1 /*******************************************************************************
2 * Vhost kernel TCM fabric driver for virtio SCSI initiators
4 * (C) Copyright 2010-2013 Datera, Inc.
5 * (C) Copyright 2010-2012 IBM Corp.
7 * Licensed to the Linux Foundation under the General Public License (GPL) version 2.
9 * Authors: Nicholas A. Bellinger <nab@daterainc.com>
10 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
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.
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.
22 ****************************************************************************/
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>
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>
52 #define VHOST_SCSI_VERSION "v0.1"
53 #define VHOST_SCSI_NAMELEN 256
54 #define VHOST_SCSI_MAX_CDB_SIZE 32
55 #define VHOST_SCSI_DEFAULT_TAGS 256
56 #define VHOST_SCSI_PREALLOC_SGLS 2048
57 #define VHOST_SCSI_PREALLOC_UPAGES 2048
58 #define VHOST_SCSI_PREALLOC_PROT_SGLS 2048
60 /* Max number of requests before requeueing the job.
61 * Using this limit prevents one virtqueue from starving others with
64 #define VHOST_SCSI_WEIGHT 256
66 struct vhost_scsi_inflight {
67 /* Wait for the flush operation to finish */
68 struct completion comp;
69 /* Refcount for the inflight reqs */
73 struct vhost_scsi_cmd {
74 /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
76 /* virtio-scsi initiator task attribute */
78 /* virtio-scsi response incoming iovecs */
80 /* virtio-scsi initiator data direction */
81 enum dma_data_direction tvc_data_direction;
82 /* Expected data transfer length from virtio-scsi header */
84 /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
86 /* The number of scatterlists associated with this cmd */
88 u32 tvc_prot_sgl_count;
89 /* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
91 /* Pointer to the SGL formatted memory from virtio-scsi */
92 struct scatterlist *tvc_sgl;
93 struct scatterlist *tvc_prot_sgl;
94 struct page **tvc_upages;
95 /* Pointer to response header iovec */
96 struct iovec tvc_resp_iov;
97 /* Pointer to vhost_scsi for our device */
98 struct vhost_scsi *tvc_vhost;
99 /* Pointer to vhost_virtqueue for the cmd */
100 struct vhost_virtqueue *tvc_vq;
101 /* Pointer to vhost nexus memory */
102 struct vhost_scsi_nexus *tvc_nexus;
103 /* The TCM I/O descriptor that is accessed via container_of() */
104 struct se_cmd tvc_se_cmd;
105 /* work item used for cmwq dispatch to vhost_scsi_submission_work() */
106 struct work_struct work;
107 /* Copy of the incoming SCSI command descriptor block (CDB) */
108 unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
109 /* Sense buffer that will be mapped into outgoing status */
110 unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
111 /* Completed commands list, serviced from vhost worker thread */
112 struct llist_node tvc_completion_list;
113 /* Used to track inflight cmd */
114 struct vhost_scsi_inflight *inflight;
117 struct vhost_scsi_nexus {
118 /* Pointer to TCM session for I_T Nexus */
119 struct se_session *tvn_se_sess;
122 struct vhost_scsi_tpg {
123 /* Vhost port target portal group tag for TCM */
125 /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
126 int tv_tpg_port_count;
127 /* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
128 int tv_tpg_vhost_count;
129 /* Used for enabling T10-PI with legacy devices */
130 int tv_fabric_prot_type;
131 /* list for vhost_scsi_list */
132 struct list_head tv_tpg_list;
133 /* Used to protect access for tpg_nexus */
134 struct mutex tv_tpg_mutex;
135 /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
136 struct vhost_scsi_nexus *tpg_nexus;
137 /* Pointer back to vhost_scsi_tport */
138 struct vhost_scsi_tport *tport;
139 /* Returned by vhost_scsi_make_tpg() */
140 struct se_portal_group se_tpg;
141 /* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
142 struct vhost_scsi *vhost_scsi;
145 struct vhost_scsi_tport {
146 /* SCSI protocol the tport is providing */
148 /* Binary World Wide unique Port Name for Vhost Target port */
150 /* ASCII formatted WWPN for Vhost Target port */
151 char tport_name[VHOST_SCSI_NAMELEN];
152 /* Returned by vhost_scsi_make_tport() */
153 struct se_wwn tport_wwn;
156 struct vhost_scsi_evt {
157 /* event to be sent to guest */
158 struct virtio_scsi_event event;
159 /* event list, serviced from vhost worker thread */
160 struct llist_node list;
164 VHOST_SCSI_VQ_CTL = 0,
165 VHOST_SCSI_VQ_EVT = 1,
166 VHOST_SCSI_VQ_IO = 2,
169 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
171 VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
172 (1ULL << VIRTIO_SCSI_F_T10_PI)
175 #define VHOST_SCSI_MAX_TARGET 256
176 #define VHOST_SCSI_MAX_VQ 128
177 #define VHOST_SCSI_MAX_EVENT 128
179 struct vhost_scsi_virtqueue {
180 struct vhost_virtqueue vq;
182 * Reference counting for inflight reqs, used for flush operation. At
183 * each time, one reference tracks new commands submitted, while we
184 * wait for another one to reach 0.
186 struct vhost_scsi_inflight inflights[2];
188 * Indicate current inflight in use, protected by vq->mutex.
189 * Writers must also take dev mutex and flush under it.
