xref: /openbmc/linux/drivers/vhost/scsi.c (revision bc5aa3a0)
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 out, in;
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 		ret = copy_from_iter(req, req_size, &out_iter);
926 		if (unlikely(ret != req_size)) {
927 			vq_err(vq, "Faulted on copy_from_iter\n");
928 			vhost_scsi_send_bad_target(vs, vq, head, out);
929 			continue;
930 		}
931 		/* virtio-scsi spec requires byte 0 of the lun to be 1 */
932 		if (unlikely(*lunp != 1)) {
933 			vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
934 			vhost_scsi_send_bad_target(vs, vq, head, out);
935 			continue;
936 		}
937 
938 		tpg = ACCESS_ONCE(vs_tpg[*target]);
939 		if (unlikely(!tpg)) {
940 			/* Target does not exist, fail the request */
941 			vhost_scsi_send_bad_target(vs, vq, head, out);
942 			continue;
943 		}
944 		/*
945 		 * Determine data_direction by calculating the total outgoing
946 		 * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
947 		 * response headers respectively.
948 		 *
949 		 * For DMA_TO_DEVICE this is out_iter, which is already pointing
950 		 * to the right place.
951 		 *
952 		 * For DMA_FROM_DEVICE, the iovec will be just past the end
953 		 * of the virtio-scsi response header in either the same
954 		 * or immediately following iovec.
955 		 *
956 		 * Any associated T10_PI bytes for the outgoing / incoming
957 		 * payloads are included in calculation of exp_data_len here.
958 		 */
959 		prot_bytes = 0;
960 
961 		if (out_size > req_size) {
962 			data_direction = DMA_TO_DEVICE;
963 			exp_data_len = out_size - req_size;
964 			data_iter = out_iter;
965 		} else if (in_size > rsp_size) {
966 			data_direction = DMA_FROM_DEVICE;
967 			exp_data_len = in_size - rsp_size;
968 
969 			iov_iter_init(&in_iter, READ, &vq->iov[out], in,
970 				      rsp_size + exp_data_len);
971 			iov_iter_advance(&in_iter, rsp_size);
972 			data_iter = in_iter;
973 		} else {
974 			data_direction = DMA_NONE;
975 			exp_data_len = 0;
976 		}
977 		/*
978 		 * If T10_PI header + payload is present, setup prot_iter values
979 		 * and recalculate data_iter for vhost_scsi_mapal() mapping to
980 		 * host scatterlists via get_user_pages_fast().
981 		 */
982 		if (t10_pi) {
983 			if (v_req_pi.pi_bytesout) {
984 				if (data_direction != DMA_TO_DEVICE) {
985 					vq_err(vq, "Received non zero pi_bytesout,"
986 						" but wrong data_direction\n");
987 					vhost_scsi_send_bad_target(vs, vq, head, out);
988 					continue;
989 				}
990 				prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
991 			} else if (v_req_pi.pi_bytesin) {
992 				if (data_direction != DMA_FROM_DEVICE) {
993 					vq_err(vq, "Received non zero pi_bytesin,"
994 						" but wrong data_direction\n");
995 					vhost_scsi_send_bad_target(vs, vq, head, out);
996 					continue;
997 				}
998 				prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
999 			}
1000 			/*
1001 			 * Set prot_iter to data_iter, and advance past any
1002 			 * preceeding prot_bytes that may be present.
1003 			 *
1004 			 * Also fix up the exp_data_len to reflect only the
1005 			 * actual data payload length.
1006 			 */
1007 			if (prot_bytes) {
1008 				exp_data_len -= prot_bytes;
1009 				prot_iter = data_iter;
1010 				iov_iter_advance(&data_iter, prot_bytes);
1011 			}
1012 			tag = vhost64_to_cpu(vq, v_req_pi.tag);
1013 			task_attr = v_req_pi.task_attr;
1014 			cdb = &v_req_pi.cdb[0];
1015 			lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1016 		} else {
1017 			tag = vhost64_to_cpu(vq, v_req.tag);
1018 			task_attr = v_req.task_attr;
1019 			cdb = &v_req.cdb[0];
1020 			lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1021 		}
1022 		/*
1023 		 * Check that the received CDB size does not exceeded our
1024 		 * hardcoded max for vhost-scsi, then get a pre-allocated
1025 		 * cmd descriptor for the new virtio-scsi tag.
1026 		 *
1027 		 * TODO what if cdb was too small for varlen cdb header?
1028 		 */
1029 		if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1030 			vq_err(vq, "Received SCSI CDB with command_size: %d that"
1031 				" exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1032 				scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1033 			vhost_scsi_send_bad_target(vs, vq, head, out);
1034 			continue;
1035 		}
1036 		cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1037 					 exp_data_len + prot_bytes,
1038 					 data_direction);
1039 		if (IS_ERR(cmd)) {
1040 			vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1041 			       PTR_ERR(cmd));
1042 			vhost_scsi_send_bad_target(vs, vq, head, out);
1043 			continue;
1044 		}
1045 		cmd->tvc_vhost = vs;
1046 		cmd->tvc_vq = vq;
1047 		cmd->tvc_resp_iov = vq->iov[out];
1048 		cmd->tvc_in_iovs = in;
1049 
1050 		pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1051 			 cmd->tvc_cdb[0], cmd->tvc_lun);
1052 		pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1053 			 " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1054 
1055 		if (data_direction != DMA_NONE) {
1056 			ret = vhost_scsi_mapal(cmd,
1057 					       prot_bytes, &prot_iter,
1058 					       exp_data_len, &data_iter);
1059 			if (unlikely(ret)) {
1060 				vq_err(vq, "Failed to map iov to sgl\n");
1061 				vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1062 				vhost_scsi_send_bad_target(vs, vq, head, out);
1063 				continue;
1064 			}
1065 		}
1066 		/*
1067 		 * Save the descriptor from vhost_get_vq_desc() to be used to
1068 		 * complete the virtio-scsi request in TCM callback context via
1069 		 * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1070 		 */
1071 		cmd->tvc_vq_desc = head;
1072 		/*
1073 		 * Dispatch cmd descriptor for cmwq execution in process
1074 		 * context provided by vhost_scsi_workqueue.  This also ensures
1075 		 * cmd is executed on the same kworker CPU as this vhost
1076 		 * thread to gain positive L2 cache locality effects.
