2 * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
3 * Copyright (C) 2014 Red Hat, Inc.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 #include <linux/spinlock.h>
20 #include <linux/module.h>
21 #include <linux/idr.h>
22 #include <linux/timer.h>
23 #include <linux/parser.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_host.h>
26 #include <linux/uio_driver.h>
27 #include <net/genetlink.h>
28 #include <target/target_core_base.h>
29 #include <target/target_core_fabric.h>
30 #include <target/target_core_backend.h>
31 #include <linux/target_core_user.h>
34 * Define a shared-memory interface for LIO to pass SCSI commands and
35 * data to userspace for processing. This is to allow backends that
36 * are too complex for in-kernel support to be possible.
38 * It uses the UIO framework to do a lot of the device-creation and
39 * introspection work for us.
41 * See the .h file for how the ring is laid out. Note that while the
42 * command ring is defined, the particulars of the data area are
43 * not. Offset values in the command entry point to other locations
44 * internal to the mmap()ed area. There is separate space outside the
45 * command ring for data buffers. This leaves maximum flexibility for
46 * moving buffer allocations, or even page flipping or other
47 * allocation techniques, without altering the command ring layout.
50 * The user process must be assumed to be malicious. There's no way to
51 * prevent it breaking the command ring protocol if it wants, but in
52 * order to prevent other issues we must only ever read *data* from
53 * the shared memory area, not offsets or sizes. This applies to
54 * command ring entries as well as the mailbox. Extra code needed for
55 * this may have a 'UAM' comment.
59 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
61 #define CMDR_SIZE (16 * 4096)
62 #define DATA_SIZE (257 * 4096)
64 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
66 static struct device *tcmu_root_device;
72 /* User wants all cmds or just some */
79 #define TCMU_CONFIG_LEN 256
82 struct se_device se_dev;
87 #define TCMU_DEV_BIT_OPEN 0
88 #define TCMU_DEV_BIT_BROKEN 1
90 enum passthru_level pass_level;
92 struct uio_info uio_info;
94 struct tcmu_mailbox *mb_addr;
97 u32 cmdr_last_cleaned;
98 /* Offset of data ring from start of mb */
101 /* Ring head + tail values. */
102 /* Must add data_off and mb_addr to get the address */
106 wait_queue_head_t wait_cmdr;
107 /* TODO should this be a mutex? */
108 spinlock_t cmdr_lock;
111 spinlock_t commands_lock;
113 struct timer_list timeout;
115 char dev_config[TCMU_CONFIG_LEN];
118 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
120 #define CMDR_OFF sizeof(struct tcmu_mailbox)
123 struct se_cmd *se_cmd;
124 struct tcmu_dev *tcmu_dev;
128 /* Can't use se_cmd->data_length when cleaning up expired cmds, because if
129 cmd has been completed then accessing se_cmd is off limits */
132 unsigned long deadline;
134 #define TCMU_CMD_BIT_EXPIRED 0
138 static struct kmem_cache *tcmu_cmd_cache;
140 /* multicast group */
141 enum tcmu_multicast_groups {
145 static const struct genl_multicast_group tcmu_mcgrps[] = {
146 [TCMU_MCGRP_CONFIG] = { .name = "config", },
149 /* Our generic netlink family */
150 static struct genl_family tcmu_genl_family = {
151 .id = GENL_ID_GENERATE,
155 .maxattr = TCMU_ATTR_MAX,
156 .mcgrps = tcmu_mcgrps,
157 .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
160 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
162 struct se_device *se_dev = se_cmd->se_dev;
163 struct tcmu_dev *udev = TCMU_DEV(se_dev);
164 struct tcmu_cmd *tcmu_cmd;
167 tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
171 tcmu_cmd->se_cmd = se_cmd;
172 tcmu_cmd->tcmu_dev = udev;
173 tcmu_cmd->data_length = se_cmd->data_length;
175 tcmu_cmd->deadline = jiffies + msecs_to_jiffies(TCMU_TIME_OUT);
177 idr_preload(GFP_KERNEL);
178 spin_lock_irq(&udev->commands_lock);
179 cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 0,
180 USHRT_MAX, GFP_NOWAIT);
181 spin_unlock_irq(&udev->commands_lock);
185 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
188 tcmu_cmd->cmd_id = cmd_id;
193 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
195 unsigned long offset = (unsigned long) vaddr & ~PAGE_MASK;
197 size = round_up(size+offset, PAGE_SIZE);
201 flush_dcache_page(virt_to_page(vaddr));
207 * Some ring helper functions. We don't assume size is a power of 2 so
208 * we can't use circ_buf.h.
