1 // SPDX-License-Identifier: GPL-2.0+
2 /*******************************************************************************
3 * Vhost kernel TCM fabric driver for virtio SCSI initiators
5 * (C) Copyright 2010-2013 Datera, Inc.
6 * (C) Copyright 2010-2012 IBM Corp.
8 * Authors: Nicholas A. Bellinger <nab@daterainc.com>
9 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
10 ****************************************************************************/
12 #include <linux/module.h>
13 #include <linux/moduleparam.h>
14 #include <generated/utsrelease.h>
15 #include <linux/utsname.h>
16 #include <linux/init.h>
17 #include <linux/slab.h>
18 #include <linux/kthread.h>
19 #include <linux/types.h>
20 #include <linux/string.h>
21 #include <linux/configfs.h>
22 #include <linux/ctype.h>
23 #include <linux/compat.h>
24 #include <linux/eventfd.h>
26 #include <linux/vmalloc.h>
27 #include <linux/miscdevice.h>
28 #include <linux/blk_types.h>
29 #include <linux/bio.h>
30 #include <asm/unaligned.h>
31 #include <scsi/scsi_common.h>
32 #include <scsi/scsi_proto.h>
33 #include <target/target_core_base.h>
34 #include <target/target_core_fabric.h>
35 #include <linux/vhost.h>
36 #include <linux/virtio_scsi.h>
37 #include <linux/llist.h>
38 #include <linux/bitmap.h>
42 #define VHOST_SCSI_VERSION "v0.1"
43 #define VHOST_SCSI_NAMELEN 256
44 #define VHOST_SCSI_MAX_CDB_SIZE 32
45 #define VHOST_SCSI_PREALLOC_SGLS 2048
46 #define VHOST_SCSI_PREALLOC_UPAGES 2048
47 #define VHOST_SCSI_PREALLOC_PROT_SGLS 2048
49 /* Max number of requests before requeueing the job.
50 * Using this limit prevents one virtqueue from starving others with
53 #define VHOST_SCSI_WEIGHT 256
55 struct vhost_scsi_inflight
{
56 /* Wait for the flush operation to finish */
57 struct completion comp
;
58 /* Refcount for the inflight reqs */
62 struct vhost_scsi_cmd
{
63 /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
65 /* virtio-scsi initiator task attribute */
67 /* virtio-scsi response incoming iovecs */
69 /* virtio-scsi initiator data direction */
70 enum dma_data_direction tvc_data_direction
;
71 /* Expected data transfer length from virtio-scsi header */
73 /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
75 /* The number of scatterlists associated with this cmd */
77 u32 tvc_prot_sgl_count
;
78 /* Saved unpacked SCSI LUN for vhost_scsi_target_queue_cmd() */
81 const void *saved_iter_addr
;
82 struct iov_iter saved_iter
;
83 /* Pointer to the SGL formatted memory from virtio-scsi */
84 struct scatterlist
*tvc_sgl
;
85 struct scatterlist
*tvc_prot_sgl
;
86 struct page
**tvc_upages
;
87 /* Pointer to response header iovec */
88 struct iovec
*tvc_resp_iov
;
89 /* Pointer to vhost_scsi for our device */
90 struct vhost_scsi
*tvc_vhost
;
91 /* Pointer to vhost_virtqueue for the cmd */
92 struct vhost_virtqueue
*tvc_vq
;
93 /* Pointer to vhost nexus memory */
94 struct vhost_scsi_nexus
*tvc_nexus
;
95 /* The TCM I/O descriptor that is accessed via container_of() */
96 struct se_cmd tvc_se_cmd
;
97 /* Copy of the incoming SCSI command descriptor block (CDB) */
98 unsigned char tvc_cdb
[VHOST_SCSI_MAX_CDB_SIZE
];
99 /* Sense buffer that will be mapped into outgoing status */
100 unsigned char tvc_sense_buf
[TRANSPORT_SENSE_BUFFER
];
101 /* Completed commands list, serviced from vhost worker thread */
102 struct llist_node tvc_completion_list
;
103 /* Used to track inflight cmd */
104 struct vhost_scsi_inflight
*inflight
;
107 struct vhost_scsi_nexus
{
108 /* Pointer to TCM session for I_T Nexus */
109 struct se_session
*tvn_se_sess
;
112 struct vhost_scsi_tpg
{
113 /* Vhost port target portal group tag for TCM */
115 /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
116 int tv_tpg_port_count
;
117 /* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
118 int tv_tpg_vhost_count
;
119 /* Used for enabling T10-PI with legacy devices */
120 int tv_fabric_prot_type
;
121 /* list for vhost_scsi_list */
122 struct list_head tv_tpg_list
;
123 /* Used to protect access for tpg_nexus */
124 struct mutex tv_tpg_mutex
;
125 /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
126 struct vhost_scsi_nexus
*tpg_nexus
;
127 /* Pointer back to vhost_scsi_tport */
128 struct vhost_scsi_tport
*tport
;
129 /* Returned by vhost_scsi_make_tpg() */
130 struct se_portal_group se_tpg
;
131 /* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
132 struct vhost_scsi
*vhost_scsi
;
135 struct vhost_scsi_tport
{
136 /* SCSI protocol the tport is providing */
138 /* Binary World Wide unique Port Name for Vhost Target port */
140 /* ASCII formatted WWPN for Vhost Target port */
141 char tport_name
[VHOST_SCSI_NAMELEN
];
142 /* Returned by vhost_scsi_make_tport() */
143 struct se_wwn tport_wwn
;
146 struct vhost_scsi_evt
{
147 /* event to be sent to guest */
148 struct virtio_scsi_event event
;
149 /* event list, serviced from vhost worker thread */
150 struct llist_node list
;
154 VHOST_SCSI_VQ_CTL
= 0,
155 VHOST_SCSI_VQ_EVT
= 1,
156 VHOST_SCSI_VQ_IO
= 2,
159 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
161 VHOST_SCSI_FEATURES
= VHOST_FEATURES
| (1ULL << VIRTIO_SCSI_F_HOTPLUG
) |
162 (1ULL << VIRTIO_SCSI_F_T10_PI
)
165 #define VHOST_SCSI_MAX_TARGET 256
166 #define VHOST_SCSI_MAX_IO_VQ 1024
167 #define VHOST_SCSI_MAX_EVENT 128
169 static unsigned vhost_scsi_max_io_vqs
= 128;
170 module_param_named(max_io_vqs
, vhost_scsi_max_io_vqs
, uint
, 0644);
171 MODULE_PARM_DESC(max_io_vqs
, "Set the max number of IO virtqueues a vhost scsi device can support. The default is 128. The max is 1024.");
173 struct vhost_scsi_virtqueue
{
174 struct vhost_virtqueue vq
;
175 struct vhost_scsi
*vs
;
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.
181 struct vhost_scsi_inflight inflights
[2];
183 * Indicate current inflight in use, protected by vq->mutex.
184 * Writers must also take dev mutex and flush under it.
187 struct vhost_scsi_cmd
*scsi_cmds
;
188 struct sbitmap scsi_tags
;
191 struct vhost_work completion_work
;
192 struct llist_head completion_list
;
196 /* Protected by vhost_scsi->dev.mutex */
197 struct vhost_scsi_tpg
**vs_tpg
;
198 char vs_vhost_wwpn
[TRANSPORT_IQN_LEN
];
200 struct vhost_dev dev
;
201 struct vhost_scsi_virtqueue
*vqs
;
202 struct vhost_scsi_inflight
**old_inflight
;
204 struct vhost_work vs_event_work
; /* evt injection work item */
205 struct llist_head vs_event_list
; /* evt injection queue */
207 bool vs_events_missed
; /* any missed events, protected by vq->mutex */
208 int vs_events_nr
; /* num of pending events, protected by vq->mutex */
211 struct vhost_scsi_tmf
{
212 struct vhost_work vwork
;
213 struct work_struct flush_work
;
214 struct vhost_scsi
*vhost
;
215 struct vhost_scsi_virtqueue
*svq
;
217 struct se_cmd se_cmd
;
219 struct vhost_scsi_inflight
*inflight
;
220 struct iovec resp_iov
;
226 * Context for processing request and control queue operations.
228 struct vhost_scsi_ctx
{
230 unsigned int out
, in
;
231 size_t req_size
, rsp_size
;
232 size_t out_size
, in_size
;
235 struct iov_iter out_iter
;
239 * Global mutex to protect vhost_scsi TPG list for vhost IOCTLs and LIO
240 * configfs management operations.
