xref: /openbmc/linux/drivers/vhost/scsi.c (revision 1c2dd16a)
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 struct vhost_scsi_inflight {
62 	/* Wait for the flush operation to finish */
63 	struct completion comp;
64 	/* Refcount for the inflight reqs */
65 	struct kref kref;
66 };
67 
68 struct vhost_scsi_cmd {
69 	/* Descriptor from vhost_get_vq_desc() for virt_queue segment */
70 	int tvc_vq_desc;
71 	/* virtio-scsi initiator task attribute */
72 	int tvc_task_attr;
73 	/* virtio-scsi response incoming iovecs */
74 	int tvc_in_iovs;
75 	/* virtio-scsi initiator data direction */
76 	enum dma_data_direction tvc_data_direction;
77 	/* Expected data transfer length from virtio-scsi header */
78 	u32 tvc_exp_data_len;
79 	/* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
80 	u64 tvc_tag;
81 	/* The number of scatterlists associated with this cmd */
82 	u32 tvc_sgl_count;
83 	u32 tvc_prot_sgl_count;
84 	/* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
85 	u32 tvc_lun;
86 	/* Pointer to the SGL formatted memory from virtio-scsi */
87 	struct scatterlist *tvc_sgl;
88 	struct scatterlist *tvc_prot_sgl;
89 	struct page **tvc_upages;
90 	/* Pointer to response header iovec */
91 	struct iovec tvc_resp_iov;
92 	/* Pointer to vhost_scsi for our device */
93 	struct vhost_scsi *tvc_vhost;
94 	/* Pointer to vhost_virtqueue for the cmd */
95 	struct vhost_virtqueue *tvc_vq;
96 	/* Pointer to vhost nexus memory */
97 	struct vhost_scsi_nexus *tvc_nexus;
98 	/* The TCM I/O descriptor that is accessed via container_of() */
99 	struct se_cmd tvc_se_cmd;
100 	/* work item used for cmwq dispatch to vhost_scsi_submission_work() */
101 	struct work_struct work;
102 	/* Copy of the incoming SCSI command descriptor block (CDB) */
103 	unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
104 	/* Sense buffer that will be mapped into outgoing status */
105 	unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
106 	/* Completed commands list, serviced from vhost worker thread */
107 	struct llist_node tvc_completion_list;
108 	/* Used to track inflight cmd */
109 	struct vhost_scsi_inflight *inflight;
110 };
111 
112 struct vhost_scsi_nexus {
113 	/* Pointer to TCM session for I_T Nexus */
114 	struct se_session *tvn_se_sess;
115 };
116 
117 struct vhost_scsi_tpg {
118 	/* Vhost port target portal group tag for TCM */
119 	u16 tport_tpgt;
120 	/* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
121 	int tv_tpg_port_count;
122 	/* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
123 	int tv_tpg_vhost_count;
124 	/* Used for enabling T10-PI with legacy devices */
125 	int tv_fabric_prot_type;
126 	/* list for vhost_scsi_list */
127 	struct list_head tv_tpg_list;
128 	/* Used to protect access for tpg_nexus */
129 	struct mutex tv_tpg_mutex;
130 	/* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
131 	struct vhost_scsi_nexus *tpg_nexus;
132 	/* Pointer back to vhost_scsi_tport */
133 	struct vhost_scsi_tport *tport;
134 	/* Returned by vhost_scsi_make_tpg() */
135 	struct se_portal_group se_tpg;
136 	/* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
137 	struct vhost_scsi *vhost_scsi;
138 };
139 
140 struct vhost_scsi_tport {
141 	/* SCSI protocol the tport is providing */
142 	u8 tport_proto_id;
143 	/* Binary World Wide unique Port Name for Vhost Target port */
144 	u64 tport_wwpn;
145 	/* ASCII formatted WWPN for Vhost Target port */
146 	char tport_name[VHOST_SCSI_NAMELEN];
147 	/* Returned by vhost_scsi_make_tport() */
148 	struct se_wwn tport_wwn;
149 };
150 
151 struct vhost_scsi_evt {
152 	/* event to be sent to guest */
153 	struct virtio_scsi_event event;
154 	/* event list, serviced from vhost worker thread */
155 	struct llist_node list;
156 };
157 
158 enum {
159 	VHOST_SCSI_VQ_CTL = 0,
160 	VHOST_SCSI_VQ_EVT = 1,
161 	VHOST_SCSI_VQ_IO = 2,
162 };
163 
164 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
165 enum {
166 	VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
167 					       (1ULL << VIRTIO_SCSI_F_T10_PI)
168 };
169 
170 #define VHOST_SCSI_MAX_TARGET	256
171 #define VHOST_SCSI_MAX_VQ	128
172 #define VHOST_SCSI_MAX_EVENT	128
173 
174 struct vhost_scsi_virtqueue {
175 	struct vhost_virtqueue vq;
176 	/*
177 	 * Reference counting for inflight reqs, used for flush operation. At
178 	 * each time, one reference tracks new commands submitted, while we
179 	 * wait for another one to reach 0.
180 	 */
181 	struct vhost_scsi_inflight inflights[2];
182 	/*
183 	 * Indicate current inflight in use, protected by vq->mutex.
184 	 * Writers must also take dev mutex and flush under it.
185 	 */
186 	int inflight_idx;
187 };
188 
189 struct vhost_scsi {
190 	/* Protected by vhost_scsi->dev.mutex */
191 	struct vhost_scsi_tpg **vs_tpg;
192 	char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
193 
194 	struct vhost_dev dev;
195 	struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
196 
197 	struct vhost_work vs_completion_work; /* cmd completion work item */
198 	struct llist_head vs_completion_list; /* cmd completion queue */
199 
200 	struct vhost_work vs_event_work; /* evt injection work item */
201 	struct llist_head vs_event_list; /* evt injection queue */
202 
203 	bool vs_events_missed; /* any missed events, protected by vq->mutex */
204 	int vs_events_nr; /* num of pending events, protected by vq->mutex */
205 };
206 
207 static struct workqueue_struct *vhost_scsi_workqueue;
208 
209 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
210 static DEFINE_MUTEX(vhost_scsi_mutex);
211 static LIST_HEAD(vhost_scsi_list);
212 
213 static int iov_num_pages(void __user *iov_base, size_t iov_len)
214 {
215 	return (PAGE_ALIGN((unsigned long)iov_base + iov_len) -
216 	       ((unsigned long)iov_base & PAGE_MASK)) >> PAGE_SHIFT;
217 }
218 
219 static void vhost_scsi_done_inflight(struct kref *kref)
220 {
221 	struct vhost_scsi_inflight *inflight;
222 
223 	inflight = container_of(kref, struct vhost_scsi_inflight, kref);
224 	complete(&inflight->comp);
225 }
226 
227 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
228 				    struct vhost_scsi_inflight *old_inflight[])
229 {
230 	struct vhost_scsi_inflight *new_inflight;
231 	struct vhost_virtqueue *vq;
232 	int idx, i;
233 
234 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
235 		vq = &vs->vqs[i].vq;
236 
237 		mutex_lock(&vq->mutex);
238 
239 		/* store old infight */
240 		idx = vs->vqs[i].inflight_idx;
241 		if (old_inflight)
242 			old_inflight[i] = &vs->vqs[i].inflights[idx];
243 
244 		/* setup new infight */
245 		vs->vqs[i].inflight_idx = idx ^ 1;
246 		new_inflight = &vs->vqs[i].inflights[idx ^ 1];
247 		kref_init(&new_inflight->kref);
248 		init_completion(&new_inflight->comp);
249 
250 		mutex_unlock(&vq->mutex);
251 	}
252 }
253 
254 static struct vhost_scsi_inflight *
255 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
256 {
257 	struct vhost_scsi_inflight *inflight;
258 	struct vhost_scsi_virtqueue *svq;
259 
260 	svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
261 	inflight = &svq->inflights[svq->inflight_idx];
262 	kref_get(&inflight->kref);
263 
264 	return inflight;
265 }
266 
267 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
268 {
269 	kref_put(&inflight->kref, vhost_scsi_done_inflight);
270 }
271 
272 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
273 {
274 	return 1;
275 }
276 
277 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
278 {
279 	return 0;
280 }
281 
282 static char *vhost_scsi_get_fabric_name(void)
283 {
284 	return "vhost";
285 }
286 
287 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
288 {
289 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
290 				struct vhost_scsi_tpg, se_tpg);
291 	struct vhost_scsi_tport *tport = tpg->tport;
292 
293 	return &tport->tport_name[0];
294 }
295 
296 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
297 {
298 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
299 				struct vhost_scsi_tpg, se_tpg);
300 	return tpg->tport_tpgt;
301 }
302 
303 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
304 {
305 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
306 				struct vhost_scsi_tpg, se_tpg);
307 
308 	return tpg->tv_fabric_prot_type;
309 }
310 
311 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
312 {
313 	return 1;
314 }
315 
316 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
317 {
318 	struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
319 				struct vhost_scsi_cmd, tvc_se_cmd);
320 	struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
321 	int i;
322 
323 	if (tv_cmd->tvc_sgl_count) {
324 		for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