1077 		 */
1078 		INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1079 		queue_work(vhost_scsi_workqueue, &cmd->work);
1080 	}
1081 out:
1082 	mutex_unlock(&vq->mutex);
1083 }
1084 
1085 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1086 {
1087 	pr_debug("%s: The handling func for control queue.\n", __func__);
1088 }
1089 
1090 static void
1091 vhost_scsi_send_evt(struct vhost_scsi *vs,
1092 		   struct vhost_scsi_tpg *tpg,
1093 		   struct se_lun *lun,
1094 		   u32 event,
1095 		   u32 reason)
1096 {
1097 	struct vhost_scsi_evt *evt;
1098 
1099 	evt = vhost_scsi_allocate_evt(vs, event, reason);
1100 	if (!evt)
1101 		return;
1102 
1103 	if (tpg && lun) {
1104 		/* TODO: share lun setup code with virtio-scsi.ko */
1105 		/*
1106 		 * Note: evt->event is zeroed when we allocate it and
1107 		 * lun[4-7] need to be zero according to virtio-scsi spec.
1108 		 */
1109 		evt->event.lun[0] = 0x01;
1110 		evt->event.lun[1] = tpg->tport_tpgt;
1111 		if (lun->unpacked_lun >= 256)
1112 			evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1113 		evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1114 	}
1115 
1116 	llist_add(&evt->list, &vs->vs_event_list);
1117 	vhost_work_queue(&vs->dev, &vs->vs_event_work);
1118 }
1119 
1120 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1121 {
1122 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1123 						poll.work);
1124 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1125 
1126 	mutex_lock(&vq->mutex);
1127 	if (!vq->private_data)
1128 		goto out;
1129 
1130 	if (vs->vs_events_missed)
1131 		vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1132 out:
1133 	mutex_unlock(&vq->mutex);
1134 }
1135 
1136 static void vhost_scsi_handle_kick(struct vhost_work *work)
1137 {
1138 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1139 						poll.work);
1140 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1141 
1142 	vhost_scsi_handle_vq(vs, vq);
1143 }
1144 
1145 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1146 {
1147 	vhost_poll_flush(&vs->vqs[index].vq.poll);
1148 }
1149 
1150 /* Callers must hold dev mutex */
1151 static void vhost_scsi_flush(struct vhost_scsi *vs)
1152 {
1153 	struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1154 	int i;
1155 
1156 	/* Init new inflight and remember the old inflight */
1157 	vhost_scsi_init_inflight(vs, old_inflight);
1158 
1159 	/*
1160 	 * The inflight->kref was initialized to 1. We decrement it here to
1161 	 * indicate the start of the flush operation so that it will reach 0
1162 	 * when all the reqs are finished.
1163 	 */
1164 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1165 		kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1166 
1167 	/* Flush both the vhost poll and vhost work */
1168 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1169 		vhost_scsi_flush_vq(vs, i);
1170 	vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1171 	vhost_work_flush(&vs->dev, &vs->vs_event_work);
1172 
1173 	/* Wait for all reqs issued before the flush to be finished */
1174 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1175 		wait_for_completion(&old_inflight[i]->comp);
1176 }
1177 
1178 /*
1179  * Called from vhost_scsi_ioctl() context to walk the list of available
1180  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1181  *
1182  *  The lock nesting rule is:
1183  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1184  */
1185 static int
1186 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1187 			struct vhost_scsi_target *t)
1188 {
1189 	struct se_portal_group *se_tpg;
1190 	struct vhost_scsi_tport *tv_tport;
1191 	struct vhost_scsi_tpg *tpg;
1192 	struct vhost_scsi_tpg **vs_tpg;
1193 	struct vhost_virtqueue *vq;
1194 	int index, ret, i, len;
1195 	bool match = false;
1196 
1197 	mutex_lock(&vhost_scsi_mutex);
1198 	mutex_lock(&vs->dev.mutex);
1199 
1200 	/* Verify that ring has been setup correctly. */
1201 	for (index = 0; index < vs->dev.nvqs; ++index) {
1202 		/* Verify that ring has been setup correctly. */
1203 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1204 			ret = -EFAULT;
1205 			goto out;
1206 		}
1207 	}
1208 
1209 	len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1210 	vs_tpg = kzalloc(len, GFP_KERNEL);
1211 	if (!vs_tpg) {
1212 		ret = -ENOMEM;
1213 		goto out;
1214 	}
1215 	if (vs->vs_tpg)
1216 		memcpy(vs_tpg, vs->vs_tpg, len);
1217 
1218 	list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1219 		mutex_lock(&tpg->tv_tpg_mutex);
1220 		if (!tpg->tpg_nexus) {
1221 			mutex_unlock(&tpg->tv_tpg_mutex);
1222 			continue;
1223 		}
1224 		if (tpg->tv_tpg_vhost_count != 0) {
1225 			mutex_unlock(&tpg->tv_tpg_mutex);
1226 			continue;
1227 		}
1228 		tv_tport = tpg->tport;
1229 
1230 		if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1231 			if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1232 				kfree(vs_tpg);
1233 				mutex_unlock(&tpg->tv_tpg_mutex);
1234 				ret = -EEXIST;
1235 				goto out;
1236 			}
1237 			/*
1238 			 * In order to ensure individual vhost-scsi configfs
1239 			 * groups cannot be removed while in use by vhost ioctl,
1240 			 * go ahead and take an explicit se_tpg->tpg_group.cg_item
1241 			 * dependency now.
