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/kernel.h>
23 #include <linux/timer.h>
24 #include <linux/parser.h>
25 #include <linux/uio_driver.h>
26 #include <net/genetlink.h>
27 #include <scsi/scsi_common.h>
28 #include <scsi/scsi_proto.h>
29 #include <target/target_core_base.h>
30 #include <target/target_core_fabric.h>
31 #include <target/target_core_backend.h>
32 #include <target/target_core_backend_configfs.h>
34 #include <linux/target_core_user.h>
37 * Define a shared-memory interface for LIO to pass SCSI commands and
38 * data to userspace for processing. This is to allow backends that
39 * are too complex for in-kernel support to be possible.
41 * It uses the UIO framework to do a lot of the device-creation and
42 * introspection work for us.
44 * See the .h file for how the ring is laid out. Note that while the
45 * command ring is defined, the particulars of the data area are
46 * not. Offset values in the command entry point to other locations
47 * internal to the mmap()ed area. There is separate space outside the
48 * command ring for data buffers. This leaves maximum flexibility for
49 * moving buffer allocations, or even page flipping or other
50 * allocation techniques, without altering the command ring layout.
53 * The user process must be assumed to be malicious. There's no way to
54 * prevent it breaking the command ring protocol if it wants, but in
55 * order to prevent other issues we must only ever read *data* from
56 * the shared memory area, not offsets or sizes. This applies to
57 * command ring entries as well as the mailbox. Extra code needed for
58 * this may have a 'UAM' comment.
62 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
64 #define CMDR_SIZE (16 * 4096)
65 #define DATA_SIZE (257 * 4096)
67 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
69 static struct device
*tcmu_root_device
;
75 /* User wants all cmds or just some */
82 #define TCMU_CONFIG_LEN 256
85 struct se_device se_dev
;
90 #define TCMU_DEV_BIT_OPEN 0
91 #define TCMU_DEV_BIT_BROKEN 1
93 enum passthru_level pass_level
;
95 struct uio_info uio_info
;
97 struct tcmu_mailbox
*mb_addr
;
100 u32 cmdr_last_cleaned
;
101 /* Offset of data ring from start of mb */
104 /* Ring head + tail values. */
105 /* Must add data_off and mb_addr to get the address */
109 wait_queue_head_t wait_cmdr
;
110 /* TODO should this be a mutex? */
111 spinlock_t cmdr_lock
;
114 spinlock_t commands_lock
;
116 struct timer_list timeout
;
118 char dev_config
[TCMU_CONFIG_LEN
];
121 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
123 #define CMDR_OFF sizeof(struct tcmu_mailbox)
126 struct se_cmd
*se_cmd
;
127 struct tcmu_dev
*tcmu_dev
;
131 /* Can't use se_cmd->data_length when cleaning up expired cmds, because if
132 cmd has been completed then accessing se_cmd is off limits */
135 unsigned long deadline
;
137 #define TCMU_CMD_BIT_EXPIRED 0
141 static struct kmem_cache
*tcmu_cmd_cache
;
143 /* multicast group */
144 enum tcmu_multicast_groups
{
148 static const struct genl_multicast_group tcmu_mcgrps
[] = {
149 [TCMU_MCGRP_CONFIG
] = { .name
= "config", },
152 /* Our generic netlink family */
153 static struct genl_family tcmu_genl_family
= {
154 .id
= GENL_ID_GENERATE
,
158 .maxattr
= TCMU_ATTR_MAX
,
159 .mcgrps
= tcmu_mcgrps
,
160 .n_mcgrps
= ARRAY_SIZE(tcmu_mcgrps
),
163 static struct tcmu_cmd
*tcmu_alloc_cmd(struct se_cmd
*se_cmd
)
165 struct se_device
*se_dev
= se_cmd
->se_dev
;
166 struct tcmu_dev
*udev
= TCMU_DEV(se_dev
);
167 struct tcmu_cmd
*tcmu_cmd
;
170 tcmu_cmd
= kmem_cache_zalloc(tcmu_cmd_cache
, GFP_KERNEL
);
174 tcmu_cmd
->se_cmd
= se_cmd
;
175 tcmu_cmd
->tcmu_dev
= udev
;
176 tcmu_cmd
->data_length
= se_cmd
->data_length
;
178 tcmu_cmd
->deadline
= jiffies
+ msecs_to_jiffies(TCMU_TIME_OUT
);
180 idr_preload(GFP_KERNEL
);
181 spin_lock_irq(&udev
->commands_lock
);
182 cmd_id
= idr_alloc(&udev
->commands
, tcmu_cmd
, 0,
183 USHRT_MAX
, GFP_NOWAIT
);
184 spin_unlock_irq(&udev
->commands_lock
);
188 kmem_cache_free(tcmu_cmd_cache
, tcmu_cmd
);
191 tcmu_cmd
->cmd_id
= cmd_id
;
196 static inline void tcmu_flush_dcache_range(void *vaddr
, size_t size
)
198 unsigned long offset
= (unsigned long) vaddr
& ~PAGE_MASK
;
200 size
= round_up(size
+offset
, PAGE_SIZE
);
204 flush_dcache_page(virt_to_page(vaddr
));
210 * Some ring helper functions. We don't assume size is a power of 2 so
211 * we can't use circ_buf.h.