210 static inline size_t spc_used(size_t head, size_t tail, size_t size)
212 int diff = head - tail;
220 static inline size_t spc_free(size_t head, size_t tail, size_t size)
222 /* Keep 1 byte unused or we can't tell full from empty */
223 return (size - spc_used(head, tail, size) - 1);
226 static inline size_t head_to_end(size_t head, size_t size)
231 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
234 * We can't queue a command until we have space available on the cmd ring *and* space
235 * space avail on the data ring.
237 * Called with ring lock held.
239 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t data_needed)
241 struct tcmu_mailbox *mb = udev->mb_addr;
246 tcmu_flush_dcache_range(mb, sizeof(*mb));
248 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
251 * If cmd end-of-ring space is too small then we need space for a NOP plus
252 * original cmd - cmds are internally contiguous.
254 if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
255 cmd_needed = cmd_size;
257 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
259 space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
260 if (space < cmd_needed) {
261 pr_debug("no cmd space: %u %u %u\n", cmd_head,
262 udev->cmdr_last_cleaned, udev->cmdr_size);
266 space = spc_free(udev->data_head, udev->data_tail, udev->data_size);
267 if (space < data_needed) {
268 pr_debug("no data space: %zu %zu %zu\n", udev->data_head,
269 udev->data_tail, udev->data_size);
276 static int tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
278 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
279 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
280 size_t base_command_size, command_size;
281 struct tcmu_mailbox *mb;
282 struct tcmu_cmd_entry *entry;
284 struct scatterlist *sg;
290 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
294 * Must be a certain minimum size for response sense info, but
295 * also may be larger if the iov array is large.
297 * iovs = sgl_nents+1, for end-of-ring case, plus another 1
298 * b/c size == offsetof one-past-element.
300 base_command_size = max(offsetof(struct tcmu_cmd_entry,
301 req.iov[se_cmd->t_data_nents + 2]),
302 sizeof(struct tcmu_cmd_entry));
303 command_size = base_command_size
304 + round_up(scsi_command_size(se_cmd->t_task_cdb), TCMU_OP_ALIGN_SIZE);
306 WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
308 spin_lock_irq(&udev->cmdr_lock);
311 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
312 if ((command_size > (udev->cmdr_size / 2))
313 || tcmu_cmd->data_length > (udev->data_size - 1))
314 pr_warn("TCMU: Request of size %zu/%zu may be too big for %u/%zu "
315 "cmd/data ring buffers\n", command_size, tcmu_cmd->data_length,
316 udev->cmdr_size, udev->data_size);
318 while (!is_ring_space_avail(udev, command_size, tcmu_cmd->data_length)) {
322 prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
324 pr_debug("sleeping for ring space\n");
325 spin_unlock_irq(&udev->cmdr_lock);
326 ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
327 finish_wait(&udev->wait_cmdr, &__wait);
329 pr_warn("tcmu: command timed out\n");
333 spin_lock_irq(&udev->cmdr_lock);
335 /* We dropped cmdr_lock, cmd_head is stale */
336 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
339 /* Insert a PAD if end-of-ring space is too small */
340 if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
341 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
343 entry = (void *) mb + CMDR_OFF + cmd_head;
344 tcmu_flush_dcache_range(entry, sizeof(*entry));
345 tcmu_hdr_set_op(&entry->hdr, TCMU_OP_PAD);
346 tcmu_hdr_set_len(&entry->hdr, pad_size);
348 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
350 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
351 WARN_ON(cmd_head != 0);
354 entry = (void *) mb + CMDR_OFF + cmd_head;
355 tcmu_flush_dcache_range(entry, sizeof(*entry));
356 tcmu_hdr_set_op(&entry->hdr, TCMU_OP_CMD);
357 tcmu_hdr_set_len(&entry->hdr, command_size);
358 entry->cmd_id = tcmu_cmd->cmd_id;
361 * Fix up iovecs, and handle if allocation in data ring wrapped.