195 /* Protected by vhost_scsi->dev.mutex */
196 struct vhost_scsi_tpg **vs_tpg;
197 char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
199 struct vhost_dev dev;
200 struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
202 struct vhost_work vs_completion_work; /* cmd completion work item */
203 struct llist_head vs_completion_list; /* cmd completion queue */
205 struct vhost_work vs_event_work; /* evt injection work item */
206 struct llist_head vs_event_list; /* evt injection queue */
208 bool vs_events_missed; /* any missed events, protected by vq->mutex */
209 int vs_events_nr; /* num of pending events, protected by vq->mutex */
212 static struct workqueue_struct *vhost_scsi_workqueue;
214 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
215 static DEFINE_MUTEX(vhost_scsi_mutex);
216 static LIST_HEAD(vhost_scsi_list);
218 static void vhost_scsi_done_inflight(struct kref *kref)
220 struct vhost_scsi_inflight *inflight;
222 inflight = container_of(kref, struct vhost_scsi_inflight, kref);
223 complete(&inflight->comp);
226 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
227 struct vhost_scsi_inflight *old_inflight[])
229 struct vhost_scsi_inflight *new_inflight;
230 struct vhost_virtqueue *vq;
233 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
236 mutex_lock(&vq->mutex);
238 /* store old infight */
239 idx = vs->vqs[i].inflight_idx;
241 old_inflight[i] = &vs->vqs[i].inflights[idx];
243 /* setup new infight */
244 vs->vqs[i].inflight_idx = idx ^ 1;
245 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
246 kref_init(&new_inflight->kref);
247 init_completion(&new_inflight->comp);
249 mutex_unlock(&vq->mutex);
253 static struct vhost_scsi_inflight *
254 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
256 struct vhost_scsi_inflight *inflight;
257 struct vhost_scsi_virtqueue *svq;
259 svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
260 inflight = &svq->inflights[svq->inflight_idx];
261 kref_get(&inflight->kref);
266 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
268 kref_put(&inflight->kref, vhost_scsi_done_inflight);
271 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
276 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
281 static char *vhost_scsi_get_fabric_name(void)
286 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
288 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
289 struct vhost_scsi_tpg, se_tpg);
290 struct vhost_scsi_tport *tport = tpg->tport;
292 return &tport->tport_name[0];
295 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
297 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
298 struct vhost_scsi_tpg, se_tpg);
299 return tpg->tport_tpgt;
302 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
304 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
305 struct vhost_scsi_tpg, se_tpg);
307 return tpg->tv_fabric_prot_type;
310 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
315 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
317 struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
318 struct vhost_scsi_cmd, tvc_se_cmd);
319 struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
322 if (tv_cmd->tvc_sgl_count) {
323 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
324 put_page(sg_page(&tv_cmd->tvc_sgl[i]));
326 if (tv_cmd->tvc_prot_sgl_count) {
327 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
328 put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
331 vhost_scsi_put_inflight(tv_cmd->inflight);
332 target_free_tag(se_sess, se_cmd);
335 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
340 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
342 /* Go ahead and process the write immediately */
343 target_execute_cmd(se_cmd);
347 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
352 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
357 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
362 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
364 struct vhost_scsi *vs = cmd->tvc_vhost;
366 llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
368 vhost_work_queue(&vs->dev, &vs->vs_completion_work);
371 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
373 struct vhost_scsi_cmd *cmd = container_of(se_cmd,
374 struct vhost_scsi_cmd, tvc_se_cmd);
375 vhost_scsi_complete_cmd(cmd);
379 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
381 struct vhost_scsi_cmd *cmd = container_of(se_cmd,
382 struct vhost_scsi_cmd, tvc_se_cmd);
383 vhost_scsi_complete_cmd(cmd);
387 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
392 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
397 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
403 static struct vhost_scsi_evt *
404 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
405 u32 event, u32 reason)
407 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
408 struct vhost_scsi_evt *evt;
410 if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
411 vs->vs_events_missed = true;
415 evt = kzalloc(sizeof(*evt), GFP_KERNEL);
417 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
418 vs->vs_events_missed = true;
422 evt->event.event = cpu_to_vhost32(vq, event);
423 evt->event.reason = cpu_to_vhost32(vq, reason);
429 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
431 struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
433 /* TODO locking against target/backend threads? */
434 transport_generic_free_cmd(se_cmd, 0);
438 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
440 return target_put_sess_cmd(se_cmd);
444 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
446 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
447 struct virtio_scsi_event *event = &evt->event;
448 struct virtio_scsi_event __user *eventp;
452 if (!vq->private_data) {
453 vs->vs_events_missed = true;
458 vhost_disable_notify(&vs->dev, vq);
459 head = vhost_get_vq_desc(vq, vq->iov,
460 ARRAY_SIZE(vq->iov), &out, &in,
463 vs->vs_events_missed = true;
466 if (head == vq->num) {
467 if (vhost_enable_notify(&vs->dev, vq))
469 vs->vs_events_missed = true;
473 if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
474 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
475 vq->iov[out].iov_len);
476 vs->vs_events_missed = true;
480 if (vs->vs_events_missed) {
481 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
482 vs->vs_events_missed = false;
485 eventp = vq->iov[out].iov_base;
486 ret = __copy_to_user(eventp, event, sizeof(*event));
488 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
490 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
493 static void vhost_scsi_evt_work(struct vhost_work *work)
495 struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
497 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
498 struct vhost_scsi_evt *evt, *t;
499 struct llist_node *llnode;
501 mutex_lock(&vq->mutex);
502 llnode = llist_del_all(&vs->vs_event_list);
503 llist_for_each_entry_safe(evt, t, llnode, list) {
504 vhost_scsi_do_evt_work(vs, evt);
505 vhost_scsi_free_evt(vs, evt);
507 mutex_unlock(&vq->mutex);
510 /* Fill in status and signal that we are done processing this command
512 * This is scheduled in the vhost work queue so we are called with the owner
513 * process mm and can access the vring.
515 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
517 struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
519 DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
520 struct virtio_scsi_cmd_resp v_rsp;
521 struct vhost_scsi_cmd *cmd, *t;
522 struct llist_node *llnode;
523 struct se_cmd *se_cmd;
524 struct iov_iter iov_iter;
527 bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
528 llnode = llist_del_all(&vs->vs_completion_list);
529 llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
530 se_cmd = &cmd->tvc_se_cmd;
532 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
533 cmd, se_cmd->residual_count, se_cmd->scsi_status);
535 memset(&v_rsp, 0, sizeof(v_rsp));
536 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
537 /* TODO is status_qualifier field needed? */
538 v_rsp.status = se_cmd->scsi_status;
539 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
540 se_cmd->scsi_sense_length);
541 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
542 se_cmd->scsi_sense_length);
544 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
545 cmd->tvc_in_iovs, sizeof(v_rsp));
546 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
547 if (likely(ret == sizeof(v_rsp))) {
548 struct vhost_scsi_virtqueue *q;
549 vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
550 q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
552 __set_bit(vq, signal);
554 pr_err("Faulted on virtio_scsi_cmd_resp\n");
556 vhost_scsi_free_cmd(cmd);
560 while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
562 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
565 static struct vhost_scsi_cmd *
566 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
567 unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
568 u32 exp_data_len, int data_direction)
570 struct vhost_scsi_cmd *cmd;
571 struct vhost_scsi_nexus *tv_nexus;
572 struct se_session *se_sess;
573 struct scatterlist *sg, *prot_sg;
577 tv_nexus = tpg->tpg_nexus;
579 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
580 return ERR_PTR(-EIO);
582 se_sess = tv_nexus->tvn_se_sess;
584 tag = sbitmap_queue_get(&se_sess->sess_tag_pool, &cpu);
586 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
587 return ERR_PTR(-ENOMEM);
590 cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
592 prot_sg = cmd->tvc_prot_sgl;
593 pages = cmd->tvc_upages;
594 memset(cmd, 0, sizeof(*cmd));
596 cmd->tvc_prot_sgl = prot_sg;
597 cmd->tvc_upages = pages;
598 cmd->tvc_se_cmd.map_tag = tag;
599 cmd->tvc_se_cmd.map_cpu = cpu;
600 cmd->tvc_tag = scsi_tag;
602 cmd->tvc_task_attr = task_attr;
603 cmd->tvc_exp_data_len = exp_data_len;
604 cmd->tvc_data_direction = data_direction;
605 cmd->tvc_nexus = tv_nexus;
606 cmd->inflight = vhost_scsi_get_inflight(vq);
608 memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
614 * Map a user memory range into a scatterlist
616 * Returns the number of scatterlist entries used or -errno on error.
