSI_MAX_TARGET   256
 175 #define VHOST_SCSI_MAX_VQ       128
 176 #define VHOST_SCSI_MAX_EVENT    128
 177
 178 struct vhost_scsi_virtqueue {
 179         struct vhost_virtqueue vq;
 180         /*
 181          * Reference counting for inflight reqs, used for flush operation. At
 182          * each time, one reference tracks new commands submitted, while we
 183          * wait for another one to reach 0.
 184          */
 185         struct vhost_scsi_inflight inflights[2];
 186         /*
 187          * Indicate current inflight in use, protected by vq->mutex.
 188          * Writers must also take dev mutex and flush under it.
 189          */
 190         int inflight_idx;
 191 };
 192
 193 struct vhost_scsi {
 194         /* Protected by vhost_scsi->dev.mutex */
 195         struct vhost_scsi_tpg **vs_tpg;
 196         char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
 197
 198         struct vhost_dev dev;
 199         struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
 200
 201         struct vhost_work vs_completion_work; /* cmd completion work item */
 202         struct llist_head vs_completion_list; /* cmd completion queue */
 203
 204         struct vhost_work vs_event_work; /* evt injection work item */
 205         struct llist_head vs_event_list; /* evt injection queue */
 206
 207         bool vs_events_missed; /* any missed events, protected by vq->mutex */
 208         int vs_events_nr; /* num of pending events, protected by vq->mutex */
 209 };
 210
 211 static struct workqueue_struct *vhost_scsi_workqueue;
 212
 213 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
 214 static DEFINE_MUTEX(vhost_scsi_mutex);
 215 static LIST_HEAD(vhost_scsi_list);
 216
 217 static int iov_num_pages(void __user *iov_base, size_t iov_len)
 218 {
 219         return (PAGE_ALIGN((unsigned long)iov_base + iov_len) -
 220                ((unsigned long)iov_base & PAGE_MASK)) >> PAGE_SHIFT;
 221 }
 222
 223 static void vhost_scsi_done_inflight(struct kref *kref)
 224 {
 225         struct vhost_scsi_inflight *inflight;
 226
 227         inflight = container_of(kref, struct vhost_scsi_inflight, kref);
 228         complete(&inflight->comp);
 229 }
 230
 231 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
 232                                     struct vhost_scsi_inflight *old_inflight[])
 233 {
 234         struct vhost_scsi_inflight *new_inflight;
 235         struct vhost_virtqueue *vq;
 236         int idx, i;
 237
 238         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
 239                 vq = &vs->vqs[i].vq;
 240
 241                 mutex_lock(&vq->mutex);
 242
 243                 /* store old infight */
 244                 idx = vs->vqs[i].inflight_idx;
 245                 if (old_inflight)
 246                         old_inflight[i] = &vs->vqs[i].inflights[idx];
 247
 248                 /* setup new infight */
 249                 vs->vqs[i].inflight_idx = idx ^ 1;
 250                 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
 251                 kref_init(&new_inflight->kref);
 252                 init_completion(&new_inflight->comp);
 253
 254                 mutex_unlock(&vq->mutex);
 255         }
 256 }
 257
 258 static struct vhost_scsi_inflight *
 259 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
 260 {
 261         struct vhost_scsi_inflight *inflight;
 262         struct vhost_scsi_virtqueue *svq;
 263
 264         svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
 265         inflight = &svq->inflights[svq->inflight_idx];
 266         kref_get(&inflight->kref);
 267
 268         return inflight;
 269 }
 270
 271 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
 272 {
 273         kref_put(&inflight->kref, vhost_scsi_done_inflight);
 274 }
 275
 276 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
 277 {
 278         return 1;
 279 }
 280
 281 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
 282 {
 283         return 0;
 284 }
 285
 286 static char *vhost_scsi_get_fabric_name(void)
 287 {
 288         return "vhost";
 289 }
 290
 291 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
 292 {
 293         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 294                                 struct vhost_scsi_tpg, se_tpg);
 295         struct vhost_scsi_tport *tport = tpg->tport;
 296
 297         return &tport->tport_name[0];
 298 }
 299
 300 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
 301 {
 302         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 303                                 struct vhost_scsi_tpg, se_tpg);
 304         return tpg->tport_tpgt;
 305 }
 306
 307 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
 308 {
 309         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 310                                 struct vhost_scsi_tpg, se_tpg);
 311
 312         return tpg->tv_fabric_prot_type;
 313 }
 314
 315 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
 316 {
 317         return 1;
 318 }
 319
 320 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
 321 {
 322         struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
 323                                 struct vhost_scsi_cmd, tvc_se_cmd);
 324         struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
 325         int i;
 326
 327         if (tv_cmd->tvc_sgl_count) {
 328                 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
 329                         put_page(sg_page(&tv_cmd->tvc_sgl[i]));
 330         }
 331         if (tv_cmd->tvc_prot_sgl_count) {
 332                 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
 333                         put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
 334         }
 335
 336         vhost_scsi_put_inflight(tv_cmd->inflight);
 337         percpu_ida_free(&se_sess->sess_tag_pool, se_cmd->map_tag);
 338 }
 339
 340 static int vhost_scsi_shutdown_session(struct se_session *se_sess)
 341 {
 342         return 0;
 343 }
 344
 345 static void vhost_scsi_close_session(struct se_session *se_sess)
 346 {
 347         return;
 348 }
 349
 350 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
 351 {
 352         return 0;
 353 }
 354
 355 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
 356 {
 357         /* Go ahead and process the write immediately */
 358         target_execute_cmd(se_cmd);
 359         return 0;
 360 }
 361
 362 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
 363 {
 364         return 0;
 365 }
 366
 367 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
 368 {
 369         return;
 370 }
 371
 372 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
 373 {
 374         return 0;
 375 }
 376
 377 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
 378 {
 379         struct vhost_scsi *vs = cmd->tvc_vhost;
 380
 381         llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
 382
 383         vhost_work_queue(&vs->dev, &vs->vs_completion_work);
 384 }
 385
 386 static int vhost_scsi_queue_data_in(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 int vhost_scsi_queue_status(struct se_cmd *se_cmd)
 395 {
 396         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 397                                 struct vhost_scsi_cmd, tvc_se_cmd);
 398         vhost_scsi_complete_cmd(cmd);
 399         return 0;
 400 }
 401
 402 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
 403 {
 404         return;
 405 }
 406
 407 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
 408 {
 409         return;
 410 }
 411
 412 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 413 {
 414         vs->vs_events_nr--;
 415         kfree(evt);
 416 }
 417
 418 static struct vhost_scsi_evt *
 419 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
 420                        u32 event, u32 reason)
 421 {
 422         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 423         struct vhost_scsi_evt *evt;
 424
 425         if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
 426                 vs->vs_events_missed = true;
 427                 return NULL;
 428         }
 429
 430         evt = kzalloc(sizeof(*evt), GFP_KERNEL);
 431         if (!evt) {
 432                 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
 433                 vs->vs_events_missed = true;
 434                 return NULL;
 435         }
 436
 437         evt->event.event = cpu_to_vhost32(vq, event);
 438         evt->event.reason = cpu_to_vhost32(vq, reason);
 439         vs->vs_events_nr++;
 440
 441         return evt;
 442 }
 443
 444 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
 445 {
 446         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 447
 448         /* TODO locking against target/backend threads? */
 449         transport_generic_free_cmd(se_cmd, 0);
 450
 451 }
 452
 453 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
 454 {
 455         return target_put_sess_cmd(se_cmd);
 456 }
 457
 458 static void
 459 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 460 {
 461         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 462         struct virtio_scsi_event *event = &evt->event;
 463         struct virtio_scsi_event __user *eventp;
 464         unsigned out, in;
 465         int head, ret;
 466
 467         if (!vq->private_data) {
 468                 vs->vs_events_missed = true;
 469                 return;
 470         }
 471
 472 again:
 473         vhost_disable_notify(&vs->dev, vq);
 474         head = vhost_get_vq_desc(vq, vq->iov,
 475                         ARRAY_SIZE(vq->iov), &out, &in,
 476                         NULL, NULL);
 477         if (head < 0) {
 478                 vs->vs_events_missed = true;
 479                 return;
 480         }
 481         if (head == vq->num) {
 482                 if (vhost_enable_notify(&vs->dev, vq))
 483                         goto again;
 484                 vs->vs_events_missed = true;
 485                 return;
 486         }
 487
 488         if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
 489                 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
 490                                 vq->iov[out].iov_len);
 491                 vs->vs_events_missed = true;
 492                 return;
 493         }
 494
 495         if (vs->vs_events_missed) {
 496                 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
 497                 vs->vs_events_missed = false;
 498         }
 499
 500         eventp = vq->iov[out].iov_base;
 501         ret = __copy_to_user(eventp, event, sizeof(*event));
 502         if (!ret)
 503                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 504         else
 505                 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
 506 }
 507
 508 static void vhost_scsi_evt_work(struct vhost_work *work)
 509 {
 510         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 511                                         vs_event_work);
 512         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 513         struct vhost_scsi_evt *evt;
 514         struct llist_node *llnode;
 515
 516         mutex_lock(&vq->mutex);
 517         llnode = llist_del_all(&vs->vs_event_list);
 518         while (llnode) {
 519                 evt = llist_entry(llnode, struct vhost_scsi_evt, list);
 520                 llnode = llist_next(llnode);
 521                 vhost_scsi_do_evt_work(vs, evt);
 522                 vhost_scsi_free_evt(vs, evt);
 523         }
 524         mutex_unlock(&vq->mutex);
 525 }
 526
 527 /* Fill in status and signal that we are done processing this command
 528  *
 529  * This is scheduled in the vhost work queue so we are called with the owner
 530  * process mm and can access the vring.