363 iov = &entry->req.iov[0];
364 for_each_sg(se_cmd->t_data_sg, sg, se_cmd->t_data_nents, i) {
365 size_t copy_bytes = min((size_t)sg->length,
366 head_to_end(udev->data_head, udev->data_size));
367 void *from = kmap_atomic(sg_page(sg)) + sg->offset;
368 void *to = (void *) mb + udev->data_off + udev->data_head;
370 if (tcmu_cmd->se_cmd->data_direction == DMA_TO_DEVICE) {
371 memcpy(to, from, copy_bytes);
372 tcmu_flush_dcache_range(to, copy_bytes);
375 /* Even iov_base is relative to mb_addr */
376 iov->iov_len = copy_bytes;
377 iov->iov_base = (void *) udev->data_off + udev->data_head;
381 UPDATE_HEAD(udev->data_head, copy_bytes, udev->data_size);
383 /* Uh oh, we wrapped the buffer. Must split sg across 2 iovs. */
384 if (sg->length != copy_bytes) {
386 copy_bytes = sg->length - copy_bytes;
388 iov->iov_len = copy_bytes;
389 iov->iov_base = (void *) udev->data_off + udev->data_head;
391 if (se_cmd->data_direction == DMA_TO_DEVICE) {
392 to = (void *) mb + udev->data_off + udev->data_head;
393 memcpy(to, from, copy_bytes);
394 tcmu_flush_dcache_range(to, copy_bytes);
400 UPDATE_HEAD(udev->data_head, copy_bytes, udev->data_size);
405 entry->req.iov_cnt = iov_cnt;
407 /* All offsets relative to mb_addr, not start of entry! */
408 cdb_off = CMDR_OFF + cmd_head + base_command_size;
409 memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
410 entry->req.cdb_off = cdb_off;
411 tcmu_flush_dcache_range(entry, sizeof(*entry));
413 UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
414 tcmu_flush_dcache_range(mb, sizeof(*mb));
416 spin_unlock_irq(&udev->cmdr_lock);
418 /* TODO: only if FLUSH and FUA? */
419 uio_event_notify(&udev->uio_info);
421 mod_timer(&udev->timeout,
422 round_jiffies_up(jiffies + msecs_to_jiffies(TCMU_TIME_OUT)));
427 static int tcmu_queue_cmd(struct se_cmd *se_cmd)
429 struct se_device *se_dev = se_cmd->se_dev;
430 struct tcmu_dev *udev = TCMU_DEV(se_dev);
431 struct tcmu_cmd *tcmu_cmd;
434 tcmu_cmd = tcmu_alloc_cmd(se_cmd);
438 ret = tcmu_queue_cmd_ring(tcmu_cmd);
440 pr_err("TCMU: Could not queue command\n");
441 spin_lock_irq(&udev->commands_lock);
442 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
443 spin_unlock_irq(&udev->commands_lock);
445 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
451 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
453 struct se_cmd *se_cmd = cmd->se_cmd;
454 struct tcmu_dev *udev = cmd->tcmu_dev;
456 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
457 /* cmd has been completed already from timeout, just reclaim data
459 UPDATE_HEAD(udev->data_tail, cmd->data_length, udev->data_size);
463 if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
464 memcpy(se_cmd->sense_buffer, entry->rsp.sense_buffer,
465 se_cmd->scsi_sense_length);
467 UPDATE_HEAD(udev->data_tail, cmd->data_length, udev->data_size);
469 else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
470 struct scatterlist *sg;
473 /* It'd be easier to look at entry's iovec again, but UAM */
474 for_each_sg(se_cmd->t_data_sg, sg, se_cmd->t_data_nents, i) {
479 copy_bytes = min((size_t)sg->length,
480 head_to_end(udev->data_tail, udev->data_size));
482 to = kmap_atomic(sg_page(sg)) + sg->offset;
483 WARN_ON(sg->length + sg->offset > PAGE_SIZE);
484 from = (void *) udev->mb_addr + udev->data_off + udev->data_tail;
485 tcmu_flush_dcache_range(from, copy_bytes);
486 memcpy(to, from, copy_bytes);
488 UPDATE_HEAD(udev->data_tail, copy_bytes, udev->data_size);
490 /* Uh oh, wrapped the data buffer for this sg's data */
491 if (sg->length != copy_bytes) {
492 from = (void *) udev->mb_addr + udev->data_off + udev->data_tail;