619 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
620 struct iov_iter *iter,
621 struct scatterlist *sgl,
624 struct page **pages = cmd->tvc_upages;
625 struct scatterlist *sg = sgl;
628 unsigned int npages = 0;
630 bytes = iov_iter_get_pages(iter, pages, LONG_MAX,
631 VHOST_SCSI_PREALLOC_UPAGES, &offset);
632 /* No pages were pinned */
634 return bytes < 0 ? bytes : -EFAULT;
636 iov_iter_advance(iter, bytes);
639 unsigned n = min_t(unsigned, PAGE_SIZE - offset, bytes);
640 sg_set_page(sg++, pages[npages++], n, offset);
648 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
652 if (!iter || !iter->iov) {
653 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
654 " present\n", __func__, bytes);
658 sgl_count = iov_iter_npages(iter, 0xffff);
659 if (sgl_count > max_sgls) {
660 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
661 " max_sgls: %d\n", __func__, sgl_count, max_sgls);
668 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
669 struct iov_iter *iter,
670 struct scatterlist *sg, int sg_count)
672 struct scatterlist *p = sg;
675 while (iov_iter_count(iter)) {
676 ret = vhost_scsi_map_to_sgl(cmd, iter, sg, write);
679 struct page *page = sg_page(p++);
691 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
692 size_t prot_bytes, struct iov_iter *prot_iter,
693 size_t data_bytes, struct iov_iter *data_iter)
696 bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
699 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
700 VHOST_SCSI_PREALLOC_PROT_SGLS);
704 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
705 cmd->tvc_prot_sgl_count = sgl_count;
706 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
707 cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
709 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
711 cmd->tvc_prot_sgl_count);
713 cmd->tvc_prot_sgl_count = 0;
717 sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
718 VHOST_SCSI_PREALLOC_SGLS);
722 sg_init_table(cmd->tvc_sgl, sgl_count);
723 cmd->tvc_sgl_count = sgl_count;
724 pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
725 cmd->tvc_sgl, cmd->tvc_sgl_count);
727 ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
728 cmd->tvc_sgl, cmd->tvc_sgl_count);
730 cmd->tvc_sgl_count = 0;
736 static int vhost_scsi_to_tcm_attr(int attr)
739 case VIRTIO_SCSI_S_SIMPLE:
740 return TCM_SIMPLE_TAG;
741 case VIRTIO_SCSI_S_ORDERED:
742 return TCM_ORDERED_TAG;
743 case VIRTIO_SCSI_S_HEAD:
745 case VIRTIO_SCSI_S_ACA:
750 return TCM_SIMPLE_TAG;
753 static void vhost_scsi_submission_work(struct work_struct *work)
755 struct vhost_scsi_cmd *cmd =
756 container_of(work, struct vhost_scsi_cmd, work);
757 struct vhost_scsi_nexus *tv_nexus;
758 struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
759 struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
762 /* FIXME: BIDI operation */
763 if (cmd->tvc_sgl_count) {
764 sg_ptr = cmd->tvc_sgl;
766 if (cmd->tvc_prot_sgl_count)
767 sg_prot_ptr = cmd->tvc_prot_sgl;
769 se_cmd->prot_pto = true;
773 tv_nexus = cmd->tvc_nexus;
776 rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
777 cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
778 cmd->tvc_lun, cmd->tvc_exp_data_len,
779 vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
780 cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
781 sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
782 cmd->tvc_prot_sgl_count);
784 transport_send_check_condition_and_sense(se_cmd,
785 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
786 transport_generic_free_cmd(se_cmd, 0);
791 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
792 struct vhost_virtqueue *vq,
793 int head, unsigned out)
795 struct virtio_scsi_cmd_resp __user *resp;
796 struct virtio_scsi_cmd_resp rsp;
799 memset(&rsp, 0, sizeof(rsp));
800 rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
801 resp = vq->iov[out].iov_base;
802 ret = __copy_to_user(resp, &rsp, sizeof(rsp));
804 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
806 pr_err("Faulted on virtio_scsi_cmd_resp\n");
810 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
812 struct vhost_scsi_tpg **vs_tpg, *tpg;
813 struct virtio_scsi_cmd_req v_req;
814 struct virtio_scsi_cmd_req_pi v_req_pi;
815 struct vhost_scsi_cmd *cmd;
816 struct iov_iter out_iter, in_iter, prot_iter, data_iter;
818 u32 exp_data_len, data_direction;
819 unsigned int out = 0, in = 0;
820 int head, ret, prot_bytes, c = 0;
821 size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
822 size_t out_size, in_size;
824 u8 *target, *lunp, task_attr;
825 bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
828 mutex_lock(&vq->mutex);
830 * We can handle the vq only after the endpoint is setup by calling the
831 * VHOST_SCSI_SET_ENDPOINT ioctl.
833 vs_tpg = vq->private_data;
837 vhost_disable_notify(&vs->dev, vq);
840 head = vhost_get_vq_desc(vq, vq->iov,
841 ARRAY_SIZE(vq->iov), &out, &in,
843 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
845 /* On error, stop handling until the next kick. */
846 if (unlikely(head < 0))
848 /* Nothing new? Wait for eventfd to tell us they refilled. */
849 if (head == vq->num) {
850 if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
851 vhost_disable_notify(&vs->dev, vq);
857 * Check for a sane response buffer so we can report early
858 * errors back to the guest.
860 if (unlikely(vq->iov[out].iov_len < rsp_size)) {
861 vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
862 " size, got %zu bytes\n", vq->iov[out].iov_len);
866 * Setup pointers and values based upon different virtio-scsi
867 * request header if T10_PI is enabled in KVM guest.