325 			put_page(sg_page(&tv_cmd->tvc_sgl[i]));
326 	}
327 	if (tv_cmd->tvc_prot_sgl_count) {
328 		for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
329 			put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
330 	}
331 
332 	vhost_scsi_put_inflight(tv_cmd->inflight);
333 	percpu_ida_free(&se_sess->sess_tag_pool, se_cmd->map_tag);
334 }
335 
336 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
337 {
338 	return 0;
339 }
340 
341 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
342 {
343 	/* Go ahead and process the write immediately */
344 	target_execute_cmd(se_cmd);
345 	return 0;
346 }
347 
348 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
349 {
350 	return 0;
351 }
352 
353 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
354 {
355 	return;
356 }
357 
358 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
359 {
360 	return 0;
361 }
362 
363 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
364 {
365 	struct vhost_scsi *vs = cmd->tvc_vhost;
366 
367 	llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
368 
369 	vhost_work_queue(&vs->dev, &vs->vs_completion_work);
370 }
371 
372 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
373 {
374 	struct vhost_scsi_cmd *cmd = container_of(se_cmd,
375 				struct vhost_scsi_cmd, tvc_se_cmd);
376 	vhost_scsi_complete_cmd(cmd);
377 	return 0;
378 }
379 
380 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
381 {
382 	struct vhost_scsi_cmd *cmd = container_of(se_cmd,
383 				struct vhost_scsi_cmd, tvc_se_cmd);
384 	vhost_scsi_complete_cmd(cmd);
385 	return 0;
386 }
387 
388 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
389 {
390 	return;
391 }
392 
393 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
394 {
395 	return;
396 }
397 
398 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
399 {
400 	vs->vs_events_nr--;
401 	kfree(evt);
402 }
403 
404 static struct vhost_scsi_evt *
405 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
406 		       u32 event, u32 reason)
407 {
408 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
409 	struct vhost_scsi_evt *evt;
410 
411 	if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
412 		vs->vs_events_missed = true;
413 		return NULL;
414 	}
415 
416 	evt = kzalloc(sizeof(*evt), GFP_KERNEL);
417 	if (!evt) {
418 		vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
419 		vs->vs_events_missed = true;
420 		return NULL;
421 	}
422 
423 	evt->event.event = cpu_to_vhost32(vq, event);
424 	evt->event.reason = cpu_to_vhost32(vq, reason);
425 	vs->vs_events_nr++;
426 
427 	return evt;
428 }
429 
430 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
431 {
432 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
433 
434 	/* TODO locking against target/backend threads? */
435 	transport_generic_free_cmd(se_cmd, 0);
436 
437 }
438 
439 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
440 {
441 	return target_put_sess_cmd(se_cmd);
442 }
443 
444 static void
445 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
446 {
447 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
448 	struct virtio_scsi_event *event = &evt->event;
449 	struct virtio_scsi_event __user *eventp;
450 	unsigned out, in;
451 	int head, ret;
452 
453 	if (!vq->private_data) {
454 		vs->vs_events_missed = true;
455 		return;
456 	}
457 
458 again:
459 	vhost_disable_notify(&vs->dev, vq);
460 	head = vhost_get_vq_desc(vq, vq->iov,
461 			ARRAY_SIZE(vq->iov), &out, &in,
462 			NULL, NULL);
463 	if (head < 0) {
464 		vs->vs_events_missed = true;
465 		return;
466 	}
467 	if (head == vq->num) {
468 		if (vhost_enable_notify(&vs->dev, vq))
469 			goto again;
470 		vs->vs_events_missed = true;
471 		return;
472 	}
473 
474 	if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
475 		vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
476 				vq->iov[out].iov_len);
477 		vs->vs_events_missed = true;
478 		return;
479 	}
480 
481 	if (vs->vs_events_missed) {
482 		event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
483 		vs->vs_events_missed = false;
484 	}
485 
486 	eventp = vq->iov[out].iov_base;
487 	ret = __copy_to_user(eventp, event, sizeof(*event));
488 	if (!ret)
489 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
490 	else
491 		vq_err(vq, "Faulted on vhost_scsi_send_event\n");
492 }
493 
494 static void vhost_scsi_evt_work(struct vhost_work *work)
495 {
496 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
497 					vs_event_work);
498 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
499 	struct vhost_scsi_evt *evt;
500 	struct llist_node *llnode;
501 
502 	mutex_lock(&vq->mutex);
503 	llnode = llist_del_all(&vs->vs_event_list);
504 	while (llnode) {
505 		evt = llist_entry(llnode, struct vhost_scsi_evt, list);
506 		llnode = llist_next(llnode);
507 		vhost_scsi_do_evt_work(vs, evt);
508 		vhost_scsi_free_evt(vs, evt);
509 	}
510 	mutex_unlock(&vq->mutex);
511 }
512 
513 /* Fill in status and signal that we are done processing this command
514  *
515  * This is scheduled in the vhost work queue so we are called with the owner
516  * process mm and can access the vring.
517  */
518 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
519 {
520 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
521 					vs_completion_work);
522 	DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
523 	struct virtio_scsi_cmd_resp v_rsp;
524 	struct vhost_scsi_cmd *cmd;
525 	struct llist_node *llnode;
526 	struct se_cmd *se_cmd;
527 	struct iov_iter iov_iter;
528 	int ret, vq;
529 
530 	bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
531 	llnode = llist_del_all(&vs->vs_completion_list);
532 	while (llnode) {
533 		cmd = llist_entry(llnode, struct vhost_scsi_cmd,
534 				     tvc_completion_list);
535 		llnode = llist_next(llnode);
536 		se_cmd = &cmd->tvc_se_cmd;
537 
538 		pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
539 			cmd, se_cmd->residual_count, se_cmd->scsi_status);
540 
541 		memset(&v_rsp, 0, sizeof(v_rsp));
542 		v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
543 		/* TODO is status_qualifier field needed? */
544 		v_rsp.status = se_cmd->scsi_status;
545 		v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
546 						 se_cmd->scsi_sense_length);
547 		memcpy(v_rsp.sense, cmd->tvc_sense_buf,
548 		       se_cmd->scsi_sense_length);
549 
550 		iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
551 			      cmd->tvc_in_iovs, sizeof(v_rsp));
552 		ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
553 		if (likely(ret == sizeof(v_rsp))) {
554 			struct vhost_scsi_virtqueue *q;
555 			vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
556 			q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
557 			vq = q - vs->vqs;
558 			__set_bit(vq, signal);
559 		} else
560 			pr_err("Faulted on virtio_scsi_cmd_resp\n");
561 
562 		vhost_scsi_free_cmd(cmd);
563 	}
564 
565 	vq = -1;
566 	while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
567 		< VHOST_SCSI_MAX_VQ)
568 		vhost_signal(&vs->dev, &vs->vqs[vq].vq);
569 }
570 
571 static struct vhost_scsi_cmd *
572 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
573 		   unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
574 		   u32 exp_data_len, int data_direction)
575 {
576 	struct vhost_scsi_cmd *cmd;
577 	struct vhost_scsi_nexus *tv_nexus;
578 	struct se_session *se_sess;
579 	struct scatterlist *sg, *prot_sg;
580 	struct page **pages;
581 	int tag;
582 
583 	tv_nexus = tpg->tpg_nexus;
584 	if (!tv_nexus) {
585 		pr_err("Unable to locate active struct vhost_scsi_nexus\n");
586 		return ERR_PTR(-EIO);
587 	}
588 	se_sess = tv_nexus->tvn_se_sess;
589 
590 	tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
591 	if (tag < 0) {
592 		pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
593 		return ERR_PTR(-ENOMEM);
594 	}
595 
596 	cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
597 	sg = cmd->tvc_sgl;
598 	prot_sg = cmd->tvc_prot_sgl;
599 	pages = cmd->tvc_upages;
600 	memset(cmd, 0, sizeof(struct vhost_scsi_cmd));
601 
602 	cmd->tvc_sgl = sg;
603 	cmd->tvc_prot_sgl = prot_sg;
604 	cmd->tvc_upages = pages;
605 	cmd->tvc_se_cmd.map_tag = tag;
606 	cmd->tvc_tag = scsi_tag;
607 	cmd->tvc_lun = lun;
608 	cmd->tvc_task_attr = task_attr;
609 	cmd->tvc_exp_data_len = exp_data_len;
610 	cmd->tvc_data_direction = data_direction;
611 	cmd->tvc_nexus = tv_nexus;
612 	cmd->inflight = vhost_scsi_get_inflight(vq);
613 
614 	memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
615 
616 	return cmd;
617 }
618 
619 /*
620  * Map a user memory range into a scatterlist
621  *
622  * Returns the number of scatterlist entries used or -errno on error.