1242 			 */
1243 			se_tpg = &tpg->se_tpg;
1244 			ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1245 			if (ret) {
1246 				pr_warn("configfs_depend_item() failed: %d\n", ret);
1247 				kfree(vs_tpg);
1248 				mutex_unlock(&tpg->tv_tpg_mutex);
1249 				goto out;
1250 			}
1251 			tpg->tv_tpg_vhost_count++;
1252 			tpg->vhost_scsi = vs;
1253 			vs_tpg[tpg->tport_tpgt] = tpg;
1254 			smp_mb__after_atomic();
1255 			match = true;
1256 		}
1257 		mutex_unlock(&tpg->tv_tpg_mutex);
1258 	}
1259 
1260 	if (match) {
1261 		memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1262 		       sizeof(vs->vs_vhost_wwpn));
1263 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1264 			vq = &vs->vqs[i].vq;
1265 			mutex_lock(&vq->mutex);
1266 			vq->private_data = vs_tpg;
1267 			vhost_vq_init_access(vq);
1268 			mutex_unlock(&vq->mutex);
1269 		}
1270 		ret = 0;
1271 	} else {
1272 		ret = -EEXIST;
1273 	}
1274 
1275 	/*
1276 	 * Act as synchronize_rcu to make sure access to
1277 	 * old vs->vs_tpg is finished.
1278 	 */
1279 	vhost_scsi_flush(vs);
1280 	kfree(vs->vs_tpg);
1281 	vs->vs_tpg = vs_tpg;
1282 
1283 out:
1284 	mutex_unlock(&vs->dev.mutex);
1285 	mutex_unlock(&vhost_scsi_mutex);
1286 	return ret;
1287 }
1288 
1289 static int
1290 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1291 			  struct vhost_scsi_target *t)
1292 {
1293 	struct se_portal_group *se_tpg;
1294 	struct vhost_scsi_tport *tv_tport;
1295 	struct vhost_scsi_tpg *tpg;
1296 	struct vhost_virtqueue *vq;
1297 	bool match = false;
1298 	int index, ret, i;
1299 	u8 target;
1300 
1301 	mutex_lock(&vhost_scsi_mutex);
1302 	mutex_lock(&vs->dev.mutex);
1303 	/* Verify that ring has been setup correctly. */
1304 	for (index = 0; index < vs->dev.nvqs; ++index) {
1305 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1306 			ret = -EFAULT;
1307 			goto err_dev;
1308 		}
1309 	}
1310 
1311 	if (!vs->vs_tpg) {
1312 		ret = 0;
1313 		goto err_dev;
1314 	}
1315 
1316 	for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1317 		target = i;
1318 		tpg = vs->vs_tpg[target];
1319 		if (!tpg)
1320 			continue;
1321 
1322 		mutex_lock(&tpg->tv_tpg_mutex);
1323 		tv_tport = tpg->tport;
1324 		if (!tv_tport) {
1325 			ret = -ENODEV;
1326 			goto err_tpg;
1327 		}
1328 
1329 		if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1330 			pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1331 				" does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1332 				tv_tport->tport_name, tpg->tport_tpgt,
1333 				t->vhost_wwpn, t->vhost_tpgt);
1334 			ret = -EINVAL;
1335 			goto err_tpg;
1336 		}
1337 		tpg->tv_tpg_vhost_count--;
1338 		tpg->vhost_scsi = NULL;
1339 		vs->vs_tpg[target] = NULL;
1340 		match = true;
1341 		mutex_unlock(&tpg->tv_tpg_mutex);
1342 		/*
1343 		 * Release se_tpg->tpg_group.cg_item configfs dependency now
1344 		 * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1345 		 */
1346 		se_tpg = &tpg->se_tpg;
1347 		target_undepend_item(&se_tpg->tpg_group.cg_item);
1348 	}
1349 	if (match) {
1350 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1351 			vq = &vs->vqs[i].vq;
1352 			mutex_lock(&vq->mutex);
1353 			vq->private_data = NULL;
1354 			mutex_unlock(&vq->mutex);
1355 		}
1356 	}
1357 	/*
1358 	 * Act as synchronize_rcu to make sure access to
1359 	 * old vs->vs_tpg is finished.