871 req_size = sizeof(v_req_pi);
872 lunp = &v_req_pi.lun[0];
873 target = &v_req_pi.lun[1];
876 req_size = sizeof(v_req);
877 lunp = &v_req.lun[0];
878 target = &v_req.lun[1];
881 * FIXME: Not correct for BIDI operation
883 out_size = iov_length(vq->iov, out);
884 in_size = iov_length(&vq->iov[out], in);
887 * Copy over the virtio-scsi request header, which for a
888 * ANY_LAYOUT enabled guest may span multiple iovecs, or a
889 * single iovec may contain both the header + outgoing
892 * copy_from_iter() will advance out_iter, so that it will
893 * point at the start of the outgoing WRITE payload, if
894 * DMA_TO_DEVICE is set.
896 iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
898 if (unlikely(!copy_from_iter_full(req, req_size, &out_iter))) {
899 vq_err(vq, "Faulted on copy_from_iter\n");
900 vhost_scsi_send_bad_target(vs, vq, head, out);
903 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
904 if (unlikely(*lunp != 1)) {
905 vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
906 vhost_scsi_send_bad_target(vs, vq, head, out);
910 tpg = READ_ONCE(vs_tpg[*target]);
911 if (unlikely(!tpg)) {
912 /* Target does not exist, fail the request */
913 vhost_scsi_send_bad_target(vs, vq, head, out);
917 * Determine data_direction by calculating the total outgoing
918 * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
919 * response headers respectively.
921 * For DMA_TO_DEVICE this is out_iter, which is already pointing
922 * to the right place.
924 * For DMA_FROM_DEVICE, the iovec will be just past the end
925 * of the virtio-scsi response header in either the same
926 * or immediately following iovec.
928 * Any associated T10_PI bytes for the outgoing / incoming
929 * payloads are included in calculation of exp_data_len here.
933 if (out_size > req_size) {
934 data_direction = DMA_TO_DEVICE;
935 exp_data_len = out_size - req_size;
936 data_iter = out_iter;
937 } else if (in_size > rsp_size) {
938 data_direction = DMA_FROM_DEVICE;
939 exp_data_len = in_size - rsp_size;
941 iov_iter_init(&in_iter, READ, &vq->iov[out], in,
942 rsp_size + exp_data_len);
943 iov_iter_advance(&in_iter, rsp_size);
946 data_direction = DMA_NONE;
950 * If T10_PI header + payload is present, setup prot_iter values
951 * and recalculate data_iter for vhost_scsi_mapal() mapping to
952 * host scatterlists via get_user_pages_fast().
955 if (v_req_pi.pi_bytesout) {
956 if (data_direction != DMA_TO_DEVICE) {
957 vq_err(vq, "Received non zero pi_bytesout,"
958 " but wrong data_direction\n");
959 vhost_scsi_send_bad_target(vs, vq, head, out);
962 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
963 } else if (v_req_pi.pi_bytesin) {
964 if (data_direction != DMA_FROM_DEVICE) {
965 vq_err(vq, "Received non zero pi_bytesin,"
966 " but wrong data_direction\n");
967 vhost_scsi_send_bad_target(vs, vq, head, out);
970 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
973 * Set prot_iter to data_iter and truncate it to
974 * prot_bytes, and advance data_iter past any
975 * preceeding prot_bytes that may be present.
977 * Also fix up the exp_data_len to reflect only the
978 * actual data payload length.
981 exp_data_len -= prot_bytes;
982 prot_iter = data_iter;
983 iov_iter_truncate(&prot_iter, prot_bytes);
984 iov_iter_advance(&data_iter, prot_bytes);
986 tag = vhost64_to_cpu(vq, v_req_pi.tag);
987 task_attr = v_req_pi.task_attr;
988 cdb = &v_req_pi.cdb[0];
989 lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
991 tag = vhost64_to_cpu(vq, v_req.tag);
992 task_attr = v_req.task_attr;
994 lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
997 * Check that the received CDB size does not exceeded our
998 * hardcoded max for vhost-scsi, then get a pre-allocated
999 * cmd descriptor for the new virtio-scsi tag.
1001 * TODO what if cdb was too small for varlen cdb header?
1003 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1004 vq_err(vq, "Received SCSI CDB with command_size: %d that"
1005 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1006 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1007 vhost_scsi_send_bad_target(vs, vq, head, out);
1010 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1011 exp_data_len + prot_bytes,
1014 vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1016 vhost_scsi_send_bad_target(vs, vq, head, out);
1019 cmd->tvc_vhost = vs;
1021 cmd->tvc_resp_iov = vq->iov[out];
1022 cmd->tvc_in_iovs = in;
1024 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1025 cmd->tvc_cdb[0], cmd->tvc_lun);
1026 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1027 " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1029 if (data_direction != DMA_NONE) {
1030 ret = vhost_scsi_mapal(cmd,
1031 prot_bytes, &prot_iter,
1032 exp_data_len, &data_iter);
1033 if (unlikely(ret)) {
1034 vq_err(vq, "Failed to map iov to sgl\n");
1035 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1036 vhost_scsi_send_bad_target(vs, vq, head, out);
1041 * Save the descriptor from vhost_get_vq_desc() to be used to
1042 * complete the virtio-scsi request in TCM callback context via
1043 * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1045 cmd->tvc_vq_desc = head;
1047 * Dispatch cmd descriptor for cmwq execution in process
1048 * context provided by vhost_scsi_workqueue. This also ensures
1049 * cmd is executed on the same kworker CPU as this vhost
1050 * thread to gain positive L2 cache locality effects.
1052 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1053 queue_work(vhost_scsi_workqueue, &cmd->work);
1054 } while (likely(!vhost_exceeds_weight(vq, ++c, 0)));
1056 mutex_unlock(&vq->mutex);
1059 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1061 pr_debug("%s: The handling func for control queue.\n", __func__);
1065 vhost_scsi_send_evt(struct vhost_scsi *vs,
1066 struct vhost_scsi_tpg *tpg,
1071 struct vhost_scsi_evt *evt;
1073 evt = vhost_scsi_allocate_evt(vs, event, reason);
1078 /* TODO: share lun setup code with virtio-scsi.ko */
1080 * Note: evt->event is zeroed when we allocate it and
1081 * lun[4-7] need to be zero according to virtio-scsi spec.