 531  */
 532 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 533 {
 534         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 535                                         vs_completion_work);
 536         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
 537         struct virtio_scsi_cmd_resp v_rsp;
 538         struct vhost_scsi_cmd *cmd;
 539         struct llist_node *llnode;
 540         struct se_cmd *se_cmd;
 541         struct iov_iter iov_iter;
 542         int ret, vq;
 543
 544         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
 545         llnode = llist_del_all(&vs->vs_completion_list);
 546         while (llnode) {
 547                 cmd = llist_entry(llnode, struct vhost_scsi_cmd,
 548                                      tvc_completion_list);
 549                 llnode = llist_next(llnode);
 550                 se_cmd = &cmd->tvc_se_cmd;
 551
 552                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
 553                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
 554
 555                 memset(&v_rsp, 0, sizeof(v_rsp));
 556                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
 557                 /* TODO is status_qualifier field needed? */
 558                 v_rsp.status = se_cmd->scsi_status;
 559                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
 560                                                  se_cmd->scsi_sense_length);
 561                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
 562                        se_cmd->scsi_sense_length);
 563
 564                 iov_iter_init(&iov_iter, READ, cmd->tvc_resp_iov,
 565                               cmd->tvc_in_iovs, sizeof(v_rsp));
 566                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
 567                 if (likely(ret == sizeof(v_rsp))) {
 568                         struct vhost_scsi_virtqueue *q;
 569                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
 570                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
 571                         vq = q - vs->vqs;
 572                         __set_bit(vq, signal);
 573                 } else
 574                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
 575
 576                 vhost_scsi_free_cmd(cmd);
 577         }
 578
 579         vq = -1;
 580         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
 581                 < VHOST_SCSI_MAX_VQ)
 582                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
 583 }
 584
 585 static struct vhost_scsi_cmd *
 586 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
 587                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
 588                    u32 exp_data_len, int data_direction)
 589 {
 590         struct vhost_scsi_cmd *cmd;
 591         struct vhost_scsi_nexus *tv_nexus;
 592         struct se_session *se_sess;
 593         struct scatterlist *sg, *prot_sg;
 594         struct page **pages;
 595         int tag;
 596
 597         tv_nexus = tpg->tpg_nexus;
 598         if (!tv_nexus) {
 599                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
 600                 return ERR_PTR(-EIO);
 601         }
 602         se_sess = tv_nexus->tvn_se_sess;
 603
 604         tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
 605         if (tag < 0) {
 606                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
 607                 return ERR_PTR(-ENOMEM);
 608         }
 609
 610         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
 611         sg = cmd->tvc_sgl;
 612         prot_sg = cmd->tvc_prot_sgl;
 613         pages = cmd->tvc_upages;
 614         memset(cmd, 0, sizeof(struct vhost_scsi_cmd));
 615
 616         cmd->tvc_sgl = sg;
 617         cmd->tvc_prot_sgl = prot_sg;
 618         cmd->tvc_upages = pages;
 619         cmd->tvc_se_cmd.map_tag = tag;
 620         cmd->tvc_tag = scsi_tag;
 621         cmd->tvc_lun = lun;
 622         cmd->tvc_task_attr = task_attr;
 623         cmd->tvc_exp_data_len = exp_data_len;
 624         cmd->tvc_data_direction = data_direction;
 625         cmd->tvc_nexus = tv_nexus;
 626         cmd->inflight = vhost_scsi_get_inflight(vq);
 627
 628         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
 629
 630         return cmd;
 631 }
 632
 633 /*
 634  * Map a user memory range into a scatterlist
 635  *
 636  * Returns the number of scatterlist entries used or -errno on error.
 637  */
 638 static int
 639 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
 640                       void __user *ptr,
 641                       size_t len,
 642                       struct scatterlist *sgl,
 643                       bool write)
 644 {
 645         unsigned int npages = 0, offset, nbytes;
 646         unsigned int pages_nr = iov_num_pages(ptr, len);
 647         struct scatterlist *sg = sgl;
 648         struct page **pages = cmd->tvc_upages;
 649         int ret, i;
 650
 651         if (pages_nr > VHOST_SCSI_PREALLOC_UPAGES) {
 652                 pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
 653                        " preallocated VHOST_SCSI_PREALLOC_UPAGES: %u\n",
 654                         pages_nr, VHOST_SCSI_PREALLOC_UPAGES);
 655                 return -ENOBUFS;
 656         }
 657
 658         ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
 659         /* No pages were pinned */
 660         if (ret < 0)
 661                 goto out;
 662         /* Less pages pinned than wanted */
 663         if (ret != pages_nr) {
 664                 for (i = 0; i < ret; i++)
 665                         put_page(pages[i]);
 666                 ret = -EFAULT;
 667                 goto out;
 668         }
 669
 670         while (len > 0) {
 671                 offset = (uintptr_t)ptr & ~PAGE_MASK;
 672                 nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
 673                 sg_set_page(sg, pages[npages], nbytes, offset);
 674                 ptr += nbytes;
 675                 len -= nbytes;
 676                 sg++;
 677                 npages++;
 678         }
 679
 680 out:
 681         return ret;
 682 }
 683
 684 static int
 685 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
 686 {
 687         int sgl_count = 0;
 688
 689         if (!iter || !iter->iov) {
 690                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
 691                        " present\n", __func__, bytes);
 692                 return -EINVAL;
 693         }
 694
 695         sgl_count = iov_iter_npages(iter, 0xffff);
 696         if (sgl_count > max_sgls) {
 697                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
 698                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
 699                 return -EINVAL;
 700         }
 701         return sgl_count;
 702 }
 703
 704 static int
 705 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
 706                       struct iov_iter *iter,
 707                       struct scatterlist *sg, int sg_count)
 708 {
 709         size_t off = iter->iov_offset;
 710         int i, ret;
 711
 712         for (i = 0; i < iter->nr_segs; i++) {
 713                 void __user *base = iter->iov[i].iov_base + off;
 714                 size_t len = iter->iov[i].iov_len - off;
 715
 716                 ret = vhost_scsi_map_to_sgl(cmd, base, len, sg, write);
 717                 if (ret < 0) {
 718                         for (i = 0; i < sg_count; i++) {
 719                                 struct page *page = sg_page(&sg[i]);
 720                                 if (page)
 721                                         put_page(page);
 722                         }
 723                         return ret;
 724                 }
 725                 sg += ret;
 726                 off = 0;
 727         }
 728         return 0;
 729 }
 730
 731 static int
 732 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
 733                  size_t prot_bytes, struct iov_iter *prot_iter,
 734                  size_t data_bytes, struct iov_iter *data_iter)
 735 {
 736         int sgl_count, ret;
 737         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
 738
 739         if (prot_bytes) {
 740                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
 741                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
 742                 if (sgl_count < 0)
 743                         return sgl_count;
 744
 745                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
 746                 cmd->tvc_prot_sgl_count = sgl_count;
 747                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
 748                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
 749
 750                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
 751                                             cmd->tvc_prot_sgl,
 752                                             cmd->tvc_prot_sgl_count);
 753                 if (ret < 0) {
 754                         cmd->tvc_prot_sgl_count = 0;
 755                         return ret;
 756                 }
 757         }
 758         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
 759                                          VHOST_SCSI_PREALLOC_SGLS);