213 static inline size_t spc_used(size_t head
, size_t tail
, size_t size
)
215 int diff
= head
- tail
;
223 static inline size_t spc_free(size_t head
, size_t tail
, size_t size
)
225 /* Keep 1 byte unused or we can't tell full from empty */
226 return (size
- spc_used(head
, tail
, size
) - 1);
229 static inline size_t head_to_end(size_t head
, size_t size
)
234 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
237 * We can't queue a command until we have space available on the cmd ring *and* space
238 * space avail on the data ring.
240 * Called with ring lock held.
242 static bool is_ring_space_avail(struct tcmu_dev
*udev
, size_t cmd_size
, size_t data_needed
)
244 struct tcmu_mailbox
*mb
= udev
->mb_addr
;
249 tcmu_flush_dcache_range(mb
, sizeof(*mb
));
251 cmd_head
= mb
->cmd_head
% udev
->cmdr_size
; /* UAM */
254 * If cmd end-of-ring space is too small then we need space for a NOP plus
255 * original cmd - cmds are internally contiguous.
257 if (head_to_end(cmd_head
, udev
->cmdr_size
) >= cmd_size
)
258 cmd_needed
= cmd_size
;
260 cmd_needed
= cmd_size
+ head_to_end(cmd_head
, udev
->cmdr_size
);
262 space
= spc_free(cmd_head
, udev
->cmdr_last_cleaned
, udev
->cmdr_size
);
263 if (space
< cmd_needed
) {
264 pr_debug("no cmd space: %u %u %u\n", cmd_head
,
265 udev
->cmdr_last_cleaned
, udev
->cmdr_size
);
269 space
= spc_free(udev
->data_head
, udev
->data_tail
, udev
->data_size
);
270 if (space
< data_needed
) {
271 pr_debug("no data space: %zu %zu %zu\n", udev
->data_head
,
272 udev
->data_tail
, udev
->data_size
);
279 static int tcmu_queue_cmd_ring(struct tcmu_cmd
*tcmu_cmd
)
281 struct tcmu_dev
*udev
= tcmu_cmd
->tcmu_dev
;
282 struct se_cmd
*se_cmd
= tcmu_cmd
->se_cmd
;
283 size_t base_command_size
, command_size
;
284 struct tcmu_mailbox
*mb
;
285 struct tcmu_cmd_entry
*entry
;
287 struct scatterlist
*sg
;
293 if (test_bit(TCMU_DEV_BIT_BROKEN
, &udev
->flags
))
297 * Must be a certain minimum size for response sense info, but
298 * also may be larger if the iov array is large.
300 * iovs = sgl_nents+1, for end-of-ring case, plus another 1
301 * b/c size == offsetof one-past-element.