493 WARN_ON(udev->data_tail);
495 copy_bytes = sg->length - copy_bytes;
496 tcmu_flush_dcache_range(from, copy_bytes);
497 memcpy(to, from, copy_bytes);
499 UPDATE_HEAD(udev->data_tail, copy_bytes, udev->data_size);
505 } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
506 UPDATE_HEAD(udev->data_tail, cmd->data_length, udev->data_size);
508 pr_warn("TCMU: data direction was %d!\n", se_cmd->data_direction);
511 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
514 kmem_cache_free(tcmu_cmd_cache, cmd);
517 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
519 struct tcmu_mailbox *mb;
524 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
525 pr_err("ring broken, not handling completions\n");
529 spin_lock_irqsave(&udev->cmdr_lock, flags);
532 tcmu_flush_dcache_range(mb, sizeof(*mb));
534 while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) {
536 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
537 struct tcmu_cmd *cmd;
539 tcmu_flush_dcache_range(entry, sizeof(*entry));
541 if (tcmu_hdr_get_op(&entry->hdr) == TCMU_OP_PAD) {
542 UPDATE_HEAD(udev->cmdr_last_cleaned, tcmu_hdr_get_len(&entry->hdr), udev->cmdr_size);
545 WARN_ON(tcmu_hdr_get_op(&entry->hdr) != TCMU_OP_CMD);
547 spin_lock(&udev->commands_lock);
548 cmd = idr_find(&udev->commands, entry->cmd_id);
550 idr_remove(&udev->commands, cmd->cmd_id);
551 spin_unlock(&udev->commands_lock);
554 pr_err("cmd_id not found, ring is broken\n");
555 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
559 tcmu_handle_completion(cmd, entry);
561 UPDATE_HEAD(udev->cmdr_last_cleaned, tcmu_hdr_get_len(&entry->hdr), udev->cmdr_size);
566 if (mb->cmd_tail == mb->cmd_head)
567 del_timer(&udev->timeout); /* no more pending cmds */
569 spin_unlock_irqrestore(&udev->cmdr_lock, flags);
571 wake_up(&udev->wait_cmdr);
576 static int tcmu_check_expired_cmd(int id, void *p, void *data)
578 struct tcmu_cmd *cmd = p;
580 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
583 if (!time_after(cmd->deadline, jiffies))
586 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
587 target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
590 kmem_cache_free(tcmu_cmd_cache, cmd);
595 static void tcmu_device_timedout(unsigned long data)
597 struct tcmu_dev *udev = (struct tcmu_dev *)data;
601 handled = tcmu_handle_completions(udev);
603 pr_warn("%d completions handled from timeout\n", handled);
605 spin_lock_irqsave(&udev->commands_lock, flags);
606 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
607 spin_unlock_irqrestore(&udev->commands_lock, flags);
610 * We don't need to wakeup threads on wait_cmdr since they have their
615 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
617 struct tcmu_hba *tcmu_hba;
619 tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
623 tcmu_hba->host_id = host_id;
624 hba->hba_ptr = tcmu_hba;
629 static void tcmu_detach_hba(struct se_hba *hba)
635 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
637 struct tcmu_dev *udev;
639 udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
643 udev->name = kstrdup(name, GFP_KERNEL);
651 init_waitqueue_head(&udev->wait_cmdr);
652 spin_lock_init(&udev->cmdr_lock);
654 idr_init(&udev->commands);
655 spin_lock_init(&udev->commands_lock);
657 setup_timer(&udev->timeout, tcmu_device_timedout,
658 (unsigned long)udev);
660 udev->pass_level = TCMU_PASS_ALL;
662 return &udev->se_dev;
665 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
667 struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
669 tcmu_handle_completions(tcmu_dev);
675 * mmap code from uio.c. Copied here because we want to hook mmap()
676 * and this stuff must come along.