 760         if (sgl_count < 0)
 761                 return sgl_count;
 762
 763         sg_init_table(cmd->tvc_sgl, sgl_count);
 764         cmd->tvc_sgl_count = sgl_count;
 765         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
 766                   cmd->tvc_sgl, cmd->tvc_sgl_count);
 767
 768         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
 769                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
 770         if (ret < 0) {
 771                 cmd->tvc_sgl_count = 0;
 772                 return ret;
 773         }
 774         return 0;
 775 }
 776
 777 static int vhost_scsi_to_tcm_attr(int attr)
 778 {
 779         switch (attr) {
 780         case VIRTIO_SCSI_S_SIMPLE:
 781                 return TCM_SIMPLE_TAG;
 782         case VIRTIO_SCSI_S_ORDERED:
 783                 return TCM_ORDERED_TAG;
 784         case VIRTIO_SCSI_S_HEAD:
 785                 return TCM_HEAD_TAG;
 786         case VIRTIO_SCSI_S_ACA:
 787                 return TCM_ACA_TAG;
 788         default:
 789                 break;
 790         }
 791         return TCM_SIMPLE_TAG;
 792 }
 793
 794 static void vhost_scsi_submission_work(struct work_struct *work)
 795 {
 796         struct vhost_scsi_cmd *cmd =
 797                 container_of(work, struct vhost_scsi_cmd, work);
 798         struct vhost_scsi_nexus *tv_nexus;
 799         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 800         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
 801         int rc;
 802
 803         /* FIXME: BIDI operation */
 804         if (cmd->tvc_sgl_count) {
 805                 sg_ptr = cmd->tvc_sgl;
 806
 807                 if (cmd->tvc_prot_sgl_count)
 808                         sg_prot_ptr = cmd->tvc_prot_sgl;
 809                 else
 810                         se_cmd->prot_pto = true;
 811         } else {
 812                 sg_ptr = NULL;
 813         }
 814         tv_nexus = cmd->tvc_nexus;
 815
 816         se_cmd->tag = 0;
 817         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
 818                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
 819                         cmd->tvc_lun, cmd->tvc_exp_data_len,
 820                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
 821                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
 822                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
 823                         cmd->tvc_prot_sgl_count);
 824         if (rc < 0) {
 825                 transport_send_check_condition_and_sense(se_cmd,
 826                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
 827                 transport_generic_free_cmd(se_cmd, 0);
 828         }
 829 }
 830
 831 static void
 832 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
 833                            struct vhost_virtqueue *vq,
 834                            int head, unsigned out)
 835 {
 836         struct virtio_scsi_cmd_resp __user *resp;
 837         struct virtio_scsi_cmd_resp rsp;
 838         int ret;
 839
 840         memset(&rsp, 0, sizeof(rsp));
 841         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
 842         resp = vq->iov[out].iov_base;
 843         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
 844         if (!ret)
 845                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 846         else
 847                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
 848 }
 849
 850 static void
 851 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
 852 {
 853         struct vhost_scsi_tpg **vs_tpg, *tpg;
 854         struct virtio_scsi_cmd_req v_req;
 855         struct virtio_scsi_cmd_req_pi v_req_pi;
 856         struct vhost_scsi_cmd *cmd;
 857         struct iov_iter out_iter, in_iter, prot_iter, data_iter;
 858         u64 tag;
 859         u32 exp_data_len, data_direction;
 860         unsigned out, in;
 861         int head, ret, prot_bytes;
 862         size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
 863         size_t out_size, in_size;
 864         u16 lun;
 865         u8 *target, *lunp, task_attr;
 866         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
 867         void *req, *cdb;
 868
 869         mutex_lock(&vq->mutex);
 870         /*
 871          * We can handle the vq only after the endpoint is setup by calling the
 872          * VHOST_SCSI_SET_ENDPOINT ioctl.
 873          */
 874         vs_tpg = vq->private_data;
 875         if (!vs_tpg)
 876                 goto out;
 877
 878         vhost_disable_notify(&vs->dev, vq);
 879
 880         for (;;) {
 881                 head = vhost_get_vq_desc(vq, vq->iov,
 882                                          ARRAY_SIZE(vq->iov), &out, &in,
 883                                          NULL, NULL);
 884                 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
 885                          head, out, in);
 886                 /* On error, stop handling until the next kick. */
 887                 if (unlikely(head < 0))
 888                         break;
 889                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 890                 if (head == vq->num) {
 891                         if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
 892                                 vhost_disable_notify(&vs->dev, vq);
 893                                 continue;
 894                         }
 895                         break;
 896                 }
 897                 /*
 898                  * Check for a sane response buffer so we can report early
 899                  * errors back to the guest.
 900                  */
 901                 if (unlikely(vq->iov[out].iov_len < rsp_size)) {
 902                         vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
 903                                 " size, got %zu bytes\n", vq->iov[out].iov_len);
 904                         break;
 905                 }
 906                 /*
 907                  * Setup pointers and values based upon different virtio-scsi
 908                  * request header if T10_PI is enabled in KVM guest.
 909                  */
 910                 if (t10_pi) {
 911                         req = &v_req_pi;
 912                         req_size = sizeof(v_req_pi);
 913                         lunp = &v_req_pi.lun[0];
 914                         target = &v_req_pi.lun[1];
 915                 } else {
 916                         req = &v_req;
 917                         req_size = sizeof(v_req);
 918                         lunp = &v_req.lun[0];
 919                         target = &v_req.lun[1];
 920                 }
 921                 /*
 922                  * FIXME: Not correct for BIDI operation
 923                  */
 924                 out_size = iov_length(vq->iov, out);
 925                 in_size = iov_length(&vq->iov[out], in);
 926
 927                 /*
 928                  * Copy over the virtio-scsi request header, which for a
 929                  * ANY_LAYOUT enabled guest may span multiple iovecs, or a
 930                  * single iovec may contain both the header + outgoing
 931                  * WRITE payloads.
 932                  *
 933                  * copy_from_iter() will advance out_iter, so that it will
 934                  * point at the start of the outgoing WRITE payload, if
 935                  * DMA_TO_DEVICE is set.
 936                  */
 937                 iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
 938
 939                 ret = copy_from_iter(req, req_size, &out_iter);
 940                 if (unlikely(ret != req_size)) {
 941                         vq_err(vq, "Faulted on copy_from_iter\n");
 942                         vhost_scsi_send_bad_target(vs, vq, head, out);
 943                         continue;
 944                 }
 945                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
 946                 if (unlikely(*lunp != 1)) {
 947                         vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
 948                         vhost_scsi_send_bad_target(vs, vq, head, out);
 949                         continue;
 950                 }
 951
 952                 tpg = ACCESS_ONCE(vs_tpg[*target]);
 953                 if (unlikely(!tpg)) {
 954                         /* Target does not exist, fail the request */
 955                         vhost_scsi_send_bad_target(vs, vq, head, out);
 956                         continue;
 957                 }
 958                 /*
 959                  * Determine data_direction by calculating the total outgoing
 960                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
 961                  * response headers respectively.
 962                  *
 963                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
 964                  * to the right place.
 965                  *
 966                  * For DMA_FROM_DEVICE, the iovec will be just past the end
 967                  * of the virtio-scsi response header in either the same
 968                  * or immediately following iovec.
 969                  *
 970                  * Any associated T10_PI bytes for the outgoing / incoming
 971                  * payloads are included in calculation of exp_data_len here.