1360 	 */
1361 	vhost_scsi_flush(vs);
1362 	kfree(vs->vs_tpg);
1363 	vs->vs_tpg = NULL;
1364 	WARN_ON(vs->vs_events_nr);
1365 	mutex_unlock(&vs->dev.mutex);
1366 	mutex_unlock(&vhost_scsi_mutex);
1367 	return 0;
1368 
1369 err_tpg:
1370 	mutex_unlock(&tpg->tv_tpg_mutex);
1371 err_dev:
1372 	mutex_unlock(&vs->dev.mutex);
1373 	mutex_unlock(&vhost_scsi_mutex);
1374 	return ret;
1375 }
1376 
1377 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1378 {
1379 	struct vhost_virtqueue *vq;
1380 	int i;
1381 
1382 	if (features & ~VHOST_SCSI_FEATURES)
1383 		return -EOPNOTSUPP;
1384 
1385 	mutex_lock(&vs->dev.mutex);
1386 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
1387 	    !vhost_log_access_ok(&vs->dev)) {
1388 		mutex_unlock(&vs->dev.mutex);
1389 		return -EFAULT;
1390 	}
1391 
1392 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1393 		vq = &vs->vqs[i].vq;
1394 		mutex_lock(&vq->mutex);
1395 		vq->acked_features = features;
1396 		mutex_unlock(&vq->mutex);
1397 	}
1398 	mutex_unlock(&vs->dev.mutex);
1399 	return 0;
1400 }
1401 
1402 static int vhost_scsi_open(struct inode *inode, struct file *f)
1403 {
1404 	struct vhost_scsi *vs;
1405 	struct vhost_virtqueue **vqs;
1406 	int r = -ENOMEM, i;
1407 
1408 	vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
1409 	if (!vs) {
1410 		vs = vzalloc(sizeof(*vs));
1411 		if (!vs)
1412 			goto err_vs;
1413 	}
1414 
1415 	vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1416 	if (!vqs)
1417 		goto err_vqs;
1418 
1419 	vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1420 	vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1421 
1422 	vs->vs_events_nr = 0;
1423 	vs->vs_events_missed = false;
1424 
1425 	vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1426 	vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1427 	vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1428 	vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1429 	for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1430 		vqs[i] = &vs->vqs[i].vq;
1431 		vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1432 	}
1433 	vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1434 
1435 	vhost_scsi_init_inflight(vs, NULL);
1436 
1437 	f->private_data = vs;
1438 	return 0;
1439 
1440 err_vqs:
1441 	kvfree(vs);
1442 err_vs:
1443 	return r;
1444 }
1445 
1446 static int vhost_scsi_release(struct inode *inode, struct file *f)
1447 {
1448 	struct vhost_scsi *vs = f->private_data;
1449 	struct vhost_scsi_target t;
1450 
1451 	mutex_lock(&vs->dev.mutex);
1452 	memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1453 	mutex_unlock(&vs->dev.mutex);
1454 	vhost_scsi_clear_endpoint(vs, &t);
1455 	vhost_dev_stop(&vs->dev);
1456 	vhost_dev_cleanup(&vs->dev, false);
1457 	/* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1458 	vhost_scsi_flush(vs);
1459 	kfree(vs->dev.vqs);
1460 	kvfree(vs);
1461 	return 0;
1462 }
1463 
1464 static long
1465 vhost_scsi_ioctl(struct file *f,
1466 		 unsigned int ioctl,
1467 		 unsigned long arg)
1468 {
1469 	struct vhost_scsi *vs = f->private_data;
1470 	struct vhost_scsi_target backend;
1471 	void __user *argp = (void __user *)arg;
1472 	u64 __user *featurep = argp;
1473 	u32 __user *eventsp = argp;
1474 	u32 events_missed;
1475 	u64 features;
1476 	int r, abi_version = VHOST_SCSI_ABI_VERSION;
1477 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1478 
1479 	switch (ioctl) {
1480 	case VHOST_SCSI_SET_ENDPOINT:
1481 		if (copy_from_user(&backend, argp, sizeof backend))
1482 			return -EFAULT;
1483 		if (backend.reserved != 0)
1484 			return -EOPNOTSUPP;
1485 
1486 		return vhost_scsi_set_endpoint(vs, &backend);
1487 	case VHOST_SCSI_CLEAR_ENDPOINT:
1488 		if (copy_from_user(&backend, argp, sizeof backend))
1489 			return -EFAULT;
1490 		if (backend.reserved != 0)
1491 			return -EOPNOTSUPP;
1492 
1493 		return vhost_scsi_clear_endpoint(vs, &backend);
1494 	case VHOST_SCSI_GET_ABI_VERSION:
1495 		if (copy_to_user(argp, &abi_version, sizeof abi_version))
1496 			return -EFAULT;
1497 		return 0;
1498 	case VHOST_SCSI_SET_EVENTS_MISSED:
1499 		if (get_user(events_missed, eventsp))
1500 			return -EFAULT;
1501 		mutex_lock(&vq->mutex);
1502 		vs->vs_events_missed = events_missed;
1503 		mutex_unlock(&vq->mutex);
1504 		return 0;
1505 	case VHOST_SCSI_GET_EVENTS_MISSED:
1506 		mutex_lock(&vq->mutex);
1507 		events_missed = vs->vs_events_missed;
1508 		mutex_unlock(&vq->mutex);
1509 		if (put_user(events_missed, eventsp))
1510 			return -EFAULT;
1511 		return 0;
1512 	case VHOST_GET_FEATURES:
1513 		features = VHOST_SCSI_FEATURES;
1514 		if (copy_to_user(featurep, &features, sizeof features))
1515 			return -EFAULT;
1516 		return 0;
1517 	case VHOST_SET_FEATURES:
1518 		if (copy_from_user(&features, featurep, sizeof features))
1519 			return -EFAULT;
1520 		return vhost_scsi_set_features(vs, features);
1521 	default:
1522 		mutex_lock(&vs->dev.mutex);
1523 		r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1524 		/* TODO: flush backend after dev ioctl. */
1525 		if (r == -ENOIOCTLCMD)
1526 			r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1527 		mutex_unlock(&vs->dev.mutex);
1528 		return r;
1529 	}
1530 }
1531 
1532 #ifdef CONFIG_COMPAT
1533 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1534 				unsigned long arg)
1535 {
1536 	return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1537 }
1538 #endif
1539 
1540 static const struct file_operations vhost_scsi_fops = {
1541 	.owner          = THIS_MODULE,
1542 	.release        = vhost_scsi_release,
1543 	.unlocked_ioctl = vhost_scsi_ioctl,
1544 #ifdef CONFIG_COMPAT
1545 	.compat_ioctl	= vhost_scsi_compat_ioctl,
1546 #endif
1547 	.open           = vhost_scsi_open,
1548 	.llseek		= noop_llseek,
1549 };
1550 
1551 static struct miscdevice vhost_scsi_misc = {
1552 	MISC_DYNAMIC_MINOR,
1553 	"vhost-scsi",
1554 	&vhost_scsi_fops,
1555 };
1556 
1557 static int __init vhost_scsi_register(void)
1558 {
1559 	return misc_register(&vhost_scsi_misc);
1560 }
1561 
1562 static void vhost_scsi_deregister(void)
1563 {