678 static int tcmu_find_mem_index(struct vm_area_struct *vma)
680 struct tcmu_dev *udev = vma->vm_private_data;
681 struct uio_info *info = &udev->uio_info;
683 if (vma->vm_pgoff < MAX_UIO_MAPS) {
684 if (info->mem[vma->vm_pgoff].size == 0)
686 return (int)vma->vm_pgoff;
691 static int tcmu_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
693 struct tcmu_dev *udev = vma->vm_private_data;
694 struct uio_info *info = &udev->uio_info;
696 unsigned long offset;
699 int mi = tcmu_find_mem_index(vma);
701 return VM_FAULT_SIGBUS;
704 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
707 offset = (vmf->pgoff - mi) << PAGE_SHIFT;
709 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
710 if (info->mem[mi].memtype == UIO_MEM_LOGICAL)
711 page = virt_to_page(addr);
713 page = vmalloc_to_page(addr);
719 static const struct vm_operations_struct tcmu_vm_ops = {
720 .fault = tcmu_vma_fault,
723 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
725 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
727 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
728 vma->vm_ops = &tcmu_vm_ops;
730 vma->vm_private_data = udev;
732 /* Ensure the mmap is exactly the right size */
733 if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
739 static int tcmu_open(struct uio_info *info, struct inode *inode)
741 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
743 /* O_EXCL not supported for char devs, so fake it? */
744 if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
752 static int tcmu_release(struct uio_info *info, struct inode *inode)
754 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
756 clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
763 static int tcmu_netlink_event(enum tcmu_genl_cmd cmd, const char *name, int minor)
769 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
773 msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
777 ret = nla_put_string(skb, TCMU_ATTR_DEVICE, name);
781 ret = nla_put_u32(skb, TCMU_ATTR_MINOR, minor);
785 ret = genlmsg_end(skb, msg_header);
789 ret = genlmsg_multicast(&tcmu_genl_family, skb, 0,
790 TCMU_MCGRP_CONFIG, GFP_KERNEL);
792 /* We don't care if no one is listening */
802 static int tcmu_configure_device(struct se_device *dev)
804 struct tcmu_dev *udev = TCMU_DEV(dev);
805 struct tcmu_hba *hba = udev->hba->hba_ptr;
806 struct uio_info *info;
807 struct tcmu_mailbox *mb;
813 info = &udev->uio_info;
815 size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
817 size += 1; /* for \0 */
818 str = kmalloc(size, GFP_KERNEL);
822 used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
824 if (udev->dev_config[0])
825 snprintf(str + used, size - used, "/%s", udev->dev_config);
829 udev->mb_addr = vzalloc(TCMU_RING_SIZE);
830 if (!udev->mb_addr) {
835 /* mailbox fits in first part of CMDR space */
836 udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
837 udev->data_off = CMDR_SIZE;
838 udev->data_size = TCMU_RING_SIZE - CMDR_SIZE;
842 mb->cmdr_off = CMDR_OFF;
843 mb->cmdr_size = udev->cmdr_size;
845 WARN_ON(!PAGE_ALIGNED(udev->data_off));
846 WARN_ON(udev->data_size % PAGE_SIZE);
850 info->mem[0].name = "tcm-user command & data buffer";