1083 evt->event.lun[0] = 0x01;
1084 evt->event.lun[1] = tpg->tport_tpgt;
1085 if (lun->unpacked_lun >= 256)
1086 evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1087 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1090 llist_add(&evt->list, &vs->vs_event_list);
1091 vhost_work_queue(&vs->dev, &vs->vs_event_work);
1094 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1096 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1098 struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1100 mutex_lock(&vq->mutex);
1101 if (!vq->private_data)
1104 if (vs->vs_events_missed)
1105 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1107 mutex_unlock(&vq->mutex);
1110 static void vhost_scsi_handle_kick(struct vhost_work *work)
1112 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1114 struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1116 vhost_scsi_handle_vq(vs, vq);
1119 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1121 vhost_poll_flush(&vs->vqs[index].vq.poll);
1124 /* Callers must hold dev mutex */
1125 static void vhost_scsi_flush(struct vhost_scsi *vs)
1127 struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1130 /* Init new inflight and remember the old inflight */
1131 vhost_scsi_init_inflight(vs, old_inflight);
1134 * The inflight->kref was initialized to 1. We decrement it here to
1135 * indicate the start of the flush operation so that it will reach 0
1136 * when all the reqs are finished.
1138 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1139 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1141 /* Flush both the vhost poll and vhost work */
1142 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1143 vhost_scsi_flush_vq(vs, i);
1144 vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1145 vhost_work_flush(&vs->dev, &vs->vs_event_work);
1147 /* Wait for all reqs issued before the flush to be finished */
1148 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1149 wait_for_completion(&old_inflight[i]->comp);
1153 * Called from vhost_scsi_ioctl() context to walk the list of available
1154 * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1156 * The lock nesting rule is:
1157 * vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1160 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1161 struct vhost_scsi_target *t)
1163 struct se_portal_group *se_tpg;
1164 struct vhost_scsi_tport *tv_tport;
1165 struct vhost_scsi_tpg *tpg;
1166 struct vhost_scsi_tpg **vs_tpg;
1167 struct vhost_virtqueue *vq;
1168 int index, ret, i, len;
1171 mutex_lock(&vhost_scsi_mutex);
1172 mutex_lock(&vs->dev.mutex);
1174 /* Verify that ring has been setup correctly. */
1175 for (index = 0; index < vs->dev.nvqs; ++index) {
1176 /* Verify that ring has been setup correctly. */
1177 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1183 len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1184 vs_tpg = kzalloc(len, GFP_KERNEL);
1190 memcpy(vs_tpg, vs->vs_tpg, len);
1192 list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1193 mutex_lock(&tpg->tv_tpg_mutex);
1194 if (!tpg->tpg_nexus) {
1195 mutex_unlock(&tpg->tv_tpg_mutex);
1198 if (tpg->tv_tpg_vhost_count != 0) {
1199 mutex_unlock(&tpg->tv_tpg_mutex);
1202 tv_tport = tpg->tport;
1204 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1205 if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1207 mutex_unlock(&tpg->tv_tpg_mutex);
1212 * In order to ensure individual vhost-scsi configfs
1213 * groups cannot be removed while in use by vhost ioctl,
1214 * go ahead and take an explicit se_tpg->tpg_group.cg_item
1217 se_tpg = &tpg->se_tpg;
1218 ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1220 pr_warn("configfs_depend_item() failed: %d\n", ret);
1222 mutex_unlock(&tpg->tv_tpg_mutex);
1225 tpg->tv_tpg_vhost_count++;
1226 tpg->vhost_scsi = vs;
1227 vs_tpg[tpg->tport_tpgt] = tpg;
1228 smp_mb__after_atomic();
1231 mutex_unlock(&tpg->tv_tpg_mutex);
1235 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1236 sizeof(vs->vs_vhost_wwpn));
1237 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1238 vq = &vs->vqs[i].vq;
1239 mutex_lock(&vq->mutex);
1240 vq->private_data = vs_tpg;
1241 vhost_vq_init_access(vq);
1242 mutex_unlock(&vq->mutex);
1250 * Act as synchronize_rcu to make sure access to
1251 * old vs->vs_tpg is finished.
1253 vhost_scsi_flush(vs);
1255 vs->vs_tpg = vs_tpg;
1258 mutex_unlock(&vs->dev.mutex);
1259 mutex_unlock(&vhost_scsi_mutex);
1264 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1265 struct vhost_scsi_target *t)
1267 struct se_portal_group *se_tpg;
1268 struct vhost_scsi_tport *tv_tport;
1269 struct vhost_scsi_tpg *tpg;
1270 struct vhost_virtqueue *vq;
1275 mutex_lock(&vhost_scsi_mutex);
1276 mutex_lock(&vs->dev.mutex);
1277 /* Verify that ring has been setup correctly. */
1278 for (index = 0; index < vs->dev.nvqs; ++index) {
1279 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1290 for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1292 tpg = vs->vs_tpg[target];
1296 mutex_lock(&tpg->tv_tpg_mutex);
1297 tv_tport = tpg->tport;
1303 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1304 pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1305 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1306 tv_tport->tport_name, tpg->tport_tpgt,
1307 t->vhost_wwpn, t->vhost_tpgt);
1311 tpg->tv_tpg_vhost_count--;
1312 tpg->vhost_scsi = NULL;
1313 vs->vs_tpg[target] = NULL;
1315 mutex_unlock(&tpg->tv_tpg_mutex);
1317 * Release se_tpg->tpg_group.cg_item configfs dependency now
1318 * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1320 se_tpg = &tpg->se_tpg;
1321 target_undepend_item(&se_tpg->tpg_group.cg_item);
1324 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1325 vq = &vs->vqs[i].vq;
1326 mutex_lock(&vq->mutex);
1327 vq->private_data = NULL;
1328 mutex_unlock(&vq->mutex);
1332 * Act as synchronize_rcu to make sure access to
1333 * old vs->vs_tpg is finished.