1564 	misc_deregister(&vhost_scsi_misc);
1565 }
1566 
1567 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1568 {
1569 	switch (tport->tport_proto_id) {
1570 	case SCSI_PROTOCOL_SAS:
1571 		return "SAS";
1572 	case SCSI_PROTOCOL_FCP:
1573 		return "FCP";
1574 	case SCSI_PROTOCOL_ISCSI:
1575 		return "iSCSI";
1576 	default:
1577 		break;
1578 	}
1579 
1580 	return "Unknown";
1581 }
1582 
1583 static void
1584 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1585 		  struct se_lun *lun, bool plug)
1586 {
1587 
1588 	struct vhost_scsi *vs = tpg->vhost_scsi;
1589 	struct vhost_virtqueue *vq;
1590 	u32 reason;
1591 
1592 	if (!vs)
1593 		return;
1594 
1595 	mutex_lock(&vs->dev.mutex);
1596 
1597 	if (plug)
1598 		reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1599 	else
1600 		reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1601 
1602 	vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1603 	mutex_lock(&vq->mutex);
1604 	if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1605 		vhost_scsi_send_evt(vs, tpg, lun,
1606 				   VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1607 	mutex_unlock(&vq->mutex);
1608 	mutex_unlock(&vs->dev.mutex);
1609 }
1610 
1611 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1612 {
1613 	vhost_scsi_do_plug(tpg, lun, true);
1614 }
1615 
1616 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1617 {
1618 	vhost_scsi_do_plug(tpg, lun, false);
1619 }
1620 
1621 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1622 			       struct se_lun *lun)
1623 {
1624 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1625 				struct vhost_scsi_tpg, se_tpg);
1626 
1627 	mutex_lock(&vhost_scsi_mutex);
1628 
1629 	mutex_lock(&tpg->tv_tpg_mutex);
1630 	tpg->tv_tpg_port_count++;
1631 	mutex_unlock(&tpg->tv_tpg_mutex);
1632 
1633 	vhost_scsi_hotplug(tpg, lun);
1634 
1635 	mutex_unlock(&vhost_scsi_mutex);
1636 
1637 	return 0;
1638 }
1639 
1640 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1641 				  struct se_lun *lun)
1642 {
1643 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1644 				struct vhost_scsi_tpg, se_tpg);
1645 
1646 	mutex_lock(&vhost_scsi_mutex);
1647 
1648 	mutex_lock(&tpg->tv_tpg_mutex);
1649 	tpg->tv_tpg_port_count--;
1650 	mutex_unlock(&tpg->tv_tpg_mutex);
1651 
1652 	vhost_scsi_hotunplug(tpg, lun);
1653 
1654 	mutex_unlock(&vhost_scsi_mutex);
1655 }
1656 
1657 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1658 {
1659 	struct vhost_scsi_cmd *tv_cmd;
1660 	unsigned int i;
1661 
1662 	if (!se_sess->sess_cmd_map)
1663 		return;
1664 
1665 	for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1666 		tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1667 
1668 		kfree(tv_cmd->tvc_sgl);
1669 		kfree(tv_cmd->tvc_prot_sgl);
1670 		kfree(tv_cmd->tvc_upages);
1671 	}
1672 }
1673 
1674 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1675 		struct config_item *item, const char *page, size_t count)
1676 {
1677 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
1678 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1679 				struct vhost_scsi_tpg, se_tpg);
1680 	unsigned long val;
1681 	int ret = kstrtoul(page, 0, &val);
1682 
1683 	if (ret) {
1684 		pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1685 		return ret;
1686 	}
1687 	if (val != 0 && val != 1 && val != 3) {
1688 		pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1689 		return -EINVAL;
1690 	}
1691 	tpg->tv_fabric_prot_type = val;
1692 
1693 	return count;
1694 }
1695 
1696 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1697 		struct config_item *item, char *page)
1698 {
1699 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
1700 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1701 				struct vhost_scsi_tpg, se_tpg);
1702 
1703 	return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1704 }
1705 
1706 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1707 
1708 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1709 	&vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1710 	NULL,
1711 };
1712 
1713 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1714 			       struct se_session *se_sess, void *p)
1715 {
1716 	struct vhost_scsi_cmd *tv_cmd;
1717 	unsigned int i;
1718 
1719 	for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1720 		tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1721 
1722 		tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1723 					VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1724 		if (!tv_cmd->tvc_sgl) {
1725 			pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1726 			goto out;
1727 		}
1728 
1729 		tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1730 				VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1731 		if (!tv_cmd->tvc_upages) {
1732 			pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1733 			goto out;
1734 		}
1735 
1736 		tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1737 				VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1738 		if (!tv_cmd->tvc_prot_sgl) {
1739 			pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1740 			goto out;
1741 		}
1742 	}
1743 	return 0;
1744 out:
1745 	vhost_scsi_free_cmd_map_res(se_sess);
1746 	return -ENOMEM;
1747 }
1748 
1749 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1750 				const char *name)
1751 {
1752 	struct se_portal_group *se_tpg;
1753 	struct vhost_scsi_nexus *tv_nexus;
1754 
1755 	mutex_lock(&tpg->tv_tpg_mutex);
1756 	if (tpg->tpg_nexus) {
1757 		mutex_unlock(&tpg->tv_tpg_mutex);
1758 		pr_debug("tpg->tpg_nexus already exists\n");
1759 		return -EEXIST;
1760 	}
1761 	se_tpg = &tpg->se_tpg;
1762 
1763 	tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1764 	if (!tv_nexus) {
1765 		mutex_unlock(&tpg->tv_tpg_mutex);
1766 		pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1767 		return -ENOMEM;
1768 	}
1769 	/*
1770 	 * Since we are running in 'demo mode' this call with generate a
1771 	 * struct se_node_acl for the vhost_scsi struct se_portal_group with
1772 	 * the SCSI Initiator port name of the passed configfs group 'name'.