242 static DEFINE_MUTEX(vhost_scsi_mutex
);
243 static LIST_HEAD(vhost_scsi_list
);
245 static void vhost_scsi_done_inflight(struct kref
*kref
)
247 struct vhost_scsi_inflight
*inflight
;
249 inflight
= container_of(kref
, struct vhost_scsi_inflight
, kref
);
250 complete(&inflight
->comp
);
253 static void vhost_scsi_init_inflight(struct vhost_scsi
*vs
,
254 struct vhost_scsi_inflight
*old_inflight
[])
256 struct vhost_scsi_inflight
*new_inflight
;
257 struct vhost_virtqueue
*vq
;
260 for (i
= 0; i
< vs
->dev
.nvqs
; i
++) {
263 mutex_lock(&vq
->mutex
);
265 /* store old infight */
266 idx
= vs
->vqs
[i
].inflight_idx
;
268 old_inflight
[i
] = &vs
->vqs
[i
].inflights
[idx
];
270 /* setup new infight */
271 vs
->vqs
[i
].inflight_idx
= idx
^ 1;
272 new_inflight
= &vs
->vqs
[i
].inflights
[idx
^ 1];
273 kref_init(&new_inflight
->kref
);
274 init_completion(&new_inflight
->comp
);
276 mutex_unlock(&vq
->mutex
);
280 static struct vhost_scsi_inflight
*
281 vhost_scsi_get_inflight(struct vhost_virtqueue
*vq
)
283 struct vhost_scsi_inflight
*inflight
;
284 struct vhost_scsi_virtqueue
*svq
;
286 svq
= container_of(vq
, struct vhost_scsi_virtqueue
, vq
);
287 inflight
= &svq
->inflights
[svq
->inflight_idx
];
288 kref_get(&inflight
->kref
);
293 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight
*inflight
)
295 kref_put(&inflight
->kref
, vhost_scsi_done_inflight
);
298 static int vhost_scsi_check_true(struct se_portal_group
*se_tpg
)
303 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group
*se_tpg
)
305 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
306 struct vhost_scsi_tpg
, se_tpg
);
307 struct vhost_scsi_tport
*tport
= tpg
->tport
;
309 return &tport
->tport_name
[0];
312 static u16
vhost_scsi_get_tpgt(struct se_portal_group
*se_tpg
)
314 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
315 struct vhost_scsi_tpg
, se_tpg
);
316 return tpg
->tport_tpgt
;
319 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group
*se_tpg
)
321 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
322 struct vhost_scsi_tpg
, se_tpg
);
324 return tpg
->tv_fabric_prot_type
;
327 static void vhost_scsi_release_cmd_res(struct se_cmd
*se_cmd
)
329 struct vhost_scsi_cmd
*tv_cmd
= container_of(se_cmd
,
330 struct vhost_scsi_cmd
, tvc_se_cmd
);
331 struct vhost_scsi_virtqueue
*svq
= container_of(tv_cmd
->tvc_vq
,
332 struct vhost_scsi_virtqueue
, vq
);
333 struct vhost_scsi_inflight
*inflight
= tv_cmd
->inflight
;
336 if (tv_cmd
->tvc_sgl_count
) {
337 for (i
= 0; i
< tv_cmd
->tvc_sgl_count
; i
++) {
338 if (tv_cmd
->copied_iov
)
339 __free_page(sg_page(&tv_cmd
->tvc_sgl
[i
]));
341 put_page(sg_page(&tv_cmd
->tvc_sgl
[i
]));
343 kfree(tv_cmd
->saved_iter_addr
);
345 if (tv_cmd
->tvc_prot_sgl_count
) {
346 for (i
= 0; i
< tv_cmd
->tvc_prot_sgl_count
; i
++)
347 put_page(sg_page(&tv_cmd
->tvc_prot_sgl
[i
]));
350 sbitmap_clear_bit(&svq
->scsi_tags
, se_cmd
->map_tag
);
351 vhost_scsi_put_inflight(inflight
);
354 static void vhost_scsi_release_tmf_res(struct vhost_scsi_tmf
*tmf
)
356 struct vhost_scsi_inflight
*inflight
= tmf
->inflight
;
359 vhost_scsi_put_inflight(inflight
);
362 static void vhost_scsi_drop_cmds(struct vhost_scsi_virtqueue
*svq
)
364 struct vhost_scsi_cmd
*cmd
, *t
;
365 struct llist_node
*llnode
;
367 llnode
= llist_del_all(&svq
->completion_list
);
368 llist_for_each_entry_safe(cmd
, t
, llnode
, tvc_completion_list
)
369 vhost_scsi_release_cmd_res(&cmd
->tvc_se_cmd
);
372 static void vhost_scsi_release_cmd(struct se_cmd
*se_cmd
)
374 if (se_cmd
->se_cmd_flags
& SCF_SCSI_TMR_CDB
) {
375 struct vhost_scsi_tmf
*tmf
= container_of(se_cmd
,
376 struct vhost_scsi_tmf
, se_cmd
);
378 schedule_work(&tmf
->flush_work
);
380 struct vhost_scsi_cmd
*cmd
= container_of(se_cmd
,
381 struct vhost_scsi_cmd
, tvc_se_cmd
);
382 struct vhost_scsi_virtqueue
*svq
= container_of(cmd
->tvc_vq
,
383 struct vhost_scsi_virtqueue
, vq
);
385 llist_add(&cmd
->tvc_completion_list
, &svq
->completion_list
);
386 if (!vhost_vq_work_queue(&svq
->vq
, &svq
->completion_work
))
387 vhost_scsi_drop_cmds(svq
);
391 static int vhost_scsi_write_pending(struct se_cmd
*se_cmd
)
393 /* Go ahead and process the write immediately */
394 target_execute_cmd(se_cmd
);
398 static int vhost_scsi_queue_data_in(struct se_cmd
*se_cmd
)
400 transport_generic_free_cmd(se_cmd
, 0);
404 static int vhost_scsi_queue_status(struct se_cmd
*se_cmd
)
406 transport_generic_free_cmd(se_cmd
, 0);
410 static void vhost_scsi_queue_tm_rsp(struct se_cmd
*se_cmd
)
412 struct vhost_scsi_tmf
*tmf
= container_of(se_cmd
, struct vhost_scsi_tmf
,
415 tmf
->scsi_resp
= se_cmd
->se_tmr_req
->response
;
416 transport_generic_free_cmd(&tmf
->se_cmd
, 0);
419 static void vhost_scsi_aborted_task(struct se_cmd
*se_cmd
)
424 static void vhost_scsi_free_evt(struct vhost_scsi
*vs
, struct vhost_scsi_evt
*evt
)
430 static struct vhost_scsi_evt
*
431 vhost_scsi_allocate_evt(struct vhost_scsi
*vs
,
432 u32 event
, u32 reason
)
434 struct vhost_virtqueue
*vq
= &vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
;
435 struct vhost_scsi_evt
*evt
;
437 if (vs
->vs_events_nr
> VHOST_SCSI_MAX_EVENT
) {
438 vs
->vs_events_missed
= true;
442 evt
= kzalloc(sizeof(*evt
), GFP_KERNEL
);
444 vq_err(vq
, "Failed to allocate vhost_scsi_evt\n");
445 vs
->vs_events_missed
= true;
449 evt
->event
.event
= cpu_to_vhost32(vq
, event
);
450 evt
->event
.reason
= cpu_to_vhost32(vq
, reason
);
456 static int vhost_scsi_check_stop_free(struct se_cmd
*se_cmd
)
458 return target_put_sess_cmd(se_cmd
);
462 vhost_scsi_do_evt_work(struct vhost_scsi
*vs
, struct vhost_scsi_evt
*evt
)
464 struct vhost_virtqueue
*vq
= &vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
;
465 struct virtio_scsi_event
*event
= &evt
->event
;
466 struct virtio_scsi_event __user
*eventp
;
470 if (!vhost_vq_get_backend(vq
)) {
471 vs
->vs_events_missed
= true;
476 vhost_disable_notify(&vs
->dev
, vq
);
477 head
= vhost_get_vq_desc(vq
, vq
->iov
,
478 ARRAY_SIZE(vq
->iov
), &out
, &in
,
481 vs
->vs_events_missed
= true;
484 if (head
== vq
->num
) {
485 if (vhost_enable_notify(&vs
->dev
, vq
))
487 vs
->vs_events_missed
= true;
491 if ((vq
->iov
[out
].iov_len
!= sizeof(struct virtio_scsi_event
))) {
492 vq_err(vq
, "Expecting virtio_scsi_event, got %zu bytes\n",
493 vq
->iov
[out
].iov_len
);
494 vs
->vs_events_missed
= true;
498 if (vs
->vs_events_missed
) {
499 event
->event
|= cpu_to_vhost32(vq
, VIRTIO_SCSI_T_EVENTS_MISSED
);
500 vs
->vs_events_missed
= false;
503 eventp
= vq
->iov
[out
].iov_base
;
504 ret
= __copy_to_user(eventp
, event
, sizeof(*event
));
506 vhost_add_used_and_signal(&vs
->dev
, vq
, head
, 0);
508 vq_err(vq
, "Faulted on vhost_scsi_send_event\n");
511 static void vhost_scsi_complete_events(struct vhost_scsi
*vs
, bool drop
)
513 struct vhost_virtqueue
*vq
= &vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
;
514 struct vhost_scsi_evt
*evt
, *t
;
515 struct llist_node
*llnode
;
517 mutex_lock(&vq
->mutex
);
518 llnode
= llist_del_all(&vs
->vs_event_list
);
519 llist_for_each_entry_safe(evt
, t
, llnode
, list
) {
521 vhost_scsi_do_evt_work(vs
, evt
);
522 vhost_scsi_free_evt(vs
, evt
);
524 mutex_unlock(&vq
->mutex
);
527 static void vhost_scsi_evt_work(struct vhost_work
*work
)
529 struct vhost_scsi
*vs
= container_of(work
, struct vhost_scsi
,
531 vhost_scsi_complete_events(vs
, false);
534 static int vhost_scsi_copy_sgl_to_iov(struct vhost_scsi_cmd
*cmd
)
536 struct iov_iter
*iter
= &cmd
->saved_iter
;
537 struct scatterlist
*sg
= cmd
->tvc_sgl
;
542 for (i
= 0; i
< cmd
->tvc_sgl_count
; i
++) {
543 page
= sg_page(&sg
[i
]);
546 if (copy_page_to_iter(page
, 0, len
, iter
) != len
) {
547 pr_err("Could not copy data while handling misaligned cmd. Error %zu\n",
556 /* Fill in status and signal that we are done processing this command
558 * This is scheduled in the vhost work queue so we are called with the owner
559 * process mm and can access the vring.
561 static void vhost_scsi_complete_cmd_work(struct vhost_work
*work
)
563 struct vhost_scsi_virtqueue
*svq
= container_of(work
,
564 struct vhost_scsi_virtqueue
, completion_work
);
565 struct virtio_scsi_cmd_resp v_rsp
;
566 struct vhost_scsi_cmd
*cmd
, *t
;
567 struct llist_node
*llnode
;
568 struct se_cmd
*se_cmd
;
569 struct iov_iter iov_iter
;
573 llnode
= llist_del_all(&svq
->completion_list
);
574 llist_for_each_entry_safe(cmd
, t
, llnode
, tvc_completion_list
) {
575 se_cmd
= &cmd
->tvc_se_cmd
;
577 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__
,
578 cmd
, se_cmd
->residual_count
, se_cmd
->scsi_status
);
579 memset(&v_rsp
, 0, sizeof(v_rsp
));
581 if (cmd
->saved_iter_addr
&& vhost_scsi_copy_sgl_to_iov(cmd
)) {
582 v_rsp
.response
= VIRTIO_SCSI_S_BAD_TARGET
;
584 v_rsp
.resid
= cpu_to_vhost32(cmd
->tvc_vq
,
585 se_cmd
->residual_count
);
586 /* TODO is status_qualifier field needed? */
587 v_rsp
.status
= se_cmd
->scsi_status
;
588 v_rsp
.sense_len
= cpu_to_vhost32(cmd
->tvc_vq
,
589 se_cmd
->scsi_sense_length
);
590 memcpy(v_rsp
.sense
, cmd
->tvc_sense_buf
,
591 se_cmd
->scsi_sense_length
);
594 iov_iter_init(&iov_iter
, ITER_DEST
, cmd
->tvc_resp_iov
,
595 cmd
->tvc_in_iovs
, sizeof(v_rsp
));
596 ret
= copy_to_iter(&v_rsp
, sizeof(v_rsp
), &iov_iter
);
597 if (likely(ret
== sizeof(v_rsp
))) {
600 vhost_add_used(cmd
->tvc_vq
, cmd
->tvc_vq_desc
, 0);
602 pr_err("Faulted on virtio_scsi_cmd_resp\n");
604 vhost_scsi_release_cmd_res(se_cmd
);
608 vhost_signal(&svq
->vs
->dev
, &svq
->vq
);
611 static struct vhost_scsi_cmd
*
612 vhost_scsi_get_cmd(struct vhost_virtqueue
*vq
, struct vhost_scsi_tpg
*tpg
,
613 unsigned char *cdb
, u64 scsi_tag
, u16 lun
, u8 task_attr
,
614 u32 exp_data_len
, int data_direction
)
616 struct vhost_scsi_virtqueue
*svq
= container_of(vq
,
617 struct vhost_scsi_virtqueue
, vq
);
618 struct vhost_scsi_cmd
*cmd
;
619 struct vhost_scsi_nexus
*tv_nexus
;
620 struct scatterlist
*sg
, *prot_sg
;
621 struct iovec
*tvc_resp_iov
;
625 tv_nexus
= tpg
->tpg_nexus
;
627 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
628 return ERR_PTR(-EIO
);
631 tag
= sbitmap_get(&svq
->scsi_tags
);
633 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
634 return ERR_PTR(-ENOMEM
);
637 cmd
= &svq
->scsi_cmds
[tag
];
639 prot_sg
= cmd
->tvc_prot_sgl
;
640 pages
= cmd
->tvc_upages
;
641 tvc_resp_iov
= cmd
->tvc_resp_iov
;
642 memset(cmd
, 0, sizeof(*cmd
));
644 cmd
->tvc_prot_sgl
= prot_sg
;
645 cmd
->tvc_upages
= pages
;
646 cmd
->tvc_se_cmd
.map_tag
= tag
;
647 cmd
->tvc_tag
= scsi_tag
;
649 cmd
->tvc_task_attr
= task_attr
;
650 cmd
->tvc_exp_data_len
= exp_data_len
;
651 cmd
->tvc_data_direction
= data_direction
;
652 cmd
->tvc_nexus
= tv_nexus
;
653 cmd
->inflight
= vhost_scsi_get_inflight(vq
);
654 cmd
->tvc_resp_iov
= tvc_resp_iov
;
656 memcpy(cmd
->tvc_cdb
, cdb
, VHOST_SCSI_MAX_CDB_SIZE
);
662 * Map a user memory range into a scatterlist
664 * Returns the number of scatterlist entries used or -errno on error.