303 base_command_size
= max(offsetof(struct tcmu_cmd_entry
,
304 req
.iov
[se_cmd
->t_data_nents
+ 2]),
305 sizeof(struct tcmu_cmd_entry
));
306 command_size
= base_command_size
307 + round_up(scsi_command_size(se_cmd
->t_task_cdb
), TCMU_OP_ALIGN_SIZE
);
309 WARN_ON(command_size
& (TCMU_OP_ALIGN_SIZE
-1));
311 spin_lock_irq(&udev
->cmdr_lock
);
314 cmd_head
= mb
->cmd_head
% udev
->cmdr_size
; /* UAM */
315 if ((command_size
> (udev
->cmdr_size
/ 2))
316 || tcmu_cmd
->data_length
> (udev
->data_size
- 1))
317 pr_warn("TCMU: Request of size %zu/%zu may be too big for %u/%zu "
318 "cmd/data ring buffers\n", command_size
, tcmu_cmd
->data_length
,
319 udev
->cmdr_size
, udev
->data_size
);
321 while (!is_ring_space_avail(udev
, command_size
, tcmu_cmd
->data_length
)) {
325 prepare_to_wait(&udev
->wait_cmdr
, &__wait
, TASK_INTERRUPTIBLE
);
327 pr_debug("sleeping for ring space\n");
328 spin_unlock_irq(&udev
->cmdr_lock
);
329 ret
= schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT
));
330 finish_wait(&udev
->wait_cmdr
, &__wait
);
332 pr_warn("tcmu: command timed out\n");
336 spin_lock_irq(&udev
->cmdr_lock
);
338 /* We dropped cmdr_lock, cmd_head is stale */
339 cmd_head
= mb
->cmd_head
% udev
->cmdr_size
; /* UAM */
342 /* Insert a PAD if end-of-ring space is too small */
343 if (head_to_end(cmd_head
, udev
->cmdr_size
) < command_size
) {
344 size_t pad_size
= head_to_end(cmd_head
, udev
->cmdr_size
);
346 entry
= (void *) mb
+ CMDR_OFF
+ cmd_head
;
347 tcmu_flush_dcache_range(entry
, sizeof(*entry
));
348 tcmu_hdr_set_op(&entry
->hdr
.len_op
, TCMU_OP_PAD
);
349 tcmu_hdr_set_len(&entry
->hdr
.len_op
, pad_size
);
350 entry
->hdr
.cmd_id
= 0; /* not used for PAD */
351 entry
->hdr
.kflags
= 0;
352 entry
->hdr
.uflags
= 0;
354 UPDATE_HEAD(mb
->cmd_head
, pad_size
, udev
->cmdr_size
);
356 cmd_head
= mb
->cmd_head
% udev
->cmdr_size
; /* UAM */
357 WARN_ON(cmd_head
!= 0);
360 entry
= (void *) mb
+ CMDR_OFF
+ cmd_head
;
361 tcmu_flush_dcache_range(entry
, sizeof(*entry
));
362 tcmu_hdr_set_op(&entry
->hdr
.len_op
, TCMU_OP_CMD
);
363 tcmu_hdr_set_len(&entry
->hdr
.len_op
, command_size
);
364 entry
->hdr
.cmd_id
= tcmu_cmd
->cmd_id
;
365 entry
->hdr
.kflags
= 0;
366 entry
->hdr
.uflags
= 0;
369 * Fix up iovecs, and handle if allocation in data ring wrapped.
