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 vhost_scsi *vhost;
214 struct vhost_scsi_virtqueue *svq;
216 struct se_cmd se_cmd;
218 struct vhost_scsi_inflight *inflight;
219 struct iovec resp_iov;
225 * Context for processing request and control queue operations.
227 struct vhost_scsi_ctx {
229 unsigned int out, in;
230 size_t req_size, rsp_size;
231 size_t out_size, in_size;
234 struct iov_iter out_iter;
238 * Global mutex to protect vhost_scsi TPG list for vhost IOCTLs and LIO
239 * configfs management operations.
241 static DEFINE_MUTEX(vhost_scsi_mutex);
242 static LIST_HEAD(vhost_scsi_list);
244 static void vhost_scsi_done_inflight(struct kref *kref)
246 struct vhost_scsi_inflight *inflight;
248 inflight = container_of(kref, struct vhost_scsi_inflight, kref);
249 complete(&inflight->comp);
252 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
253 struct vhost_scsi_inflight *old_inflight[])
255 struct vhost_scsi_inflight *new_inflight;
256 struct vhost_virtqueue *vq;
259 for (i = 0; i < vs->dev.nvqs; i++) {
262 mutex_lock(&vq->mutex);
264 /* store old infight */
265 idx = vs->vqs[i].inflight_idx;
267 old_inflight[i] = &vs->vqs[i].inflights[idx];
269 /* setup new infight */
270 vs->vqs[i].inflight_idx = idx ^ 1;
271 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
272 kref_init(&new_inflight->kref);
273 init_completion(&new_inflight->comp);
275 mutex_unlock(&vq->mutex);
279 static struct vhost_scsi_inflight *
280 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
282 struct vhost_scsi_inflight *inflight;
283 struct vhost_scsi_virtqueue *svq;
285 svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
286 inflight = &svq->inflights[svq->inflight_idx];
287 kref_get(&inflight->kref);
292 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
294 kref_put(&inflight->kref, vhost_scsi_done_inflight);
297 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
302 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
304 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
305 struct vhost_scsi_tpg, se_tpg);
306 struct vhost_scsi_tport *tport = tpg->tport;
308 return &tport->tport_name[0];
311 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
313 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
314 struct vhost_scsi_tpg, se_tpg);
315 return tpg->tport_tpgt;
318 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
320 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
321 struct vhost_scsi_tpg, se_tpg);
323 return tpg->tv_fabric_prot_type;
326 static void vhost_scsi_release_cmd_res(struct se_cmd *se_cmd)
328 struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
329 struct vhost_scsi_cmd, tvc_se_cmd);
330 struct vhost_scsi_virtqueue *svq = container_of(tv_cmd->tvc_vq,
331 struct vhost_scsi_virtqueue, vq);
332 struct vhost_scsi_inflight *inflight = tv_cmd->inflight;
335 if (tv_cmd->tvc_sgl_count) {
336 for (i = 0; i < tv_cmd->tvc_sgl_count; i++) {
337 if (tv_cmd->copied_iov)
338 __free_page(sg_page(&tv_cmd->tvc_sgl[i]));
340 put_page(sg_page(&tv_cmd->tvc_sgl[i]));
342 kfree(tv_cmd->saved_iter_addr);
344 if (tv_cmd->tvc_prot_sgl_count) {
345 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
346 put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
349 sbitmap_clear_bit(&svq->scsi_tags, se_cmd->map_tag);
350 vhost_scsi_put_inflight(inflight);
353 static void vhost_scsi_release_tmf_res(struct vhost_scsi_tmf *tmf)
355 struct vhost_scsi_inflight *inflight = tmf->inflight;
358 vhost_scsi_put_inflight(inflight);
361 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
363 if (se_cmd->se_cmd_flags & SCF_SCSI_TMR_CDB) {
364 struct vhost_scsi_tmf *tmf = container_of(se_cmd,
365 struct vhost_scsi_tmf, se_cmd);
366 struct vhost_virtqueue *vq = &tmf->svq->vq;
368 vhost_vq_work_queue(vq, &tmf->vwork);
370 struct vhost_scsi_cmd *cmd = container_of(se_cmd,
371 struct vhost_scsi_cmd, tvc_se_cmd);
372 struct vhost_scsi_virtqueue *svq = container_of(cmd->tvc_vq,
373 struct vhost_scsi_virtqueue, vq);
375 llist_add(&cmd->tvc_completion_list, &svq->completion_list);
376 vhost_vq_work_queue(&svq->vq, &svq->completion_work);
380 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
382 /* Go ahead and process the write immediately */
383 target_execute_cmd(se_cmd);
387 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
389 transport_generic_free_cmd(se_cmd, 0);
393 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
395 transport_generic_free_cmd(se_cmd, 0);
399 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
401 struct vhost_scsi_tmf *tmf = container_of(se_cmd, struct vhost_scsi_tmf,
404 tmf->scsi_resp = se_cmd->se_tmr_req->response;
405 transport_generic_free_cmd(&tmf->se_cmd, 0);
408 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
413 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
419 static struct vhost_scsi_evt *
420 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
421 u32 event, u32 reason)
423 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
424 struct vhost_scsi_evt *evt;
426 if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
427 vs->vs_events_missed = true;
431 evt = kzalloc(sizeof(*evt), GFP_KERNEL);
433 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
434 vs->vs_events_missed = true;
438 evt->event.event = cpu_to_vhost32(vq, event);
439 evt->event.reason = cpu_to_vhost32(vq, reason);
445 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
447 return target_put_sess_cmd(se_cmd);
451 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
453 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
454 struct virtio_scsi_event *event = &evt->event;
455 struct virtio_scsi_event __user *eventp;
459 if (!vhost_vq_get_backend(vq)) {
460 vs->vs_events_missed = true;
465 vhost_disable_notify(&vs->dev, vq);
466 head = vhost_get_vq_desc(vq, vq->iov,
467 ARRAY_SIZE(vq->iov), &out, &in,
470 vs->vs_events_missed = true;
473 if (head == vq->num) {
474 if (vhost_enable_notify(&vs->dev, vq))
476 vs->vs_events_missed = true;
480 if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
481 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
482 vq->iov[out].iov_len);
483 vs->vs_events_missed = true;
487 if (vs->vs_events_missed) {
488 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
489 vs->vs_events_missed = false;
492 eventp = vq->iov[out].iov_base;
493 ret = __copy_to_user(eventp, event, sizeof(*event));
495 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
497 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
500 static void vhost_scsi_evt_work(struct vhost_work *work)
502 struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
504 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
505 struct vhost_scsi_evt *evt, *t;
506 struct llist_node *llnode;
508 mutex_lock(&vq->mutex);
509 llnode = llist_del_all(&vs->vs_event_list);
510 llist_for_each_entry_safe(evt, t, llnode, list) {
511 vhost_scsi_do_evt_work(vs, evt);
512 vhost_scsi_free_evt(vs, evt);
514 mutex_unlock(&vq->mutex);
517 static int vhost_scsi_copy_sgl_to_iov(struct vhost_scsi_cmd *cmd)
519 struct iov_iter *iter = &cmd->saved_iter;
520 struct scatterlist *sg = cmd->tvc_sgl;
525 for (i = 0; i < cmd->tvc_sgl_count; i++) {
526 page = sg_page(&sg[i]);
529 if (copy_page_to_iter(page, 0, len, iter) != len) {
530 pr_err("Could not copy data while handling misaligned cmd. Error %zu\n",
539 /* Fill in status and signal that we are done processing this command
541 * This is scheduled in the vhost work queue so we are called with the owner
542 * process mm and can access the vring.