667 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd
*cmd
,
668 struct iov_iter
*iter
,
669 struct scatterlist
*sgl
,
672 struct page
**pages
= cmd
->tvc_upages
;
673 struct scatterlist
*sg
= sgl
;
674 ssize_t bytes
, mapped_bytes
;
675 size_t offset
, mapped_offset
;
676 unsigned int npages
= 0;
678 bytes
= iov_iter_get_pages2(iter
, pages
, LONG_MAX
,
679 VHOST_SCSI_PREALLOC_UPAGES
, &offset
);
680 /* No pages were pinned */
682 return bytes
< 0 ? bytes
: -EFAULT
;
684 mapped_bytes
= bytes
;
685 mapped_offset
= offset
;
688 unsigned n
= min_t(unsigned, PAGE_SIZE
- offset
, bytes
);
690 * The block layer requires bios/requests to be a multiple of
691 * 512 bytes, but Windows can send us vecs that are misaligned.
692 * This can result in bios and later requests with misaligned
693 * sizes if we have to break up a cmd/scatterlist into multiple
696 * We currently only break up a command into multiple bios if
697 * we hit the vec/seg limit, so check if our sgl_count is
698 * greater than the max and if a vec in the cmd has a
699 * misaligned offset/size.
702 (offset
& (SECTOR_SIZE
- 1) || n
& (SECTOR_SIZE
- 1)) &&
703 cmd
->tvc_sgl_count
> BIO_MAX_VECS
) {
705 "vhost-scsi detected misaligned IO. Performance may be degraded.");
706 goto revert_iter_get_pages
;
709 sg_set_page(sg
++, pages
[npages
++], n
, offset
);
716 revert_iter_get_pages
:
717 iov_iter_revert(iter
, mapped_bytes
);
720 while (mapped_bytes
) {
721 unsigned int n
= min_t(unsigned int, PAGE_SIZE
- mapped_offset
,
724 put_page(pages
[npages
++]);
734 vhost_scsi_calc_sgls(struct iov_iter
*iter
, size_t bytes
, int max_sgls
)
738 if (!iter
|| !iter_iov(iter
)) {
739 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
740 " present\n", __func__
, bytes
);
744 sgl_count
= iov_iter_npages(iter
, 0xffff);
745 if (sgl_count
> max_sgls
) {
746 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
747 " max_sgls: %d\n", __func__
, sgl_count
, max_sgls
);
754 vhost_scsi_copy_iov_to_sgl(struct vhost_scsi_cmd
*cmd
, struct iov_iter
*iter
,
755 struct scatterlist
*sg
, int sg_count
)
757 size_t len
= iov_iter_count(iter
);
758 unsigned int nbytes
= 0;
762 if (cmd
->tvc_data_direction
== DMA_FROM_DEVICE
) {
763 cmd
->saved_iter_addr
= dup_iter(&cmd
->saved_iter
, iter
,
765 if (!cmd
->saved_iter_addr
)
769 for (i
= 0; i
< sg_count
; i
++) {
770 page
= alloc_page(GFP_KERNEL
);
776 nbytes
= min_t(unsigned int, PAGE_SIZE
, len
);
777 sg_set_page(&sg
[i
], page
, nbytes
, 0);
779 if (cmd
->tvc_data_direction
== DMA_TO_DEVICE
&&
780 copy_page_from_iter(page
, 0, nbytes
, iter
) != nbytes
)
790 pr_err("Could not read %u bytes while handling misaligned cmd\n",
794 __free_page(sg_page(&sg
[i
]));
795 kfree(cmd
->saved_iter_addr
);
800 vhost_scsi_map_iov_to_sgl(struct vhost_scsi_cmd
*cmd
, struct iov_iter
*iter
,
801 struct scatterlist
*sg
, int sg_count
, bool is_prot
)
803 struct scatterlist
*p
= sg
;
807 while (iov_iter_count(iter
)) {
808 ret
= vhost_scsi_map_to_sgl(cmd
, iter
, sg
, is_prot
);
813 struct page
*page
= sg_page(p
);
817 revert_bytes
+= p
->length
;
822 iov_iter_revert(iter
, revert_bytes
);
832 vhost_scsi_mapal(struct vhost_scsi_cmd
*cmd
,
833 size_t prot_bytes
, struct iov_iter
*prot_iter
,
834 size_t data_bytes
, struct iov_iter
*data_iter
)
839 sgl_count
= vhost_scsi_calc_sgls(prot_iter
, prot_bytes
,
840 VHOST_SCSI_PREALLOC_PROT_SGLS
);
844 sg_init_table(cmd
->tvc_prot_sgl
, sgl_count
);
845 cmd
->tvc_prot_sgl_count
= sgl_count
;
846 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__
,
847 cmd
->tvc_prot_sgl
, cmd
->tvc_prot_sgl_count
);
849 ret
= vhost_scsi_map_iov_to_sgl(cmd
, prot_iter
,
851 cmd
->tvc_prot_sgl_count
, true);
853 cmd
->tvc_prot_sgl_count
= 0;
857 sgl_count
= vhost_scsi_calc_sgls(data_iter
, data_bytes
,
858 VHOST_SCSI_PREALLOC_SGLS
);
862 sg_init_table(cmd
->tvc_sgl
, sgl_count
);
863 cmd
->tvc_sgl_count
= sgl_count
;
864 pr_debug("%s data_sg %p data_sgl_count %u\n", __func__
,
865 cmd
->tvc_sgl
, cmd
->tvc_sgl_count
);
867 ret
= vhost_scsi_map_iov_to_sgl(cmd
, data_iter
, cmd
->tvc_sgl
,
868 cmd
->tvc_sgl_count
, false);
869 if (ret
== -EINVAL
) {
870 sg_init_table(cmd
->tvc_sgl
, cmd
->tvc_sgl_count
);
871 ret
= vhost_scsi_copy_iov_to_sgl(cmd
, data_iter
, cmd
->tvc_sgl
,
876 cmd
->tvc_sgl_count
= 0;
882 static int vhost_scsi_to_tcm_attr(int attr
)
885 case VIRTIO_SCSI_S_SIMPLE
:
886 return TCM_SIMPLE_TAG
;
887 case VIRTIO_SCSI_S_ORDERED
:
888 return TCM_ORDERED_TAG
;
889 case VIRTIO_SCSI_S_HEAD
:
891 case VIRTIO_SCSI_S_ACA
:
896 return TCM_SIMPLE_TAG
;
899 static void vhost_scsi_target_queue_cmd(struct vhost_scsi_cmd
*cmd
)
901 struct se_cmd
*se_cmd
= &cmd
->tvc_se_cmd
;
902 struct vhost_scsi_nexus
*tv_nexus
;
903 struct scatterlist
*sg_ptr
, *sg_prot_ptr
= NULL
;
905 /* FIXME: BIDI operation */
906 if (cmd
->tvc_sgl_count
) {
907 sg_ptr
= cmd
->tvc_sgl
;
909 if (cmd
->tvc_prot_sgl_count
)
910 sg_prot_ptr
= cmd
->tvc_prot_sgl
;
912 se_cmd
->prot_pto
= true;
916 tv_nexus
= cmd
->tvc_nexus
;
919 target_init_cmd(se_cmd
, tv_nexus
->tvn_se_sess
, &cmd
->tvc_sense_buf
[0],
920 cmd
->tvc_lun
, cmd
->tvc_exp_data_len
,
921 vhost_scsi_to_tcm_attr(cmd
->tvc_task_attr
),
922 cmd
->tvc_data_direction
, TARGET_SCF_ACK_KREF
);
924 if (target_submit_prep(se_cmd
, cmd
->tvc_cdb
, sg_ptr
,
925 cmd
->tvc_sgl_count
, NULL
, 0, sg_prot_ptr
,
926 cmd
->tvc_prot_sgl_count
, GFP_KERNEL
))
929 target_submit(se_cmd
);
933 vhost_scsi_send_bad_target(struct vhost_scsi
*vs
,
934 struct vhost_virtqueue
*vq
,
935 int head
, unsigned out
)
937 struct virtio_scsi_cmd_resp __user
*resp
;
938 struct virtio_scsi_cmd_resp rsp
;
941 memset(&rsp
, 0, sizeof(rsp
));
942 rsp
.response
= VIRTIO_SCSI_S_BAD_TARGET
;
943 resp
= vq
->iov
[out
].iov_base
;
944 ret
= __copy_to_user(resp
, &rsp
, sizeof(rsp
));
946 vhost_add_used_and_signal(&vs
->dev
, vq
, head
, 0);
948 pr_err("Faulted on virtio_scsi_cmd_resp\n");
952 vhost_scsi_get_desc(struct vhost_scsi
*vs
, struct vhost_virtqueue
*vq
,
953 struct vhost_scsi_ctx
*vc
)
957 vc
->head
= vhost_get_vq_desc(vq
, vq
->iov
,
958 ARRAY_SIZE(vq
->iov
), &vc
->out
, &vc
->in
,
961 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
962 vc
->head
, vc
->out
, vc
->in
);
964 /* On error, stop handling until the next kick. */
965 if (unlikely(vc
->head
< 0))
968 /* Nothing new? Wait for eventfd to tell us they refilled. */
969 if (vc
->head
== vq
->num
) {
970 if (unlikely(vhost_enable_notify(&vs
->dev
, vq
))) {
971 vhost_disable_notify(&vs
->dev
, vq
);
978 * Get the size of request and response buffers.
979 * FIXME: Not correct for BIDI operation
981 vc
->out_size
= iov_length(vq
->iov
, vc
->out
);
982 vc
->in_size
= iov_length(&vq
->iov
[vc
->out
], vc
->in
);
985 * Copy over the virtio-scsi request header, which for a
986 * ANY_LAYOUT enabled guest may span multiple iovecs, or a
987 * single iovec may contain both the header + outgoing
990 * copy_from_iter() will advance out_iter, so that it will
991 * point at the start of the outgoing WRITE payload, if
992 * DMA_TO_DEVICE is set.