623  */
624 static int
625 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
626 		      void __user *ptr,
627 		      size_t len,
628 		      struct scatterlist *sgl,
629 		      bool write)
630 {
631 	unsigned int npages = 0, offset, nbytes;
632 	unsigned int pages_nr = iov_num_pages(ptr, len);
633 	struct scatterlist *sg = sgl;
634 	struct page **pages = cmd->tvc_upages;
635 	int ret, i;
636 
637 	if (pages_nr > VHOST_SCSI_PREALLOC_UPAGES) {
638 		pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
639 		       " preallocated VHOST_SCSI_PREALLOC_UPAGES: %u\n",
640 			pages_nr, VHOST_SCSI_PREALLOC_UPAGES);
641 		return -ENOBUFS;
642 	}
643 
644 	ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
645 	/* No pages were pinned */
646 	if (ret < 0)
647 		goto out;
648 	/* Less pages pinned than wanted */
649 	if (ret != pages_nr) {
650 		for (i = 0; i < ret; i++)
651 			put_page(pages[i]);
652 		ret = -EFAULT;
653 		goto out;
654 	}
655 
656 	while (len > 0) {
657 		offset = (uintptr_t)ptr & ~PAGE_MASK;
658 		nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
659 		sg_set_page(sg, pages[npages], nbytes, offset);
660 		ptr += nbytes;
661 		len -= nbytes;
662 		sg++;
663 		npages++;
664 	}
665 
666 out:
667 	return ret;
668 }
669 
670 static int
671 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
672 {
673 	int sgl_count = 0;
674 
675 	if (!iter || !iter->iov) {
676 		pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
677 		       " present\n", __func__, bytes);
678 		return -EINVAL;
679 	}
680 
681 	sgl_count = iov_iter_npages(iter, 0xffff);
682 	if (sgl_count > max_sgls) {
683 		pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
684 		       " max_sgls: %d\n", __func__, sgl_count, max_sgls);
685 		return -EINVAL;
686 	}
687 	return sgl_count;
688 }
689 
690 static int
691 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
692 		      struct iov_iter *iter,
693 		      struct scatterlist *sg, int sg_count)
694 {
695 	size_t off = iter->iov_offset;
696 	int i, ret;
697 
698 	for (i = 0; i < iter->nr_segs; i++) {
699 		void __user *base = iter->iov[i].iov_base + off;
700 		size_t len = iter->iov[i].iov_len - off;
701 
702 		ret = vhost_scsi_map_to_sgl(cmd, base, len, sg, write);
703 		if (ret < 0) {
704 			for (i = 0; i < sg_count; i++) {
705 				struct page *page = sg_page(&sg[i]);
706 				if (page)
707 					put_page(page);
708 			}
709 			return ret;
710 		}
711 		sg += ret;
712 		off = 0;
713 	}
714 	return 0;
715 }
716 
717 static int
718 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
719 		 size_t prot_bytes, struct iov_iter *prot_iter,
720 		 size_t data_bytes, struct iov_iter *data_iter)
721 {
722 	int sgl_count, ret;
723 	bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
724 
725 	if (prot_bytes) {
726 		sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
727 						 VHOST_SCSI_PREALLOC_PROT_SGLS);
728 		if (sgl_count < 0)
729 			return sgl_count;
730 
731 		sg_init_table(cmd->tvc_prot_sgl, sgl_count);
732 		cmd->tvc_prot_sgl_count = sgl_count;
733 		pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
734 			 cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
735 
736 		ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
737 					    cmd->tvc_prot_sgl,
738 					    cmd->tvc_prot_sgl_count);
739 		if (ret < 0) {
740 			cmd->tvc_prot_sgl_count = 0;
741 			return ret;
742 		}
743 	}
744 	sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
745 					 VHOST_SCSI_PREALLOC_SGLS);
746 	if (sgl_count < 0)
747 		return sgl_count;
748 
749 	sg_init_table(cmd->tvc_sgl, sgl_count);
750 	cmd->tvc_sgl_count = sgl_count;
751 	pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
752 		  cmd->tvc_sgl, cmd->tvc_sgl_count);
753 
754 	ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
755 				    cmd->tvc_sgl, cmd->tvc_sgl_count);
756 	if (ret < 0) {
757 		cmd->tvc_sgl_count = 0;
758 		return ret;
759 	}
760 	return 0;
761 }
762 
763 static int vhost_scsi_to_tcm_attr(int attr)
764 {
765 	switch (attr) {
766 	case VIRTIO_SCSI_S_SIMPLE:
767 		return TCM_SIMPLE_TAG;
768 	case VIRTIO_SCSI_S_ORDERED:
769 		return TCM_ORDERED_TAG;
770 	case VIRTIO_SCSI_S_HEAD:
771 		return TCM_HEAD_TAG;
772 	case VIRTIO_SCSI_S_ACA:
773 		return TCM_ACA_TAG;
774 	default:
775 		break;
776 	}
777 	return TCM_SIMPLE_TAG;
778 }
779 
780 static void vhost_scsi_submission_work(struct work_struct *work)
781 {
782 	struct vhost_scsi_cmd *cmd =
783 		container_of(work, struct vhost_scsi_cmd, work);
784 	struct vhost_scsi_nexus *tv_nexus;
785 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
786 	struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
787 	int rc;
788 
789 	/* FIXME: BIDI operation */
790 	if (cmd->tvc_sgl_count) {
791 		sg_ptr = cmd->tvc_sgl;
792 
793 		if (cmd->tvc_prot_sgl_count)
794 			sg_prot_ptr = cmd->tvc_prot_sgl;
795 		else
796 			se_cmd->prot_pto = true;
797 	} else {
798 		sg_ptr = NULL;
799 	}
800 	tv_nexus = cmd->tvc_nexus;
801 
802 	se_cmd->tag = 0;
803 	rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
804 			cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
805 			cmd->tvc_lun, cmd->tvc_exp_data_len,
806 			vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
807 			cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
808 			sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
809 			cmd->tvc_prot_sgl_count);
810 	if (rc < 0) {
811 		transport_send_check_condition_and_sense(se_cmd,
812 				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
813 		transport_generic_free_cmd(se_cmd, 0);
814 	}
815 }
816 
817 static void
818 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
819 			   struct vhost_virtqueue *vq,
820 			   int head, unsigned out)
821 {
822 	struct virtio_scsi_cmd_resp __user *resp;
823 	struct virtio_scsi_cmd_resp rsp;
824 	int ret;
825 
826 	memset(&rsp, 0, sizeof(rsp));
827 	rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
828 	resp = vq->iov[out].iov_base;
829 	ret = __copy_to_user(resp, &rsp, sizeof(rsp));
830 	if (!ret)
831 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
832 	else
833 		pr_err("Faulted on virtio_scsi_cmd_resp\n");
834 }
835 
836 static void
837 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
838 {
839 	struct vhost_scsi_tpg **vs_tpg, *tpg;
840 	struct virtio_scsi_cmd_req v_req;
841 	struct virtio_scsi_cmd_req_pi v_req_pi;
842 	struct vhost_scsi_cmd *cmd;
843 	struct iov_iter out_iter, in_iter, prot_iter, data_iter;
844 	u64 tag;
845 	u32 exp_data_len, data_direction;
846 	unsigned int out = 0, in = 0;
847 	int head, ret, prot_bytes;
848 	size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
849 	size_t out_size, in_size;
850 	u16 lun;
851 	u8 *target, *lunp, task_attr;
852 	bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
853 	void *req, *cdb;
854 
855 	mutex_lock(&vq->mutex);
856 	/*
857 	 * We can handle the vq only after the endpoint is setup by calling the
858 	 * VHOST_SCSI_SET_ENDPOINT ioctl.
859 	 */
860 	vs_tpg = vq->private_data;
861 	if (!vs_tpg)
862 		goto out;
863 
864 	vhost_disable_notify(&vs->dev, vq);
865 
866 	for (;;) {
867 		head = vhost_get_vq_desc(vq, vq->iov,
868 					 ARRAY_SIZE(vq->iov), &out, &in,
869 					 NULL, NULL);
870 		pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
871 			 head, out, in);
872 		/* On error, stop handling until the next kick. */
873 		if (unlikely(head < 0))
874 			break;
875 		/* Nothing new?  Wait for eventfd to tell us they refilled. */
876 		if (head == vq->num) {
877 			if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
878 				vhost_disable_notify(&vs->dev, vq);
879 				continue;
880 			}
881 			break;
882 		}
883 		/*
884 		 * Check for a sane response buffer so we can report early
885 		 * errors back to the guest.
886 		 */
887 		if (unlikely(vq->iov[out].iov_len < rsp_size)) {
888 			vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
889 				" size, got %zu bytes\n", vq->iov[out].iov_len);
890 			break;
891 		}
892 		/*
893 		 * Setup pointers and values based upon different virtio-scsi
894 		 * request header if T10_PI is enabled in KVM guest.