544 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
546 struct vhost_scsi_virtqueue *svq = container_of(work,
547 struct vhost_scsi_virtqueue, completion_work);
548 struct virtio_scsi_cmd_resp v_rsp;
549 struct vhost_scsi_cmd *cmd, *t;
550 struct llist_node *llnode;
551 struct se_cmd *se_cmd;
552 struct iov_iter iov_iter;
556 llnode = llist_del_all(&svq->completion_list);
557 llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
558 se_cmd = &cmd->tvc_se_cmd;
560 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
561 cmd, se_cmd->residual_count, se_cmd->scsi_status);
562 memset(&v_rsp, 0, sizeof(v_rsp));
564 if (cmd->saved_iter_addr && vhost_scsi_copy_sgl_to_iov(cmd)) {
565 v_rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
567 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq,
568 se_cmd->residual_count);
569 /* TODO is status_qualifier field needed? */
570 v_rsp.status = se_cmd->scsi_status;
571 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
572 se_cmd->scsi_sense_length);
573 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
574 se_cmd->scsi_sense_length);
577 iov_iter_init(&iov_iter, ITER_DEST, cmd->tvc_resp_iov,
578 cmd->tvc_in_iovs, sizeof(v_rsp));
579 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
580 if (likely(ret == sizeof(v_rsp))) {
583 vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
585 pr_err("Faulted on virtio_scsi_cmd_resp\n");
587 vhost_scsi_release_cmd_res(se_cmd);
591 vhost_signal(&svq->vs->dev, &svq->vq);
594 static struct vhost_scsi_cmd *
595 vhost_scsi_get_cmd(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
596 unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
597 u32 exp_data_len, int data_direction)
599 struct vhost_scsi_virtqueue *svq = container_of(vq,
600 struct vhost_scsi_virtqueue, vq);
601 struct vhost_scsi_cmd *cmd;
602 struct vhost_scsi_nexus *tv_nexus;
603 struct scatterlist *sg, *prot_sg;
604 struct iovec *tvc_resp_iov;
608 tv_nexus = tpg->tpg_nexus;
610 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
611 return ERR_PTR(-EIO);
614 tag = sbitmap_get(&svq->scsi_tags);
616 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
617 return ERR_PTR(-ENOMEM);
620 cmd = &svq->scsi_cmds[tag];
622 prot_sg = cmd->tvc_prot_sgl;
623 pages = cmd->tvc_upages;
624 tvc_resp_iov = cmd->tvc_resp_iov;
625 memset(cmd, 0, sizeof(*cmd));
627 cmd->tvc_prot_sgl = prot_sg;
628 cmd->tvc_upages = pages;
629 cmd->tvc_se_cmd.map_tag = tag;
630 cmd->tvc_tag = scsi_tag;
632 cmd->tvc_task_attr = task_attr;
633 cmd->tvc_exp_data_len = exp_data_len;
634 cmd->tvc_data_direction = data_direction;
635 cmd->tvc_nexus = tv_nexus;
636 cmd->inflight = vhost_scsi_get_inflight(vq);
637 cmd->tvc_resp_iov = tvc_resp_iov;
639 memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
645 * Map a user memory range into a scatterlist
647 * Returns the number of scatterlist entries used or -errno on error.
650 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
651 struct iov_iter *iter,
652 struct scatterlist *sgl,
655 struct page **pages = cmd->tvc_upages;
656 struct scatterlist *sg = sgl;
657 ssize_t bytes, mapped_bytes;
658 size_t offset, mapped_offset;
659 unsigned int npages = 0;
661 bytes = iov_iter_get_pages2(iter, pages, LONG_MAX,
662 VHOST_SCSI_PREALLOC_UPAGES, &offset);
663 /* No pages were pinned */
665 return bytes < 0 ? bytes : -EFAULT;
667 mapped_bytes = bytes;
668 mapped_offset = offset;
671 unsigned n = min_t(unsigned, PAGE_SIZE - offset, bytes);
673 * The block layer requires bios/requests to be a multiple of
674 * 512 bytes, but Windows can send us vecs that are misaligned.
675 * This can result in bios and later requests with misaligned
676 * sizes if we have to break up a cmd/scatterlist into multiple
679 * We currently only break up a command into multiple bios if
680 * we hit the vec/seg limit, so check if our sgl_count is
681 * greater than the max and if a vec in the cmd has a
682 * misaligned offset/size.
685 (offset & (SECTOR_SIZE - 1) || n & (SECTOR_SIZE - 1)) &&
686 cmd->tvc_sgl_count > BIO_MAX_VECS) {
688 "vhost-scsi detected misaligned IO. Performance may be degraded.");
689 goto revert_iter_get_pages;
692 sg_set_page(sg++, pages[npages++], n, offset);
699 revert_iter_get_pages:
700 iov_iter_revert(iter, mapped_bytes);
703 while (mapped_bytes) {
704 unsigned int n = min_t(unsigned int, PAGE_SIZE - mapped_offset,
707 put_page(pages[npages++]);
717 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
721 if (!iter || !iter_iov(iter)) {
722 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
723 " present\n", __func__, bytes);
727 sgl_count = iov_iter_npages(iter, 0xffff);
728 if (sgl_count > max_sgls) {
729 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
730 " max_sgls: %d\n", __func__, sgl_count, max_sgls);
737 vhost_scsi_copy_iov_to_sgl(struct vhost_scsi_cmd *cmd, struct iov_iter *iter,
738 struct scatterlist *sg, int sg_count)
740 size_t len = iov_iter_count(iter);
741 unsigned int nbytes = 0;
745 if (cmd->tvc_data_direction == DMA_FROM_DEVICE) {
746 cmd->saved_iter_addr = dup_iter(&cmd->saved_iter, iter,
748 if (!cmd->saved_iter_addr)
752 for (i = 0; i < sg_count; i++) {
753 page = alloc_page(GFP_KERNEL);
759 nbytes = min_t(unsigned int, PAGE_SIZE, len);
760 sg_set_page(&sg[i], page, nbytes, 0);
762 if (cmd->tvc_data_direction == DMA_TO_DEVICE &&
763 copy_page_from_iter(page, 0, nbytes, iter) != nbytes)
773 pr_err("Could not read %u bytes while handling misaligned cmd\n",
777 __free_page(sg_page(&sg[i]));
778 kfree(cmd->saved_iter_addr);
783 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, struct iov_iter *iter,
784 struct scatterlist *sg, int sg_count, bool is_prot)
786 struct scatterlist *p = sg;
790 while (iov_iter_count(iter)) {
791 ret = vhost_scsi_map_to_sgl(cmd, iter, sg, is_prot);
796 struct page *page = sg_page(p);
800 revert_bytes += p->length;
805 iov_iter_revert(iter, revert_bytes);
815 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
816 size_t prot_bytes, struct iov_iter *prot_iter,
817 size_t data_bytes, struct iov_iter *data_iter)
822 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
823 VHOST_SCSI_PREALLOC_PROT_SGLS);
827 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
828 cmd->tvc_prot_sgl_count = sgl_count;
829 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
830 cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
832 ret = vhost_scsi_iov_to_sgl(cmd, prot_iter, cmd->tvc_prot_sgl,
833 cmd->tvc_prot_sgl_count, true);
835 cmd->tvc_prot_sgl_count = 0;
839 sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
840 VHOST_SCSI_PREALLOC_SGLS);
844 sg_init_table(cmd->tvc_sgl, sgl_count);
845 cmd->tvc_sgl_count = sgl_count;
846 pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
847 cmd->tvc_sgl, cmd->tvc_sgl_count);
849 ret = vhost_scsi_iov_to_sgl(cmd, data_iter, cmd->tvc_sgl,
850 cmd->tvc_sgl_count, false);
851 if (ret == -EINVAL) {
852 sg_init_table(cmd->tvc_sgl, cmd->tvc_sgl_count);
853 ret = vhost_scsi_copy_iov_to_sgl(cmd, data_iter, cmd->tvc_sgl,
858 cmd->tvc_sgl_count = 0;
864 static int vhost_scsi_to_tcm_attr(int attr)
867 case VIRTIO_SCSI_S_SIMPLE:
868 return TCM_SIMPLE_TAG;
869 case VIRTIO_SCSI_S_ORDERED:
870 return TCM_ORDERED_TAG;
871 case VIRTIO_SCSI_S_HEAD:
873 case VIRTIO_SCSI_S_ACA:
878 return TCM_SIMPLE_TAG;
881 static void vhost_scsi_target_queue_cmd(struct vhost_scsi_cmd *cmd)
883 struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
884 struct vhost_scsi_nexus *tv_nexus;
885 struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
887 /* FIXME: BIDI operation */
888 if (cmd->tvc_sgl_count) {
889 sg_ptr = cmd->tvc_sgl;
891 if (cmd->tvc_prot_sgl_count)
892 sg_prot_ptr = cmd->tvc_prot_sgl;
894 se_cmd->prot_pto = true;
898 tv_nexus = cmd->tvc_nexus;
901 target_init_cmd(se_cmd, tv_nexus->tvn_se_sess, &cmd->tvc_sense_buf[0],
902 cmd->tvc_lun, cmd->tvc_exp_data_len,
903 vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
904 cmd->tvc_data_direction, TARGET_SCF_ACK_KREF);
906 if (target_submit_prep(se_cmd, cmd->tvc_cdb, sg_ptr,
907 cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
908 cmd->tvc_prot_sgl_count, GFP_KERNEL))
911 target_queue_submission(se_cmd);
915 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
916 struct vhost_virtqueue *vq,
917 int head, unsigned out)
919 struct virtio_scsi_cmd_resp __user *resp;
920 struct virtio_scsi_cmd_resp rsp;
923 memset(&rsp, 0, sizeof(rsp));
924 rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
925 resp = vq->iov[out].iov_base;
926 ret = __copy_to_user(resp, &rsp, sizeof(rsp));
928 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
930 pr_err("Faulted on virtio_scsi_cmd_resp\n");
934 vhost_scsi_get_desc(struct vhost_scsi *vs, struct vhost_virtqueue *vq,
935 struct vhost_scsi_ctx *vc)
939 vc->head = vhost_get_vq_desc(vq, vq->iov,
940 ARRAY_SIZE(vq->iov), &vc->out, &vc->in,
943 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
944 vc->head, vc->out, vc->in);
946 /* On error, stop handling until the next kick. */
947 if (unlikely(vc->head < 0))
950 /* Nothing new? Wait for eventfd to tell us they refilled. */
951 if (vc->head == vq->num) {
952 if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
953 vhost_disable_notify(&vs->dev, vq);
960 * Get the size of request and response buffers.