 972                  */
 973                 prot_bytes = 0;
 974
 975                 if (out_size > req_size) {
 976                         data_direction = DMA_TO_DEVICE;
 977                         exp_data_len = out_size - req_size;
 978                         data_iter = out_iter;
 979                 } else if (in_size > rsp_size) {
 980                         data_direction = DMA_FROM_DEVICE;
 981                         exp_data_len = in_size - rsp_size;
 982
 983                         iov_iter_init(&in_iter, READ, &vq->iov[out], in,
 984                                       rsp_size + exp_data_len);
 985                         iov_iter_advance(&in_iter, rsp_size);
 986                         data_iter = in_iter;
 987                 } else {
 988                         data_direction = DMA_NONE;
 989                         exp_data_len = 0;
 990                 }
 991                 /*
 992                  * If T10_PI header + payload is present, setup prot_iter values
 993                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
 994                  * host scatterlists via get_user_pages_fast().
 995                  */
 996                 if (t10_pi) {
 997                         if (v_req_pi.pi_bytesout) {
 998                                 if (data_direction != DMA_TO_DEVICE) {
 999                                         vq_err(vq, "Received non zero pi_bytesout,"
1000                                                 " but wrong data_direction\n");
1001                                         vhost_scsi_send_bad_target(vs, vq, head, out);
1002                                         continue;
1003                                 }
1004                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
1005                         } else if (v_req_pi.pi_bytesin) {
1006                                 if (data_direction != DMA_FROM_DEVICE) {
1007                                         vq_err(vq, "Received non zero pi_bytesin,"
1008                                                 " but wrong data_direction\n");
1009                                         vhost_scsi_send_bad_target(vs, vq, head, out);
1010                                         continue;
1011                                 }
1012                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1013                         }
1014                         /*
1015                          * Set prot_iter to data_iter, and advance past any
1016                          * preceeding prot_bytes that may be present.
1017                          *
1018                          * Also fix up the exp_data_len to reflect only the
1019                          * actual data payload length.
1020                          */
1021                         if (prot_bytes) {
1022                                 exp_data_len -= prot_bytes;
1023                                 prot_iter = data_iter;
1024                                 iov_iter_advance(&data_iter, prot_bytes);
1025                         }
1026                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
1027                         task_attr = v_req_pi.task_attr;
1028                         cdb = &v_req_pi.cdb[0];
1029                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1030                 } else {
1031                         tag = vhost64_to_cpu(vq, v_req.tag);
1032                         task_attr = v_req.task_attr;
1033                         cdb = &v_req.cdb[0];
1034                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1035                 }
1036                 /*
1037                  * Check that the received CDB size does not exceeded our
1038                  * hardcoded max for vhost-scsi, then get a pre-allocated
1039                  * cmd descriptor for the new virtio-scsi tag.
1040                  *
1041                  * TODO what if cdb was too small for varlen cdb header?
1042                  */
1043                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1044                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
1045                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1046                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1047                         vhost_scsi_send_bad_target(vs, vq, head, out);
1048                         continue;
1049                 }
1050                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1051                                          exp_data_len + prot_bytes,
1052                                          data_direction);
1053                 if (IS_ERR(cmd)) {
1054                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1055                                PTR_ERR(cmd));
1056                         vhost_scsi_send_bad_target(vs, vq, head, out);
1057                         continue;
1058                 }
1059                 cmd->tvc_vhost = vs;
1060                 cmd->tvc_vq = vq;
1061                 cmd->tvc_resp_iov = &vq->iov[out];
1062                 cmd->tvc_in_iovs = in;
1063
1064                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1065                          cmd->tvc_cdb[0], cmd->tvc_lun);
1066                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1067                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1068
1069                 if (data_direction != DMA_NONE) {
1070                         ret = vhost_scsi_mapal(cmd,
1071                                                prot_bytes, &prot_iter,
1072                                                exp_data_len, &data_iter);
1073                         if (unlikely(ret)) {
1074                                 vq_err(vq, "Failed to map iov to sgl\n");
1075                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1076                                 vhost_scsi_send_bad_target(vs, vq, head, out);
1077                                 continue;
1078                         }
1079                 }
1080                 /*
1081                  * Save the descriptor from vhost_get_vq_desc() to be used to
1082                  * complete the virtio-scsi request in TCM callback context via
1083                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1084                  */
1085                 cmd->tvc_vq_desc = head;
1086                 /*
1087                  * Dispatch cmd descriptor for cmwq execution in process
1088                  * context provided by vhost_scsi_workqueue.  This also ensures
1089                  * cmd is executed on the same kworker CPU as this vhost
1090                  * thread to gain positive L2 cache locality effects.
1091                  */
1092                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1093                 queue_work(vhost_scsi_workqueue, &cmd->work);
1094         }
1095 out:
1096         mutex_unlock(&vq->mutex);
1097 }
1098
1099 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1100 {
1101         pr_debug("%s: The handling func for control queue.\n", __func__);
1102 }
1103
1104 static void
1105 vhost_scsi_send_evt(struct vhost_scsi *vs,
1106                    struct vhost_scsi_tpg *tpg,
1107                    struct se_lun *lun,
1108                    u32 event,
1109                    u32 reason)
1110 {
1111         struct vhost_scsi_evt *evt;
1112
1113         evt = vhost_scsi_allocate_evt(vs, event, reason);
1114         if (!evt)
1115                 return;
1116
1117         if (tpg && lun) {
1118                 /* TODO: share lun setup code with virtio-scsi.ko */
1119                 /*
1120                  * Note: evt->event is zeroed when we allocate it and
1121                  * lun[4-7] need to be zero according to virtio-scsi spec.
1122                  */
1123                 evt->event.lun[0] = 0x01;
1124                 evt->event.lun[1] = tpg->tport_tpgt;
1125                 if (lun->unpacked_lun >= 256)
1126                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1127                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1128         }
1129
1130         llist_add(&evt->list, &vs->vs_event_list);
1131         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1132 }
1133
1134 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1135 {
1136         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1137                                                 poll.work);
1138         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1139
1140         mutex_lock(&vq->mutex);
1141         if (!vq->private_data)
1142                 goto out;
1143
1144         if (vs->vs_events_missed)
1145                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1146 out:
1147         mutex_unlock(&vq->mutex);
1148 }
1149
1150 static void vhost_scsi_handle_kick(struct vhost_work *work)
1151 {
1152         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1153                                                 poll.work);
1154         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1155
1156         vhost_scsi_handle_vq(vs, vq);
1157 }
1158
1159 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1160 {
1161         vhost_poll_flush(&vs->vqs[index].vq.poll);
1162 }
1163
1164 /* Callers must hold dev mutex */
1165 static void vhost_scsi_flush(struct vhost_scsi *vs)
1166 {
1167         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1168         int i;
1169
1170         /* Init new inflight and remember the old inflight */
1171         vhost_scsi_init_inflight(vs, old_inflight);
1172
1173         /*
1174          * The inflight->kref was initialized to 1. We decrement it here to
1175          * indicate the start of the flush operation so that it will reach 0
1176          * when all the reqs are finished.