895 		 */
896 		if (t10_pi) {
897 			req = &v_req_pi;
898 			req_size = sizeof(v_req_pi);
899 			lunp = &v_req_pi.lun[0];
900 			target = &v_req_pi.lun[1];
901 		} else {
902 			req = &v_req;
903 			req_size = sizeof(v_req);
904 			lunp = &v_req.lun[0];
905 			target = &v_req.lun[1];
906 		}
907 		/*
908 		 * FIXME: Not correct for BIDI operation
909 		 */
910 		out_size = iov_length(vq->iov, out);
911 		in_size = iov_length(&vq->iov[out], in);
912 
913 		/*
914 		 * Copy over the virtio-scsi request header, which for a
915 		 * ANY_LAYOUT enabled guest may span multiple iovecs, or a
916 		 * single iovec may contain both the header + outgoing
917 		 * WRITE payloads.
918 		 *
919 		 * copy_from_iter() will advance out_iter, so that it will
920 		 * point at the start of the outgoing WRITE payload, if
921 		 * DMA_TO_DEVICE is set.
922 		 */
923 		iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
924 
925 		if (unlikely(!copy_from_iter_full(req, req_size, &out_iter))) {
926 			vq_err(vq, "Faulted on copy_from_iter\n");
927 			vhost_scsi_send_bad_target(vs, vq, head, out);
928 			continue;
929 		}
930 		/* virtio-scsi spec requires byte 0 of the lun to be 1 */
931 		if (unlikely(*lunp != 1)) {
932 			vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
933 			vhost_scsi_send_bad_target(vs, vq, head, out);
934 			continue;
935 		}
936 
937 		tpg = ACCESS_ONCE(vs_tpg[*target]);
938 		if (unlikely(!tpg)) {
939 			/* Target does not exist, fail the request */
940 			vhost_scsi_send_bad_target(vs, vq, head, out);
941 			continue;
942 		}
943 		/*
944 		 * Determine data_direction by calculating the total outgoing
945 		 * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
946 		 * response headers respectively.
947 		 *
948 		 * For DMA_TO_DEVICE this is out_iter, which is already pointing
949 		 * to the right place.
950 		 *
951 		 * For DMA_FROM_DEVICE, the iovec will be just past the end
952 		 * of the virtio-scsi response header in either the same
953 		 * or immediately following iovec.
954 		 *
955 		 * Any associated T10_PI bytes for the outgoing / incoming
956 		 * payloads are included in calculation of exp_data_len here.
957 		 */
958 		prot_bytes = 0;
959 
960 		if (out_size > req_size) {
961 			data_direction = DMA_TO_DEVICE;
962 			exp_data_len = out_size - req_size;
963 			data_iter = out_iter;
964 		} else if (in_size > rsp_size) {
965 			data_direction = DMA_FROM_DEVICE;
966 			exp_data_len = in_size - rsp_size;
967 
968 			iov_iter_init(&in_iter, READ, &vq->iov[out], in,
969 				      rsp_size + exp_data_len);
970 			iov_iter_advance(&in_iter, rsp_size);
971 			data_iter = in_iter;
972 		} else {
973 			data_direction = DMA_NONE;
974 			exp_data_len = 0;
975 		}
976 		/*
977 		 * If T10_PI header + payload is present, setup prot_iter values
978 		 * and recalculate data_iter for vhost_scsi_mapal() mapping to
979 		 * host scatterlists via get_user_pages_fast().
980 		 */
981 		if (t10_pi) {
982 			if (v_req_pi.pi_bytesout) {
983 				if (data_direction != DMA_TO_DEVICE) {
984 					vq_err(vq, "Received non zero pi_bytesout,"
985 						" but wrong data_direction\n");
986 					vhost_scsi_send_bad_target(vs, vq, head, out);
987 					continue;
988 				}
989 				prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
990 			} else if (v_req_pi.pi_bytesin) {
991 				if (data_direction != DMA_FROM_DEVICE) {
992 					vq_err(vq, "Received non zero pi_bytesin,"
993 						" but wrong data_direction\n");
994 					vhost_scsi_send_bad_target(vs, vq, head, out);
995 					continue;
996 				}
997 				prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
998 			}
999 			/*
1000 			 * Set prot_iter to data_iter, and advance past any
1001 			 * preceeding prot_bytes that may be present.
1002 			 *
1003 			 * Also fix up the exp_data_len to reflect only the
1004 			 * actual data payload length.
1005 			 */
1006 			if (prot_bytes) {
1007 				exp_data_len -= prot_bytes;
1008 				prot_iter = data_iter;
1009 				iov_iter_advance(&data_iter, prot_bytes);
1010 			}
1011 			tag = vhost64_to_cpu(vq, v_req_pi.tag);
1012 			task_attr = v_req_pi.task_attr;
1013 			cdb = &v_req_pi.cdb[0];
1014 			lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1015 		} else {
1016 			tag = vhost64_to_cpu(vq, v_req.tag);
1017 			task_attr = v_req.task_attr;
1018 			cdb = &v_req.cdb[0];
1019 			lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1020 		}
1021 		/*
1022 		 * Check that the received CDB size does not exceeded our
1023 		 * hardcoded max for vhost-scsi, then get a pre-allocated
1024 		 * cmd descriptor for the new virtio-scsi tag.
1025 		 *
1026 		 * TODO what if cdb was too small for varlen cdb header?
1027 		 */
1028 		if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1029 			vq_err(vq, "Received SCSI CDB with command_size: %d that"
1030 				" exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1031 				scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1032 			vhost_scsi_send_bad_target(vs, vq, head, out);
1033 			continue;
1034 		}
1035 		cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1036 					 exp_data_len + prot_bytes,
1037 					 data_direction);
1038 		if (IS_ERR(cmd)) {
1039 			vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1040 			       PTR_ERR(cmd));
1041 			vhost_scsi_send_bad_target(vs, vq, head, out);
1042 			continue;
1043 		}
1044 		cmd->tvc_vhost = vs;
1045 		cmd->tvc_vq = vq;
1046 		cmd->tvc_resp_iov = vq->iov[out];
1047 		cmd->tvc_in_iovs = in;
1048 
1049 		pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1050 			 cmd->tvc_cdb[0], cmd->tvc_lun);
1051 		pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1052 			 " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1053 
1054 		if (data_direction != DMA_NONE) {
1055 			ret = vhost_scsi_mapal(cmd,
1056 					       prot_bytes, &prot_iter,
1057 					       exp_data_len, &data_iter);
1058 			if (unlikely(ret)) {
1059 				vq_err(vq, "Failed to map iov to sgl\n");
1060 				vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1061 				vhost_scsi_send_bad_target(vs, vq, head, out);
1062 				continue;
1063 			}
1064 		}
1065 		/*
1066 		 * Save the descriptor from vhost_get_vq_desc() to be used to
1067 		 * complete the virtio-scsi request in TCM callback context via
1068 		 * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1069 		 */
1070 		cmd->tvc_vq_desc = head;
1071 		/*
1072 		 * Dispatch cmd descriptor for cmwq execution in process
1073 		 * context provided by vhost_scsi_workqueue.  This also ensures
1074 		 * cmd is executed on the same kworker CPU as this vhost
1075 		 * thread to gain positive L2 cache locality effects.
1076 		 */
1077 		INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1078 		queue_work(vhost_scsi_workqueue, &cmd->work);
1079 	}
1080 out:
1081 	mutex_unlock(&vq->mutex);
1082 }
1083 
1084 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1085 {
1086 	pr_debug("%s: The handling func for control queue.\n", __func__);
1087 }
1088 
1089 static void
1090 vhost_scsi_send_evt(struct vhost_scsi *vs,
1091 		   struct vhost_scsi_tpg *tpg,
1092 		   struct se_lun *lun,
1093 		   u32 event,
1094 		   u32 reason)
1095 {
1096 	struct vhost_scsi_evt *evt;
1097 
1098 	evt = vhost_scsi_allocate_evt(vs, event, reason);
1099 	if (!evt)
1100 		return;
1101 
1102 	if (tpg && lun) {
1103 		/* TODO: share lun setup code with virtio-scsi.ko */
1104 		/*
1105 		 * Note: evt->event is zeroed when we allocate it and
1106 		 * lun[4-7] need to be zero according to virtio-scsi spec.