1773 	 */
1774 	tv_nexus->tvn_se_sess = target_alloc_session(&tpg->se_tpg,
1775 					VHOST_SCSI_DEFAULT_TAGS,
1776 					sizeof(struct vhost_scsi_cmd),
1777 					TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1778 					(unsigned char *)name, tv_nexus,
1779 					vhost_scsi_nexus_cb);
1780 	if (IS_ERR(tv_nexus->tvn_se_sess)) {
1781 		mutex_unlock(&tpg->tv_tpg_mutex);
1782 		kfree(tv_nexus);
1783 		return -ENOMEM;
1784 	}
1785 	tpg->tpg_nexus = tv_nexus;
1786 
1787 	mutex_unlock(&tpg->tv_tpg_mutex);
1788 	return 0;
1789 }
1790 
1791 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1792 {
1793 	struct se_session *se_sess;
1794 	struct vhost_scsi_nexus *tv_nexus;
1795 
1796 	mutex_lock(&tpg->tv_tpg_mutex);
1797 	tv_nexus = tpg->tpg_nexus;
1798 	if (!tv_nexus) {
1799 		mutex_unlock(&tpg->tv_tpg_mutex);
1800 		return -ENODEV;
1801 	}
1802 
1803 	se_sess = tv_nexus->tvn_se_sess;
1804 	if (!se_sess) {
1805 		mutex_unlock(&tpg->tv_tpg_mutex);
1806 		return -ENODEV;
1807 	}
1808 
1809 	if (tpg->tv_tpg_port_count != 0) {
1810 		mutex_unlock(&tpg->tv_tpg_mutex);
1811 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1812 			" active TPG port count: %d\n",
1813 			tpg->tv_tpg_port_count);
1814 		return -EBUSY;
1815 	}
1816 
1817 	if (tpg->tv_tpg_vhost_count != 0) {
1818 		mutex_unlock(&tpg->tv_tpg_mutex);
1819 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1820 			" active TPG vhost count: %d\n",
1821 			tpg->tv_tpg_vhost_count);
1822 		return -EBUSY;
1823 	}
1824 
1825 	pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1826 		" %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1827 		tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1828 
1829 	vhost_scsi_free_cmd_map_res(se_sess);
1830 	/*
1831 	 * Release the SCSI I_T Nexus to the emulated vhost Target Port
1832 	 */
1833 	transport_deregister_session(tv_nexus->tvn_se_sess);
1834 	tpg->tpg_nexus = NULL;
1835 	mutex_unlock(&tpg->tv_tpg_mutex);
1836 
1837 	kfree(tv_nexus);
1838 	return 0;
1839 }
1840 
1841 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1842 {
1843 	struct se_portal_group *se_tpg = to_tpg(item);
1844 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1845 				struct vhost_scsi_tpg, se_tpg);
1846 	struct vhost_scsi_nexus *tv_nexus;
1847 	ssize_t ret;
1848 
1849 	mutex_lock(&tpg->tv_tpg_mutex);
1850 	tv_nexus = tpg->tpg_nexus;
1851 	if (!tv_nexus) {
1852 		mutex_unlock(&tpg->tv_tpg_mutex);
1853 		return -ENODEV;
1854 	}
1855 	ret = snprintf(page, PAGE_SIZE, "%s\n",
1856 			tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1857 	mutex_unlock(&tpg->tv_tpg_mutex);
1858 
1859 	return ret;
1860 }
1861 
1862 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1863 		const char *page, size_t count)
1864 {
1865 	struct se_portal_group *se_tpg = to_tpg(item);
1866 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1867 				struct vhost_scsi_tpg, se_tpg);
1868 	struct vhost_scsi_tport *tport_wwn = tpg->tport;
1869 	unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1870 	int ret;
1871 	/*
1872 	 * Shutdown the active I_T nexus if 'NULL' is passed..