371 iov
= &entry
->req
.iov
[0];
372 for_each_sg(se_cmd
->t_data_sg
, sg
, se_cmd
->t_data_nents
, i
) {
373 size_t copy_bytes
= min((size_t)sg
->length
,
374 head_to_end(udev
->data_head
, udev
->data_size
));
375 void *from
= kmap_atomic(sg_page(sg
)) + sg
->offset
;
376 void *to
= (void *) mb
+ udev
->data_off
+ udev
->data_head
;
378 if (tcmu_cmd
->se_cmd
->data_direction
== DMA_TO_DEVICE
) {
379 memcpy(to
, from
, copy_bytes
);
380 tcmu_flush_dcache_range(to
, copy_bytes
);
383 /* Even iov_base is relative to mb_addr */
384 iov
->iov_len
= copy_bytes
;
385 iov
->iov_base
= (void __user
*) udev
->data_off
+
390 UPDATE_HEAD(udev
->data_head
, copy_bytes
, udev
->data_size
);
392 /* Uh oh, we wrapped the buffer. Must split sg across 2 iovs. */
393 if (sg
->length
!= copy_bytes
) {
395 copy_bytes
= sg
->length
- copy_bytes
;
397 iov
->iov_len
= copy_bytes
;
398 iov
->iov_base
= (void __user
*) udev
->data_off
+
401 if (se_cmd
->data_direction
== DMA_TO_DEVICE
) {
402 to
= (void *) mb
+ udev
->data_off
+ udev
->data_head
;
403 memcpy(to
, from
, copy_bytes
);
404 tcmu_flush_dcache_range(to
, copy_bytes
);
410 UPDATE_HEAD(udev
->data_head
, copy_bytes
, udev
->data_size
);
415 entry
->req
.iov_cnt
= iov_cnt
;
416 entry
->req
.iov_bidi_cnt
= 0;
417 entry
->req
.iov_dif_cnt
= 0;
419 /* All offsets relative to mb_addr, not start of entry! */
420 cdb_off
= CMDR_OFF
+ cmd_head
+ base_command_size
;
421 memcpy((void *) mb
+ cdb_off
, se_cmd
->t_task_cdb
, scsi_command_size(se_cmd
->t_task_cdb
));
422 entry
->req
.cdb_off
= cdb_off
;
423 tcmu_flush_dcache_range(entry
, sizeof(*entry
));
425 UPDATE_HEAD(mb
->cmd_head
, command_size
, udev
->cmdr_size
);
426 tcmu_flush_dcache_range(mb
, sizeof(*mb
));
428 spin_unlock_irq(&udev
->cmdr_lock
);
430 /* TODO: only if FLUSH and FUA? */
431 uio_event_notify(&udev
->uio_info
);
433 mod_timer(&udev
->timeout
,
434 round_jiffies_up(jiffies
+ msecs_to_jiffies(TCMU_TIME_OUT
)));
439 static int tcmu_queue_cmd(struct se_cmd
*se_cmd
)
441 struct se_device
*se_dev
= se_cmd
->se_dev
;
442 struct tcmu_dev
*udev
= TCMU_DEV(se_dev
);
443 struct tcmu_cmd
*tcmu_cmd
;
446 tcmu_cmd
= tcmu_alloc_cmd(se_cmd
);
450 ret
= tcmu_queue_cmd_ring(tcmu_cmd
);
452 pr_err("TCMU: Could not queue command\n");
453 spin_lock_irq(&udev
->commands_lock
);
454 idr_remove(&udev
->commands
, tcmu_cmd
->cmd_id
);
455 spin_unlock_irq(&udev
->commands_lock
);
457 kmem_cache_free(tcmu_cmd_cache
, tcmu_cmd
);
463 static void tcmu_handle_completion(struct tcmu_cmd
*cmd
, struct tcmu_cmd_entry
*entry
)
465 struct se_cmd
*se_cmd
= cmd
->se_cmd
;
466 struct tcmu_dev
*udev
= cmd
->tcmu_dev
;
468 if (test_bit(TCMU_CMD_BIT_EXPIRED
, &cmd
->flags
)) {
469 /* cmd has been completed already from timeout, just reclaim data
471 UPDATE_HEAD(udev
->data_tail
, cmd
->data_length
, udev
->data_size
);
475 if (entry
->hdr
.uflags
& TCMU_UFLAG_UNKNOWN_OP
) {
476 UPDATE_HEAD(udev
->data_tail
, cmd
->data_length
, udev
->data_size
);
477 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
479 transport_generic_request_failure(cmd
->se_cmd
,
480 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
);
482 kmem_cache_free(tcmu_cmd_cache
, cmd
);
486 if (entry
->rsp
.scsi_status
== SAM_STAT_CHECK_CONDITION