1177          */
1178         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1179                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1180
1181         /* Flush both the vhost poll and vhost work */
1182         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1183                 vhost_scsi_flush_vq(vs, i);
1184         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1185         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1186
1187         /* Wait for all reqs issued before the flush to be finished */
1188         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1189                 wait_for_completion(&old_inflight[i]->comp);
1190 }
1191
1192 /*
1193  * Called from vhost_scsi_ioctl() context to walk the list of available
1194  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1195  *
1196  *  The lock nesting rule is:
1197  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1198  */
1199 static int
1200 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1201                         struct vhost_scsi_target *t)
1202 {
1203         struct se_portal_group *se_tpg;
1204         struct vhost_scsi_tport *tv_tport;
1205         struct vhost_scsi_tpg *tpg;
1206         struct vhost_scsi_tpg **vs_tpg;
1207         struct vhost_virtqueue *vq;
1208         int index, ret, i, len;
1209         bool match = false;
1210
1211         mutex_lock(&vhost_scsi_mutex);
1212         mutex_lock(&vs->dev.mutex);
1213
1214         /* Verify that ring has been setup correctly. */
1215         for (index = 0; index < vs->dev.nvqs; ++index) {
1216                 /* Verify that ring has been setup correctly. */
1217                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1218                         ret = -EFAULT;
1219                         goto out;
1220                 }
1221         }
1222
1223         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1224         vs_tpg = kzalloc(len, GFP_KERNEL);
1225         if (!vs_tpg) {
1226                 ret = -ENOMEM;
1227                 goto out;
1228         }
1229         if (vs->vs_tpg)
1230                 memcpy(vs_tpg, vs->vs_tpg, len);
1231
1232         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1233                 mutex_lock(&tpg->tv_tpg_mutex);
1234                 if (!tpg->tpg_nexus) {
1235                         mutex_unlock(&tpg->tv_tpg_mutex);
1236                         continue;
1237                 }
1238                 if (tpg->tv_tpg_vhost_count != 0) {
1239                         mutex_unlock(&tpg->tv_tpg_mutex);
1240                         continue;
1241                 }
1242                 tv_tport = tpg->tport;
1243
1244                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1245                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1246                                 kfree(vs_tpg);
1247                                 mutex_unlock(&tpg->tv_tpg_mutex);
1248                                 ret = -EEXIST;
1249                                 goto out;
1250                         }
1251                         /*
1252                          * In order to ensure individual vhost-scsi configfs
1253                          * groups cannot be removed while in use by vhost ioctl,
1254                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1255                          * dependency now.
1256                          */
1257                         se_tpg = &tpg->se_tpg;
1258                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1259                         if (ret) {
1260                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1261                                 kfree(vs_tpg);
1262                                 mutex_unlock(&tpg->tv_tpg_mutex);
1263                                 goto out;
1264                         }
1265                         tpg->tv_tpg_vhost_count++;
1266                         tpg->vhost_scsi = vs;
1267                         vs_tpg[tpg->tport_tpgt] = tpg;
1268                         smp_mb__after_atomic();
1269                         match = true;
1270                 }
1271                 mutex_unlock(&tpg->tv_tpg_mutex);
1272         }
1273
1274         if (match) {
1275                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1276                        sizeof(vs->vs_vhost_wwpn));
1277                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1278                         vq = &vs->vqs[i].vq;
1279                         mutex_lock(&vq->mutex);
1280                         vq->private_data = vs_tpg;
1281                         vhost_init_used(vq);
1282                         mutex_unlock(&vq->mutex);
1283                 }
1284                 ret = 0;
1285         } else {
1286                 ret = -EEXIST;
1287         }
1288
1289         /*
1290          * Act as synchronize_rcu to make sure access to
1291          * old vs->vs_tpg is finished.
1292          */
1293         vhost_scsi_flush(vs);
1294         kfree(vs->vs_tpg);
1295         vs->vs_tpg = vs_tpg;
1296
1297 out:
1298         mutex_unlock(&vs->dev.mutex);
1299         mutex_unlock(&vhost_scsi_mutex);
1300         return ret;
1301 }
1302
1303 static int
1304 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1305                           struct vhost_scsi_target *t)
1306 {
1307         struct se_portal_group *se_tpg;
1308         struct vhost_scsi_tport *tv_tport;
1309         struct vhost_scsi_tpg *tpg;
1310         struct vhost_virtqueue *vq;
1311         bool match = false;
1312         int index, ret, i;
1313         u8 target;
1314
1315         mutex_lock(&vhost_scsi_mutex);
1316         mutex_lock(&vs->dev.mutex);
1317         /* Verify that ring has been setup correctly. */
1318         for (index = 0; index < vs->dev.nvqs; ++index) {
1319                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1320                         ret = -EFAULT;
1321                         goto err_dev;
1322                 }
1323         }
1324
1325         if (!vs->vs_tpg) {
1326                 ret = 0;
1327                 goto err_dev;
1328         }
1329
1330         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1331                 target = i;
1332                 tpg = vs->vs_tpg[target];
1333                 if (!tpg)
1334                         continue;
1335
1336                 mutex_lock(&tpg->tv_tpg_mutex);
1337                 tv_tport = tpg->tport;
1338                 if (!tv_tport) {
1339                         ret = -ENODEV;
1340                         goto err_tpg;
1341                 }
1342
1343                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1344                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1345                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1346                                 tv_tport->tport_name, tpg->tport_tpgt,
1347                                 t->vhost_wwpn, t->vhost_tpgt);
1348                         ret = -EINVAL;
1349                         goto err_tpg;
1350                 }
1351                 tpg->tv_tpg_vhost_count--;
1352                 tpg->vhost_scsi = NULL;
1353                 vs->vs_tpg[target] = NULL;
1354                 match = true;
1355                 mutex_unlock(&tpg->tv_tpg_mutex);
1356                 /*
1357                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1358                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1359                  */
1360                 se_tpg = &tpg->se_tpg;
1361                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1362         }
1363         if (match) {
1364                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1365                         vq = &vs->vqs[i].vq;
1366                         mutex_lock(&vq->mutex);
1367                         vq->private_data = NULL;
1368                         mutex_unlock(&vq->mutex);
1369                 }
1370         }
1371         /*
1372          * Act as synchronize_rcu to make sure access to
1373          * old vs->vs_tpg is finished.
1374          */
1375         vhost_scsi_flush(vs);
1376         kfree(vs->vs_tpg);
1377         vs->vs_tpg = NULL;
1378         WARN_ON(vs->vs_events_nr);
1379         mutex_unlock(&vs->dev.mutex);
1380         mutex_unlock(&vhost_scsi_mutex);
1381         return 0;
1382
1383 err_tpg:
1384         mutex_unlock(&tpg->tv_tpg_mutex);
1385 err_dev:
1386         mutex_unlock(&vs->dev.mutex);
1387         mutex_unlock(&vhost_scsi_mutex);
1388         return ret;
1389 }
1390
1391 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1392 {
1393         struct vhost_virtqueue *vq;
1394         int i;
1395
1396         if (features & ~VHOST_SCSI_FEATURES)
1397                 return -EOPNOTSUPP;
1398
1399         mutex_lock(&vs->dev.mutex);
1400         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1401             !vhost_log_access_ok(&vs->dev)) {
1402                 mutex_unlock(&vs->dev.mutex);
1403                 return -EFAULT;
1404         }
1405
1406         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1407                 vq = &vs->vqs[i].vq;
1408                 mutex_lock(&vq->mutex);
1409                 vq->acked_features = features;
1410                 mutex_unlock(&vq->mutex);
1411         }
1412         mutex_unlock(&vs->dev.mutex);
1413         return 0;
1414 }
1415
1416 static int vhost_scsi_open(struct inode *inode, struct file *f)
1417 {
1418         struct vhost_scsi *vs;
1419         struct vhost_virtqueue **vqs;
1420         int r = -ENOMEM, i;
1421
1422         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
1423         if (!vs) {
1424                 vs = vzalloc(sizeof(*vs));
1425                 if (!vs)
1426                         goto err_vs;
1427         }
1428
1429         vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1430         if (!vqs)
1431                 goto err_vqs;
1432
1433         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1434         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1435
1436         vs->vs_events_nr = 0;
1437         vs->vs_events_missed = false;
1438
1439         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1440         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1441         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1442         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1443         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1444                 vqs[i] = &vs->vqs[i].vq;
1445                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1446         }
1447         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1448
1449         vhost_scsi_init_inflight(vs, NULL);
1450
1451         f->private_data = vs;
1452         return 0;
1453
1454 err_vqs:
1455         kvfree(vs);
1456 err_vs:
1457         return r;
1458 }
1459
1460 static int vhost_scsi_release(struct inode *inode, struct file *f)
1461 {
1462         struct vhost_scsi *vs = f->private_data;
1463         struct vhost_scsi_target t;
1464
1465         mutex_lock(&vs->dev.mutex);
1466         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1467         mutex_unlock(&vs->dev.mutex);
1468         vhost_scsi_clear_endpoint(vs, &t);
1469         vhost_dev_stop(&vs->dev);
1470         vhost_dev_cleanup(&vs->dev, false);
1471         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1472         vhost_scsi_flush(vs);
1473         kfree(vs->dev.vqs);
1474         kvfree(vs);
1475         return 0;
1476 }
1477
1478 static long
1479 vhost_scsi_ioctl(struct file *f,
1480                  unsigned int ioctl,
1481                  unsigned long arg)
1482 {
1483         struct vhost_scsi *vs = f->private_data;
1484         struct vhost_scsi_target backend;