1873 	 */
1874 	if (!strncmp(page, "NULL", 4)) {
1875 		ret = vhost_scsi_drop_nexus(tpg);
1876 		return (!ret) ? count : ret;
1877 	}
1878 	/*
1879 	 * Otherwise make sure the passed virtual Initiator port WWN matches
1880 	 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1881 	 * vhost_scsi_make_nexus().
1882 	 */
1883 	if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1884 		pr_err("Emulated NAA Sas Address: %s, exceeds"
1885 				" max: %d\n", page, VHOST_SCSI_NAMELEN);
1886 		return -EINVAL;
1887 	}
1888 	snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1889 
1890 	ptr = strstr(i_port, "naa.");
1891 	if (ptr) {
1892 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1893 			pr_err("Passed SAS Initiator Port %s does not"
1894 				" match target port protoid: %s\n", i_port,
1895 				vhost_scsi_dump_proto_id(tport_wwn));
1896 			return -EINVAL;
1897 		}
1898 		port_ptr = &i_port[0];
1899 		goto check_newline;
1900 	}
1901 	ptr = strstr(i_port, "fc.");
1902 	if (ptr) {
1903 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1904 			pr_err("Passed FCP Initiator Port %s does not"
1905 				" match target port protoid: %s\n", i_port,
1906 				vhost_scsi_dump_proto_id(tport_wwn));
1907 			return -EINVAL;
1908 		}
1909 		port_ptr = &i_port[3]; /* Skip over "fc." */
1910 		goto check_newline;
1911 	}
1912 	ptr = strstr(i_port, "iqn.");
1913 	if (ptr) {
1914 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1915 			pr_err("Passed iSCSI Initiator Port %s does not"
1916 				" match target port protoid: %s\n", i_port,
1917 				vhost_scsi_dump_proto_id(tport_wwn));
1918 			return -EINVAL;
1919 		}
1920 		port_ptr = &i_port[0];
1921 		goto check_newline;
1922 	}
1923 	pr_err("Unable to locate prefix for emulated Initiator Port:"
1924 			" %s\n", i_port);
1925 	return -EINVAL;
1926 	/*
1927 	 * Clear any trailing newline for the NAA WWN
1928 	 */
1929 check_newline:
1930 	if (i_port[strlen(i_port)-1] == '\n')
1931 		i_port[strlen(i_port)-1] = '\0';
1932 
1933 	ret = vhost_scsi_make_nexus(tpg, port_ptr);
1934 	if (ret < 0)
1935 		return ret;
1936 
1937 	return count;
1938 }
1939 
1940 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1941 
1942 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1943 	&vhost_scsi_tpg_attr_nexus,
1944 	NULL,
1945 };
1946 
1947 static struct se_portal_group *
1948 vhost_scsi_make_tpg(struct se_wwn *wwn,
1949 		   struct config_group *group,
1950 		   const char *name)
1951 {
1952 	struct vhost_scsi_tport *tport = container_of(wwn,
1953 			struct vhost_scsi_tport, tport_wwn);
1954 
1955 	struct vhost_scsi_tpg *tpg;
1956 	u16 tpgt;
1957 	int ret;
1958 
1959 	if (strstr(name, "tpgt_") != name)
1960 		return ERR_PTR(-EINVAL);
1961 	if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1962 		return ERR_PTR(-EINVAL);
1963 
1964 	tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1965 	if (!tpg) {
1966 		pr_err("Unable to allocate struct vhost_scsi_tpg");
1967 		return ERR_PTR(-ENOMEM);
1968 	}
1969 	mutex_init(&tpg->tv_tpg_mutex);
1970 	INIT_LIST_HEAD(&tpg->tv_tpg_list);
1971 	tpg->tport = tport;
1972 	tpg->tport_tpgt = tpgt;
1973 
1974 	ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1975 	if (ret < 0) {
1976 		kfree(tpg);
1977 		return NULL;
1978 	}
1979 	mutex_lock(&vhost_scsi_mutex);
1980 	list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1981 	mutex_unlock(&vhost_scsi_mutex);
1982 
1983 	return &tpg->se_tpg;
1984 }
1985 
1986 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1987 {
1988 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1989 				struct vhost_scsi_tpg, se_tpg);
1990 
1991 	mutex_lock(&vhost_scsi_mutex);
1992 	list_del(&tpg->tv_tpg_list);
1993 	mutex_unlock(&vhost_scsi_mutex);
1994 	/*
1995 	 * Release the virtual I_T Nexus for this vhost TPG
1996 	 */
1997 	vhost_scsi_drop_nexus(tpg);
1998 	/*
1999 	 * Deregister the se_tpg from TCM..
2000 	 */
2001 	core_tpg_deregister(se_tpg);
2002 	kfree(tpg);
2003 }
2004 
2005 static struct se_wwn *
2006 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2007 		     struct config_group *group,
2008 		     const char *name)
2009 {
2010 	struct vhost_scsi_tport *tport;
2011 	char *ptr;
2012 	u64 wwpn = 0;
2013 	int off = 0;
2014 
2015 	/* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2016 		return ERR_PTR(-EINVAL); */
2017 
2018 	tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2019 	if (!tport) {
2020 		pr_err("Unable to allocate struct vhost_scsi_tport");
2021 		return ERR_PTR(-ENOMEM);
2022 	}
2023 	tport->tport_wwpn = wwpn;
2024 	/*
2025 	 * Determine the emulated Protocol Identifier and Target Port Name
2026 	 * based on the incoming configfs directory name.