) {
487 memcpy(se_cmd
->sense_buffer
, entry
->rsp
.sense_buffer
,
488 se_cmd
->scsi_sense_length
);
490 UPDATE_HEAD(udev
->data_tail
, cmd
->data_length
, udev
->data_size
);
492 else if (se_cmd
->data_direction
== DMA_FROM_DEVICE
) {
493 struct scatterlist
*sg
;
496 /* It'd be easier to look at entry's iovec again, but UAM */
497 for_each_sg(se_cmd
->t_data_sg
, sg
, se_cmd
->t_data_nents
, i
) {
502 copy_bytes
= min((size_t)sg
->length
,
503 head_to_end(udev
->data_tail
, udev
->data_size
));
505 to
= kmap_atomic(sg_page(sg
)) + sg
->offset
;
506 WARN_ON(sg
->length
+ sg
->offset
> PAGE_SIZE
);
507 from
= (void *) udev
->mb_addr
+ udev
->data_off
+ udev
->data_tail
;
508 tcmu_flush_dcache_range(from
, copy_bytes
);
509 memcpy(to
, from
, copy_bytes
);
511 UPDATE_HEAD(udev
->data_tail
, copy_bytes
, udev
->data_size
);
513 /* Uh oh, wrapped the data buffer for this sg's data */
514 if (sg
->length
!= copy_bytes
) {
515 from
= (void *) udev
->mb_addr
+ udev
->data_off
+ udev
->data_tail
;
516 WARN_ON(udev
->data_tail
);
518 copy_bytes
= sg
->length
- copy_bytes
;
519 tcmu_flush_dcache_range(from
, copy_bytes
);
520 memcpy(to
, from
, copy_bytes
);
522 UPDATE_HEAD(udev
->data_tail
, copy_bytes
, udev
->data_size
);
528 } else if (se_cmd
->data_direction
== DMA_TO_DEVICE
) {
529 UPDATE_HEAD(udev
->data_tail
, cmd
->data_length
, udev
->data_size
);
531 pr_warn("TCMU: data direction was %d!\n", se_cmd
->data_direction
);
534 target_complete_cmd(cmd
->se_cmd
, entry
->rsp
.scsi_status
);
537 kmem_cache_free(tcmu_cmd_cache
, cmd
);
540 static unsigned int tcmu_handle_completions(struct tcmu_dev
*udev
)
542 struct tcmu_mailbox
*mb
;
547 if (test_bit(TCMU_DEV_BIT_BROKEN
, &udev
->flags
)) {
548 pr_err("ring broken, not handling completions\n");
552 spin_lock_irqsave(&udev
->cmdr_lock
, flags
);
555 tcmu_flush_dcache_range(mb
, sizeof(*mb
));
557 while (udev
->cmdr_last_cleaned
!= ACCESS_ONCE(mb
->cmd_tail
)) {
559 struct tcmu_cmd_entry
*entry
= (void *) mb
+ CMDR_OFF
+ udev
->cmdr_last_cleaned
;
560 struct tcmu_cmd
*cmd
;
562 tcmu_flush_dcache_range(entry
, sizeof(*entry
));
564 if (tcmu_hdr_get_op(entry
->hdr
.len_op
) == TCMU_OP_PAD
) {
565 UPDATE_HEAD(udev
->cmdr_last_cleaned
,
566 tcmu_hdr_get_len(entry
->hdr
.len_op
),
570 WARN_ON(tcmu_hdr_get_op(entry
->hdr
.len_op
) != TCMU_OP_CMD
);
572 spin_lock(&udev
->commands_lock
);
573 cmd
= idr_find(&udev
->commands
, entry
->hdr
.cmd_id
);
575 idr_remove(&udev
->commands
, cmd
->cmd_id
);
576 spin_unlock(&udev
->commands_lock
);
579 pr_err("cmd_id not found, ring is broken\n");
580 set_bit(TCMU_DEV_BIT_BROKEN
, &udev
->flags
);
584 tcmu_handle_completion(cmd
, entry
);
586 UPDATE_HEAD(udev
->cmdr_last_cleaned
,
587 tcmu_hdr_get_len(entry
->hdr
.len_op
),
593 if (mb
->cmd_tail
== mb
->cmd_head
)
594 del_timer(&udev
->timeout
); /* no more pending cmds */
596 spin_unlock_irqrestore(&udev
->cmdr_lock
, flags
);
598 wake_up(&udev
->wait_cmdr
);
603 static int tcmu_check_expired_cmd(int id
, void *p
, void *data
)
605 struct tcmu_cmd
*cmd
= p
;
607 if (test_bit(TCMU_CMD_BIT_EXPIRED
, &cmd
->flags
))
610 if (!time_after(cmd
->deadline
, jiffies
))
613 set_bit(TCMU_CMD_BIT_EXPIRED
, &cmd
->flags
);
614 target_complete_cmd(cmd