961 * FIXME: Not correct for BIDI operation
963 vc->out_size = iov_length(vq->iov, vc->out);
964 vc->in_size = iov_length(&vq->iov[vc->out], vc->in);
967 * Copy over the virtio-scsi request header, which for a
968 * ANY_LAYOUT enabled guest may span multiple iovecs, or a
969 * single iovec may contain both the header + outgoing
972 * copy_from_iter() will advance out_iter, so that it will
973 * point at the start of the outgoing WRITE payload, if
974 * DMA_TO_DEVICE is set.
976 iov_iter_init(&vc->out_iter, ITER_SOURCE, vq->iov, vc->out, vc->out_size);
984 vhost_scsi_chk_size(struct vhost_virtqueue *vq, struct vhost_scsi_ctx *vc)
986 if (unlikely(vc->in_size < vc->rsp_size)) {
988 "Response buf too small, need min %zu bytes got %zu",
989 vc->rsp_size, vc->in_size);
991 } else if (unlikely(vc->out_size < vc->req_size)) {
993 "Request buf too small, need min %zu bytes got %zu",
994 vc->req_size, vc->out_size);
1002 vhost_scsi_get_req(struct vhost_virtqueue *vq, struct vhost_scsi_ctx *vc,
1003 struct vhost_scsi_tpg **tpgp)
1007 if (unlikely(!copy_from_iter_full(vc->req, vc->req_size,
1009 vq_err(vq, "Faulted on copy_from_iter_full\n");
1010 } else if (unlikely(*vc->lunp != 1)) {
1011 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
1012 vq_err(vq, "Illegal virtio-scsi lun: %u\n", *vc->lunp);
1014 struct vhost_scsi_tpg **vs_tpg, *tpg;
1016 vs_tpg = vhost_vq_get_backend(vq); /* validated at handler entry */
1018 tpg = READ_ONCE(vs_tpg[*vc->target]);
1019 if (unlikely(!tpg)) {
1020 vq_err(vq, "Target 0x%x does not exist\n", *vc->target);
1031 static u16 vhost_buf_to_lun(u8 *lun_buf)
1033 return ((lun_buf[2] << 8) | lun_buf[3]) & 0x3FFF;
1037 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
1039 struct vhost_scsi_tpg **vs_tpg, *tpg;
1040 struct virtio_scsi_cmd_req v_req;
1041 struct virtio_scsi_cmd_req_pi v_req_pi;
1042 struct vhost_scsi_ctx vc;
1043 struct vhost_scsi_cmd *cmd;
1044 struct iov_iter in_iter, prot_iter, data_iter;
1046 u32 exp_data_len, data_direction;
1047 int ret, prot_bytes, i, c = 0;
1050 bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
1053 mutex_lock(&vq->mutex);
1055 * We can handle the vq only after the endpoint is setup by calling the
1056 * VHOST_SCSI_SET_ENDPOINT ioctl.
1058 vs_tpg = vhost_vq_get_backend(vq);
1062 memset(&vc, 0, sizeof(vc));
1063 vc.rsp_size = sizeof(struct virtio_scsi_cmd_resp);
1065 vhost_disable_notify(&vs->dev, vq);
1068 ret = vhost_scsi_get_desc(vs, vq, &vc);
1073 * Setup pointers and values based upon different virtio-scsi
1074 * request header if T10_PI is enabled in KVM guest.
1078 vc.req_size = sizeof(v_req_pi);
1079 vc.lunp = &v_req_pi.lun[0];
1080 vc.target = &v_req_pi.lun[1];
1083 vc.req_size = sizeof(v_req);
1084 vc.lunp = &v_req.lun[0];
1085 vc.target = &v_req.lun[1];
1089 * Validate the size of request and response buffers.
1090 * Check for a sane response buffer so we can report
1091 * early errors back to the guest.
1093 ret = vhost_scsi_chk_size(vq, &vc);
1097 ret = vhost_scsi_get_req(vq, &vc, &tpg);
1101 ret = -EIO; /* bad target on any error from here on */
1104 * Determine data_direction by calculating the total outgoing
1105 * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
1106 * response headers respectively.
1108 * For DMA_TO_DEVICE this is out_iter, which is already pointing
1109 * to the right place.
1111 * For DMA_FROM_DEVICE, the iovec will be just past the end
1112 * of the virtio-scsi response header in either the same
1113 * or immediately following iovec.
1115 * Any associated T10_PI bytes for the outgoing / incoming
1116 * payloads are included in calculation of exp_data_len here.
1120 if (vc.out_size > vc.req_size) {
1121 data_direction = DMA_TO_DEVICE;
1122 exp_data_len = vc.out_size - vc.req_size;
1123 data_iter = vc.out_iter;
1124 } else if (vc.in_size > vc.rsp_size) {
1125 data_direction = DMA_FROM_DEVICE;
1126 exp_data_len = vc.in_size - vc.rsp_size;
1128 iov_iter_init(&in_iter, ITER_DEST, &vq->iov[vc.out], vc.in,
1129 vc.rsp_size + exp_data_len);
1130 iov_iter_advance(&in_iter, vc.rsp_size);
1131 data_iter = in_iter;
1133 data_direction = DMA_NONE;
1137 * If T10_PI header + payload is present, setup prot_iter values
1138 * and recalculate data_iter for vhost_scsi_mapal() mapping to
1139 * host scatterlists via get_user_pages_fast().
1142 if (v_req_pi.pi_bytesout) {
1143 if (data_direction != DMA_TO_DEVICE) {
1144 vq_err(vq, "Received non zero pi_bytesout,"
1145 " but wrong data_direction\n");
1148 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
1149 } else if (v_req_pi.pi_bytesin) {
1150 if (data_direction != DMA_FROM_DEVICE) {
1151 vq_err(vq, "Received non zero pi_bytesin,"
1152 " but wrong data_direction\n");
1155 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1158 * Set prot_iter to data_iter and truncate it to
1159 * prot_bytes, and advance data_iter past any
1160 * preceeding prot_bytes that may be present.
1162 * Also fix up the exp_data_len to reflect only the
1163 * actual data payload length.