2027 	 */
2028 	ptr = strstr(name, "naa.");
2029 	if (ptr) {
2030 		tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2031 		goto check_len;
2032 	}
2033 	ptr = strstr(name, "fc.");
2034 	if (ptr) {
2035 		tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2036 		off = 3; /* Skip over "fc." */
2037 		goto check_len;
2038 	}
2039 	ptr = strstr(name, "iqn.");
2040 	if (ptr) {
2041 		tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2042 		goto check_len;
2043 	}
2044 
2045 	pr_err("Unable to locate prefix for emulated Target Port:"
2046 			" %s\n", name);
2047 	kfree(tport);
2048 	return ERR_PTR(-EINVAL);
2049 
2050 check_len:
2051 	if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2052 		pr_err("Emulated %s Address: %s, exceeds"
2053 			" max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2054 			VHOST_SCSI_NAMELEN);
2055 		kfree(tport);
2056 		return ERR_PTR(-EINVAL);
2057 	}
2058 	snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2059 
2060 	pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2061 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2062 
2063 	return &tport->tport_wwn;
2064 }
2065 
2066 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2067 {
2068 	struct vhost_scsi_tport *tport = container_of(wwn,
2069 				struct vhost_scsi_tport, tport_wwn);
2070 
2071 	pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2072 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2073 		tport->tport_name);
2074 
2075 	kfree(tport);
2076 }
2077 
2078 static ssize_t
2079 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2080 {
2081 	return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2082 		"on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2083 		utsname()->machine);
2084 }
2085 
2086 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2087 
2088 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2089 	&vhost_scsi_wwn_attr_version,
2090 	NULL,
2091 };
2092 
2093 static struct target_core_fabric_ops vhost_scsi_ops = {
2094 	.module				= THIS_MODULE,
2095 	.name				= "vhost",
2096 	.get_fabric_name		= vhost_scsi_get_fabric_name,
2097 	.tpg_get_wwn			= vhost_scsi_get_fabric_wwn,
2098 	.tpg_get_tag			= vhost_scsi_get_tpgt,
2099 	.tpg_check_demo_mode		= vhost_scsi_check_true,
2100 	.tpg_check_demo_mode_cache	= vhost_scsi_check_true,
2101 	.tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2102 	.tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2103 	.tpg_check_prot_fabric_only	= vhost_scsi_check_prot_fabric_only,
2104 	.tpg_get_inst_index		= vhost_scsi_tpg_get_inst_index,
2105 	.release_cmd			= vhost_scsi_release_cmd,
2106 	.check_stop_free		= vhost_scsi_check_stop_free,
2107 	.sess_get_index			= vhost_scsi_sess_get_index,
2108 	.sess_get_initiator_sid		= NULL,
2109 	.write_pending			= vhost_scsi_write_pending,
2110 	.write_pending_status		= vhost_scsi_write_pending_status,
2111 	.set_default_node_attributes	= vhost_scsi_set_default_node_attrs,
2112 	.get_cmd_state			= vhost_scsi_get_cmd_state,
2113 	.queue_data_in			= vhost_scsi_queue_data_in,
2114 	.queue_status			= vhost_scsi_queue_status,
2115 	.queue_tm_rsp			= vhost_scsi_queue_tm_rsp,
2116 	.aborted_task			= vhost_scsi_aborted_task,
2117 	/*
2118 	 * Setup callers for generic logic in target_core_fabric_configfs.c
2119 	 */
2120 	.fabric_make_wwn		= vhost_scsi_make_tport,
2121 	.fabric_drop_wwn		= vhost_scsi_drop_tport,
2122 	.fabric_make_tpg		= vhost_scsi_make_tpg,
2123 	.fabric_drop_tpg		= vhost_scsi_drop_tpg,
2124 	.fabric_post_link		= vhost_scsi_port_link,
2125 	.fabric_pre_unlink		= vhost_scsi_port_unlink,
2126 
2127 	.tfc_wwn_attrs			= vhost_scsi_wwn_attrs,
2128 	.tfc_tpg_base_attrs		= vhost_scsi_tpg_attrs,
2129 	.tfc_tpg_attrib_attrs		= vhost_scsi_tpg_attrib_attrs,
2130 };
2131 
2132 static int __init vhost_scsi_init(void)
2133 {
2134 	int ret = -ENOMEM;
2135 
2136 	pr_debug("TCM_VHOST fabric module %s on %s/%s"
2137 		" on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2138 		utsname()->machine);
2139 
2140 	/*
2141 	 * Use our own dedicated workqueue for submitting I/O into
2142 	 * target core to avoid contention within system_wq.
2143 	 */
2144 	vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2145 	if (!vhost_scsi_workqueue)
2146 		goto out;
2147 
2148 	ret = vhost_scsi_register();
2149 	if (ret < 0)
2150 		goto out_destroy_workqueue;
2151 
2152 	ret = target_register_template(&vhost_scsi_ops);
2153 	if (ret < 0)
2154 		goto out_vhost_scsi_deregister;
2155 
2156 	return 0;
2157 
2158 out_vhost_scsi_deregister:
2159 	vhost_scsi_deregister();
2160 out_destroy_workqueue:
2161 	destroy_workqueue(vhost_scsi_workqueue);
2162 out:
2163 	return ret;
2164 };
2165 
2166 static void vhost_scsi_exit(void)
2167 {
2168 	target_unregister_template(&vhost_scsi_ops);
2169 	vhost_scsi_deregister();
2170 	destroy_workqueue(vhost_scsi_workqueue);
2171 };
2172 
2173 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2174 MODULE_ALIAS("tcm_vhost");
2175 MODULE_LICENSE("GPL");
2176 module_init(vhost_scsi_init);
2177 module_exit(vhost_scsi_exit);
2178