1485         void __user *argp = (void __user *)arg;
1486         u64 __user *featurep = argp;
1487         u32 __user *eventsp = argp;
1488         u32 events_missed;
1489         u64 features;
1490         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1491         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1492
1493         switch (ioctl) {
1494         case VHOST_SCSI_SET_ENDPOINT:
1495                 if (copy_from_user(&backend, argp, sizeof backend))
1496                         return -EFAULT;
1497                 if (backend.reserved != 0)
1498                         return -EOPNOTSUPP;
1499
1500                 return vhost_scsi_set_endpoint(vs, &backend);
1501         case VHOST_SCSI_CLEAR_ENDPOINT:
1502                 if (copy_from_user(&backend, argp, sizeof backend))
1503                         return -EFAULT;
1504                 if (backend.reserved != 0)
1505                         return -EOPNOTSUPP;
1506
1507                 return vhost_scsi_clear_endpoint(vs, &backend);
1508         case VHOST_SCSI_GET_ABI_VERSION:
1509                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1510                         return -EFAULT;
1511                 return 0;
1512         case VHOST_SCSI_SET_EVENTS_MISSED:
1513                 if (get_user(events_missed, eventsp))
1514                         return -EFAULT;
1515                 mutex_lock(&vq->mutex);
1516                 vs->vs_events_missed = events_missed;
1517                 mutex_unlock(&vq->mutex);
1518                 return 0;
1519         case VHOST_SCSI_GET_EVENTS_MISSED:
1520                 mutex_lock(&vq->mutex);
1521                 events_missed = vs->vs_events_missed;
1522                 mutex_unlock(&vq->mutex);
1523                 if (put_user(events_missed, eventsp))
1524                         return -EFAULT;
1525                 return 0;
1526         case VHOST_GET_FEATURES:
1527                 features = VHOST_SCSI_FEATURES;
1528                 if (copy_to_user(featurep, &features, sizeof features))
1529                         return -EFAULT;
1530                 return 0;
1531         case VHOST_SET_FEATURES:
1532                 if (copy_from_user(&features, featurep, sizeof features))
1533                         return -EFAULT;
1534                 return vhost_scsi_set_features(vs, features);
1535         default:
1536                 mutex_lock(&vs->dev.mutex);
1537                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1538                 /* TODO: flush backend after dev ioctl. */
1539                 if (r == -ENOIOCTLCMD)
1540                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1541                 mutex_unlock(&vs->dev.mutex);
1542                 return r;
1543         }
1544 }
1545
1546 #ifdef CONFIG_COMPAT
1547 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1548                                 unsigned long arg)
1549 {
1550         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1551 }
1552 #endif
1553
1554 static const struct file_operations vhost_scsi_fops = {
1555         .owner          = THIS_MODULE,
1556         .release        = vhost_scsi_release,
1557         .unlocked_ioctl = vhost_scsi_ioctl,
1558 #ifdef CONFIG_COMPAT
1559         .compat_ioctl   = vhost_scsi_compat_ioctl,
1560 #endif
1561         .open           = vhost_scsi_open,
1562         .llseek         = noop_llseek,
1563 };
1564
1565 static struct miscdevice vhost_scsi_misc = {
1566         MISC_DYNAMIC_MINOR,
1567         "vhost-scsi",
1568         &vhost_scsi_fops,
1569 };
1570
1571 static int __init vhost_scsi_register(void)
1572 {
1573         return misc_register(&vhost_scsi_misc);
1574 }
1575
1576 static int vhost_scsi_deregister(void)
1577 {
1578         return misc_deregister(&vhost_scsi_misc);
1579 }
1580
1581 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1582 {
1583         switch (tport->tport_proto_id) {
1584         case SCSI_PROTOCOL_SAS:
1585                 return "SAS";
1586         case SCSI_PROTOCOL_FCP:
1587                 return "FCP";
1588         case SCSI_PROTOCOL_ISCSI:
1589                 return "iSCSI";
1590         default:
1591                 break;
1592         }
1593
1594         return "Unknown";
1595 }
1596
1597 static void
1598 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1599                   struct se_lun *lun, bool plug)
1600 {
1601
1602         struct vhost_scsi *vs = tpg->vhost_scsi;
1603         struct vhost_virtqueue *vq;
1604         u32 reason;
1605
1606         if (!vs)
1607                 return;
1608
1609         mutex_lock(&vs->dev.mutex);
1610
1611         if (plug)
1612                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1613         else
1614                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1615
1616         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1617         mutex_lock(&vq->mutex);
1618         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1619                 vhost_scsi_send_evt(vs, tpg, lun,
1620                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1621         mutex_unlock(&vq->mutex);
1622         mutex_unlock(&vs->dev.mutex);
1623 }
1624
1625 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1626 {
1627         vhost_scsi_do_plug(tpg, lun, true);
1628 }
1629
1630 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1631 {
1632         vhost_scsi_do_plug(tpg, lun, false);
1633 }
1634
1635 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1636                                struct se_lun *lun)
1637 {
1638         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1639                                 struct vhost_scsi_tpg, se_tpg);
1640
1641         mutex_lock(&vhost_scsi_mutex);
1642
1643         mutex_lock(&tpg->tv_tpg_mutex);
1644         tpg->tv_tpg_port_count++;
1645         mutex_unlock(&tpg->tv_tpg_mutex);
1646
1647         vhost_scsi_hotplug(tpg, lun);
1648
1649         mutex_unlock(&vhost_scsi_mutex);
1650
1651         return 0;
1652 }
1653
1654 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1655                                   struct se_lun *lun)
1656 {
1657         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1658                                 struct vhost_scsi_tpg, se_tpg);
1659
1660         mutex_lock(&vhost_scsi_mutex);
1661
1662         mutex_lock(&tpg->tv_tpg_mutex);
1663         tpg->tv_tpg_port_count--;
1664         mutex_unlock(&tpg->tv_tpg_mutex);
1665
1666         vhost_scsi_hotunplug(tpg, lun);
1667
1668         mutex_unlock(&vhost_scsi_mutex);
1669 }
1670
1671 static void vhost_scsi_free_cmd_map_res(struct vhost_scsi_nexus *nexus,
1672                                        struct se_session *se_sess)
1673 {
1674         struct vhost_scsi_cmd *tv_cmd;
1675         unsigned int i;
1676
1677         if (!se_sess->sess_cmd_map)
1678                 return;
1679
1680         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1681                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1682
1683                 kfree(tv_cmd->tvc_sgl);
1684                 kfree(tv_cmd->tvc_prot_sgl);
1685                 kfree(tv_cmd->tvc_upages);
1686         }
1687 }
1688
1689 static ssize_t vhost_scsi_tpg_attrib_store_fabric_prot_type(
1690         struct se_portal_group *se_tpg,
1691         const char *page,
1692         size_t count)
1693 {
1694         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1695                                 struct vhost_scsi_tpg, se_tpg);
1696         unsigned long val;
1697         int ret = kstrtoul(page, 0, &val);
1698
1699         if (ret) {
1700                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1701                 return ret;
1702         }
1703         if (val != 0 && val != 1 && val != 3) {
1704                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1705                 return -EINVAL;
1706         }
1707         tpg->tv_fabric_prot_type = val;
1708
1709         return count;
1710 }
1711
1712 static ssize_t vhost_scsi_tpg_attrib_show_fabric_prot_type(
1713         struct se_portal_group *se_tpg,
1714         char *page)
1715 {
1716         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1717                                 struct vhost_scsi_tpg, se_tpg);
1718
1719         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1720 }
1721 TF_TPG_ATTRIB_ATTR(vhost_scsi, fabric_prot_type, S_IRUGO | S_IWUSR);
1722
1723 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1724         &vhost_scsi_tpg_attrib_fabric_prot_type.attr,
1725         NULL,
1726 };
1727
1728 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1729                                 const char *name)
1730 {
1731         struct se_portal_group *se_tpg;
1732         struct se_session *se_sess;
1733         struct vhost_scsi_nexus *tv_nexus;
1734         struct vhost_scsi_cmd *tv_cmd;
1735         unsigned int i;
1736
1737         mutex_lock(&tpg->tv_tpg_mutex);
1738         if (tpg->tpg_nexus) {
1739                 mutex_unlock(&tpg->tv_tpg_mutex);
1740                 pr_debug("tpg->tpg_nexus already exists\n");
1741                 return -EEXIST;
1742         }
1743         se_tpg = &tpg->se_tpg;
1744
1745         tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1746         if (!tv_nexus) {
1747                 mutex_unlock(&tpg->tv_tpg_mutex);
1748                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1749                 return -ENOMEM;
1750         }
1751         /*
1752          *  Initialize the struct se_session pointer and setup tagpool
1753          *  for struct vhost_scsi_cmd descriptors
1754          */
1755         tv_nexus->tvn_se_sess = transport_init_session_tags(
1756                                         VHOST_SCSI_DEFAULT_TAGS,
1757                                         sizeof(struct vhost_scsi_cmd),
1758                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS);
1759         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1760                 mutex_unlock(&tpg->tv_tpg_mutex);
1761                 kfree(tv_nexus);
1762                 return -ENOMEM;
1763         }
1764         se_sess = tv_nexus->tvn_se_sess;
1765         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1766                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1767
1768                 tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1769                                         VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1770                 if (!tv_cmd->tvc_sgl) {
1771                         mutex_unlock(&tpg->tv_tpg_mutex);
1772                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1773                         goto out;
1774                 }
1775
1776                 tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1777                                         VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1778                 if (!tv_cmd->tvc_upages) {
1779                         mutex_unlock(&tpg->tv_tpg_mutex);
1780                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1781                         goto out;
1782                 }
1783
1784                 tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1785                                         VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1786                 if (!tv_cmd->tvc_prot_sgl) {
1787                         mutex_unlock(&tpg->tv_tpg_mutex);
1788                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1789                         goto out;
1790                 }
1791         }
1792         /*
1793          * Since we are running in 'demo mode' this call with generate a
1794          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1795          * the SCSI Initiator port name of the passed configfs group 'name'.