1166 exp_data_len -= prot_bytes;
1167 prot_iter = data_iter;
1168 iov_iter_truncate(&prot_iter, prot_bytes);
1169 iov_iter_advance(&data_iter, prot_bytes);
1171 tag = vhost64_to_cpu(vq, v_req_pi.tag);
1172 task_attr = v_req_pi.task_attr;
1173 cdb = &v_req_pi.cdb[0];
1174 lun = vhost_buf_to_lun(v_req_pi.lun);
1176 tag = vhost64_to_cpu(vq, v_req.tag);
1177 task_attr = v_req.task_attr;
1178 cdb = &v_req.cdb[0];
1179 lun = vhost_buf_to_lun(v_req.lun);
1182 * Check that the received CDB size does not exceeded our
1183 * hardcoded max for vhost-scsi, then get a pre-allocated
1184 * cmd descriptor for the new virtio-scsi tag.
1186 * TODO what if cdb was too small for varlen cdb header?
1188 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1189 vq_err(vq, "Received SCSI CDB with command_size: %d that"
1190 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1191 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1194 cmd = vhost_scsi_get_cmd(vq, tpg, cdb, tag, lun, task_attr,
1195 exp_data_len + prot_bytes,
1198 vq_err(vq, "vhost_scsi_get_cmd failed %ld\n",
1202 cmd->tvc_vhost = vs;
1204 for (i = 0; i < vc.in ; i++)
1205 cmd->tvc_resp_iov[i] = vq->iov[vc.out + i];
1206 cmd->tvc_in_iovs = vc.in;
1208 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1209 cmd->tvc_cdb[0], cmd->tvc_lun);
1210 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1211 " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1213 if (data_direction != DMA_NONE) {
1214 if (unlikely(vhost_scsi_mapal(cmd, prot_bytes,
1215 &prot_iter, exp_data_len,
1217 vq_err(vq, "Failed to map iov to sgl\n");
1218 vhost_scsi_release_cmd_res(&cmd->tvc_se_cmd);
1223 * Save the descriptor from vhost_get_vq_desc() to be used to
1224 * complete the virtio-scsi request in TCM callback context via
1225 * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1227 cmd->tvc_vq_desc = vc.head;
1228 vhost_scsi_target_queue_cmd(cmd);
1232 * ENXIO: No more requests, or read error, wait for next kick
1233 * EINVAL: Invalid response buffer, drop the request
1234 * EIO: Respond with bad target
1235 * EAGAIN: Pending request
1239 else if (ret == -EIO)
1240 vhost_scsi_send_bad_target(vs, vq, vc.head, vc.out);
1241 } while (likely(!vhost_exceeds_weight(vq, ++c, 0)));
1243 mutex_unlock(&vq->mutex);
1247 vhost_scsi_send_tmf_resp(struct vhost_scsi *vs, struct vhost_virtqueue *vq,
1248 int in_iovs, int vq_desc, struct iovec *resp_iov,
1251 struct virtio_scsi_ctrl_tmf_resp rsp;
1252 struct iov_iter iov_iter;
1255 pr_debug("%s\n", __func__);
1256 memset(&rsp, 0, sizeof(rsp));
1257 rsp.response = tmf_resp_code;
1259 iov_iter_init(&iov_iter, ITER_DEST, resp_iov, in_iovs, sizeof(rsp));
1261 ret = copy_to_iter(&rsp, sizeof(rsp), &iov_iter);
1262 if (likely(ret == sizeof(rsp)))
1263 vhost_add_used_and_signal(&vs->dev, vq, vq_desc, 0);
1265 pr_err("Faulted on virtio_scsi_ctrl_tmf_resp\n");
1268 static void vhost_scsi_tmf_resp_work(struct vhost_work *work)
1270 struct vhost_scsi_tmf *tmf = container_of(work, struct vhost_scsi_tmf,
1272 struct vhost_virtqueue *ctl_vq, *vq;
1275 if (tmf->scsi_resp == TMR_FUNCTION_COMPLETE) {
1277 * Flush IO vqs that don't share a worker with the ctl to make
1278 * sure they have sent their responses before us.
1280 ctl_vq = &tmf->vhost->vqs[VHOST_SCSI_VQ_CTL].vq;
1281 for (i = VHOST_SCSI_VQ_IO; i < tmf->vhost->dev.nvqs; i++) {
1282 vq = &tmf->vhost->vqs[i].vq;
1284 if (vhost_vq_is_setup(vq) &&
1285 vq->worker != ctl_vq->worker)
1289 resp_code = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
1291 resp_code = VIRTIO_SCSI_S_FUNCTION_REJECTED;
1294 vhost_scsi_send_tmf_resp(tmf->vhost, &tmf->svq->vq, tmf->in_iovs,
1295 tmf->vq_desc, &tmf->resp_iov, resp_code);
1296 vhost_scsi_release_tmf_res(tmf);
1300 vhost_scsi_handle_tmf(struct vhost_scsi *vs, struct vhost_scsi_tpg *tpg,
1301 struct vhost_virtqueue *vq,
1302 struct virtio_scsi_ctrl_tmf_req *vtmf,
1303 struct vhost_scsi_ctx *vc)
1305 struct vhost_scsi_virtqueue *svq = container_of(vq,
1306 struct vhost_scsi_virtqueue, vq);
1307 struct vhost_scsi_tmf *tmf;
1309 if (vhost32_to_cpu(vq, vtmf->subtype) !=
1310 VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET)
1313 if (!tpg->tpg_nexus || !tpg->tpg_nexus->tvn_se_sess) {
1314 pr_err("Unable to locate active struct vhost_scsi_nexus for LUN RESET.\n");
1318 tmf = kzalloc(sizeof(*tmf), GFP_KERNEL);
1322 vhost_work_init(&tmf->vwork, vhost_scsi_tmf_resp_work);
1325 tmf->resp_iov = vq->iov[vc->out];
1326 tmf->vq_desc = vc->head;
1327 tmf->in_iovs = vc->in;
1328 tmf->inflight = vhost_scsi_get_inflight(vq);
1330 if (target_submit_tmr(&tmf->se_cmd, tpg->tpg_nexus->tvn_se_sess, NULL,
1331 vhost_buf_to_lun(vtmf->lun), NULL,
1332 TMR_LUN_RESET, GFP_KERNEL, 0,
1333 TARGET_SCF_ACK_KREF) < 0) {
1334 vhost_scsi_release_tmf_res(tmf);
1341 vhost_scsi_send_tmf_resp(vs, vq, vc->in, vc->head, &vq->iov[vc->out],
1342 VIRTIO_SCSI_S_FUNCTION_REJECTED);
1346 vhost_scsi_send_an_resp(struct vhost_scsi *vs,
1347 struct vhost_virtqueue *vq,
1348 struct vhost_scsi_ctx *vc)
1350 struct virtio_scsi_ctrl_an_resp rsp;
1351 struct iov_iter iov_iter;
1354 pr_debug("%s\n", __func__);
1355 memset(&rsp, 0, sizeof(rsp)); /* event_actual = 0 */
1356 rsp.response = VIRTIO_SCSI_S_OK;
1358 iov_iter_init(&iov_iter, ITER_DEST, &vq->iov[vc->out], vc->in, sizeof(rsp));
1360 ret = copy_to_iter(&rsp, sizeof(rsp), &iov_iter);
1361 if (likely(ret == sizeof(rsp)))
1362 vhost_add_used_and_signal(&vs->dev, vq, vc->head, 0);
1364 pr_err("Faulted on virtio_scsi_ctrl_an_resp\n");
1368 vhost_scsi_ctl_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
1370 struct vhost_scsi_tpg *tpg;
1373 struct virtio_scsi_ctrl_an_req an;
1374 struct virtio_scsi_ctrl_tmf_req tmf;
1376 struct vhost_scsi_ctx vc;
1380 mutex_lock(&vq->mutex);
1382 * We can handle the vq only after the endpoint is setup by calling the
1383 * VHOST_SCSI_SET_ENDPOINT ioctl.
1385 if (!vhost_vq_get_backend(vq))
1388 memset(&vc, 0, sizeof(vc));
1390 vhost_disable_notify(&vs->dev, vq);
1393 ret = vhost_scsi_get_desc(vs, vq, &vc);
1398 * Get the request type first in order to setup
1399 * other parameters dependent on the type.
1401 vc.req = &v_req.type;
1402 typ_size = sizeof(v_req.type);
1404 if (unlikely(!copy_from_iter_full(vc.req, typ_size,
1406 vq_err(vq, "Faulted on copy_from_iter tmf type\n");
1408 * The size of the response buffer depends on the
1409 * request type and must be validated against it.