1335 vhost_scsi_flush(vs);
1338 WARN_ON(vs->vs_events_nr);
1339 mutex_unlock(&vs->dev.mutex);
1340 mutex_unlock(&vhost_scsi_mutex);
1344 mutex_unlock(&tpg->tv_tpg_mutex);
1346 mutex_unlock(&vs->dev.mutex);
1347 mutex_unlock(&vhost_scsi_mutex);
1351 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1353 struct vhost_virtqueue *vq;
1356 if (features & ~VHOST_SCSI_FEATURES)
1359 mutex_lock(&vs->dev.mutex);
1360 if ((features & (1 << VHOST_F_LOG_ALL)) &&
1361 !vhost_log_access_ok(&vs->dev)) {
1362 mutex_unlock(&vs->dev.mutex);
1366 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1367 vq = &vs->vqs[i].vq;
1368 mutex_lock(&vq->mutex);
1369 vq->acked_features = features;
1370 mutex_unlock(&vq->mutex);
1372 mutex_unlock(&vs->dev.mutex);
1376 static int vhost_scsi_open(struct inode *inode, struct file *f)
1378 struct vhost_scsi *vs;
1379 struct vhost_virtqueue **vqs;
1382 vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_RETRY_MAYFAIL);
1384 vs = vzalloc(sizeof(*vs));
1389 vqs = kmalloc_array(VHOST_SCSI_MAX_VQ, sizeof(*vqs), GFP_KERNEL);
1393 vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1394 vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1396 vs->vs_events_nr = 0;
1397 vs->vs_events_missed = false;
1399 vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1400 vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1401 vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1402 vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1403 for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1404 vqs[i] = &vs->vqs[i].vq;
1405 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1407 vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ, UIO_MAXIOV,
1408 VHOST_SCSI_WEIGHT, 0);
1410 vhost_scsi_init_inflight(vs, NULL);
1412 f->private_data = vs;
1421 static int vhost_scsi_release(struct inode *inode, struct file *f)
1423 struct vhost_scsi *vs = f->private_data;
1424 struct vhost_scsi_target t;
1426 mutex_lock(&vs->dev.mutex);
1427 memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1428 mutex_unlock(&vs->dev.mutex);
1429 vhost_scsi_clear_endpoint(vs, &t);
1430 vhost_dev_stop(&vs->dev);
1431 vhost_dev_cleanup(&vs->dev);
1432 /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1433 vhost_scsi_flush(vs);
1440 vhost_scsi_ioctl(struct file *f,
1444 struct vhost_scsi *vs = f->private_data;
1445 struct vhost_scsi_target backend;
1446 void __user *argp = (void __user *)arg;
1447 u64 __user *featurep = argp;
1448 u32 __user *eventsp = argp;
1451 int r, abi_version = VHOST_SCSI_ABI_VERSION;
1452 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1455 case VHOST_SCSI_SET_ENDPOINT:
1456 if (copy_from_user(&backend, argp, sizeof backend))
1458 if (backend.reserved != 0)
1461 return vhost_scsi_set_endpoint(vs, &backend);
1462 case VHOST_SCSI_CLEAR_ENDPOINT:
1463 if (copy_from_user(&backend, argp, sizeof backend))
1465 if (backend.reserved != 0)
1468 return vhost_scsi_clear_endpoint(vs, &backend);
1469 case VHOST_SCSI_GET_ABI_VERSION:
1470 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1473 case VHOST_SCSI_SET_EVENTS_MISSED:
1474 if (get_user(events_missed, eventsp))
1476 mutex_lock(&vq->mutex);
1477 vs->vs_events_missed = events_missed;
1478 mutex_unlock(&vq->mutex);
1480 case VHOST_SCSI_GET_EVENTS_MISSED:
1481 mutex_lock(&vq->mutex);
1482 events_missed = vs->vs_events_missed;
1483 mutex_unlock(&vq->mutex);
1484 if (put_user(events_missed, eventsp))
1487 case VHOST_GET_FEATURES:
1488 features = VHOST_SCSI_FEATURES;
1489 if (copy_to_user(featurep, &features, sizeof features))
1492 case VHOST_SET_FEATURES:
1493 if (copy_from_user(&features, featurep, sizeof features))
1495 return vhost_scsi_set_features(vs, features);
1497 mutex_lock(&vs->dev.mutex);
1498 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1499 /* TODO: flush backend after dev ioctl. */
1500 if (r == -ENOIOCTLCMD)
1501 r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1502 mutex_unlock(&vs->dev.mutex);
1507 #ifdef CONFIG_COMPAT
1508 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1511 return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1515 static const struct file_operations vhost_scsi_fops = {
1516 .owner = THIS_MODULE,
1517 .release = vhost_scsi_release,
1518 .unlocked_ioctl = vhost_scsi_ioctl,
1519 #ifdef CONFIG_COMPAT
1520 .compat_ioctl = vhost_scsi_compat_ioctl,
1522 .open = vhost_scsi_open,
1523 .llseek = noop_llseek,
1526 static struct miscdevice vhost_scsi_misc = {
1532 static int __init vhost_scsi_register(void)
1534 return misc_register(&vhost_scsi_misc);
1537 static void vhost_scsi_deregister(void)
1539 misc_deregister(&vhost_scsi_misc);
1542 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1544 switch (tport->tport_proto_id) {
1545 case SCSI_PROTOCOL_SAS:
1547 case SCSI_PROTOCOL_FCP:
1549 case SCSI_PROTOCOL_ISCSI:
1559 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1560 struct se_lun *lun, bool plug)
1563 struct vhost_scsi *vs = tpg->vhost_scsi;
1564 struct vhost_virtqueue *vq;
1570 mutex_lock(&vs->dev.mutex);
1573 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1575 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1577 vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1578 mutex_lock(&vq->mutex);
1579 if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1580 vhost_scsi_send_evt(vs, tpg, lun,
1581 VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1582 mutex_unlock(&vq->mutex);
1583 mutex_unlock(&vs->dev.mutex);
1586 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1588 vhost_scsi_do_plug(tpg, lun, true);
1591 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1593 vhost_scsi_do_plug(tpg, lun, false);
1596 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1599 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1600 struct vhost_scsi_tpg, se_tpg);
1602 mutex_lock(&vhost_scsi_mutex);
1604 mutex_lock(&tpg->tv_tpg_mutex);
1605 tpg->tv_tpg_port_count++;
1606 mutex_unlock(&tpg->tv_tpg_mutex);
1608 vhost_scsi_hotplug(tpg, lun);