1796          */
1797         tv_nexus->tvn_se_sess->se_node_acl = core_tpg_check_initiator_node_acl(
1798                                 se_tpg, (unsigned char *)name);
1799         if (!tv_nexus->tvn_se_sess->se_node_acl) {
1800                 mutex_unlock(&tpg->tv_tpg_mutex);
1801                 pr_debug("core_tpg_check_initiator_node_acl() failed"
1802                                 " for %s\n", name);
1803                 goto out;
1804         }
1805         /*
1806          * Now register the TCM vhost virtual I_T Nexus as active.
1807          */
1808         transport_register_session(se_tpg, tv_nexus->tvn_se_sess->se_node_acl,
1809                         tv_nexus->tvn_se_sess, tv_nexus);
1810         tpg->tpg_nexus = tv_nexus;
1811
1812         mutex_unlock(&tpg->tv_tpg_mutex);
1813         return 0;
1814
1815 out:
1816         vhost_scsi_free_cmd_map_res(tv_nexus, se_sess);
1817         transport_free_session(se_sess);
1818         kfree(tv_nexus);
1819         return -ENOMEM;
1820 }
1821
1822 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1823 {
1824         struct se_session *se_sess;
1825         struct vhost_scsi_nexus *tv_nexus;
1826
1827         mutex_lock(&tpg->tv_tpg_mutex);
1828         tv_nexus = tpg->tpg_nexus;
1829         if (!tv_nexus) {
1830                 mutex_unlock(&tpg->tv_tpg_mutex);
1831                 return -ENODEV;
1832         }
1833
1834         se_sess = tv_nexus->tvn_se_sess;
1835         if (!se_sess) {
1836                 mutex_unlock(&tpg->tv_tpg_mutex);
1837                 return -ENODEV;
1838         }
1839
1840         if (tpg->tv_tpg_port_count != 0) {
1841                 mutex_unlock(&tpg->tv_tpg_mutex);
1842                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1843                         " active TPG port count: %d\n",
1844                         tpg->tv_tpg_port_count);
1845                 return -EBUSY;
1846         }
1847
1848         if (tpg->tv_tpg_vhost_count != 0) {
1849                 mutex_unlock(&tpg->tv_tpg_mutex);
1850                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1851                         " active TPG vhost count: %d\n",
1852                         tpg->tv_tpg_vhost_count);
1853                 return -EBUSY;
1854         }
1855
1856         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1857                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1858                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1859
1860         vhost_scsi_free_cmd_map_res(tv_nexus, se_sess);
1861         /*
1862          * Release the SCSI I_T Nexus to the emulated vhost Target Port
1863          */
1864         transport_deregister_session(tv_nexus->tvn_se_sess);
1865         tpg->tpg_nexus = NULL;
1866         mutex_unlock(&tpg->tv_tpg_mutex);
1867
1868         kfree(tv_nexus);
1869         return 0;
1870 }
1871
1872 static ssize_t vhost_scsi_tpg_show_nexus(struct se_portal_group *se_tpg,
1873                                         char *page)
1874 {
1875         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1876                                 struct vhost_scsi_tpg, se_tpg);
1877         struct vhost_scsi_nexus *tv_nexus;
1878         ssize_t ret;
1879
1880         mutex_lock(&tpg->tv_tpg_mutex);
1881         tv_nexus = tpg->tpg_nexus;
1882         if (!tv_nexus) {
1883                 mutex_unlock(&tpg->tv_tpg_mutex);
1884                 return -ENODEV;
1885         }
1886         ret = snprintf(page, PAGE_SIZE, "%s\n",
1887                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1888         mutex_unlock(&tpg->tv_tpg_mutex);
1889
1890         return ret;
1891 }
1892
1893 static ssize_t vhost_scsi_tpg_store_nexus(struct se_portal_group *se_tpg,
1894                                          const char *page,
1895                                          size_t count)
1896 {
1897         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1898                                 struct vhost_scsi_tpg, se_tpg);
1899         struct vhost_scsi_tport *tport_wwn = tpg->tport;
1900         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1901         int ret;
1902         /*
1903          * Shutdown the active I_T nexus if 'NULL' is passed..
1904          */
1905         if (!strncmp(page, "NULL", 4)) {
1906                 ret = vhost_scsi_drop_nexus(tpg);
1907                 return (!ret) ? count : ret;
1908         }
1909         /*
1910          * Otherwise make sure the passed virtual Initiator port WWN matches
1911          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1912          * vhost_scsi_make_nexus().