994 iov_iter_init(&vc
->out_iter
, ITER_SOURCE
, vq
->iov
, vc
->out
, vc
->out_size
);
1002 vhost_scsi_chk_size(struct vhost_virtqueue
*vq
, struct vhost_scsi_ctx
*vc
)
1004 if (unlikely(vc
->in_size
< vc
->rsp_size
)) {
1006 "Response buf too small, need min %zu bytes got %zu",
1007 vc
->rsp_size
, vc
->in_size
);
1009 } else if (unlikely(vc
->out_size
< vc
->req_size
)) {
1011 "Request buf too small, need min %zu bytes got %zu",
1012 vc
->req_size
, vc
->out_size
);
1020 vhost_scsi_get_req(struct vhost_virtqueue
*vq
, struct vhost_scsi_ctx
*vc
,
1021 struct vhost_scsi_tpg
**tpgp
)
1025 if (unlikely(!copy_from_iter_full(vc
->req
, vc
->req_size
,
1027 vq_err(vq
, "Faulted on copy_from_iter_full\n");
1028 } else if (unlikely(*vc
->lunp
!= 1)) {
1029 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
1030 vq_err(vq
, "Illegal virtio-scsi lun: %u\n", *vc
->lunp
);
1032 struct vhost_scsi_tpg
**vs_tpg
, *tpg
;
1034 vs_tpg
= vhost_vq_get_backend(vq
); /* validated at handler entry */
1036 tpg
= READ_ONCE(vs_tpg
[*vc
->target
]);
1037 if (unlikely(!tpg
)) {
1038 vq_err(vq
, "Target 0x%x does not exist\n", *vc
->target
);
1049 static u16
vhost_buf_to_lun(u8
*lun_buf
)
1051 return ((lun_buf
[2] << 8) | lun_buf
[3]) & 0x3FFF;
1055 vhost_scsi_handle_vq(struct vhost_scsi
*vs
, struct vhost_virtqueue
*vq
)
1057 struct vhost_scsi_tpg
**vs_tpg
, *tpg
;
1058 struct virtio_scsi_cmd_req v_req
;
1059 struct virtio_scsi_cmd_req_pi v_req_pi
;
1060 struct vhost_scsi_ctx vc
;
1061 struct vhost_scsi_cmd
*cmd
;
1062 struct iov_iter in_iter
, prot_iter
, data_iter
;
1064 u32 exp_data_len
, data_direction
;
1065 int ret
, prot_bytes
, i
, c
= 0;
1068 bool t10_pi
= vhost_has_feature(vq
, VIRTIO_SCSI_F_T10_PI
);
1071 mutex_lock(&vq
->mutex
);
1073 * We can handle the vq only after the endpoint is setup by calling the
1074 * VHOST_SCSI_SET_ENDPOINT ioctl.
1076 vs_tpg
= vhost_vq_get_backend(vq
);
1080 memset(&vc
, 0, sizeof(vc
));
1081 vc
.rsp_size
= sizeof(struct virtio_scsi_cmd_resp
);
1083 vhost_disable_notify(&vs
->dev
, vq
);
1086 ret
= vhost_scsi_get_desc(vs
, vq
, &vc
);
1091 * Setup pointers and values based upon different virtio-scsi
1092 * request header if T10_PI is enabled in KVM guest.
1096 vc
.req_size
= sizeof(v_req_pi
);
1097 vc
.lunp
= &v_req_pi
.lun
[0];
1098 vc
.target
= &v_req_pi
.lun
[1];
1101 vc
.req_size
= sizeof(v_req
);
1102 vc
.lunp
= &v_req
.lun
[0];
1103 vc
.target
= &v_req
.lun
[1];
1107 * Validate the size of request and response buffers.
1108 * Check for a sane response buffer so we can report
1109 * early errors back to the guest.
1111 ret
= vhost_scsi_chk_size(vq
, &vc
);
1115 ret
= vhost_scsi_get_req(vq
, &vc
, &tpg
);
1119 ret
= -EIO
; /* bad target on any error from here on */
1122 * Determine data_direction by calculating the total outgoing
1123 * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
1124 * response headers respectively.
1126 * For DMA_TO_DEVICE this is out_iter, which is already pointing
1127 * to the right place.
1129 * For DMA_FROM_DEVICE, the iovec will be just past the end
1130 * of the virtio-scsi response header in either the same
1131 * or immediately following iovec.
1133 * Any associated T10_PI bytes for the outgoing / incoming
1134 * payloads are included in calculation of exp_data_len here.
1138 if (vc
.out_size
> vc
.req_size
) {
1139 data_direction
= DMA_TO_DEVICE
;
1140 exp_data_len
= vc
.out_size
- vc
.req_size
;
1141 data_iter
= vc
.out_iter
;
1142 } else if (vc
.in_size
> vc
.rsp_size
) {
1143 data_direction
= DMA_FROM_DEVICE
;
1144 exp_data_len
= vc
.in_size
- vc
.rsp_size
;
1146 iov_iter_init(&in_iter
, ITER_DEST
, &vq
->iov
[vc
.out
], vc
.in
,
1147 vc
.rsp_size
+ exp_data_len
);
1148 iov_iter_advance(&in_iter
, vc
.rsp_size
);
1149 data_iter
= in_iter
;
1151 data_direction
= DMA_NONE
;
1155 * If T10_PI header + payload is present, setup prot_iter values
1156 * and recalculate data_iter for vhost_scsi_mapal() mapping to
1157 * host scatterlists via get_user_pages_fast().
1160 if (v_req_pi
.pi_bytesout
) {
1161 if (data_direction
!= DMA_TO_DEVICE
) {
1162 vq_err(vq
, "Received non zero pi_bytesout,"
1163 " but wrong data_direction\n");
1166 prot_bytes
= vhost32_to_cpu(vq
, v_req_pi
.pi_bytesout
);
1167 } else if (v_req_pi
.pi_bytesin
) {
1168 if (data_direction
!= DMA_FROM_DEVICE
) {
1169 vq_err(vq
, "Received non zero pi_bytesin,"
1170 " but wrong data_direction\n");
1173 prot_bytes
= vhost32_to_cpu(vq
, v_req_pi
.pi_bytesin
);
1176 * Set prot_iter to data_iter and truncate it to
1177 * prot_bytes, and advance data_iter past any
1178 * preceding prot_bytes that may be present.
1180 * Also fix up the exp_data_len to reflect only the
1181 * actual data payload length.
1184 exp_data_len
-= prot_bytes
;
1185 prot_iter
= data_iter
;
1186 iov_iter_truncate(&prot_iter
, prot_bytes
);
1187 iov_iter_advance(&data_iter
, prot_bytes
);
1189 tag
= vhost64_to_cpu(vq
, v_req_pi
.tag
);
1190 task_attr
= v_req_pi
.task_attr
;
1191 cdb
= &v_req_pi
.cdb
[0];
1192 lun
= vhost_buf_to_lun(v_req_pi
.lun
);
1194 tag
= vhost64_to_cpu(vq
, v_req
.tag
);
1195 task_attr
= v_req
.task_attr
;
1196 cdb
= &v_req
.cdb
[0];
1197 lun
= vhost_buf_to_lun(v_req
.lun
);
1200 * Check that the received CDB size does not exceeded our
1201 * hardcoded max for vhost-scsi, then get a pre-allocated
1202 * cmd descriptor for the new virtio-scsi tag.
1204 * TODO what if cdb was too small for varlen cdb header?
1206 if (unlikely(scsi_command_size(cdb
) > VHOST_SCSI_MAX_CDB_SIZE
)) {
1207 vq_err(vq
, "Received SCSI CDB with command_size: %d that"
1208 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1209 scsi_command_size(cdb
), VHOST_SCSI_MAX_CDB_SIZE
);
1212 cmd
= vhost_scsi_get_cmd(vq
, tpg
, cdb
, tag
, lun
, task_attr
,
1213 exp_data_len
+ prot_bytes
,
1216 vq_err(vq
, "vhost_scsi_get_cmd failed %ld\n",
1220 cmd
->tvc_vhost
= vs
;
1222 for (i
= 0; i
< vc
.in
; i
++)
1223 cmd
->tvc_resp_iov
[i
] = vq
->iov
[vc
.out
+ i
];
1224 cmd
->tvc_in_iovs
= vc
.in
;
1226 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1227 cmd
->tvc_cdb
[0], cmd
->tvc_lun
);
1228 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1229 " %d\n", cmd
, exp_data_len
, prot_bytes
, data_direction
);
1231 if (data_direction
!= DMA_NONE
) {
1232 if (unlikely(vhost_scsi_mapal(cmd
, prot_bytes
,
1233 &prot_iter
, exp_data_len
,
1235 vq_err(vq
, "Failed to map iov to sgl\n");
1236 vhost_scsi_release_cmd_res(&cmd
->tvc_se_cmd
);
1241 * Save the descriptor from vhost_get_vq_desc() to be used to
1242 * complete the virtio-scsi request in TCM callback context via
1243 * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1245 cmd
->tvc_vq_desc
= vc
.head
;
1246 vhost_scsi_target_queue_cmd(cmd
);
1250 * ENXIO: No more requests, or read error, wait for next kick
1251 * EINVAL: Invalid response buffer, drop the request
1252 * EIO: Respond with bad target
1253 * EAGAIN: Pending request
1257 else if (ret
== -EIO
)
1258 vhost_scsi_send_bad_target(vs
, vq
, vc
.head
, vc
.out
);
1259 } while (likely(!vhost_exceeds_weight(vq
, ++c
, 0)));
1261 mutex_unlock(&vq
->mutex
);
1265 vhost_scsi_send_tmf_resp(struct vhost_scsi
*vs
, struct vhost_virtqueue
*vq
,
1266 int in_iovs
, int vq_desc
, struct iovec
*resp_iov
,
1269 struct virtio_scsi_ctrl_tmf_resp rsp
;
1270 struct iov_iter iov_iter
;
1273 pr_debug("%s\n", __func__
);
1274 memset(&rsp
, 0, sizeof(rsp
));
1275 rsp
.response
= tmf_resp_code
;
1277 iov_iter_init(&iov_iter
, ITER_DEST
, resp_iov
, in_iovs
, sizeof(rsp
));
1279 ret
= copy_to_iter(&rsp
, sizeof(rsp
), &iov_iter
);
1280 if (likely(ret
== sizeof(rsp
)))
1281 vhost_add_used_and_signal(&vs
->dev
, vq
, vq_desc
, 0);
1283 pr_err("Faulted on virtio_scsi_ctrl_tmf_resp\n");
1286 static void vhost_scsi_tmf_resp_work(struct vhost_work
*work
)
1288 struct vhost_scsi_tmf
*tmf
= container_of(work
, struct vhost_scsi_tmf
,
1292 if (tmf
->scsi_resp
== TMR_FUNCTION_COMPLETE
)
1293 resp_code
= VIRTIO_SCSI_S_FUNCTION_SUCCEEDED
;
1295 resp_code
= VIRTIO_SCSI_S_FUNCTION_REJECTED
;
1297 vhost_scsi_send_tmf_resp(tmf
->vhost
, &tmf
->svq
->vq
, tmf
->in_iovs
,
1298 tmf
->vq_desc
, &tmf
->resp_iov
, resp_code
);
1299 vhost_scsi_release_tmf_res(tmf
);
1302 static void vhost_scsi_tmf_flush_work(struct work_struct
*work
)
1304 struct vhost_scsi_tmf
*tmf
= container_of(work
, struct vhost_scsi_tmf
,
1306 struct vhost_virtqueue
*vq
= &tmf
->svq
->vq
;
1308 * Make sure we have sent responses for other commands before we
1309 * send our response.