1107 		 */
1108 		evt->event.lun[0] = 0x01;
1109 		evt->event.lun[1] = tpg->tport_tpgt;
1110 		if (lun->unpacked_lun >= 256)
1111 			evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1112 		evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1113 	}
1114 
1115 	llist_add(&evt->list, &vs->vs_event_list);
1116 	vhost_work_queue(&vs->dev, &vs->vs_event_work);
1117 }
1118 
1119 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1120 {
1121 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1122 						poll.work);
1123 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1124 
1125 	mutex_lock(&vq->mutex);
1126 	if (!vq->private_data)
1127 		goto out;
1128 
1129 	if (vs->vs_events_missed)
1130 		vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1131 out:
1132 	mutex_unlock(&vq->mutex);
1133 }
1134 
1135 static void vhost_scsi_handle_kick(struct vhost_work *work)
1136 {
1137 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1138 						poll.work);
1139 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1140 
1141 	vhost_scsi_handle_vq(vs, vq);
1142 }
1143 
1144 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1145 {
1146 	vhost_poll_flush(&vs->vqs[index].vq.poll);
1147 }
1148 
1149 /* Callers must hold dev mutex */
1150 static void vhost_scsi_flush(struct vhost_scsi *vs)
1151 {
1152 	struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1153 	int i;
1154 
1155 	/* Init new inflight and remember the old inflight */
1156 	vhost_scsi_init_inflight(vs, old_inflight);
1157 
1158 	/*
1159 	 * The inflight->kref was initialized to 1. We decrement it here to
1160 	 * indicate the start of the flush operation so that it will reach 0
1161 	 * when all the reqs are finished.
1162 	 */
1163 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1164 		kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1165 
1166 	/* Flush both the vhost poll and vhost work */
1167 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1168 		vhost_scsi_flush_vq(vs, i);
1169 	vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1170 	vhost_work_flush(&vs->dev, &vs->vs_event_work);
1171 
1172 	/* Wait for all reqs issued before the flush to be finished */
1173 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1174 		wait_for_completion(&old_inflight[i]->comp);
1175 }
1176 
1177 /*
1178  * Called from vhost_scsi_ioctl() context to walk the list of available
1179  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1180  *
1181  *  The lock nesting rule is:
1182  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1183  */
1184 static int
1185 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1186 			struct vhost_scsi_target *t)
1187 {
1188 	struct se_portal_group *se_tpg;
1189 	struct vhost_scsi_tport *tv_tport;
1190 	struct vhost_scsi_tpg *tpg;
1191 	struct vhost_scsi_tpg **vs_tpg;
1192 	struct vhost_virtqueue *vq;
1193 	int index, ret, i, len;
1194 	bool match = false;
1195 
1196 	mutex_lock(&vhost_scsi_mutex);
1197 	mutex_lock(&vs->dev.mutex);
1198 
1199 	/* Verify that ring has been setup correctly. */
1200 	for (index = 0; index < vs->dev.nvqs; ++index) {
1201 		/* Verify that ring has been setup correctly. */
1202 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1203 			ret = -EFAULT;
1204 			goto out;
1205 		}
1206 	}
1207 
1208 	len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1209 	vs_tpg = kzalloc(len, GFP_KERNEL);
1210 	if (!vs_tpg) {
1211 		ret = -ENOMEM;
1212 		goto out;
1213 	}
1214 	if (vs->vs_tpg)
1215 		memcpy(vs_tpg, vs->vs_tpg, len);
1216 
1217 	list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1218 		mutex_lock(&tpg->tv_tpg_mutex);
1219 		if (!tpg->tpg_nexus) {
1220 			mutex_unlock(&tpg->tv_tpg_mutex);
1221 			continue;
1222 		}
1223 		if (tpg->tv_tpg_vhost_count != 0) {
1224 			mutex_unlock(&tpg->tv_tpg_mutex);
1225 			continue;
1226 		}
1227 		tv_tport = tpg->tport;
1228 
1229 		if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1230 			if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1231 				kfree(vs_tpg);
1232 				mutex_unlock(&tpg->tv_tpg_mutex);
1233 				ret = -EEXIST;
1234 				goto out;
1235 			}
1236 			/*
1237 			 * In order to ensure individual vhost-scsi configfs
1238 			 * groups cannot be removed while in use by vhost ioctl,
1239 			 * go ahead and take an explicit se_tpg->tpg_group.cg_item
1240 			 * dependency now.
1241 			 */
1242 			se_tpg = &tpg->se_tpg;
1243 			ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1244 			if (ret) {
1245 				pr_warn("configfs_depend_item() failed: %d\n", ret);
1246 				kfree(vs_tpg);
1247 				mutex_unlock(&tpg->tv_tpg_mutex);
1248 				goto out;
1249 			}
1250 			tpg->tv_tpg_vhost_count++;
1251 			tpg->vhost_scsi = vs;
1252 			vs_tpg[tpg->tport_tpgt] = tpg;
1253 			smp_mb__after_atomic();
1254 			match = true;
1255 		}
1256 		mutex_unlock(&tpg->tv_tpg_mutex);
1257 	}
1258 
1259 	if (match) {
1260 		memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1261 		       sizeof(vs->vs_vhost_wwpn));
1262 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1263 			vq = &vs->vqs[i].vq;
1264 			mutex_lock(&vq->mutex);
1265 			vq->private_data = vs_tpg;
1266 			vhost_vq_init_access(vq);
1267 			mutex_unlock(&vq->mutex);
1268 		}
1269 		ret = 0;
1270 	} else {
1271 		ret = -EEXIST;
1272 	}
1273 
1274 	/*
1275 	 * Act as synchronize_rcu to make sure access to
1276 	 * old vs->vs_tpg is finished.
1277 	 */
1278 	vhost_scsi_flush(vs);
1279 	kfree(vs->vs_tpg);
1280 	vs->vs_tpg = vs_tpg;
1281 
1282 out:
1283 	mutex_unlock(&vs->dev.mutex);
1284 	mutex_unlock(&vhost_scsi_mutex);
1285 	return ret;
1286 }
1287 
1288 static int
1289 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1290 			  struct vhost_scsi_target *t)
1291 {
1292 	struct se_portal_group *se_tpg;
1293 	struct vhost_scsi_tport *tv_tport;
1294 	struct vhost_scsi_tpg *tpg;
1295 	struct vhost_virtqueue *vq;
1296 	bool match = false;
1297 	int index, ret, i;
1298 	u8 target;
1299 
1300 	mutex_lock(&vhost_scsi_mutex);
1301 	mutex_lock(&vs->dev.mutex);
1302 	/* Verify that ring has been setup correctly. */
1303 	for (index = 0; index < vs->dev.nvqs; ++index) {
1304 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1305 			ret = -EFAULT;
1306 			goto err_dev;
1307 		}
1308 	}
1309 
1310 	if (!vs->vs_tpg) {
1311 		ret = 0;
1312 		goto err_dev;
1313 	}
1314 
1315 	for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1316 		target = i;
1317 		tpg = vs->vs_tpg[target];
1318 		if (!tpg)
1319 			continue;
1320 
1321 		mutex_lock(&tpg->tv_tpg_mutex);
1322 		tv_tport = tpg->tport;
1323 		if (!tv_tport) {
1324 			ret = -ENODEV;
1325 			goto err_tpg;
1326 		}
1327 
1328 		if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1329 			pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1330 				" does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1331 				tv_tport->tport_name, tpg->tport_tpgt,
1332 				t->vhost_wwpn, t->vhost_tpgt);
1333 			ret = -EINVAL;
1334 			goto err_tpg;
1335 		}
1336 		tpg->tv_tpg_vhost_count--;
1337 		tpg->vhost_scsi = NULL;
1338 		vs->vs_tpg[target] = NULL;
1339 		match = true;
1340 		mutex_unlock(&tpg->tv_tpg_mutex);
1341 		/*
1342 		 * Release se_tpg->tpg_group.cg_item configfs dependency now
1343 		 * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1344 		 */
1345 		se_tpg = &tpg->se_tpg;
1346 		target_undepend_item(&se_tpg->tpg_group.cg_item);
1347 	}
1348 	if (match) {
1349 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1350 			vq = &vs->vqs[i].vq;
1351 			mutex_lock(&vq->mutex);
1352 			vq->private_data = NULL;
1353 			mutex_unlock(&vq->mutex);
1354 		}
1355 	}
1356 	/*
1357 	 * Act as synchronize_rcu to make sure access to
1358 	 * old vs->vs_tpg is finished.