851 info->mem[0].addr = (phys_addr_t) udev->mb_addr;
852 info->mem[0].size = TCMU_RING_SIZE;
853 info->mem[0].memtype = UIO_MEM_VIRTUAL;
855 info->irqcontrol = tcmu_irqcontrol;
856 info->irq = UIO_IRQ_CUSTOM;
858 info->mmap = tcmu_mmap;
859 info->open = tcmu_open;
860 info->release = tcmu_release;
862 ret = uio_register_device(tcmu_root_device, info);
866 /* Other attributes can be configured in userspace */
867 dev->dev_attrib.hw_block_size = 512;
868 dev->dev_attrib.hw_max_sectors = 128;
869 dev->dev_attrib.hw_queue_depth = 128;
871 ret = tcmu_netlink_event(TCMU_CMD_ADDED_DEVICE, udev->uio_info.name,
872 udev->uio_info.uio_dev->minor);
879 uio_unregister_device(&udev->uio_info);
881 vfree(udev->mb_addr);
888 static int tcmu_check_pending_cmd(int id, void *p, void *data)
890 struct tcmu_cmd *cmd = p;
892 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
897 static void tcmu_free_device(struct se_device *dev)
899 struct tcmu_dev *udev = TCMU_DEV(dev);
902 del_timer_sync(&udev->timeout);
904 vfree(udev->mb_addr);
906 /* Upper layer should drain all requests before calling this */
907 spin_lock_irq(&udev->commands_lock);
908 i = idr_for_each(&udev->commands, tcmu_check_pending_cmd, NULL);
909 idr_destroy(&udev->commands);
910 spin_unlock_irq(&udev->commands_lock);
913 /* Device was configured */
914 if (udev->uio_info.uio_dev) {
915 tcmu_netlink_event(TCMU_CMD_REMOVED_DEVICE, udev->uio_info.name,
916 udev->uio_info.uio_dev->minor);
918 uio_unregister_device(&udev->uio_info);
919 kfree(udev->uio_info.name);
927 Opt_dev_config, Opt_dev_size, Opt_err, Opt_pass_level,
930 static match_table_t tokens = {
931 {Opt_dev_config, "dev_config=%s"},
932 {Opt_dev_size, "dev_size=%u"},
933 {Opt_pass_level, "pass_level=%u"},
937 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
938 const char *page, ssize_t count)
940 struct tcmu_dev *udev = TCMU_DEV(dev);
941 char *orig, *ptr, *opts, *arg_p;
942 substring_t args[MAX_OPT_ARGS];
946 opts = kstrdup(page, GFP_KERNEL);
952 while ((ptr = strsep(&opts, ",\n")) != NULL) {
956 token = match_token(ptr, tokens, args);
959 if (match_strlcpy(udev->dev_config, &args[0],
960 TCMU_CONFIG_LEN) == 0) {
964 pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
967 arg_p = match_strdup(&args[0]);
972 ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
975 pr_err("kstrtoul() failed for dev_size=\n");
978 match_int(args, &arg);
979 if (arg >= TCMU_PASS_INVALID) {
980 pr_warn("TCMU: Invalid pass_level: %d\n", arg);
984 pr_debug("TCMU: Setting pass_level to %d\n", arg);
985 udev->pass_level = arg;
993 return (!ret) ? count : ret;
996 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
998 struct tcmu_dev *udev = TCMU_DEV(dev);
1001 bl = sprintf(b + bl, "Config: %s ",
1002 udev->dev_config[0] ? udev->dev_config : "NULL");
1003 bl += sprintf(b + bl, "Size: %zu PassLevel: %u\n",
1004 udev->dev_size, udev->pass_level);
1009 static sector_t tcmu_get_blocks(struct se_device *dev)
1011 struct tcmu_dev *udev = TCMU_DEV(dev);
1013 return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1014 dev->dev_attrib.block_size);
1017 static sense_reason_t