1610 mutex_unlock(&vhost_scsi_mutex);
1615 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1618 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1619 struct vhost_scsi_tpg, se_tpg);
1621 mutex_lock(&vhost_scsi_mutex);
1623 mutex_lock(&tpg->tv_tpg_mutex);
1624 tpg->tv_tpg_port_count--;
1625 mutex_unlock(&tpg->tv_tpg_mutex);
1627 vhost_scsi_hotunplug(tpg, lun);
1629 mutex_unlock(&vhost_scsi_mutex);
1632 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1634 struct vhost_scsi_cmd *tv_cmd;
1637 if (!se_sess->sess_cmd_map)
1640 for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1641 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1643 kfree(tv_cmd->tvc_sgl);
1644 kfree(tv_cmd->tvc_prot_sgl);
1645 kfree(tv_cmd->tvc_upages);
1649 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1650 struct config_item *item, const char *page, size_t count)
1652 struct se_portal_group *se_tpg = attrib_to_tpg(item);
1653 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1654 struct vhost_scsi_tpg, se_tpg);
1656 int ret = kstrtoul(page, 0, &val);
1659 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1662 if (val != 0 && val != 1 && val != 3) {
1663 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1666 tpg->tv_fabric_prot_type = val;
1671 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1672 struct config_item *item, char *page)
1674 struct se_portal_group *se_tpg = attrib_to_tpg(item);
1675 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1676 struct vhost_scsi_tpg, se_tpg);
1678 return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1681 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1683 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1684 &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1688 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1689 struct se_session *se_sess, void *p)
1691 struct vhost_scsi_cmd *tv_cmd;
1694 for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1695 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1697 tv_cmd->tvc_sgl = kcalloc(VHOST_SCSI_PREALLOC_SGLS,
1698 sizeof(struct scatterlist),
1700 if (!tv_cmd->tvc_sgl) {
1701 pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1705 tv_cmd->tvc_upages = kcalloc(VHOST_SCSI_PREALLOC_UPAGES,
1706 sizeof(struct page *),
1708 if (!tv_cmd->tvc_upages) {
1709 pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1713 tv_cmd->tvc_prot_sgl = kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS,
1714 sizeof(struct scatterlist),
1716 if (!tv_cmd->tvc_prot_sgl) {
1717 pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1723 vhost_scsi_free_cmd_map_res(se_sess);
1727 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1730 struct vhost_scsi_nexus *tv_nexus;
1732 mutex_lock(&tpg->tv_tpg_mutex);
1733 if (tpg->tpg_nexus) {
1734 mutex_unlock(&tpg->tv_tpg_mutex);
1735 pr_debug("tpg->tpg_nexus already exists\n");
1739 tv_nexus = kzalloc(sizeof(*tv_nexus), GFP_KERNEL);
1741 mutex_unlock(&tpg->tv_tpg_mutex);
1742 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1746 * Since we are running in 'demo mode' this call with generate a
1747 * struct se_node_acl for the vhost_scsi struct se_portal_group with
1748 * the SCSI Initiator port name of the passed configfs group 'name'.
1750 tv_nexus->tvn_se_sess = target_setup_session(&tpg->se_tpg,
1751 VHOST_SCSI_DEFAULT_TAGS,
1752 sizeof(struct vhost_scsi_cmd),
1753 TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1754 (unsigned char *)name, tv_nexus,
1755 vhost_scsi_nexus_cb);
1756 if (IS_ERR(tv_nexus->tvn_se_sess)) {
1757 mutex_unlock(&tpg->tv_tpg_mutex);
1761 tpg->tpg_nexus = tv_nexus;
1763 mutex_unlock(&tpg->tv_tpg_mutex);
1767 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1769 struct se_session *se_sess;
1770 struct vhost_scsi_nexus *tv_nexus;
1772 mutex_lock(&tpg->tv_tpg_mutex);
1773 tv_nexus = tpg->tpg_nexus;
1775 mutex_unlock(&tpg->tv_tpg_mutex);
1779 se_sess = tv_nexus->tvn_se_sess;
1781 mutex_unlock(&tpg->tv_tpg_mutex);
1785 if (tpg->tv_tpg_port_count != 0) {
1786 mutex_unlock(&tpg->tv_tpg_mutex);
1787 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1788 " active TPG port count: %d\n",
1789 tpg->tv_tpg_port_count);
1793 if (tpg->tv_tpg_vhost_count != 0) {
1794 mutex_unlock(&tpg->tv_tpg_mutex);
1795 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1796 " active TPG vhost count: %d\n",
1797 tpg->tv_tpg_vhost_count);
1801 pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1802 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1803 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1805 vhost_scsi_free_cmd_map_res(se_sess);
1807 * Release the SCSI I_T Nexus to the emulated vhost Target Port
1809 target_remove_session(se_sess);
1810 tpg->tpg_nexus = NULL;
1811 mutex_unlock(&tpg->tv_tpg_mutex);
1817 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1819 struct se_portal_group *se_tpg = to_tpg(item);
1820 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1821 struct vhost_scsi_tpg, se_tpg);
1822 struct vhost_scsi_nexus *tv_nexus;
1825 mutex_lock(&tpg->tv_tpg_mutex);
1826 tv_nexus = tpg->tpg_nexus;
1828 mutex_unlock(&tpg->tv_tpg_mutex);
1831 ret = snprintf(page, PAGE_SIZE, "%s\n",
1832 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1833 mutex_unlock(&tpg->tv_tpg_mutex);
1838 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1839 const char *page, size_t count)
1841 struct se_portal_group *se_tpg = to_tpg(item);
1842 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1843 struct vhost_scsi_tpg, se_tpg);
1844 struct vhost_scsi_tport *tport_wwn = tpg->tport;
1845 unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1848 * Shutdown the active I_T nexus if 'NULL' is passed..
1850 if (!strncmp(page, "NULL", 4)) {
1851 ret = vhost_scsi_drop_nexus(tpg);
1852 return (!ret) ? count : ret;
1855 * Otherwise make sure the passed virtual Initiator port WWN matches
1856 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1857 * vhost_scsi_make_nexus().