1410 * Since the request type is not known, don't send
1416 switch (vhost32_to_cpu(vq, v_req.type)) {
1417 case VIRTIO_SCSI_T_TMF:
1418 vc.req = &v_req.tmf;
1419 vc.req_size = sizeof(struct virtio_scsi_ctrl_tmf_req);
1420 vc.rsp_size = sizeof(struct virtio_scsi_ctrl_tmf_resp);
1421 vc.lunp = &v_req.tmf.lun[0];
1422 vc.target = &v_req.tmf.lun[1];
1424 case VIRTIO_SCSI_T_AN_QUERY:
1425 case VIRTIO_SCSI_T_AN_SUBSCRIBE:
1427 vc.req_size = sizeof(struct virtio_scsi_ctrl_an_req);
1428 vc.rsp_size = sizeof(struct virtio_scsi_ctrl_an_resp);
1429 vc.lunp = &v_req.an.lun[0];
1433 vq_err(vq, "Unknown control request %d", v_req.type);
1438 * Validate the size of request and response buffers.
1439 * Check for a sane response buffer so we can report
1440 * early errors back to the guest.
1442 ret = vhost_scsi_chk_size(vq, &vc);
1447 * Get the rest of the request now that its size is known.
1450 vc.req_size -= typ_size;
1452 ret = vhost_scsi_get_req(vq, &vc, &tpg);
1456 if (v_req.type == VIRTIO_SCSI_T_TMF)
1457 vhost_scsi_handle_tmf(vs, tpg, vq, &v_req.tmf, &vc);
1459 vhost_scsi_send_an_resp(vs, vq, &vc);
1462 * ENXIO: No more requests, or read error, wait for next kick
1463 * EINVAL: Invalid response buffer, drop the request
1464 * EIO: Respond with bad target
1465 * EAGAIN: Pending request
1469 else if (ret == -EIO)
1470 vhost_scsi_send_bad_target(vs, vq, vc.head, vc.out);
1471 } while (likely(!vhost_exceeds_weight(vq, ++c, 0)));
1473 mutex_unlock(&vq->mutex);
1476 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1478 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1480 struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1482 pr_debug("%s: The handling func for control queue.\n", __func__);
1483 vhost_scsi_ctl_handle_vq(vs, vq);
1487 vhost_scsi_send_evt(struct vhost_scsi *vs, struct vhost_virtqueue *vq,
1488 struct vhost_scsi_tpg *tpg, struct se_lun *lun,
1489 u32 event, u32 reason)
1491 struct vhost_scsi_evt *evt;
1493 evt = vhost_scsi_allocate_evt(vs, event, reason);
1498 /* TODO: share lun setup code with virtio-scsi.ko */
1500 * Note: evt->event is zeroed when we allocate it and
1501 * lun[4-7] need to be zero according to virtio-scsi spec.
1503 evt->event.lun[0] = 0x01;
1504 evt->event.lun[1] = tpg->tport_tpgt;
1505 if (lun->unpacked_lun >= 256)
1506 evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1507 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1510 llist_add(&evt->list, &vs->vs_event_list);
1511 vhost_vq_work_queue(vq, &vs->vs_event_work);
1514 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1516 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1518 struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1520 mutex_lock(&vq->mutex);
1521 if (!vhost_vq_get_backend(vq))
1524 if (vs->vs_events_missed)
1525 vhost_scsi_send_evt(vs, vq, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT,
1528 mutex_unlock(&vq->mutex);
1531 static void vhost_scsi_handle_kick(struct vhost_work *work)
1533 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1535 struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1537 vhost_scsi_handle_vq(vs, vq);
1540 /* Callers must hold dev mutex */
1541 static void vhost_scsi_flush(struct vhost_scsi *vs)
1545 /* Init new inflight and remember the old inflight */
1546 vhost_scsi_init_inflight(vs, vs->old_inflight);
1549 * The inflight->kref was initialized to 1. We decrement it here to
1550 * indicate the start of the flush operation so that it will reach 0
1551 * when all the reqs are finished.
1553 for (i = 0; i < vs->dev.nvqs; i++)
1554 kref_put(&vs->old_inflight[i]->kref, vhost_scsi_done_inflight);
1556 /* Flush both the vhost poll and vhost work */
1557 vhost_dev_flush(&vs->dev);
1559 /* Wait for all reqs issued before the flush to be finished */
1560 for (i = 0; i < vs->dev.nvqs; i++)
1561 wait_for_completion(&vs->old_inflight[i]->comp);
1564 static void vhost_scsi_destroy_vq_cmds(struct vhost_virtqueue *vq)
1566 struct vhost_scsi_virtqueue *svq = container_of(vq,
1567 struct vhost_scsi_virtqueue, vq);
1568 struct vhost_scsi_cmd *tv_cmd;
1571 if (!svq->scsi_cmds)
1574 for (i = 0; i < svq->max_cmds; i++) {
1575 tv_cmd = &svq->scsi_cmds[i];
1577 kfree(tv_cmd->tvc_sgl);
1578 kfree(tv_cmd->tvc_prot_sgl);
1579 kfree(tv_cmd->tvc_upages);
1580 kfree(tv_cmd->tvc_resp_iov);
1583 sbitmap_free(&svq->scsi_tags);
1584 kfree(svq->scsi_cmds);
1585 svq->scsi_cmds = NULL;
1588 static int vhost_scsi_setup_vq_cmds(struct vhost_virtqueue *vq, int max_cmds)
1590 struct vhost_scsi_virtqueue *svq = container_of(vq,
1591 struct vhost_scsi_virtqueue, vq);
1592 struct vhost_scsi_cmd *tv_cmd;
1598 if (sbitmap_init_node(&svq->scsi_tags, max_cmds, -1, GFP_KERNEL,
1599 NUMA_NO_NODE, false, true))
1601 svq->max_cmds = max_cmds;
1603 svq->scsi_cmds = kcalloc(max_cmds, sizeof(*tv_cmd), GFP_KERNEL);
1604 if (!svq->scsi_cmds) {
1605 sbitmap_free(&svq->scsi_tags);
1609 for (i = 0; i < max_cmds; i++) {
1610 tv_cmd = &svq->scsi_cmds[i];
1612 tv_cmd->tvc_sgl = kcalloc(VHOST_SCSI_PREALLOC_SGLS,
1613 sizeof(struct scatterlist),
1615 if (!tv_cmd->tvc_sgl) {
1616 pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1620 tv_cmd->tvc_upages = kcalloc(VHOST_SCSI_PREALLOC_UPAGES,
1621 sizeof(struct page *),
1623 if (!tv_cmd->tvc_upages) {
1624 pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1628 tv_cmd->tvc_resp_iov = kcalloc(UIO_MAXIOV,
1629 sizeof(struct iovec),
1631 if (!tv_cmd->tvc_resp_iov) {
1632 pr_err("Unable to allocate tv_cmd->tvc_resp_iov\n");
1636 tv_cmd->tvc_prot_sgl = kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS,
1637 sizeof(struct scatterlist),
1639 if (!tv_cmd->tvc_prot_sgl) {
1640 pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1646 vhost_scsi_destroy_vq_cmds(vq);
1651 * Called from vhost_scsi_ioctl() context to walk the list of available
1652 * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1654 * The lock nesting rule is:
1655 * vs->dev.mutex -> vhost_scsi_mutex -> tpg->tv_tpg_mutex -> vq->mutex
1658 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1659 struct vhost_scsi_target *t)
1661 struct se_portal_group *se_tpg;
1662 struct vhost_scsi_tport *tv_tport;
1663 struct vhost_scsi_tpg *tpg;
1664 struct vhost_scsi_tpg **vs_tpg;
1665 struct vhost_virtqueue *vq;
1666 int index, ret, i, len;
1669 mutex_lock(&vs->dev.mutex);
1671 /* Verify that ring has been setup correctly. */
1672 for (index = 0; index < vs->dev.nvqs; ++index) {
1673 /* Verify that ring has been setup correctly. */
1674 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1680 len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1681 vs_tpg = kzalloc(len, GFP_KERNEL);
1687 memcpy(vs_tpg, vs->vs_tpg, len);
1689 mutex_lock(&vhost_scsi_mutex);
1690 list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1691 mutex_lock(&tpg->tv_tpg_mutex);
1692 if (!tpg->tpg_nexus) {
1693 mutex_unlock(&tpg->tv_tpg_mutex);
1696 if (tpg->tv_tpg_vhost_count != 0) {
1697 mutex_unlock(&tpg->tv_tpg_mutex);
1700 tv_tport = tpg->tport;
1702 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1703 if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1704 mutex_unlock(&tpg->tv_tpg_mutex);
1705 mutex_unlock(&vhost_scsi_mutex);
1710 * In order to ensure individual vhost-scsi configfs
1711 * groups cannot be removed while in use by vhost ioctl,
1712 * go ahead and take an explicit se_tpg->tpg_group.cg_item
1715 se_tpg = &tpg->se_tpg;
1716 ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1718 pr_warn("target_depend_item() failed: %d\n", ret);
1719 mutex_unlock(&tpg->tv_tpg_mutex);
1720 mutex_unlock(&vhost_scsi_mutex);
1723 tpg->tv_tpg_vhost_count++;
1724 tpg->vhost_scsi = vs;
1725 vs_tpg[tpg->tport_tpgt] = tpg;
1728 mutex_unlock(&tpg->tv_tpg_mutex);
1730 mutex_unlock(&vhost_scsi_mutex);
1733 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1734 sizeof(vs->vs_vhost_wwpn));
1736 for (i = VHOST_SCSI_VQ_IO; i < vs->dev.nvqs; i++) {
1737 vq = &vs->vqs[i].vq;
1738 if (!vhost_vq_is_setup(vq))
1741 ret = vhost_scsi_setup_vq_cmds(vq, vq->num);
1743 goto destroy_vq_cmds;
1746 for (i = 0; i < vs->dev.nvqs; i++) {
1747 vq = &vs->vqs[i].vq;
1748 mutex_lock(&vq->mutex);
1749 vhost_vq_set_backend(vq, vs_tpg);
1750 vhost_vq_init_access(vq);
1751 mutex_unlock(&vq->mutex);
1759 * Act as synchronize_rcu to make sure access to
1760 * old vs->vs_tpg is finished.