1913          */
1914         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1915                 pr_err("Emulated NAA Sas Address: %s, exceeds"
1916                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
1917                 return -EINVAL;
1918         }
1919         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1920
1921         ptr = strstr(i_port, "naa.");
1922         if (ptr) {
1923                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1924                         pr_err("Passed SAS Initiator Port %s does not"
1925                                 " match target port protoid: %s\n", i_port,
1926                                 vhost_scsi_dump_proto_id(tport_wwn));
1927                         return -EINVAL;
1928                 }
1929                 port_ptr = &i_port[0];
1930                 goto check_newline;
1931         }
1932         ptr = strstr(i_port, "fc.");
1933         if (ptr) {
1934                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1935                         pr_err("Passed FCP Initiator Port %s does not"
1936                                 " match target port protoid: %s\n", i_port,
1937                                 vhost_scsi_dump_proto_id(tport_wwn));
1938                         return -EINVAL;
1939                 }
1940                 port_ptr = &i_port[3]; /* Skip over "fc." */
1941                 goto check_newline;
1942         }
1943         ptr = strstr(i_port, "iqn.");
1944         if (ptr) {
1945                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1946                         pr_err("Passed iSCSI Initiator Port %s does not"
1947                                 " match target port protoid: %s\n", i_port,
1948                                 vhost_scsi_dump_proto_id(tport_wwn));
1949                         return -EINVAL;
1950                 }
1951                 port_ptr = &i_port[0];
1952                 goto check_newline;
1953         }
1954         pr_err("Unable to locate prefix for emulated Initiator Port:"
1955                         " %s\n", i_port);
1956         return -EINVAL;
1957         /*
1958          * Clear any trailing newline for the NAA WWN
1959          */
1960 check_newline:
1961         if (i_port[strlen(i_port)-1] == '\n')
1962                 i_port[strlen(i_port)-1] = '\0';
1963
1964         ret = vhost_scsi_make_nexus(tpg, port_ptr);
1965         if (ret < 0)
1966                 return ret;
1967
1968         return count;
1969 }
1970
1971 TF_TPG_BASE_ATTR(vhost_scsi, nexus, S_IRUGO | S_IWUSR);
1972
1973 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1974         &vhost_scsi_tpg_nexus.attr,
1975         NULL,
1976 };
1977
1978 static struct se_portal_group *
1979 vhost_scsi_make_tpg(struct se_wwn *wwn,
1980                    struct config_group *group,
1981                    const char *name)
1982 {
1983         struct vhost_scsi_tport *tport = container_of(wwn,
1984                         struct vhost_scsi_tport, tport_wwn);
1985
1986         struct vhost_scsi_tpg *tpg;
1987         u16 tpgt;
1988         int ret;
1989
1990         if (strstr(name, "tpgt_") != name)
1991                 return ERR_PTR(-EINVAL);
1992         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1993                 return ERR_PTR(-EINVAL);
1994
1995         tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1996         if (!tpg) {
1997                 pr_err("Unable to allocate struct vhost_scsi_tpg");
1998                 return ERR_PTR(-ENOMEM);
1999         }
2000         mutex_init(&tpg->tv_tpg_mutex);
2001         INIT_LIST_HEAD(&tpg->tv_tpg_list);
2002         tpg->tport = tport;
2003         tpg->tport_tpgt = tpgt;
2004
2005         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
2006         if (ret < 0) {
2007                 kfree(tpg);
2008                 return NULL;
2009         }
2010         mutex_lock(&vhost_scsi_mutex);
2011         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
2012         mutex_unlock(&vhost_scsi_mutex);
2013
2014         return &tpg->se_tpg;
2015 }
2016
2017 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
2018 {
2019         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2020                                 struct vhost_scsi_tpg, se_tpg);
2021
2022         mutex_lock(&vhost_scsi_mutex);
2023         list_del(&tpg->tv_tpg_list);
2024         mutex_unlock(&vhost_scsi_mutex);
2025         /*
2026          * Release the virtual I_T Nexus for this vhost TPG
2027          */
2028         vhost_scsi_drop_nexus(tpg);
2029         /*
2030          * Deregister the se_tpg from TCM..
2031          */
2032         core_tpg_deregister(se_tpg);
2033         kfree(tpg);
2034 }
2035
2036 static struct se_wwn *
2037 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2038                      struct config_group *group,
2039                      const char *name)
2040 {
2041         struct vhost_scsi_tport *tport;
2042         char *ptr;
2043         u64 wwpn = 0;
2044         int off = 0;
2045
2046         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2047                 return ERR_PTR(-EINVAL); */
2048
2049         tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2050         if (!tport) {
2051                 pr_err("Unable to allocate struct vhost_scsi_tport");
2052                 return ERR_PTR(-ENOMEM);
2053         }
2054         tport->tport_wwpn = wwpn;
2055         /*
2056          * Determine the emulated Protocol Identifier and Target Port Name
2057          * based on the incoming configfs directory name.
2058          */
2059         ptr = strstr(name, "naa.");
2060         if (ptr) {
2061                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2062                 goto check_len;
2063         }
2064         ptr = strstr(name, "fc.");
2065         if (ptr) {
2066                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2067                 off = 3; /* Skip over "fc." */
2068                 goto check_len;
2069         }
2070         ptr = strstr(name, "iqn.");
2071         if (ptr) {
2072                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2073                 goto check_len;
2074         }
2075
2076         pr_err("Unable to locate prefix for emulated Target Port:"
2077                         " %s\n", name);
2078         kfree(tport);
2079         return ERR_PTR(-EINVAL);
2080
2081 check_len:
2082         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2083                 pr_err("Emulated %s Address: %s, exceeds"
2084                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2085                         VHOST_SCSI_NAMELEN);
2086                 kfree(tport);
2087                 return ERR_PTR(-EINVAL);
2088         }
2089         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2090
2091         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2092                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2093
2094         return &tport->tport_wwn;
2095 }
2096
2097 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2098 {
2099         struct vhost_scsi_tport *tport = container_of(wwn,
2100                                 struct vhost_scsi_tport, tport_wwn);
2101
2102         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2103                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2104                 tport->tport_name);
2105
2106         kfree(tport);
2107 }
2108
2109 static ssize_t
2110 vhost_scsi_wwn_show_attr_version(struct target_fabric_configfs *tf,
2111                                 char *page)
2112 {
2113         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2114                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2115                 utsname()->machine);
2116 }
2117
2118 TF_WWN_ATTR_RO(vhost_scsi, version);
2119
2120 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2121         &vhost_scsi_wwn_version.attr,
2122         NULL,
2123 };
2124
2125 static struct target_core_fabric_ops vhost_scsi_ops = {
2126         .module                         = THIS_MODULE,
2127         .name                           = "vhost",
2128         .get_fabric_name                = vhost_scsi_get_fabric_name,
2129         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2130         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2131         .tpg_check_demo_mode            = vhost_scsi_check_true,
2132         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2133         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2134         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2135         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2136         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2137         .release_cmd                    = vhost_scsi_release_cmd,
2138         .check_stop_free                = vhost_scsi_check_stop_free,
2139         .shutdown_session               = vhost_scsi_shutdown_session,
2140         .close_session                  = vhost_scsi_close_session,
2141         .sess_get_index                 = vhost_scsi_sess_get_index,
2142         .sess_get_initiator_sid         = NULL,
2143         .write_pending                  = vhost_scsi_write_pending,
2144         .write_pending_status           = vhost_scsi_write_pending_status,
2145         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2146         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2147         .queue_data_in                  = vhost_scsi_queue_data_in,
2148         .queue_status                   = vhost_scsi_queue_status,
2149         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2150         .aborted_task                   = vhost_scsi_aborted_task,
2151         /*
2152          * Setup callers for generic logic in target_core_fabric_configfs.c
2153          */
2154         .fabric_make_wwn                = vhost_scsi_make_tport,
2155         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2156         .fabric_make_tpg                = vhost_scsi_make_tpg,
2157         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2158         .fabric_post_link               = vhost_scsi_port_link,
2159         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2160
2161         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2162         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2163         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2164 };
2165
2166 static int __init vhost_scsi_init(void)
2167 {
2168         int ret = -ENOMEM;
2169
2170         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2171                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2172                 utsname()->machine);
2173
2174         /*
2175          * Use our own dedicated workqueue for submitting I/O into
2176          * target core to avoid contention within system_wq.
2177          */
2178         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2179         if (!vhost_scsi_workqueue)
2180                 goto out;
2181
2182         ret = vhost_scsi_register();
2183         if (ret < 0)
2184                 goto out_destroy_workqueue;
2185
2186         ret = target_register_template(&vhost_scsi_ops);
2187         if (ret < 0)
2188                 goto out_vhost_scsi_deregister;
2189
2190         return 0;
2191
2192 out_vhost_scsi_deregister:
2193         vhost_scsi_deregister();
2194 out_destroy_workqueue:
2195         destroy_workqueue(vhost_scsi_workqueue);
2196 out:
2197         return ret;
2198 };
2199
2200 static void vhost_scsi_exit(void)
2201 {
2202         target_unregister_template(&vhost_scsi_ops);
2203         vhost_scsi_deregister();
2204         destroy_workqueue(vhost_scsi_workqueue);
2205 };
2206
2207 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2208 MODULE_ALIAS("tcm_vhost");
2209 MODULE_LICENSE("GPL");
2210 module_init(vhost_scsi_init);
2211 module_exit(vhost_scsi_exit);