1311 vhost_dev_flush(vq
->dev
);
1312 if (!vhost_vq_work_queue(vq
, &tmf
->vwork
))
1313 vhost_scsi_release_tmf_res(tmf
);
1317 vhost_scsi_handle_tmf(struct vhost_scsi
*vs
, struct vhost_scsi_tpg
*tpg
,
1318 struct vhost_virtqueue
*vq
,
1319 struct virtio_scsi_ctrl_tmf_req
*vtmf
,
1320 struct vhost_scsi_ctx
*vc
)
1322 struct vhost_scsi_virtqueue
*svq
= container_of(vq
,
1323 struct vhost_scsi_virtqueue
, vq
);
1324 struct vhost_scsi_tmf
*tmf
;
1326 if (vhost32_to_cpu(vq
, vtmf
->subtype
) !=
1327 VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET
)
1330 if (!tpg
->tpg_nexus
|| !tpg
->tpg_nexus
->tvn_se_sess
) {
1331 pr_err("Unable to locate active struct vhost_scsi_nexus for LUN RESET.\n");
1335 tmf
= kzalloc(sizeof(*tmf
), GFP_KERNEL
);
1339 INIT_WORK(&tmf
->flush_work
, vhost_scsi_tmf_flush_work
);
1340 vhost_work_init(&tmf
->vwork
, vhost_scsi_tmf_resp_work
);
1343 tmf
->resp_iov
= vq
->iov
[vc
->out
];
1344 tmf
->vq_desc
= vc
->head
;
1345 tmf
->in_iovs
= vc
->in
;
1346 tmf
->inflight
= vhost_scsi_get_inflight(vq
);
1348 if (target_submit_tmr(&tmf
->se_cmd
, tpg
->tpg_nexus
->tvn_se_sess
, NULL
,
1349 vhost_buf_to_lun(vtmf
->lun
), NULL
,
1350 TMR_LUN_RESET
, GFP_KERNEL
, 0,
1351 TARGET_SCF_ACK_KREF
) < 0) {
1352 vhost_scsi_release_tmf_res(tmf
);
1359 vhost_scsi_send_tmf_resp(vs
, vq
, vc
->in
, vc
->head
, &vq
->iov
[vc
->out
],
1360 VIRTIO_SCSI_S_FUNCTION_REJECTED
);
1364 vhost_scsi_send_an_resp(struct vhost_scsi
*vs
,
1365 struct vhost_virtqueue
*vq
,
1366 struct vhost_scsi_ctx
*vc
)
1368 struct virtio_scsi_ctrl_an_resp rsp
;
1369 struct iov_iter iov_iter
;
1372 pr_debug("%s\n", __func__
);
1373 memset(&rsp
, 0, sizeof(rsp
)); /* event_actual = 0 */
1374 rsp
.response
= VIRTIO_SCSI_S_OK
;
1376 iov_iter_init(&iov_iter
, ITER_DEST
, &vq
->iov
[vc
->out
], vc
->in
, sizeof(rsp
));
1378 ret
= copy_to_iter(&rsp
, sizeof(rsp
), &iov_iter
);
1379 if (likely(ret
== sizeof(rsp
)))
1380 vhost_add_used_and_signal(&vs
->dev
, vq
, vc
->head
, 0);
1382 pr_err("Faulted on virtio_scsi_ctrl_an_resp\n");
1386 vhost_scsi_ctl_handle_vq(struct vhost_scsi
*vs
, struct vhost_virtqueue
*vq
)
1388 struct vhost_scsi_tpg
*tpg
;
1391 struct virtio_scsi_ctrl_an_req an
;
1392 struct virtio_scsi_ctrl_tmf_req tmf
;
1394 struct vhost_scsi_ctx vc
;
1398 mutex_lock(&vq
->mutex
);
1400 * We can handle the vq only after the endpoint is setup by calling the
1401 * VHOST_SCSI_SET_ENDPOINT ioctl.
1403 if (!vhost_vq_get_backend(vq
))
1406 memset(&vc
, 0, sizeof(vc
));
1408 vhost_disable_notify(&vs
->dev
, vq
);
1411 ret
= vhost_scsi_get_desc(vs
, vq
, &vc
);
1416 * Get the request type first in order to setup
1417 * other parameters dependent on the type.
1419 vc
.req
= &v_req
.type
;
1420 typ_size
= sizeof(v_req
.type
);
1422 if (unlikely(!copy_from_iter_full(vc
.req
, typ_size
,
1424 vq_err(vq
, "Faulted on copy_from_iter tmf type\n");
1426 * The size of the response buffer depends on the
1427 * request type and must be validated against it.
1428 * Since the request type is not known, don't send
1434 switch (vhost32_to_cpu(vq
, v_req
.type
)) {
1435 case VIRTIO_SCSI_T_TMF
:
1436 vc
.req
= &v_req
.tmf
;
1437 vc
.req_size
= sizeof(struct virtio_scsi_ctrl_tmf_req
);
1438 vc
.rsp_size
= sizeof(struct virtio_scsi_ctrl_tmf_resp
);
1439 vc
.lunp
= &v_req
.tmf
.lun
[0];
1440 vc
.target
= &v_req
.tmf
.lun
[1];
1442 case VIRTIO_SCSI_T_AN_QUERY
:
1443 case VIRTIO_SCSI_T_AN_SUBSCRIBE
:
1445 vc
.req_size
= sizeof(struct virtio_scsi_ctrl_an_req
);
1446 vc
.rsp_size
= sizeof(struct virtio_scsi_ctrl_an_resp
);
1447 vc
.lunp
= &v_req
.an
.lun
[0];
1451 vq_err(vq
, "Unknown control request %d", v_req
.type
);
1456 * Validate the size of request and response buffers.
1457 * Check for a sane response buffer so we can report
1458 * early errors back to the guest.
1460 ret
= vhost_scsi_chk_size(vq
, &vc
);
1465 * Get the rest of the request now that its size is known.
1468 vc
.req_size
-= typ_size
;
1470 ret
= vhost_scsi_get_req(vq
, &vc
, &tpg
);
1474 if (v_req
.type
== VIRTIO_SCSI_T_TMF
)
1475 vhost_scsi_handle_tmf(vs
, tpg
, vq
, &v_req
.tmf
, &vc
);
1477 vhost_scsi_send_an_resp(vs
, vq
, &vc
);
1480 * ENXIO: No more requests, or read error, wait for next kick
1481 * EINVAL: Invalid response buffer, drop the request
1482 * EIO: Respond with bad target
1483 * EAGAIN: Pending request
1487 else if (ret
== -EIO
)
1488 vhost_scsi_send_bad_target(vs
, vq
, vc
.head
, vc
.out
);
1489 } while (likely(!vhost_exceeds_weight(vq
, ++c
, 0)));
1491 mutex_unlock(&vq
->mutex
);
1494 static void vhost_scsi_ctl_handle_kick(struct vhost_work
*work
)
1496 struct vhost_virtqueue
*vq
= container_of(work
, struct vhost_virtqueue
,
1498 struct vhost_scsi
*vs
= container_of(vq
->dev
, struct vhost_scsi
, dev
);
1500 pr_debug("%s: The handling func for control queue.\n", __func__
);
1501 vhost_scsi_ctl_handle_vq(vs
, vq
);
1505 vhost_scsi_send_evt(struct vhost_scsi
*vs
, struct vhost_virtqueue
*vq
,
1506 struct vhost_scsi_tpg
*tpg
, struct se_lun
*lun
,
1507 u32 event
, u32 reason
)
1509 struct vhost_scsi_evt
*evt
;
1511 evt
= vhost_scsi_allocate_evt(vs
, event
, reason
);
1516 /* TODO: share lun setup code with virtio-scsi.ko */
1518 * Note: evt->event is zeroed when we allocate it and
1519 * lun[4-7] need to be zero according to virtio-scsi spec.
1521 evt
->event
.lun
[0] = 0x01;
1522 evt
->event
.lun
[1] = tpg
->tport_tpgt
;
1523 if (lun
->unpacked_lun
>= 256)
1524 evt
->event
.lun
[2] = lun
->unpacked_lun
>> 8 | 0x40 ;
1525 evt
->event
.lun
[3] = lun
->unpacked_lun
& 0xFF;
1528 llist_add(&evt
->list
, &vs
->vs_event_list
);
1529 if (!vhost_vq_work_queue(vq
, &vs
->vs_event_work
))
1530 vhost_scsi_complete_events(vs
, true);
1533 static void vhost_scsi_evt_handle_kick(struct vhost_work
*work
)
1535 struct vhost_virtqueue
*vq
= container_of(work
, struct vhost_virtqueue
,
1537 struct vhost_scsi
*vs
= container_of(vq
->dev
, struct vhost_scsi
, dev
);
1539 mutex_lock(&vq
->mutex
);
1540 if (!vhost_vq_get_backend(vq
))
1543 if (vs
->vs_events_missed
)
1544 vhost_scsi_send_evt(vs
, vq
, NULL
, NULL
, VIRTIO_SCSI_T_NO_EVENT
,
1547 mutex_unlock(&vq
->mutex
);
1550 static void vhost_scsi_handle_kick(struct vhost_work
*work
)
1552 struct vhost_virtqueue
*vq
= container_of(work
, struct vhost_virtqueue
,
1554 struct vhost_scsi
*vs
= container_of(vq
->dev
, struct vhost_scsi
, dev
);
1556 vhost_scsi_handle_vq(vs
, vq
);
1559 /* Callers must hold dev mutex */
1560 static void vhost_scsi_flush(struct vhost_scsi
*vs
)
1564 /* Init new inflight and remember the old inflight */
1565 vhost_scsi_init_inflight(vs
, vs
->old_inflight
);
1568 * The inflight->kref was initialized to 1. We decrement it here to
1569 * indicate the start of the flush operation so that it will reach 0
1570 * when all the reqs are finished.