1762 vhost_scsi_flush(vs);
1764 vs->vs_tpg = vs_tpg;
1768 for (i--; i >= VHOST_SCSI_VQ_IO; i--) {
1769 if (!vhost_vq_get_backend(&vs->vqs[i].vq))
1770 vhost_scsi_destroy_vq_cmds(&vs->vqs[i].vq);
1773 for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1776 mutex_lock(&tpg->tv_tpg_mutex);
1777 tpg->vhost_scsi = NULL;
1778 tpg->tv_tpg_vhost_count--;
1779 mutex_unlock(&tpg->tv_tpg_mutex);
1780 target_undepend_item(&tpg->se_tpg.tpg_group.cg_item);
1785 mutex_unlock(&vs->dev.mutex);
1790 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1791 struct vhost_scsi_target *t)
1793 struct se_portal_group *se_tpg;
1794 struct vhost_scsi_tport *tv_tport;
1795 struct vhost_scsi_tpg *tpg;
1796 struct vhost_virtqueue *vq;
1801 mutex_lock(&vs->dev.mutex);
1802 /* Verify that ring has been setup correctly. */
1803 for (index = 0; index < vs->dev.nvqs; ++index) {
1804 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1815 for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1817 tpg = vs->vs_tpg[target];
1821 tv_tport = tpg->tport;
1827 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1828 pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1829 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1830 tv_tport->tport_name, tpg->tport_tpgt,
1831 t->vhost_wwpn, t->vhost_tpgt);
1840 /* Prevent new cmds from starting and accessing the tpgs/sessions */
1841 for (i = 0; i < vs->dev.nvqs; i++) {
1842 vq = &vs->vqs[i].vq;
1843 mutex_lock(&vq->mutex);
1844 vhost_vq_set_backend(vq, NULL);
1845 mutex_unlock(&vq->mutex);
1847 /* Make sure cmds are not running before tearing them down. */
1848 vhost_scsi_flush(vs);
1850 for (i = 0; i < vs->dev.nvqs; i++) {
1851 vq = &vs->vqs[i].vq;
1852 vhost_scsi_destroy_vq_cmds(vq);
1856 * We can now release our hold on the tpg and sessions and userspace
1857 * can free them after this point.
1859 for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1861 tpg = vs->vs_tpg[target];
1865 mutex_lock(&tpg->tv_tpg_mutex);
1867 tpg->tv_tpg_vhost_count--;
1868 tpg->vhost_scsi = NULL;
1869 vs->vs_tpg[target] = NULL;
1871 mutex_unlock(&tpg->tv_tpg_mutex);
1873 se_tpg = &tpg->se_tpg;
1874 target_undepend_item(&se_tpg->tpg_group.cg_item);
1879 * Act as synchronize_rcu to make sure access to
1880 * old vs->vs_tpg is finished.
1882 vhost_scsi_flush(vs);
1885 WARN_ON(vs->vs_events_nr);
1886 mutex_unlock(&vs->dev.mutex);
1890 mutex_unlock(&vs->dev.mutex);
1894 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1896 struct vhost_virtqueue *vq;
1899 if (features & ~VHOST_SCSI_FEATURES)
1902 mutex_lock(&vs->dev.mutex);
1903 if ((features & (1 << VHOST_F_LOG_ALL)) &&
1904 !vhost_log_access_ok(&vs->dev)) {
1905 mutex_unlock(&vs->dev.mutex);
1909 for (i = 0; i < vs->dev.nvqs; i++) {
1910 vq = &vs->vqs[i].vq;
1911 mutex_lock(&vq->mutex);
1912 vq->acked_features = features;
1913 mutex_unlock(&vq->mutex);
1915 mutex_unlock(&vs->dev.mutex);
1919 static int vhost_scsi_open(struct inode *inode, struct file *f)
1921 struct vhost_scsi_virtqueue *svq;
1922 struct vhost_scsi *vs;
1923 struct vhost_virtqueue **vqs;
1924 int r = -ENOMEM, i, nvqs = vhost_scsi_max_io_vqs;
1926 vs = kvzalloc(sizeof(*vs), GFP_KERNEL);
1930 if (nvqs > VHOST_SCSI_MAX_IO_VQ) {
1931 pr_err("Invalid max_io_vqs of %d. Using %d.\n", nvqs,
1932 VHOST_SCSI_MAX_IO_VQ);
1933 nvqs = VHOST_SCSI_MAX_IO_VQ;
1934 } else if (nvqs == 0) {
1935 pr_err("Invalid max_io_vqs of %d. Using 1.\n", nvqs);
1938 nvqs += VHOST_SCSI_VQ_IO;
1940 vs->old_inflight = kmalloc_array(nvqs, sizeof(*vs->old_inflight),
1941 GFP_KERNEL | __GFP_ZERO);
1942 if (!vs->old_inflight)
1945 vs->vqs = kmalloc_array(nvqs, sizeof(*vs->vqs),
1946 GFP_KERNEL | __GFP_ZERO);
1950 vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL);
1954 vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1956 vs->vs_events_nr = 0;
1957 vs->vs_events_missed = false;
1959 vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1960 vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1961 vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1962 vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1963 for (i = VHOST_SCSI_VQ_IO; i < nvqs; i++) {
1968 init_llist_head(&svq->completion_list);
1969 vhost_work_init(&svq->completion_work,
1970 vhost_scsi_complete_cmd_work);
1971 svq->vq.handle_kick = vhost_scsi_handle_kick;
1973 vhost_dev_init(&vs->dev, vqs, nvqs, UIO_MAXIOV,
1974 VHOST_SCSI_WEIGHT, 0, true, NULL);
1976 vhost_scsi_init_inflight(vs, NULL);
1978 f->private_data = vs;
1984 kfree(vs->old_inflight);
1991 static int vhost_scsi_release(struct inode *inode, struct file *f)
1993 struct vhost_scsi *vs = f->private_data;
1994 struct vhost_scsi_target t;
1996 mutex_lock(&vs->dev.mutex);
1997 memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1998 mutex_unlock(&vs->dev.mutex);
1999 vhost_scsi_clear_endpoint(vs, &t);
2000 vhost_dev_stop(&vs->dev);
2001 vhost_dev_cleanup(&vs->dev);
2004 kfree(vs->old_inflight);
2010 vhost_scsi_ioctl(struct file *f,
2014 struct vhost_scsi *vs = f->private_data;
2015 struct vhost_scsi_target backend;
2016 void __user *argp = (void __user *)arg;
2017 u64 __user *featurep = argp;
2018 u32 __user *eventsp = argp;
2021 int r, abi_version = VHOST_SCSI_ABI_VERSION;
2022 struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
2025 case VHOST_SCSI_SET_ENDPOINT:
2026 if (copy_from_user(&backend, argp, sizeof backend))
2028 if (backend.reserved != 0)
2031 return vhost_scsi_set_endpoint(vs, &backend);
2032 case VHOST_SCSI_CLEAR_ENDPOINT:
2033 if (copy_from_user(&backend, argp, sizeof backend))
2035 if (backend.reserved != 0)
2038 return vhost_scsi_clear_endpoint(vs, &backend);
2039 case VHOST_SCSI_GET_ABI_VERSION:
2040 if (copy_to_user(argp, &abi_version, sizeof abi_version))
2043 case VHOST_SCSI_SET_EVENTS_MISSED:
2044 if (get_user(events_missed, eventsp))
2046 mutex_lock(&vq->mutex);
2047 vs->vs_events_missed = events_missed;
2048 mutex_unlock(&vq->mutex);
2050 case VHOST_SCSI_GET_EVENTS_MISSED:
2051 mutex_lock(&vq->mutex);
2052 events_missed = vs->vs_events_missed;
2053 mutex_unlock(&vq->mutex);
2054 if (put_user(events_missed, eventsp))
2057 case VHOST_GET_FEATURES:
2058 features = VHOST_SCSI_FEATURES;
2059 if (copy_to_user(featurep, &features, sizeof features))
2062 case VHOST_SET_FEATURES:
2063 if (copy_from_user(&features, featurep, sizeof features))
2065 return vhost_scsi_set_features(vs, features);
2066 case VHOST_NEW_WORKER:
2067 case VHOST_FREE_WORKER:
2068 case VHOST_ATTACH_VRING_WORKER:
2069 case VHOST_GET_VRING_WORKER:
2070 mutex_lock(&vs->dev.mutex);