1018 tcmu_execute_rw(struct se_cmd *se_cmd, struct scatterlist *sgl, u32 sgl_nents,
1019 enum dma_data_direction data_direction)
1023 ret = tcmu_queue_cmd(se_cmd);
1026 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1028 return TCM_NO_SENSE;
1031 static sense_reason_t
1032 tcmu_pass_op(struct se_cmd *se_cmd)
1034 int ret = tcmu_queue_cmd(se_cmd);
1037 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1039 return TCM_NO_SENSE;
1042 static struct sbc_ops tcmu_sbc_ops = {
1043 .execute_rw = tcmu_execute_rw,
1044 .execute_sync_cache = tcmu_pass_op,
1045 .execute_write_same = tcmu_pass_op,
1046 .execute_write_same_unmap = tcmu_pass_op,
1047 .execute_unmap = tcmu_pass_op,
1050 static sense_reason_t
1051 tcmu_parse_cdb(struct se_cmd *cmd)
1053 unsigned char *cdb = cmd->t_task_cdb;
1054 struct tcmu_dev *udev = TCMU_DEV(cmd->se_dev);
1057 switch (udev->pass_level) {
1059 /* We're just like pscsi, then */
1061 * For REPORT LUNS we always need to emulate the response, for everything
1066 cmd->execute_cmd = spc_emulate_report_luns;
1077 cmd->se_cmd_flags |= SCF_SCSI_DATA_CDB;
1080 cmd->execute_cmd = tcmu_pass_op;
1085 ret = sbc_parse_cdb(cmd, &tcmu_sbc_ops);
1088 pr_err("Unknown tcm-user pass level %d\n", udev->pass_level);
1089 ret = TCM_CHECK_CONDITION_ABORT_CMD;
1095 static struct se_subsystem_api tcmu_template = {
1097 .inquiry_prod = "USER",
1098 .inquiry_rev = TCMU_VERSION,
1099 .owner = THIS_MODULE,
1100 .transport_type = TRANSPORT_PLUGIN_VHBA_PDEV,
1101 .attach_hba = tcmu_attach_hba,
1102 .detach_hba = tcmu_detach_hba,
1103 .alloc_device = tcmu_alloc_device,
1104 .configure_device = tcmu_configure_device,
1105 .free_device = tcmu_free_device,
1106 .parse_cdb = tcmu_parse_cdb,
1107 .set_configfs_dev_params = tcmu_set_configfs_dev_params,
1108 .show_configfs_dev_params = tcmu_show_configfs_dev_params,
1109 .get_device_type = sbc_get_device_type,
1110 .get_blocks = tcmu_get_blocks,
1113 static int __init tcmu_module_init(void)
1117 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
1119 tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
1120 sizeof(struct tcmu_cmd),
1121 __alignof__(struct tcmu_cmd),
1123 if (!tcmu_cmd_cache)
1126 tcmu_root_device = root_device_register("tcm_user");
1127 if (IS_ERR(tcmu_root_device)) {
1128 ret = PTR_ERR(tcmu_root_device);
1129 goto out_free_cache;
1132 ret = genl_register_family(&tcmu_genl_family);
1134 goto out_unreg_device;
1137 ret = transport_subsystem_register(&tcmu_template);
1139 goto out_unreg_genl;
1144 genl_unregister_family(&tcmu_genl_family);
1146 root_device_unregister(tcmu_root_device);
1148 kmem_cache_destroy(tcmu_cmd_cache);
1153 static void __exit tcmu_module_exit(void)
1155 transport_subsystem_release(&tcmu_template);
1156 genl_unregister_family(&tcmu_genl_family);
1157 root_device_unregister(tcmu_root_device);
1158 kmem_cache_destroy(tcmu_cmd_cache);
1161 MODULE_DESCRIPTION("TCM USER subsystem plugin");
1162 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
1163 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
1164 MODULE_LICENSE("GPL");
1166 module_init(tcmu_module_init);
1167 module_exit(tcmu_module_exit);