1572 for (i
= 0; i
< vs
->dev
.nvqs
; i
++)
1573 kref_put(&vs
->old_inflight
[i
]->kref
, vhost_scsi_done_inflight
);
1575 /* Flush both the vhost poll and vhost work */
1576 vhost_dev_flush(&vs
->dev
);
1578 /* Wait for all reqs issued before the flush to be finished */
1579 for (i
= 0; i
< vs
->dev
.nvqs
; i
++)
1580 wait_for_completion(&vs
->old_inflight
[i
]->comp
);
1583 static void vhost_scsi_destroy_vq_cmds(struct vhost_virtqueue
*vq
)
1585 struct vhost_scsi_virtqueue
*svq
= container_of(vq
,
1586 struct vhost_scsi_virtqueue
, vq
);
1587 struct vhost_scsi_cmd
*tv_cmd
;
1590 if (!svq
->scsi_cmds
)
1593 for (i
= 0; i
< svq
->max_cmds
; i
++) {
1594 tv_cmd
= &svq
->scsi_cmds
[i
];
1596 kfree(tv_cmd
->tvc_sgl
);
1597 kfree(tv_cmd
->tvc_prot_sgl
);
1598 kfree(tv_cmd
->tvc_upages
);
1599 kfree(tv_cmd
->tvc_resp_iov
);
1602 sbitmap_free(&svq
->scsi_tags
);
1603 kfree(svq
->scsi_cmds
);
1604 svq
->scsi_cmds
= NULL
;
1607 static int vhost_scsi_setup_vq_cmds(struct vhost_virtqueue
*vq
, int max_cmds
)
1609 struct vhost_scsi_virtqueue
*svq
= container_of(vq
,
1610 struct vhost_scsi_virtqueue
, vq
);
1611 struct vhost_scsi_cmd
*tv_cmd
;
1617 if (sbitmap_init_node(&svq
->scsi_tags
, max_cmds
, -1, GFP_KERNEL
,
1618 NUMA_NO_NODE
, false, true))
1620 svq
->max_cmds
= max_cmds
;
1622 svq
->scsi_cmds
= kcalloc(max_cmds
, sizeof(*tv_cmd
), GFP_KERNEL
);
1623 if (!svq
->scsi_cmds
) {
1624 sbitmap_free(&svq
->scsi_tags
);
1628 for (i
= 0; i
< max_cmds
; i
++) {
1629 tv_cmd
= &svq
->scsi_cmds
[i
];
1631 tv_cmd
->tvc_sgl
= kcalloc(VHOST_SCSI_PREALLOC_SGLS
,
1632 sizeof(struct scatterlist
),
1634 if (!tv_cmd
->tvc_sgl
) {
1635 pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1639 tv_cmd
->tvc_upages
= kcalloc(VHOST_SCSI_PREALLOC_UPAGES
,
1640 sizeof(struct page
*),
1642 if (!tv_cmd
->tvc_upages
) {
1643 pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1647 tv_cmd
->tvc_resp_iov
= kcalloc(UIO_MAXIOV
,
1648 sizeof(struct iovec
),
1650 if (!tv_cmd
->tvc_resp_iov
) {
1651 pr_err("Unable to allocate tv_cmd->tvc_resp_iov\n");
1655 tv_cmd
->tvc_prot_sgl
= kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS
,
1656 sizeof(struct scatterlist
),
1658 if (!tv_cmd
->tvc_prot_sgl
) {
1659 pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1665 vhost_scsi_destroy_vq_cmds(vq
);
1670 * Called from vhost_scsi_ioctl() context to walk the list of available
1671 * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1673 * The lock nesting rule is:
1674 * vs->dev.mutex -> vhost_scsi_mutex -> tpg->tv_tpg_mutex -> vq->mutex
1677 vhost_scsi_set_endpoint(struct vhost_scsi
*vs
,
1678 struct vhost_scsi_target
*t
)
1680 struct se_portal_group
*se_tpg
;
1681 struct vhost_scsi_tport
*tv_tport
;
1682 struct vhost_scsi_tpg
*tpg
;
1683 struct vhost_scsi_tpg
**vs_tpg
;
1684 struct vhost_virtqueue
*vq
;
1685 int index
, ret
, i
, len
;
1688 mutex_lock(&vs
->dev
.mutex
);
1690 /* Verify that ring has been setup correctly. */
1691 for (index
= 0; index
< vs
->dev
.nvqs
; ++index
) {
1692 /* Verify that ring has been setup correctly. */
1693 if (!vhost_vq_access_ok(&vs
->vqs
[index
].vq
)) {
1699 len
= sizeof(vs_tpg
[0]) * VHOST_SCSI_MAX_TARGET
;
1700 vs_tpg
= kzalloc(len
, GFP_KERNEL
);
1706 memcpy(vs_tpg
, vs
->vs_tpg
, len
);
1708 mutex_lock(&vhost_scsi_mutex
);
1709 list_for_each_entry(tpg
, &vhost_scsi_list
, tv_tpg_list
) {
1710 mutex_lock(&tpg
->tv_tpg_mutex
);
1711 if (!tpg
->tpg_nexus
) {
1712 mutex_unlock(&tpg
->tv_tpg_mutex
);
1715 if (tpg
->tv_tpg_vhost_count
!= 0) {
1716 mutex_unlock(&tpg
->tv_tpg_mutex
);
1719 tv_tport
= tpg
->tport
;
1721 if (!strcmp(tv_tport
->tport_name
, t
->vhost_wwpn
)) {
1722 if (vs
->vs_tpg
&& vs
->vs_tpg
[tpg
->tport_tpgt
]) {
1723 mutex_unlock(&tpg
->tv_tpg_mutex
);
1724 mutex_unlock(&vhost_scsi_mutex
);
1729 * In order to ensure individual vhost-scsi configfs
1730 * groups cannot be removed while in use by vhost ioctl,
1731 * go ahead and take an explicit se_tpg->tpg_group.cg_item
1734 se_tpg
= &tpg
->se_tpg
;
1735 ret
= target_depend_item(&se_tpg
->tpg_group
.cg_item
);
1737 pr_warn("target_depend_item() failed: %d\n", ret
);
1738 mutex_unlock(&tpg
->tv_tpg_mutex
);
1739 mutex_unlock(&vhost_scsi_mutex
);
1742 tpg
->tv_tpg_vhost_count
++;
1743 tpg
->vhost_scsi
= vs
;
1744 vs_tpg
[tpg
->tport_tpgt
] = tpg
;
1747 mutex_unlock(&tpg
->tv_tpg_mutex
);
1749 mutex_unlock(&vhost_scsi_mutex
);
1752 memcpy(vs
->vs_vhost_wwpn
, t
->vhost_wwpn
,
1753 sizeof(vs
->vs_vhost_wwpn
));
1755 for (i
= VHOST_SCSI_VQ_IO
; i
< vs
->dev
.nvqs
; i
++) {
1756 vq
= &vs
->vqs
[i
].vq
;
1757 if (!vhost_vq_is_setup(vq
))
1760 ret
= vhost_scsi_setup_vq_cmds(vq
, vq
->num
);
1762 goto destroy_vq_cmds
;
1765 for (i
= 0; i
< vs
->dev
.nvqs
; i
++) {
1766 vq
= &vs
->vqs
[i
].vq
;
1767 mutex_lock(&vq
->mutex
);
1768 vhost_vq_set_backend(vq
, vs_tpg
);
1769 vhost_vq_init_access(vq
);
1770 mutex_unlock(&vq
->mutex
);
1778 * Act as synchronize_rcu to make sure access to
1779 * old vs->vs_tpg is finished.
1781 vhost_scsi_flush(vs
);
1783 vs
->vs_tpg
= vs_tpg
;
1787 for (i
--; i
>= VHOST_SCSI_VQ_IO
; i
--) {
1788 if (!vhost_vq_get_backend(&vs
->vqs
[i
].vq
))
1789 vhost_scsi_destroy_vq_cmds(&vs
->vqs
[i
].vq
);
1792 for (i
= 0; i
< VHOST_SCSI_MAX_TARGET
; i
++) {
1795 mutex_lock(&tpg
->tv_tpg_mutex
);
1796 tpg
->vhost_scsi
= NULL
;
1797 tpg
->tv_tpg_vhost_count
--;
1798 mutex_unlock(&tpg
->tv_tpg_mutex
);
1799 target_undepend_item(&tpg
->se_tpg
.tpg_group
.cg_item
);
1804 mutex_unlock(&vs
->dev
.mutex
);
1809 vhost_scsi_clear_endpoint(struct vhost_scsi
*vs
,
1810 struct vhost_scsi_target
*t
)
1812 struct se_portal_group
*se_tpg
;
1813 struct vhost_scsi_tport
*tv_tport
;
1814 struct vhost_scsi_tpg
*tpg
;
1815 struct vhost_virtqueue
*vq
;
1820 mutex_lock(&vs
->dev
.mutex
);
1821 /* Verify that ring has been setup correctly. */
1822 for (index
= 0; index
< vs
->dev
.nvqs
; ++index
) {
1823 if (!vhost_vq_access_ok(&vs
->vqs
[index
].vq
)) {
1834 for (i
= 0; i
< VHOST_SCSI_MAX_TARGET
; i
++) {
1836 tpg
= vs
->vs_tpg
[target
];
1840 tv_tport
= tpg
->tport
;
1846 if (strcmp(tv_tport
->tport_name
, t
->vhost_wwpn
)) {
1847 pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1848 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1849 tv_tport
->tport_name
, tpg
->tport_tpgt
,
1850 t
->vhost_wwpn
, t
->vhost_tpgt
);
1859 /* Prevent new cmds from starting and accessing the tpgs/sessions */
1860 for (i
= 0; i
< vs
->dev
.nvqs
; i
++) {
1861 vq
= &vs
->vqs
[i
].vq
;
1862 mutex_lock(&vq
->mutex
);
1863 vhost_vq_set_backend(vq
, NULL
);
1864 mutex_unlock(&vq
->mutex
);
1866 /* Make sure cmds are not running before tearing them down. */
1867 vhost_scsi_flush(vs
);
1869 for (i
= 0; i
< vs
->dev
.nvqs
; i
++) {
1870 vq
= &vs
->vqs
[i
].vq
;
1871 vhost_scsi_destroy_vq_cmds(vq
);
1875 * We can now release our hold on the tpg and sessions and userspace
1876 * can free them after this point.
1878 for (i
= 0; i
< VHOST_SCSI_MAX_TARGET
; i
++) {
1880 tpg
= vs
->vs_tpg
[target
];
1884 mutex_lock(&tpg
->tv_tpg_mutex
);
1886 tpg
->tv_tpg_vhost_count
--;
1887 tpg
->vhost_scsi
= NULL
;
1888 vs
->vs_tpg
[target
] = NULL
;
1890 mutex_unlock(&tpg
->tv_tpg_mutex
);
1892 se_tpg
= &tpg
->se_tpg
;
1893 target_undepend_item(&se_tpg
->tpg_group
.cg_item
);
1898 * Act as synchronize_rcu to make sure access to
1899 * old vs->vs_tpg is finished.