1359 	 */
1360 	vhost_scsi_flush(vs);
1361 	kfree(vs->vs_tpg);
1362 	vs->vs_tpg = NULL;
1363 	WARN_ON(vs->vs_events_nr);
1364 	mutex_unlock(&vs->dev.mutex);
1365 	mutex_unlock(&vhost_scsi_mutex);
1366 	return 0;
1367 
1368 err_tpg:
1369 	mutex_unlock(&tpg->tv_tpg_mutex);
1370 err_dev:
1371 	mutex_unlock(&vs->dev.mutex);
1372 	mutex_unlock(&vhost_scsi_mutex);
1373 	return ret;
1374 }
1375 
1376 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1377 {
1378 	struct vhost_virtqueue *vq;
1379 	int i;
1380 
1381 	if (features & ~VHOST_SCSI_FEATURES)
1382 		return -EOPNOTSUPP;
1383 
1384 	mutex_lock(&vs->dev.mutex);
1385 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
1386 	    !vhost_log_access_ok(&vs->dev)) {
1387 		mutex_unlock(&vs->dev.mutex);
1388 		return -EFAULT;
1389 	}
1390 
1391 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1392 		vq = &vs->vqs[i].vq;
1393 		mutex_lock(&vq->mutex);
1394 		vq->acked_features = features;
1395 		mutex_unlock(&vq->mutex);
1396 	}
1397 	mutex_unlock(&vs->dev.mutex);
1398 	return 0;
1399 }
1400 
1401 static int vhost_scsi_open(struct inode *inode, struct file *f)
1402 {
1403 	struct vhost_scsi *vs;
1404 	struct vhost_virtqueue **vqs;
1405 	int r = -ENOMEM, i;
1406 
1407 	vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
1408 	if (!vs) {
1409 		vs = vzalloc(sizeof(*vs));
1410 		if (!vs)
1411 			goto err_vs;
1412 	}
1413 
1414 	vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1415 	if (!vqs)
1416 		goto err_vqs;
1417 
1418 	vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1419 	vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1420 
1421 	vs->vs_events_nr = 0;
1422 	vs->vs_events_missed = false;
1423 
1424 	vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1425 	vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1426 	vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1427 	vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1428 	for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1429 		vqs[i] = &vs->vqs[i].vq;
1430 		vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1431 	}
1432 	vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1433 
1434 	vhost_scsi_init_inflight(vs, NULL);
1435 
1436 	f->private_data = vs;
1437 	return 0;
1438 
1439 err_vqs:
1440 	kvfree(vs);
1441 err_vs:
1442 	return r;
1443 }
1444 
1445 static int vhost_scsi_release(struct inode *inode, struct file *f)
1446 {
1447 	struct vhost_scsi *vs = f->private_data;
1448 	struct vhost_scsi_target t;
1449 
1450 	mutex_lock(&vs->dev.mutex);
1451 	memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1452 	mutex_unlock(&vs->dev.mutex);
1453 	vhost_scsi_clear_endpoint(vs, &t);
1454 	vhost_dev_stop(&vs->dev);
1455 	vhost_dev_cleanup(&vs->dev, false);
1456 	/* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1457 	vhost_scsi_flush(vs);
1458 	kfree(vs->dev.vqs);
1459 	kvfree(vs);
1460 	return 0;
1461 }
1462 
1463 static long
1464 vhost_scsi_ioctl(struct file *f,
1465 		 unsigned int ioctl,
1466 		 unsigned long arg)
1467 {
1468 	struct vhost_scsi *vs = f->private_data;
1469 	struct vhost_scsi_target backend;
1470 	void __user *argp = (void __user *)arg;
1471 	u64 __user *featurep = argp;
1472 	u32 __user *eventsp = argp;
1473 	u32 events_missed;
1474 	u64 features;
1475 	int r, abi_version = VHOST_SCSI_ABI_VERSION;
1476 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1477 
1478 	switch (ioctl) {
1479 	case VHOST_SCSI_SET_ENDPOINT:
1480 		if (copy_from_user(&backend, argp, sizeof backend))
1481 			return -EFAULT;
1482 		if (backend.reserved != 0)
1483 			return -EOPNOTSUPP;
1484 
1485 		return vhost_scsi_set_endpoint(vs, &backend);
1486 	case VHOST_SCSI_CLEAR_ENDPOINT:
1487 		if (copy_from_user(&backend, argp, sizeof backend))
1488 			return -EFAULT;
1489 		if (backend.reserved != 0)
1490 			return -EOPNOTSUPP;
1491 
1492 		return vhost_scsi_clear_endpoint(vs, &backend);
1493 	case VHOST_SCSI_GET_ABI_VERSION:
1494 		if (copy_to_user(argp, &abi_version, sizeof abi_version))
1495 			return -EFAULT;
1496 		return 0;
1497 	case VHOST_SCSI_SET_EVENTS_MISSED:
1498 		if (get_user(events_missed, eventsp))
1499 			return -EFAULT;
1500 		mutex_lock(&vq->mutex);
1501 		vs->vs_events_missed = events_missed;
1502 		mutex_unlock(&vq->mutex);
1503 		return 0;
1504 	case VHOST_SCSI_GET_EVENTS_MISSED:
1505 		mutex_lock(&vq->mutex);
1506 		events_missed = vs->vs_events_missed;
1507 		mutex_unlock(&vq->mutex);
1508 		if (put_user(events_missed, eventsp))
1509 			return -EFAULT;
1510 		return 0;
1511 	case VHOST_GET_FEATURES:
1512 		features = VHOST_SCSI_FEATURES;
1513 		if (copy_to_user(featurep, &features, sizeof features))
1514 			return -EFAULT;
1515 		return 0;
1516 	case VHOST_SET_FEATURES:
1517 		if (copy_from_user(&features, featurep, sizeof features))
1518 			return -EFAULT;
1519 		return vhost_scsi_set_features(vs, features);
1520 	default:
1521 		mutex_lock(&vs->dev.mutex);
1522 		r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1523 		/* TODO: flush backend after dev ioctl. */
1524 		if (r == -ENOIOCTLCMD)
1525 			r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1526 		mutex_unlock(&vs->dev.mutex);
1527 		return r;
1528 	}
1529 }
1530 
1531 #ifdef CONFIG_COMPAT
1532 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1533 				unsigned long arg)
1534 {
1535 	return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1536 }
1537 #endif
1538 
1539 static const struct file_operations vhost_scsi_fops = {
1540 	.owner          = THIS_MODULE,
1541 	.release        = vhost_scsi_release,
1542 	.unlocked_ioctl = vhost_scsi_ioctl,
1543 #ifdef CONFIG_COMPAT
1544 	.compat_ioctl	= vhost_scsi_compat_ioctl,
1545 #endif
1546 	.open           = vhost_scsi_open,
1547 	.llseek		= noop_llseek,
1548 };
1549 
1550 static struct miscdevice vhost_scsi_misc = {
1551 	MISC_DYNAMIC_MINOR,
1552 	"vhost-scsi",
1553 	&vhost_scsi_fops,
1554 };
1555 
1556 static int __init vhost_scsi_register(void)
1557 {
1558 	return misc_register(&vhost_scsi_misc);
1559 }
1560 
1561 static void vhost_scsi_deregister(void)
1562 {
1563 	misc_deregister(&vhost_scsi_misc);
1564 }
1565 
1566 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1567 {
1568 	switch (tport->tport_proto_id) {
1569 	case SCSI_PROTOCOL_SAS:
1570 		return "SAS";
1571 	case SCSI_PROTOCOL_FCP:
1572 		return "FCP";
1573 	case SCSI_PROTOCOL_ISCSI:
1574 		return "iSCSI";
1575 	default:
1576 		break;
1577 	}
1578 
1579 	return "Unknown";
1580 }
1581 
1582 static void
1583 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1584 		  struct se_lun *lun, bool plug)
1585 {
1586 
1587 	struct vhost_scsi *vs = tpg->vhost_scsi;
1588 	struct vhost_virtqueue *vq;
1589 	u32 reason;
1590 
1591 	if (!vs)
1592 		return;
1593 
1594 	mutex_lock(&vs->dev.mutex);
1595 
1596 	if (plug)
1597 		reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1598 	else
1599 		reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1600 
1601 	vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1602 	mutex_lock(&vq->mutex);
1603 	if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1604 		vhost_scsi_send_evt(vs, tpg, lun,
1605 				   VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1606 	mutex_unlock(&vq->mutex);
1607 	mutex_unlock(&vs->dev.mutex);
1608 }
1609 
1610 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1611 {
1612 	vhost_scsi_do_plug(tpg, lun, true);
1613 }
1614 
1615 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1616 {
1617 	vhost_scsi_do_plug(tpg, lun, false);
1618 }
1619 
1620 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1621 			       struct se_lun *lun)
1622 {
1623 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1624 				struct vhost_scsi_tpg, se_tpg);
1625 
1626 	mutex_lock(&vhost_scsi_mutex);