2071 r = vhost_worker_ioctl(&vs->dev, ioctl, argp);
2072 mutex_unlock(&vs->dev.mutex);
2075 mutex_lock(&vs->dev.mutex);
2076 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
2077 /* TODO: flush backend after dev ioctl. */
2078 if (r == -ENOIOCTLCMD)
2079 r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
2080 mutex_unlock(&vs->dev.mutex);
2085 static const struct file_operations vhost_scsi_fops = {
2086 .owner = THIS_MODULE,
2087 .release = vhost_scsi_release,
2088 .unlocked_ioctl = vhost_scsi_ioctl,
2089 .compat_ioctl = compat_ptr_ioctl,
2090 .open = vhost_scsi_open,
2091 .llseek = noop_llseek,
2094 static struct miscdevice vhost_scsi_misc = {
2100 static int __init vhost_scsi_register(void)
2102 return misc_register(&vhost_scsi_misc);
2105 static void vhost_scsi_deregister(void)
2107 misc_deregister(&vhost_scsi_misc);
2110 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
2112 switch (tport->tport_proto_id) {
2113 case SCSI_PROTOCOL_SAS:
2115 case SCSI_PROTOCOL_FCP:
2117 case SCSI_PROTOCOL_ISCSI:
2127 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
2128 struct se_lun *lun, bool plug)
2131 struct vhost_scsi *vs = tpg->vhost_scsi;
2132 struct vhost_virtqueue *vq;
2139 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
2141 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
2143 vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
2144 mutex_lock(&vq->mutex);
2146 * We can't queue events if the backend has been cleared, because
2147 * we could end up queueing an event after the flush.
2149 if (!vhost_vq_get_backend(vq))
2152 if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
2153 vhost_scsi_send_evt(vs, vq, tpg, lun,
2154 VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
2156 mutex_unlock(&vq->mutex);
2159 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
2161 vhost_scsi_do_plug(tpg, lun, true);
2164 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
2166 vhost_scsi_do_plug(tpg, lun, false);
2169 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
2172 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2173 struct vhost_scsi_tpg, se_tpg);
2175 mutex_lock(&tpg->tv_tpg_mutex);
2176 tpg->tv_tpg_port_count++;
2177 vhost_scsi_hotplug(tpg, lun);
2178 mutex_unlock(&tpg->tv_tpg_mutex);
2183 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
2186 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2187 struct vhost_scsi_tpg, se_tpg);
2189 mutex_lock(&tpg->tv_tpg_mutex);
2190 tpg->tv_tpg_port_count--;
2191 vhost_scsi_hotunplug(tpg, lun);
2192 mutex_unlock(&tpg->tv_tpg_mutex);
2195 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
2196 struct config_item *item, const char *page, size_t count)
2198 struct se_portal_group *se_tpg = attrib_to_tpg(item);
2199 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2200 struct vhost_scsi_tpg, se_tpg);
2202 int ret = kstrtoul(page, 0, &val);
2205 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
2208 if (val != 0 && val != 1 && val != 3) {
2209 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
2212 tpg->tv_fabric_prot_type = val;
2217 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
2218 struct config_item *item, char *page)
2220 struct se_portal_group *se_tpg = attrib_to_tpg(item);
2221 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2222 struct vhost_scsi_tpg, se_tpg);
2224 return sysfs_emit(page, "%d\n", tpg->tv_fabric_prot_type);
2227 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
2229 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
2230 &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
2234 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
2237 struct vhost_scsi_nexus *tv_nexus;
2239 mutex_lock(&tpg->tv_tpg_mutex);
2240 if (tpg->tpg_nexus) {
2241 mutex_unlock(&tpg->tv_tpg_mutex);
2242 pr_debug("tpg->tpg_nexus already exists\n");
2246 tv_nexus = kzalloc(sizeof(*tv_nexus), GFP_KERNEL);
2248 mutex_unlock(&tpg->tv_tpg_mutex);
2249 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
2253 * Since we are running in 'demo mode' this call with generate a
2254 * struct se_node_acl for the vhost_scsi struct se_portal_group with
2255 * the SCSI Initiator port name of the passed configfs group 'name'.
2257 tv_nexus->tvn_se_sess = target_setup_session(&tpg->se_tpg, 0, 0,
2258 TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
2259 (unsigned char *)name, tv_nexus, NULL);
2260 if (IS_ERR(tv_nexus->tvn_se_sess)) {
2261 mutex_unlock(&tpg->tv_tpg_mutex);
2265 tpg->tpg_nexus = tv_nexus;
2267 mutex_unlock(&tpg->tv_tpg_mutex);
2271 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
2273 struct se_session *se_sess;
2274 struct vhost_scsi_nexus *tv_nexus;
2276 mutex_lock(&tpg->tv_tpg_mutex);
2277 tv_nexus = tpg->tpg_nexus;
2279 mutex_unlock(&tpg->tv_tpg_mutex);
2283 se_sess = tv_nexus->tvn_se_sess;
2285 mutex_unlock(&tpg->tv_tpg_mutex);
2289 if (tpg->tv_tpg_port_count != 0) {
2290 mutex_unlock(&tpg->tv_tpg_mutex);
2291 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2292 " active TPG port count: %d\n",
2293 tpg->tv_tpg_port_count);
2297 if (tpg->tv_tpg_vhost_count != 0) {
2298 mutex_unlock(&tpg->tv_tpg_mutex);
2299 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2300 " active TPG vhost count: %d\n",
2301 tpg->tv_tpg_vhost_count);
2305 pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
2306 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
2307 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
2310 * Release the SCSI I_T Nexus to the emulated vhost Target Port
2312 target_remove_session(se_sess);
2313 tpg->tpg_nexus = NULL;
2314 mutex_unlock(&tpg->tv_tpg_mutex);
2320 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
2322 struct se_portal_group *se_tpg = to_tpg(item);
2323 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2324 struct vhost_scsi_tpg, se_tpg);
2325 struct vhost_scsi_nexus *tv_nexus;
2328 mutex_lock(&tpg->tv_tpg_mutex);
2329 tv_nexus = tpg->tpg_nexus;
2331 mutex_unlock(&tpg->tv_tpg_mutex);
2334 ret = sysfs_emit(page, "%s\n",
2335 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
2336 mutex_unlock(&tpg->tv_tpg_mutex);
2341 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
2342 const char *page, size_t count)
2344 struct se_portal_group *se_tpg = to_tpg(item);
2345 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2346 struct vhost_scsi_tpg, se_tpg);
2347 struct vhost_scsi_tport *tport_wwn = tpg->tport;
2348 unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
2351 * Shutdown the active I_T nexus if 'NULL' is passed..