1901 vhost_scsi_flush(vs
);
1904 WARN_ON(vs
->vs_events_nr
);
1905 mutex_unlock(&vs
->dev
.mutex
);
1909 mutex_unlock(&vs
->dev
.mutex
);
1913 static int vhost_scsi_set_features(struct vhost_scsi
*vs
, u64 features
)
1915 struct vhost_virtqueue
*vq
;
1918 if (features
& ~VHOST_SCSI_FEATURES
)
1921 mutex_lock(&vs
->dev
.mutex
);
1922 if ((features
& (1 << VHOST_F_LOG_ALL
)) &&
1923 !vhost_log_access_ok(&vs
->dev
)) {
1924 mutex_unlock(&vs
->dev
.mutex
);
1928 for (i
= 0; i
< vs
->dev
.nvqs
; i
++) {
1929 vq
= &vs
->vqs
[i
].vq
;
1930 mutex_lock(&vq
->mutex
);
1931 vq
->acked_features
= features
;
1932 mutex_unlock(&vq
->mutex
);
1934 mutex_unlock(&vs
->dev
.mutex
);
1938 static int vhost_scsi_open(struct inode
*inode
, struct file
*f
)
1940 struct vhost_scsi_virtqueue
*svq
;
1941 struct vhost_scsi
*vs
;
1942 struct vhost_virtqueue
**vqs
;
1943 int r
= -ENOMEM
, i
, nvqs
= vhost_scsi_max_io_vqs
;
1945 vs
= kvzalloc(sizeof(*vs
), GFP_KERNEL
);
1949 if (nvqs
> VHOST_SCSI_MAX_IO_VQ
) {
1950 pr_err("Invalid max_io_vqs of %d. Using %d.\n", nvqs
,
1951 VHOST_SCSI_MAX_IO_VQ
);
1952 nvqs
= VHOST_SCSI_MAX_IO_VQ
;
1953 } else if (nvqs
== 0) {
1954 pr_err("Invalid max_io_vqs of %d. Using 1.\n", nvqs
);
1957 nvqs
+= VHOST_SCSI_VQ_IO
;
1959 vs
->old_inflight
= kmalloc_array(nvqs
, sizeof(*vs
->old_inflight
),
1960 GFP_KERNEL
| __GFP_ZERO
);
1961 if (!vs
->old_inflight
)
1964 vs
->vqs
= kmalloc_array(nvqs
, sizeof(*vs
->vqs
),
1965 GFP_KERNEL
| __GFP_ZERO
);
1969 vqs
= kmalloc_array(nvqs
, sizeof(*vqs
), GFP_KERNEL
);
1973 vhost_work_init(&vs
->vs_event_work
, vhost_scsi_evt_work
);
1975 vs
->vs_events_nr
= 0;
1976 vs
->vs_events_missed
= false;
1978 vqs
[VHOST_SCSI_VQ_CTL
] = &vs
->vqs
[VHOST_SCSI_VQ_CTL
].vq
;
1979 vqs
[VHOST_SCSI_VQ_EVT
] = &vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
;
1980 vs
->vqs
[VHOST_SCSI_VQ_CTL
].vq
.handle_kick
= vhost_scsi_ctl_handle_kick
;
1981 vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
.handle_kick
= vhost_scsi_evt_handle_kick
;
1982 for (i
= VHOST_SCSI_VQ_IO
; i
< nvqs
; i
++) {
1987 init_llist_head(&svq
->completion_list
);
1988 vhost_work_init(&svq
->completion_work
,
1989 vhost_scsi_complete_cmd_work
);
1990 svq
->vq
.handle_kick
= vhost_scsi_handle_kick
;
1992 vhost_dev_init(&vs
->dev
, vqs
, nvqs
, UIO_MAXIOV
,
1993 VHOST_SCSI_WEIGHT
, 0, true, NULL
);
1995 vhost_scsi_init_inflight(vs
, NULL
);
1997 f
->private_data
= vs
;
2003 kfree(vs
->old_inflight
);
2010 static int vhost_scsi_release(struct inode
*inode
, struct file
*f
)
2012 struct vhost_scsi
*vs
= f
->private_data
;
2013 struct vhost_scsi_target t
;
2015 mutex_lock(&vs
->dev
.mutex
);
2016 memcpy(t
.vhost_wwpn
, vs
->vs_vhost_wwpn
, sizeof(t
.vhost_wwpn
));
2017 mutex_unlock(&vs
->dev
.mutex
);
2018 vhost_scsi_clear_endpoint(vs
, &t
);
2019 vhost_dev_stop(&vs
->dev
);
2020 vhost_dev_cleanup(&vs
->dev
);
2023 kfree(vs
->old_inflight
);
2029 vhost_scsi_ioctl(struct file
*f
,
2033 struct vhost_scsi
*vs
= f
->private_data
;
2034 struct vhost_scsi_target backend
;
2035 void __user
*argp
= (void __user
*)arg
;
2036 u64 __user
*featurep
= argp
;
2037 u32 __user
*eventsp
= argp
;
2040 int r
, abi_version
= VHOST_SCSI_ABI_VERSION
;
2041 struct vhost_virtqueue
*vq
= &vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
;
2044 case VHOST_SCSI_SET_ENDPOINT
:
2045 if (copy_from_user(&backend
, argp
, sizeof backend
))
2047 if (backend
.reserved
!= 0)
2050 return vhost_scsi_set_endpoint(vs
, &backend
);
2051 case VHOST_SCSI_CLEAR_ENDPOINT
:
2052 if (copy_from_user(&backend
, argp
, sizeof backend
))
2054 if (backend
.reserved
!= 0)
2057 return vhost_scsi_clear_endpoint(vs
, &backend
);
2058 case VHOST_SCSI_GET_ABI_VERSION
:
2059 if (copy_to_user(argp
, &abi_version
, sizeof abi_version
))
2062 case VHOST_SCSI_SET_EVENTS_MISSED
:
2063 if (get_user(events_missed
, eventsp
))
2065 mutex_lock(&vq
->mutex
);
2066 vs
->vs_events_missed
= events_missed
;
2067 mutex_unlock(&vq
->mutex
);
2069 case VHOST_SCSI_GET_EVENTS_MISSED
:
2070 mutex_lock(&vq
->mutex
);
2071 events_missed
= vs
->vs_events_missed
;
2072 mutex_unlock(&vq
->mutex
);
2073 if (put_user(events_missed
, eventsp
))
2076 case VHOST_GET_FEATURES
:
2077 features
= VHOST_SCSI_FEATURES
;
2078 if (copy_to_user(featurep
, &features
, sizeof features
))
2081 case VHOST_SET_FEATURES
:
2082 if (copy_from_user(&features
, featurep
, sizeof features
))
2084 return vhost_scsi_set_features(vs
, features
);
2085 case VHOST_NEW_WORKER
:
2086 case VHOST_FREE_WORKER
:
2087 case VHOST_ATTACH_VRING_WORKER
:
2088 case VHOST_GET_VRING_WORKER
:
2089 mutex_lock(&vs
->dev
.mutex
);
2090 r
= vhost_worker_ioctl(&vs
->dev
, ioctl
, argp
);
2091 mutex_unlock(&vs
->dev
.mutex
);
2094 mutex_lock(&vs
->dev
.mutex
);
2095 r
= vhost_dev_ioctl(&vs
->dev
, ioctl
, argp
);
2096 /* TODO: flush backend after dev ioctl. */
2097 if (r
== -ENOIOCTLCMD
)
2098 r
= vhost_vring_ioctl(&vs
->dev
, ioctl
, argp
);
2099 mutex_unlock(&vs
->dev
.mutex
);
2104 static const struct file_operations vhost_scsi_fops
= {
2105 .owner
= THIS_MODULE
,
2106 .release
= vhost_scsi_release
,
2107 .unlocked_ioctl
= vhost_scsi_ioctl
,
2108 .compat_ioctl
= compat_ptr_ioctl
,
2109 .open
= vhost_scsi_open
,
2110 .llseek
= noop_llseek
,
2113 static struct miscdevice vhost_scsi_misc
= {
2119 static int __init
vhost_scsi_register(void)
2121 return misc_register(&vhost_scsi_misc
);
2124 static void vhost_scsi_deregister(void)
2126 misc_deregister(&vhost_scsi_misc
);
2129 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport
*tport
)
2131 switch (tport
->tport_proto_id
) {
2132 case SCSI_PROTOCOL_SAS
:
2134 case SCSI_PROTOCOL_FCP
:
2136 case SCSI_PROTOCOL_ISCSI
:
2146 vhost_scsi_do_plug(struct vhost_scsi_tpg
*tpg
,
2147 struct se_lun
*lun
, bool plug
)
2150 struct vhost_scsi
*vs
= tpg
->vhost_scsi
;
2151 struct vhost_virtqueue
*vq
;
2158 reason
= VIRTIO_SCSI_EVT_RESET_RESCAN
;
2160 reason
= VIRTIO_SCSI_EVT_RESET_REMOVED
;
2162 vq
= &vs
->vqs
[VHOST_SCSI_VQ_EVT
].vq
;
2163 mutex_lock(&vq
->mutex
);
2165 * We can't queue events if the backend has been cleared, because
2166 * we could end up queueing an event after the flush.
2168 if (!vhost_vq_get_backend(vq
))
2171 if (vhost_has_feature(vq
, VIRTIO_SCSI_F_HOTPLUG
))
2172 vhost_scsi_send_evt(vs
, vq
, tpg
, lun
,
2173 VIRTIO_SCSI_T_TRANSPORT_RESET
, reason
);
2175 mutex_unlock(&vq
->mutex
);
2178 static void vhost_scsi_hotplug(struct vhost_scsi_tpg
*tpg
, struct se_lun
*lun
)
2180 vhost_scsi_do_plug(tpg
, lun
, true);
2183 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg
*tpg
, struct se_lun
*lun
)
2185 vhost_scsi_do_plug(tpg
, lun
, false);
2188 static int vhost_scsi_port_link(struct se_portal_group
*se_tpg
,
2191 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2192 struct vhost_scsi_tpg
, se_tpg
);
2194 mutex_lock(&tpg
->tv_tpg_mutex
);
2195 tpg
->tv_tpg_port_count
++;
2196 vhost_scsi_hotplug(tpg
, lun
);
2197 mutex_unlock(&tpg
->tv_tpg_mutex
);
2202 static void vhost_scsi_port_unlink(struct se_portal_group
*se_tpg
,
2205 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2206 struct vhost_scsi_tpg
, se_tpg
);
2208 mutex_lock(&tpg
->tv_tpg_mutex
);
2209 tpg
->tv_tpg_port_count
--;
2210 vhost_scsi_hotunplug(tpg
, lun
);
2211 mutex_unlock(&tpg
->tv_tpg_mutex
);
2214 static ssize_t
vhost_scsi_tpg_attrib_fabric_prot_type_store(
2215 struct config_item
*item
, const char *page
, size_t count
)
2217 struct se_portal_group
*se_tpg
= attrib_to_tpg(item
);
2218 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2219 struct vhost_scsi_tpg
, se_tpg
);
2221 int ret
= kstrtoul(page
, 0, &val
);
2224 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret
);
2227 if (val
!= 0 && val
!= 1 && val
!= 3) {
2228 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val
);
2231 tpg
->tv_fabric_prot_type
= val
;
2236 static ssize_t
vhost_scsi_tpg_attrib_fabric_prot_type_show(
2237 struct config_item
*item
, char *page
)
2239 struct se_portal_group
*se_tpg
= attrib_to_tpg(item
);
2240 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2241 struct vhost_scsi_tpg
, se_tpg
);
2243 return sysfs_emit(page
, "%d\n", tpg
->tv_fabric_prot_type
);
2246 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_
, fabric_prot_type
);
2248 static struct configfs_attribute
*vhost_scsi_tpg_attrib_attrs
[] = {
2249 &vhost_scsi_tpg_attrib_attr_fabric_prot_type
,
2253 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg
*tpg
,
2256 struct vhost_scsi_nexus
*tv_nexus
;
2258 mutex_lock(&tpg
->tv_tpg_mutex
);
2259 if (tpg
->tpg_nexus
) {
2260 mutex_unlock(&tpg
->tv_tpg_mutex
);
2261 pr_debug("tpg->tpg_nexus already exists\n");
2265 tv_nexus
= kzalloc(sizeof(*tv_nexus
), GFP_KERNEL
);
2267 mutex_unlock(&tpg
->tv_tpg_mutex
);
2268 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
2272 * Since we are running in 'demo mode' this call with generate a
2273 * struct se_node_acl for the vhost_scsi struct se_portal_group with
2274 * the SCSI Initiator port name of the passed configfs group 'name'.
2276 tv_nexus
->tvn_se_sess
= target_setup_session(&tpg
->se_tpg
, 0, 0,
2277 TARGET_PROT_DIN_PASS
| TARGET_PROT_DOUT_PASS
,
2278 (unsigned char *)name
, tv_nexus
, NULL
);
2279 if (IS_ERR(tv_nexus
->tvn_se_sess
)) {
2280 mutex_unlock(&tpg
->tv_tpg_mutex
);
2284 tpg
->tpg_nexus
= tv_nexus
;
2286 mutex_unlock(&tpg
->tv_tpg_mutex
);
2290 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg
*tpg
)
2292 struct se_session
*se_sess
;
2293 struct vhost_scsi_nexus
*tv_nexus
;
2295 mutex_lock(&tpg
->tv_tpg_mutex
);
2296 tv_nexus
= tpg
->tpg_nexus
;
2298 mutex_unlock(&tpg
->tv_tpg_mutex
);
2302 se_sess
= tv_nexus
->tvn_se_sess
;
2304 mutex_unlock(&tpg
->tv_tpg_mutex
);
2308 if (tpg
->tv_tpg_port_count
!= 0) {
2309 mutex_unlock(&tpg
->tv_tpg_mutex
);
2310 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2311 " active TPG port count: %d\n",
2312 tpg
->tv_tpg_port_count
);
2316 if (tpg
->tv_tpg_vhost_count
!= 0) {
2317 mutex_unlock(&tpg
->tv_tpg_mutex
);
2318 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2319 " active TPG vhost count: %d\n",
2320 tpg
->tv_tpg_vhost_count
);
2324 pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
2325 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg
->tport
),
2326 tv_nexus
->tvn_se_sess
->se_node_acl
->initiatorname
);
2329 * Release the SCSI I_T Nexus to the emulated vhost Target Port
2331 target_remove_session(se_sess
);
2332 tpg
->tpg_nexus
= NULL
;
2333 mutex_unlock(&tpg
->tv_tpg_mutex
);
2339 static ssize_t
vhost_scsi_tpg_nexus_show(struct config_item
*item
, char *page
)
2341 struct se_portal_group
*se_tpg
= to_tpg(item
);
2342 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2343 struct vhost_scsi_tpg
, se_tpg
);
2344 struct vhost_scsi_nexus
*tv_nexus
;
2347 mutex_lock(&tpg
->tv_tpg_mutex
);
2348 tv_nexus
= tpg
->tpg_nexus
;
2350 mutex_unlock(&tpg
->tv_tpg_mutex
);
2353 ret
= sysfs_emit(page
, "%s\n",
2354 tv_nexus
->tvn_se_sess
->se_node_acl
->initiatorname
);
2355 mutex_unlock(&tpg
->tv_tpg_mutex
);
2360 static ssize_t
vhost_scsi_tpg_nexus_store(struct config_item
*item
,
2361 const char *page
, size_t count
)
2363 struct se_portal_group
*se_tpg
= to_tpg(item
);
2364 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2365 struct vhost_scsi_tpg
, se_tpg
);
2366 struct vhost_scsi_tport
*tport_wwn
= tpg
->tport
;
2367 unsigned char i_port
[VHOST_SCSI_NAMELEN
], *ptr
, *port_ptr
;
2370 * Shutdown the active I_T nexus if 'NULL' is passed..