1859 if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1860 pr_err("Emulated NAA Sas Address: %s, exceeds"
1861 " max: %d\n", page, VHOST_SCSI_NAMELEN);
1864 snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1866 ptr = strstr(i_port, "naa.");
1868 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1869 pr_err("Passed SAS Initiator Port %s does not"
1870 " match target port protoid: %s\n", i_port,
1871 vhost_scsi_dump_proto_id(tport_wwn));
1874 port_ptr = &i_port[0];
1877 ptr = strstr(i_port, "fc.");
1879 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1880 pr_err("Passed FCP Initiator Port %s does not"
1881 " match target port protoid: %s\n", i_port,
1882 vhost_scsi_dump_proto_id(tport_wwn));
1885 port_ptr = &i_port[3]; /* Skip over "fc." */
1888 ptr = strstr(i_port, "iqn.");
1890 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1891 pr_err("Passed iSCSI Initiator Port %s does not"
1892 " match target port protoid: %s\n", i_port,
1893 vhost_scsi_dump_proto_id(tport_wwn));
1896 port_ptr = &i_port[0];
1899 pr_err("Unable to locate prefix for emulated Initiator Port:"
1903 * Clear any trailing newline for the NAA WWN
1906 if (i_port[strlen(i_port)-1] == '\n')
1907 i_port[strlen(i_port)-1] = '\0';
1909 ret = vhost_scsi_make_nexus(tpg, port_ptr);
1916 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1918 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1919 &vhost_scsi_tpg_attr_nexus,
1923 static struct se_portal_group *
1924 vhost_scsi_make_tpg(struct se_wwn *wwn, const char *name)
1926 struct vhost_scsi_tport *tport = container_of(wwn,
1927 struct vhost_scsi_tport, tport_wwn);
1929 struct vhost_scsi_tpg *tpg;
1933 if (strstr(name, "tpgt_") != name)
1934 return ERR_PTR(-EINVAL);
1935 if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1936 return ERR_PTR(-EINVAL);
1938 tpg = kzalloc(sizeof(*tpg), GFP_KERNEL);
1940 pr_err("Unable to allocate struct vhost_scsi_tpg");
1941 return ERR_PTR(-ENOMEM);
1943 mutex_init(&tpg->tv_tpg_mutex);
1944 INIT_LIST_HEAD(&tpg->tv_tpg_list);
1946 tpg->tport_tpgt = tpgt;
1948 ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1953 mutex_lock(&vhost_scsi_mutex);
1954 list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1955 mutex_unlock(&vhost_scsi_mutex);
1957 return &tpg->se_tpg;
1960 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1962 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1963 struct vhost_scsi_tpg, se_tpg);
1965 mutex_lock(&vhost_scsi_mutex);
1966 list_del(&tpg->tv_tpg_list);
1967 mutex_unlock(&vhost_scsi_mutex);
1969 * Release the virtual I_T Nexus for this vhost TPG
1971 vhost_scsi_drop_nexus(tpg);
1973 * Deregister the se_tpg from TCM..
1975 core_tpg_deregister(se_tpg);
1979 static struct se_wwn *
1980 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
1981 struct config_group *group,
1984 struct vhost_scsi_tport *tport;
1989 /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
1990 return ERR_PTR(-EINVAL); */
1992 tport = kzalloc(sizeof(*tport), GFP_KERNEL);
1994 pr_err("Unable to allocate struct vhost_scsi_tport");
1995 return ERR_PTR(-ENOMEM);
1997 tport->tport_wwpn = wwpn;
1999 * Determine the emulated Protocol Identifier and Target Port Name
2000 * based on the incoming configfs directory name.
2002 ptr = strstr(name, "naa.");
2004 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2007 ptr = strstr(name, "fc.");
2009 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2010 off = 3; /* Skip over "fc." */
2013 ptr = strstr(name, "iqn.");
2015 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2019 pr_err("Unable to locate prefix for emulated Target Port:"
2022 return ERR_PTR(-EINVAL);
2025 if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2026 pr_err("Emulated %s Address: %s, exceeds"
2027 " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2028 VHOST_SCSI_NAMELEN);
2030 return ERR_PTR(-EINVAL);
2032 snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2034 pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2035 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2037 return &tport->tport_wwn;
2040 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2042 struct vhost_scsi_tport *tport = container_of(wwn,
2043 struct vhost_scsi_tport, tport_wwn);
2045 pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2046 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2053 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2055 return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2056 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2057 utsname()->machine);
2060 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2062 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2063 &vhost_scsi_wwn_attr_version,
2067 static const struct target_core_fabric_ops vhost_scsi_ops = {
2068 .module = THIS_MODULE,
2070 .get_fabric_name = vhost_scsi_get_fabric_name,
2071 .tpg_get_wwn = vhost_scsi_get_fabric_wwn,
2072 .tpg_get_tag = vhost_scsi_get_tpgt,
2073 .tpg_check_demo_mode = vhost_scsi_check_true,
2074 .tpg_check_demo_mode_cache = vhost_scsi_check_true,
2075 .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2076 .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2077 .tpg_check_prot_fabric_only = vhost_scsi_check_prot_fabric_only,
2078 .tpg_get_inst_index = vhost_scsi_tpg_get_inst_index,
2079 .release_cmd = vhost_scsi_release_cmd,
2080 .check_stop_free = vhost_scsi_check_stop_free,
2081 .sess_get_index = vhost_scsi_sess_get_index,
2082 .sess_get_initiator_sid = NULL,
2083 .write_pending = vhost_scsi_write_pending,
2084 .write_pending_status = vhost_scsi_write_pending_status,
2085 .set_default_node_attributes = vhost_scsi_set_default_node_attrs,
2086 .get_cmd_state = vhost_scsi_get_cmd_state,
2087 .queue_data_in = vhost_scsi_queue_data_in,
2088 .queue_status = vhost_scsi_queue_status,
2089 .queue_tm_rsp = vhost_scsi_queue_tm_rsp,
2090 .aborted_task = vhost_scsi_aborted_task,
2092 * Setup callers for generic logic in target_core_fabric_configfs.c
2094 .fabric_make_wwn = vhost_scsi_make_tport,
2095 .fabric_drop_wwn = vhost_scsi_drop_tport,
2096 .fabric_make_tpg = vhost_scsi_make_tpg,
2097 .fabric_drop_tpg = vhost_scsi_drop_tpg,
2098 .fabric_post_link = vhost_scsi_port_link,
2099 .fabric_pre_unlink = vhost_scsi_port_unlink,
2101 .tfc_wwn_attrs = vhost_scsi_wwn_attrs,
2102 .tfc_tpg_base_attrs = vhost_scsi_tpg_attrs,
2103 .tfc_tpg_attrib_attrs = vhost_scsi_tpg_attrib_attrs,
2106 static int __init vhost_scsi_init(void)
2110 pr_debug("TCM_VHOST fabric module %s on %s/%s"
2111 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2112 utsname()->machine);
2115 * Use our own dedicated workqueue for submitting I/O into
2116 * target core to avoid contention within system_wq.
2118 vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2119 if (!vhost_scsi_workqueue)
2122 ret = vhost_scsi_register();
2124 goto out_destroy_workqueue;
2126 ret = target_register_template(&vhost_scsi_ops);
2128 goto out_vhost_scsi_deregister;
2132 out_vhost_scsi_deregister:
2133 vhost_scsi_deregister();
2134 out_destroy_workqueue:
2135 destroy_workqueue(vhost_scsi_workqueue);
2140 static void vhost_scsi_exit(void)
2142 target_unregister_template(&vhost_scsi_ops);
2143 vhost_scsi_deregister();
2144 destroy_workqueue(vhost_scsi_workqueue);
2147 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2148 MODULE_ALIAS("tcm_vhost");
2149 MODULE_LICENSE("GPL");
2150 module_init(vhost_scsi_init);
2151 module_exit(vhost_scsi_exit);