2353 if (!strncmp(page, "NULL", 4)) {
2354 ret = vhost_scsi_drop_nexus(tpg);
2355 return (!ret) ? count : ret;
2358 * Otherwise make sure the passed virtual Initiator port WWN matches
2359 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
2360 * vhost_scsi_make_nexus().
2362 if (strlen(page) >= VHOST_SCSI_NAMELEN) {
2363 pr_err("Emulated NAA Sas Address: %s, exceeds"
2364 " max: %d\n", page, VHOST_SCSI_NAMELEN);
2367 snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
2369 ptr = strstr(i_port, "naa.");
2371 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
2372 pr_err("Passed SAS Initiator Port %s does not"
2373 " match target port protoid: %s\n", i_port,
2374 vhost_scsi_dump_proto_id(tport_wwn));
2377 port_ptr = &i_port[0];
2380 ptr = strstr(i_port, "fc.");
2382 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
2383 pr_err("Passed FCP Initiator Port %s does not"
2384 " match target port protoid: %s\n", i_port,
2385 vhost_scsi_dump_proto_id(tport_wwn));
2388 port_ptr = &i_port[3]; /* Skip over "fc." */
2391 ptr = strstr(i_port, "iqn.");
2393 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
2394 pr_err("Passed iSCSI Initiator Port %s does not"
2395 " match target port protoid: %s\n", i_port,
2396 vhost_scsi_dump_proto_id(tport_wwn));
2399 port_ptr = &i_port[0];
2402 pr_err("Unable to locate prefix for emulated Initiator Port:"
2406 * Clear any trailing newline for the NAA WWN
2409 if (i_port[strlen(i_port)-1] == '\n')
2410 i_port[strlen(i_port)-1] = '\0';
2412 ret = vhost_scsi_make_nexus(tpg, port_ptr);
2419 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
2421 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
2422 &vhost_scsi_tpg_attr_nexus,
2426 static struct se_portal_group *
2427 vhost_scsi_make_tpg(struct se_wwn *wwn, const char *name)
2429 struct vhost_scsi_tport *tport = container_of(wwn,
2430 struct vhost_scsi_tport, tport_wwn);
2432 struct vhost_scsi_tpg *tpg;
2436 if (strstr(name, "tpgt_") != name)
2437 return ERR_PTR(-EINVAL);
2438 if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
2439 return ERR_PTR(-EINVAL);
2441 tpg = kzalloc(sizeof(*tpg), GFP_KERNEL);
2443 pr_err("Unable to allocate struct vhost_scsi_tpg");
2444 return ERR_PTR(-ENOMEM);
2446 mutex_init(&tpg->tv_tpg_mutex);
2447 INIT_LIST_HEAD(&tpg->tv_tpg_list);
2449 tpg->tport_tpgt = tpgt;
2451 ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
2456 mutex_lock(&vhost_scsi_mutex);
2457 list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
2458 mutex_unlock(&vhost_scsi_mutex);
2460 return &tpg->se_tpg;
2463 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
2465 struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2466 struct vhost_scsi_tpg, se_tpg);
2468 mutex_lock(&vhost_scsi_mutex);
2469 list_del(&tpg->tv_tpg_list);
2470 mutex_unlock(&vhost_scsi_mutex);
2472 * Release the virtual I_T Nexus for this vhost TPG
2474 vhost_scsi_drop_nexus(tpg);
2476 * Deregister the se_tpg from TCM..
2478 core_tpg_deregister(se_tpg);
2482 static struct se_wwn *
2483 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2484 struct config_group *group,
2487 struct vhost_scsi_tport *tport;
2492 /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2493 return ERR_PTR(-EINVAL); */
2495 tport = kzalloc(sizeof(*tport), GFP_KERNEL);
2497 pr_err("Unable to allocate struct vhost_scsi_tport");
2498 return ERR_PTR(-ENOMEM);
2500 tport->tport_wwpn = wwpn;
2502 * Determine the emulated Protocol Identifier and Target Port Name
2503 * based on the incoming configfs directory name.
2505 ptr = strstr(name, "naa.");
2507 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2510 ptr = strstr(name, "fc.");
2512 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2513 off = 3; /* Skip over "fc." */
2516 ptr = strstr(name, "iqn.");
2518 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2522 pr_err("Unable to locate prefix for emulated Target Port:"
2525 return ERR_PTR(-EINVAL);
2528 if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2529 pr_err("Emulated %s Address: %s, exceeds"
2530 " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2531 VHOST_SCSI_NAMELEN);
2533 return ERR_PTR(-EINVAL);
2535 snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2537 pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2538 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2540 return &tport->tport_wwn;
2543 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2545 struct vhost_scsi_tport *tport = container_of(wwn,
2546 struct vhost_scsi_tport, tport_wwn);
2548 pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2549 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2556 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2558 return sysfs_emit(page, "TCM_VHOST fabric module %s on %s/%s"
2559 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2560 utsname()->machine);
2563 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2565 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2566 &vhost_scsi_wwn_attr_version,
2570 static const struct target_core_fabric_ops vhost_scsi_ops = {
2571 .module = THIS_MODULE,
2572 .fabric_name = "vhost",
2573 .max_data_sg_nents = VHOST_SCSI_PREALLOC_SGLS,
2574 .tpg_get_wwn = vhost_scsi_get_fabric_wwn,
2575 .tpg_get_tag = vhost_scsi_get_tpgt,
2576 .tpg_check_demo_mode = vhost_scsi_check_true,
2577 .tpg_check_demo_mode_cache = vhost_scsi_check_true,
2578 .tpg_check_prot_fabric_only = vhost_scsi_check_prot_fabric_only,
2579 .release_cmd = vhost_scsi_release_cmd,
2580 .check_stop_free = vhost_scsi_check_stop_free,
2581 .sess_get_initiator_sid = NULL,
2582 .write_pending = vhost_scsi_write_pending,
2583 .queue_data_in = vhost_scsi_queue_data_in,
2584 .queue_status = vhost_scsi_queue_status,
2585 .queue_tm_rsp = vhost_scsi_queue_tm_rsp,
2586 .aborted_task = vhost_scsi_aborted_task,
2588 * Setup callers for generic logic in target_core_fabric_configfs.c
2590 .fabric_make_wwn = vhost_scsi_make_tport,
2591 .fabric_drop_wwn = vhost_scsi_drop_tport,
2592 .fabric_make_tpg = vhost_scsi_make_tpg,
2593 .fabric_drop_tpg = vhost_scsi_drop_tpg,
2594 .fabric_post_link = vhost_scsi_port_link,
2595 .fabric_pre_unlink = vhost_scsi_port_unlink,
2597 .tfc_wwn_attrs = vhost_scsi_wwn_attrs,
2598 .tfc_tpg_base_attrs = vhost_scsi_tpg_attrs,
2599 .tfc_tpg_attrib_attrs = vhost_scsi_tpg_attrib_attrs,
2602 static int __init vhost_scsi_init(void)
2606 pr_debug("TCM_VHOST fabric module %s on %s/%s"
2607 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2608 utsname()->machine);
2610 ret = vhost_scsi_register();
2614 ret = target_register_template(&vhost_scsi_ops);
2616 goto out_vhost_scsi_deregister;
2620 out_vhost_scsi_deregister:
2621 vhost_scsi_deregister();
2626 static void vhost_scsi_exit(void)
2628 target_unregister_template(&vhost_scsi_ops);
2629 vhost_scsi_deregister();
2632 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2633 MODULE_ALIAS("tcm_vhost");
2634 MODULE_LICENSE("GPL");
2635 module_init(vhost_scsi_init);
2636 module_exit(vhost_scsi_exit);