2372 if (!strncmp(page
, "NULL", 4)) {
2373 ret
= vhost_scsi_drop_nexus(tpg
);
2374 return (!ret
) ? count
: ret
;
2377 * Otherwise make sure the passed virtual Initiator port WWN matches
2378 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
2379 * vhost_scsi_make_nexus().
2381 if (strlen(page
) >= VHOST_SCSI_NAMELEN
) {
2382 pr_err("Emulated NAA Sas Address: %s, exceeds"
2383 " max: %d\n", page
, VHOST_SCSI_NAMELEN
);
2386 snprintf(&i_port
[0], VHOST_SCSI_NAMELEN
, "%s", page
);
2388 ptr
= strstr(i_port
, "naa.");
2390 if (tport_wwn
->tport_proto_id
!= SCSI_PROTOCOL_SAS
) {
2391 pr_err("Passed SAS Initiator Port %s does not"
2392 " match target port protoid: %s\n", i_port
,
2393 vhost_scsi_dump_proto_id(tport_wwn
));
2396 port_ptr
= &i_port
[0];
2399 ptr
= strstr(i_port
, "fc.");
2401 if (tport_wwn
->tport_proto_id
!= SCSI_PROTOCOL_FCP
) {
2402 pr_err("Passed FCP Initiator Port %s does not"
2403 " match target port protoid: %s\n", i_port
,
2404 vhost_scsi_dump_proto_id(tport_wwn
));
2407 port_ptr
= &i_port
[3]; /* Skip over "fc." */
2410 ptr
= strstr(i_port
, "iqn.");
2412 if (tport_wwn
->tport_proto_id
!= SCSI_PROTOCOL_ISCSI
) {
2413 pr_err("Passed iSCSI Initiator Port %s does not"
2414 " match target port protoid: %s\n", i_port
,
2415 vhost_scsi_dump_proto_id(tport_wwn
));
2418 port_ptr
= &i_port
[0];
2421 pr_err("Unable to locate prefix for emulated Initiator Port:"
2425 * Clear any trailing newline for the NAA WWN
2428 if (i_port
[strlen(i_port
)-1] == '\n')
2429 i_port
[strlen(i_port
)-1] = '\0';
2431 ret
= vhost_scsi_make_nexus(tpg
, port_ptr
);
2438 CONFIGFS_ATTR(vhost_scsi_tpg_
, nexus
);
2440 static struct configfs_attribute
*vhost_scsi_tpg_attrs
[] = {
2441 &vhost_scsi_tpg_attr_nexus
,
2445 static struct se_portal_group
*
2446 vhost_scsi_make_tpg(struct se_wwn
*wwn
, const char *name
)
2448 struct vhost_scsi_tport
*tport
= container_of(wwn
,
2449 struct vhost_scsi_tport
, tport_wwn
);
2451 struct vhost_scsi_tpg
*tpg
;
2455 if (strstr(name
, "tpgt_") != name
)
2456 return ERR_PTR(-EINVAL
);
2457 if (kstrtou16(name
+ 5, 10, &tpgt
) || tpgt
>= VHOST_SCSI_MAX_TARGET
)
2458 return ERR_PTR(-EINVAL
);
2460 tpg
= kzalloc(sizeof(*tpg
), GFP_KERNEL
);
2462 pr_err("Unable to allocate struct vhost_scsi_tpg");
2463 return ERR_PTR(-ENOMEM
);
2465 mutex_init(&tpg
->tv_tpg_mutex
);
2466 INIT_LIST_HEAD(&tpg
->tv_tpg_list
);
2468 tpg
->tport_tpgt
= tpgt
;
2470 ret
= core_tpg_register(wwn
, &tpg
->se_tpg
, tport
->tport_proto_id
);
2475 mutex_lock(&vhost_scsi_mutex
);
2476 list_add_tail(&tpg
->tv_tpg_list
, &vhost_scsi_list
);
2477 mutex_unlock(&vhost_scsi_mutex
);
2479 return &tpg
->se_tpg
;
2482 static void vhost_scsi_drop_tpg(struct se_portal_group
*se_tpg
)
2484 struct vhost_scsi_tpg
*tpg
= container_of(se_tpg
,
2485 struct vhost_scsi_tpg
, se_tpg
);
2487 mutex_lock(&vhost_scsi_mutex
);
2488 list_del(&tpg
->tv_tpg_list
);
2489 mutex_unlock(&vhost_scsi_mutex
);
2491 * Release the virtual I_T Nexus for this vhost TPG
2493 vhost_scsi_drop_nexus(tpg
);
2495 * Deregister the se_tpg from TCM..
2497 core_tpg_deregister(se_tpg
);
2501 static struct se_wwn
*
2502 vhost_scsi_make_tport(struct target_fabric_configfs
*tf
,
2503 struct config_group
*group
,
2506 struct vhost_scsi_tport
*tport
;
2511 /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2512 return ERR_PTR(-EINVAL); */
2514 tport
= kzalloc(sizeof(*tport
), GFP_KERNEL
);
2516 pr_err("Unable to allocate struct vhost_scsi_tport");
2517 return ERR_PTR(-ENOMEM
);
2519 tport
->tport_wwpn
= wwpn
;
2521 * Determine the emulated Protocol Identifier and Target Port Name
2522 * based on the incoming configfs directory name.
2524 ptr
= strstr(name
, "naa.");
2526 tport
->tport_proto_id
= SCSI_PROTOCOL_SAS
;
2529 ptr
= strstr(name
, "fc.");
2531 tport
->tport_proto_id
= SCSI_PROTOCOL_FCP
;
2532 off
= 3; /* Skip over "fc." */
2535 ptr
= strstr(name
, "iqn.");
2537 tport
->tport_proto_id
= SCSI_PROTOCOL_ISCSI
;
2541 pr_err("Unable to locate prefix for emulated Target Port:"
2544 return ERR_PTR(-EINVAL
);
2547 if (strlen(name
) >= VHOST_SCSI_NAMELEN
) {
2548 pr_err("Emulated %s Address: %s, exceeds"
2549 " max: %d\n", name
, vhost_scsi_dump_proto_id(tport
),
2550 VHOST_SCSI_NAMELEN
);
2552 return ERR_PTR(-EINVAL
);
2554 snprintf(&tport
->tport_name
[0], VHOST_SCSI_NAMELEN
, "%s", &name
[off
]);
2556 pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2557 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport
), name
);
2559 return &tport
->tport_wwn
;
2562 static void vhost_scsi_drop_tport(struct se_wwn
*wwn
)
2564 struct vhost_scsi_tport
*tport
= container_of(wwn
,
2565 struct vhost_scsi_tport
, tport_wwn
);
2567 pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2568 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport
),
2575 vhost_scsi_wwn_version_show(struct config_item
*item
, char *page
)
2577 return sysfs_emit(page
, "TCM_VHOST fabric module %s on %s/%s"
2578 "on "UTS_RELEASE
"\n", VHOST_SCSI_VERSION
, utsname()->sysname
,
2579 utsname()->machine
);
2582 CONFIGFS_ATTR_RO(vhost_scsi_wwn_
, version
);
2584 static struct configfs_attribute
*vhost_scsi_wwn_attrs
[] = {
2585 &vhost_scsi_wwn_attr_version
,
2589 static const struct target_core_fabric_ops vhost_scsi_ops
= {
2590 .module
= THIS_MODULE
,
2591 .fabric_name
= "vhost",
2592 .max_data_sg_nents
= VHOST_SCSI_PREALLOC_SGLS
,
2593 .tpg_get_wwn
= vhost_scsi_get_fabric_wwn
,
2594 .tpg_get_tag
= vhost_scsi_get_tpgt
,
2595 .tpg_check_demo_mode
= vhost_scsi_check_true
,
2596 .tpg_check_demo_mode_cache
= vhost_scsi_check_true
,
2597 .tpg_check_prot_fabric_only
= vhost_scsi_check_prot_fabric_only
,
2598 .release_cmd
= vhost_scsi_release_cmd
,
2599 .check_stop_free
= vhost_scsi_check_stop_free
,
2600 .sess_get_initiator_sid
= NULL
,
2601 .write_pending
= vhost_scsi_write_pending
,
2602 .queue_data_in
= vhost_scsi_queue_data_in
,
2603 .queue_status
= vhost_scsi_queue_status
,
2604 .queue_tm_rsp
= vhost_scsi_queue_tm_rsp
,
2605 .aborted_task
= vhost_scsi_aborted_task
,
2607 * Setup callers for generic logic in target_core_fabric_configfs.c
2609 .fabric_make_wwn
= vhost_scsi_make_tport
,
2610 .fabric_drop_wwn
= vhost_scsi_drop_tport
,
2611 .fabric_make_tpg
= vhost_scsi_make_tpg
,
2612 .fabric_drop_tpg
= vhost_scsi_drop_tpg
,
2613 .fabric_post_link
= vhost_scsi_port_link
,
2614 .fabric_pre_unlink
= vhost_scsi_port_unlink
,
2616 .tfc_wwn_attrs
= vhost_scsi_wwn_attrs
,
2617 .tfc_tpg_base_attrs
= vhost_scsi_tpg_attrs
,
2618 .tfc_tpg_attrib_attrs
= vhost_scsi_tpg_attrib_attrs
,
2620 .default_submit_type
= TARGET_QUEUE_SUBMIT
,
2621 .direct_submit_supp
= 1,
2624 static int __init
vhost_scsi_init(void)
2628 pr_debug("TCM_VHOST fabric module %s on %s/%s"
2629 " on "UTS_RELEASE
"\n", VHOST_SCSI_VERSION
, utsname()->sysname
,
2630 utsname()->machine
);
2632 ret
= vhost_scsi_register();
2636 ret
= target_register_template(&vhost_scsi_ops
);
2638 goto out_vhost_scsi_deregister
;
2642 out_vhost_scsi_deregister
:
2643 vhost_scsi_deregister();
2648 static void vhost_scsi_exit(void)
2650 target_unregister_template(&vhost_scsi_ops
);
2651 vhost_scsi_deregister();
2654 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2655 MODULE_ALIAS("tcm_vhost");
2656 MODULE_LICENSE("GPL");
2657 module_init(vhost_scsi_init
);
2658 module_exit(vhost_scsi_exit
);