1627 
1628 	mutex_lock(&tpg->tv_tpg_mutex);
1629 	tpg->tv_tpg_port_count++;
1630 	mutex_unlock(&tpg->tv_tpg_mutex);
1631 
1632 	vhost_scsi_hotplug(tpg, lun);
1633 
1634 	mutex_unlock(&vhost_scsi_mutex);
1635 
1636 	return 0;
1637 }
1638 
1639 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1640 				  struct se_lun *lun)
1641 {
1642 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1643 				struct vhost_scsi_tpg, se_tpg);
1644 
1645 	mutex_lock(&vhost_scsi_mutex);
1646 
1647 	mutex_lock(&tpg->tv_tpg_mutex);
1648 	tpg->tv_tpg_port_count--;
1649 	mutex_unlock(&tpg->tv_tpg_mutex);
1650 
1651 	vhost_scsi_hotunplug(tpg, lun);
1652 
1653 	mutex_unlock(&vhost_scsi_mutex);
1654 }
1655 
1656 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1657 {
1658 	struct vhost_scsi_cmd *tv_cmd;
1659 	unsigned int i;
1660 
1661 	if (!se_sess->sess_cmd_map)
1662 		return;
1663 
1664 	for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1665 		tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1666 
1667 		kfree(tv_cmd->tvc_sgl);
1668 		kfree(tv_cmd->tvc_prot_sgl);
1669 		kfree(tv_cmd->tvc_upages);
1670 	}
1671 }
1672 
1673 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1674 		struct config_item *item, const char *page, size_t count)
1675 {
1676 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
1677 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1678 				struct vhost_scsi_tpg, se_tpg);
1679 	unsigned long val;
1680 	int ret = kstrtoul(page, 0, &val);
1681 
1682 	if (ret) {
1683 		pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1684 		return ret;
1685 	}
1686 	if (val != 0 && val != 1 && val != 3) {
1687 		pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1688 		return -EINVAL;
1689 	}
1690 	tpg->tv_fabric_prot_type = val;
1691 
1692 	return count;
1693 }
1694 
1695 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1696 		struct config_item *item, char *page)
1697 {
1698 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
1699 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1700 				struct vhost_scsi_tpg, se_tpg);
1701 
1702 	return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1703 }
1704 
1705 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1706 
1707 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1708 	&vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1709 	NULL,
1710 };
1711 
1712 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1713 			       struct se_session *se_sess, void *p)
1714 {
1715 	struct vhost_scsi_cmd *tv_cmd;
1716 	unsigned int i;
1717 
1718 	for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1719 		tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1720 
1721 		tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1722 					VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1723 		if (!tv_cmd->tvc_sgl) {
1724 			pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1725 			goto out;
1726 		}
1727 
1728 		tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1729 				VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1730 		if (!tv_cmd->tvc_upages) {
1731 			pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1732 			goto out;
1733 		}
1734 
1735 		tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1736 				VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1737 		if (!tv_cmd->tvc_prot_sgl) {
1738 			pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1739 			goto out;
1740 		}
1741 	}
1742 	return 0;
1743 out:
1744 	vhost_scsi_free_cmd_map_res(se_sess);
1745 	return -ENOMEM;
1746 }
1747 
1748 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1749 				const char *name)
1750 {
1751 	struct vhost_scsi_nexus *tv_nexus;
1752 
1753 	mutex_lock(&tpg->tv_tpg_mutex);
1754 	if (tpg->tpg_nexus) {
1755 		mutex_unlock(&tpg->tv_tpg_mutex);
1756 		pr_debug("tpg->tpg_nexus already exists\n");
1757 		return -EEXIST;
1758 	}
1759 
1760 	tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1761 	if (!tv_nexus) {
1762 		mutex_unlock(&tpg->tv_tpg_mutex);
1763 		pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1764 		return -ENOMEM;
1765 	}
1766 	/*
1767 	 * Since we are running in 'demo mode' this call with generate a
1768 	 * struct se_node_acl for the vhost_scsi struct se_portal_group with
1769 	 * the SCSI Initiator port name of the passed configfs group 'name'.
1770 	 */
1771 	tv_nexus->tvn_se_sess = target_alloc_session(&tpg->se_tpg,
1772 					VHOST_SCSI_DEFAULT_TAGS,
1773 					sizeof(struct vhost_scsi_cmd),
1774 					TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1775 					(unsigned char *)name, tv_nexus,
1776 					vhost_scsi_nexus_cb);
1777 	if (IS_ERR(tv_nexus->tvn_se_sess)) {
1778 		mutex_unlock(&tpg->tv_tpg_mutex);
1779 		kfree(tv_nexus);
1780 		return -ENOMEM;
1781 	}
1782 	tpg->tpg_nexus = tv_nexus;
1783 
1784 	mutex_unlock(&tpg->tv_tpg_mutex);
1785 	return 0;
1786 }
1787 
1788 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1789 {
1790 	struct se_session *se_sess;
1791 	struct vhost_scsi_nexus *tv_nexus;
1792 
1793 	mutex_lock(&tpg->tv_tpg_mutex);
1794 	tv_nexus = tpg->tpg_nexus;
1795 	if (!tv_nexus) {
1796 		mutex_unlock(&tpg->tv_tpg_mutex);
1797 		return -ENODEV;
1798 	}
1799 
1800 	se_sess = tv_nexus->tvn_se_sess;
1801 	if (!se_sess) {
1802 		mutex_unlock(&tpg->tv_tpg_mutex);
1803 		return -ENODEV;
1804 	}
1805 
1806 	if (tpg->tv_tpg_port_count != 0) {
1807 		mutex_unlock(&tpg->tv_tpg_mutex);
1808 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1809 			" active TPG port count: %d\n",
1810 			tpg->tv_tpg_port_count);
1811 		return -EBUSY;
1812 	}
1813 
1814 	if (tpg->tv_tpg_vhost_count != 0) {
1815 		mutex_unlock(&tpg->tv_tpg_mutex);
1816 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1817 			" active TPG vhost count: %d\n",
1818 			tpg->tv_tpg_vhost_count);
1819 		return -EBUSY;
1820 	}
1821 
1822 	pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1823 		" %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1824 		tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1825 
1826 	vhost_scsi_free_cmd_map_res(se_sess);
1827 	/*
1828 	 * Release the SCSI I_T Nexus to the emulated vhost Target Port
1829 	 */
1830 	transport_deregister_session(tv_nexus->tvn_se_sess);
1831 	tpg->tpg_nexus = NULL;
1832 	mutex_unlock(&tpg->tv_tpg_mutex);
1833 
1834 	kfree(tv_nexus);
1835 	return 0;
1836 }
1837 
1838 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1839 {
1840 	struct se_portal_group *se_tpg = to_tpg(item);
1841 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1842 				struct vhost_scsi_tpg, se_tpg);
1843 	struct vhost_scsi_nexus *tv_nexus;
1844 	ssize_t ret;
1845 
1846 	mutex_lock(&tpg->tv_tpg_mutex);
1847 	tv_nexus = tpg->tpg_nexus;
1848 	if (!tv_nexus) {
1849 		mutex_unlock(&tpg->tv_tpg_mutex);
1850 		return -ENODEV;
1851 	}
1852 	ret = snprintf(page, PAGE_SIZE, "%s\n",
1853 			tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1854 	mutex_unlock(&tpg->tv_tpg_mutex);
1855 
1856 	return ret;
1857 }
1858 
1859 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1860 		const char *page, size_t count)
1861 {
1862 	struct se_portal_group *se_tpg = to_tpg(item);
1863 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1864 				struct vhost_scsi_tpg, se_tpg);
1865 	struct vhost_scsi_tport *tport_wwn = tpg->tport;
1866 	unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1867 	int ret;
1868 	/*
1869 	 * Shutdown the active I_T nexus if 'NULL' is passed..
1870 	 */
1871 	if (!strncmp(page, "NULL", 4)) {
1872 		ret = vhost_scsi_drop_nexus(tpg);
1873 		return (!ret) ? count : ret;
1874 	}
1875 	/*
1876 	 * Otherwise make sure the passed virtual Initiator port WWN matches
1877 	 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1878 	 * vhost_scsi_make_nexus().
1879 	 */
1880 	if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1881 		pr_err("Emulated NAA Sas Address: %s, exceeds"
1882 				" max: %d\n", page, VHOST_SCSI_NAMELEN);
1883 		return -EINVAL;
1884 	}
1885 	snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1886 
1887 	ptr = strstr(i_port, "naa.");
1888 	if (ptr) {
1889 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1890 			pr_err("Passed SAS Initiator Port %s does not"
1891 				" match target port protoid: %s\n", i_port,
1892 				vhost_scsi_dump_proto_id(tport_wwn));
1893 			return -EINVAL;
1894 		}
1895 		port_ptr = &i_port[0];
1896 		goto check_newline;
1897 	}
1898 	ptr = strstr(i_port, "fc.");
1899 	if (ptr) {
1900 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1901 			pr_err("Passed FCP Initiator Port %s does not"
1902 				" match target port protoid: %s\n", i_port,
1903 				vhost_scsi_dump_proto_id(tport_wwn));
1904 			return -EINVAL;
1905 		}
1906 		port_ptr = &i_port[3]; /* Skip over "fc." */
1907 		goto check_newline;
1908 	}
1909 	ptr = strstr(i_port, "iqn.");
1910 	if (ptr) {
1911 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1912 			pr_err("Passed iSCSI Initiator Port %s does not"
1913 				" match target port protoid: %s\n", i_port,
1914 				vhost_scsi_dump_proto_id(tport_wwn));
1915 			return -EINVAL;
1916 		}
1917 		port_ptr = &i_port[0];
1918 		goto check_newline;
1919 	}
1920 	pr_err("Unable to locate prefix for emulated Initiator Port:"
1921 			" %s\n", i_port);
1922 	return -EINVAL;
1923 	/*
1924 	 * Clear any trailing newline for the NAA WWN
1925 	 */
1926 check_newline:
1927 	if (i_port[strlen(i_port)-1] == '\n')
1928 		i_port[strlen(i_port)-1] = '\0';
1929 
1930 	ret = vhost_scsi_make_nexus(tpg, port_ptr);
1931 	if (ret < 0)
1932 		return ret;
1933 
1934 	return count;
1935 }
1936 
1937 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1938 
1939 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1940 	&vhost_scsi_tpg_attr_nexus,
1941 	NULL,
1942 };
1943 
1944 static struct se_portal_group *
1945 vhost_scsi_make_tpg(struct se_wwn *wwn,
1946 		   struct config_group *group,
1947 		   const char *name)
1948 {
1949 	struct vhost_scsi_tport *tport = container_of(wwn,
1950 			struct vhost_scsi_tport, tport_wwn);
1951 
1952 	struct vhost_scsi_tpg *tpg;
1953 	u16 tpgt;
1954 	int ret;
1955 
1956 	if (strstr(name, "tpgt_") != name)
1957 		return ERR_PTR(-EINVAL);
1958 	if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1959 		return ERR_PTR(-EINVAL);
1960 
1961 	tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1962 	if (!tpg) {
1963 		pr_err("Unable to allocate struct vhost_scsi_tpg");
1964 		return ERR_PTR(-ENOMEM);
1965 	}
1966 	mutex_init(&tpg->tv_tpg_mutex);
1967 	INIT_LIST_HEAD(&tpg->tv_tpg_list);
1968 	tpg->tport = tport;
1969 	tpg->tport_tpgt = tpgt;
1970 
1971 	ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1972 	if (ret < 0) {
1973 		kfree(tpg);
1974 		return NULL;
1975 	}
1976 	mutex_lock(&vhost_scsi_mutex);
1977 	list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1978 	mutex_unlock(&vhost_scsi_mutex);
1979 
1980 	return &tpg->se_tpg;
1981 }
1982 
1983 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1984 {
1985 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1986 				struct vhost_scsi_tpg, se_tpg);
1987 
1988 	mutex_lock(&vhost_scsi_mutex);
1989 	list_del(&tpg->tv_tpg_list);
1990 	mutex_unlock(&vhost_scsi_mutex);
1991 	/*
1992 	 * Release the virtual I_T Nexus for this vhost TPG
1993 	 */
1994 	vhost_scsi_drop_nexus(tpg);
1995 	/*
1996 	 * Deregister the se_tpg from TCM..
1997 	 */
1998 	core_tpg_deregister(se_tpg);
1999 	kfree(tpg);
2000 }
2001 
2002 static struct se_wwn *
2003 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2004 		     struct config_group *group,
2005 		     const char *name)
2006 {
2007 	struct vhost_scsi_tport *tport;
2008 	char *ptr;
2009 	u64 wwpn = 0;
2010 	int off = 0;
2011 
2012 	/* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2013 		return ERR_PTR(-EINVAL); */
2014 
2015 	tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2016 	if (!tport) {
2017 		pr_err("Unable to allocate struct vhost_scsi_tport");
2018 		return ERR_PTR(-ENOMEM);
2019 	}
2020 	tport->tport_wwpn = wwpn;
2021 	/*
2022 	 * Determine the emulated Protocol Identifier and Target Port Name
2023 	 * based on the incoming configfs directory name.
2024 	 */
2025 	ptr = strstr(name, "naa.");
2026 	if (ptr) {
2027 		tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2028 		goto check_len;
2029 	}
2030 	ptr = strstr(name, "fc.");
2031 	if (ptr) {
2032 		tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2033 		off = 3; /* Skip over "fc." */
2034 		goto check_len;
2035 	}
2036 	ptr = strstr(name, "iqn.");
2037 	if (ptr) {
2038 		tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2039 		goto check_len;
2040 	}
2041 
2042 	pr_err("Unable to locate prefix for emulated Target Port:"
2043 			" %s\n", name);
2044 	kfree(tport);
2045 	return ERR_PTR(-EINVAL);
2046 
2047 check_len:
2048 	if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2049 		pr_err("Emulated %s Address: %s, exceeds"
2050 			" max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2051 			VHOST_SCSI_NAMELEN);
2052 		kfree(tport);
2053 		return ERR_PTR(-EINVAL);
2054 	}
2055 	snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2056 
2057 	pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2058 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2059 
2060 	return &tport->tport_wwn;
2061 }
2062 
2063 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2064 {
2065 	struct vhost_scsi_tport *tport = container_of(wwn,
2066 				struct vhost_scsi_tport, tport_wwn);
2067 
2068 	pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2069 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2070 		tport->tport_name);
2071 
2072 	kfree(tport);
2073 }
2074 
2075 static ssize_t
2076 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2077 {
2078 	return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2079 		"on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2080 		utsname()->machine);
2081 }
2082 
2083 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2084 
2085 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2086 	&vhost_scsi_wwn_attr_version,
2087 	NULL,
2088 };
2089 
2090 static const struct target_core_fabric_ops vhost_scsi_ops = {
2091 	.module				= THIS_MODULE,
2092 	.name				= "vhost",
2093 	.get_fabric_name		= vhost_scsi_get_fabric_name,
2094 	.tpg_get_wwn			= vhost_scsi_get_fabric_wwn,
2095 	.tpg_get_tag			= vhost_scsi_get_tpgt,
2096 	.tpg_check_demo_mode		= vhost_scsi_check_true,
2097 	.tpg_check_demo_mode_cache	= vhost_scsi_check_true,
2098 	.tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2099 	.tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2100 	.tpg_check_prot_fabric_only	= vhost_scsi_check_prot_fabric_only,
2101 	.tpg_get_inst_index		= vhost_scsi_tpg_get_inst_index,
2102 	.release_cmd			= vhost_scsi_release_cmd,
2103 	.check_stop_free		= vhost_scsi_check_stop_free,
2104 	.sess_get_index			= vhost_scsi_sess_get_index,
2105 	.sess_get_initiator_sid		= NULL,
2106 	.write_pending			= vhost_scsi_write_pending,
2107 	.write_pending_status		= vhost_scsi_write_pending_status,
2108 	.set_default_node_attributes	= vhost_scsi_set_default_node_attrs,
2109 	.get_cmd_state			= vhost_scsi_get_cmd_state,
2110 	.queue_data_in			= vhost_scsi_queue_data_in,
2111 	.queue_status			= vhost_scsi_queue_status,
2112 	.queue_tm_rsp			= vhost_scsi_queue_tm_rsp,
2113 	.aborted_task			= vhost_scsi_aborted_task,
2114 	/*
2115 	 * Setup callers for generic logic in target_core_fabric_configfs.c
2116 	 */
2117 	.fabric_make_wwn		= vhost_scsi_make_tport,
2118 	.fabric_drop_wwn		= vhost_scsi_drop_tport,
2119 	.fabric_make_tpg		= vhost_scsi_make_tpg,
2120 	.fabric_drop_tpg		= vhost_scsi_drop_tpg,
2121 	.fabric_post_link		= vhost_scsi_port_link,
2122 	.fabric_pre_unlink		= vhost_scsi_port_unlink,
2123 
2124 	.tfc_wwn_attrs			= vhost_scsi_wwn_attrs,
2125 	.tfc_tpg_base_attrs		= vhost_scsi_tpg_attrs,
2126 	.tfc_tpg_attrib_attrs		= vhost_scsi_tpg_attrib_attrs,
2127 };
2128 
2129 static int __init vhost_scsi_init(void)
2130 {
2131 	int ret = -ENOMEM;
2132 
2133 	pr_debug("TCM_VHOST fabric module %s on %s/%s"
2134 		" on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2135 		utsname()->machine);
2136 
2137 	/*
2138 	 * Use our own dedicated workqueue for submitting I/O into
2139 	 * target core to avoid contention within system_wq.
2140 	 */
2141 	vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2142 	if (!vhost_scsi_workqueue)
2143 		goto out;
2144 
2145 	ret = vhost_scsi_register();
2146 	if (ret < 0)
2147 		goto out_destroy_workqueue;
2148 
2149 	ret = target_register_template(&vhost_scsi_ops);
2150 	if (ret < 0)
2151 		goto out_vhost_scsi_deregister;
2152 
2153 	return 0;
2154 
2155 out_vhost_scsi_deregister:
2156 	vhost_scsi_deregister();
2157 out_destroy_workqueue:
2158 	destroy_workqueue(vhost_scsi_workqueue);
2159 out:
2160 	return ret;
2161 };
2162 
2163 static void vhost_scsi_exit(void)
2164 {
2165 	target_unregister_template(&vhost_scsi_ops);
2166 	vhost_scsi_deregister();
2167 	destroy_workqueue(vhost_scsi_workqueue);
2168 };
2169 
2170 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2171 MODULE_ALIAS("tcm_vhost");
2172 MODULE_LICENSE("GPL");
2173 module_init(vhost_scsi_init);
2174 module_exit(vhost_scsi_exit);
2175