->se_cmd
, SAM_STAT_CHECK_CONDITION
);
617 kmem_cache_free(tcmu_cmd_cache
, cmd
);
622 static void tcmu_device_timedout(unsigned long data
)
624 struct tcmu_dev
*udev
= (struct tcmu_dev
*)data
;
628 handled
= tcmu_handle_completions(udev
);
630 pr_warn("%d completions handled from timeout\n", handled
);
632 spin_lock_irqsave(&udev
->commands_lock
, flags
);
633 idr_for_each(&udev
->commands
, tcmu_check_expired_cmd
, NULL
);
634 spin_unlock_irqrestore(&udev
->commands_lock
, flags
);
637 * We don't need to wakeup threads on wait_cmdr since they have their
642 static int tcmu_attach_hba(struct se_hba
*hba
, u32 host_id
)
644 struct tcmu_hba
*tcmu_hba
;
646 tcmu_hba
= kzalloc(sizeof(struct tcmu_hba
), GFP_KERNEL
);
650 tcmu_hba
->host_id
= host_id
;
651 hba
->hba_ptr
= tcmu_hba
;
656 static void tcmu_detach_hba(struct se_hba
*hba
)
662 static struct se_device
*tcmu_alloc_device(struct se_hba
*hba
, const char *name
)
664 struct tcmu_dev
*udev
;
666 udev
= kzalloc(sizeof(struct tcmu_dev
), GFP_KERNEL
);
670 udev
->name
= kstrdup(name
, GFP_KERNEL
);
678 init_waitqueue_head(&udev
->wait_cmdr
);
679 spin_lock_init(&udev
->cmdr_lock
);
681 idr_init(&udev
->commands
);
682 spin_lock_init(&udev
->commands_lock
);
684 setup_timer(&udev
->timeout
, tcmu_device_timedout
,
685 (unsigned long)udev
);
687 udev
->pass_level
= TCMU_PASS_ALL
;
689 return &udev
->se_dev
;
692 static int tcmu_irqcontrol(struct uio_info
*info
, s32 irq_on
)
694 struct tcmu_dev
*tcmu_dev
= container_of(info
, struct tcmu_dev
, uio_info
);
696 tcmu_handle_completions(tcmu_dev
);
702 * mmap code from uio.c. Copied here because we want to hook mmap()
703 * and this stuff must come along.
705 static int tcmu_find_mem_index(struct vm_area_struct
*vma
)
707 struct tcmu_dev
*udev
= vma
->vm_private_data
;
708 struct uio_info
*info
= &udev
->uio_info
;
710 if (vma
->vm_pgoff
< MAX_UIO_MAPS
) {
711 if (info
->mem
[vma
->vm_pgoff
].size
== 0)
713 return (int)vma
->vm_pgoff
;
718 static int tcmu_vma_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
720 struct tcmu_dev
*udev
= vma
->vm_private_data
;
721 struct uio_info
*info
= &udev
->uio_info
;
723 unsigned long offset
;
726 int mi
= tcmu_find_mem_index(vma
);
728 return VM_FAULT_SIGBUS
;
731 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
734 offset
= (vmf
->pgoff
- mi
) << PAGE_SHIFT
;
736 addr
= (void *)(unsigned long)info
->mem
[mi
].addr
+ offset
;
737 if (info
->mem
[mi
].memtype
== UIO_MEM_LOGICAL
)
738 page
= virt_to_page(addr
);
740 page
= vmalloc_to_page(addr
);
746 static const struct vm_operations_struct tcmu_vm_ops
= {
747 .fault
= tcmu_vma_fault
,
750 static int tcmu_mmap(struct uio_info
*info
, struct vm_area_struct
*vma
)
752 struct tcmu_dev
*udev
= container_of(info
, struct tcmu_dev
, uio_info
);
754 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
755 vma
->vm_ops
= &tcmu_vm_ops
;
757 vma
->vm_private_data
= udev
;
759 /* Ensure the mmap is exactly the right size */
760 if (vma_pages(vma
) != (TCMU_RING_SIZE
>> PAGE_SHIFT
))
766 static int tcmu_open(struct uio_info
*info
, struct inode
*inode
)
768 struct tcmu_dev
*udev
= container_of(info
, struct tcmu_dev
, uio_info
);
770 /* O_EXCL not supported for char devs, so fake it? */
771 if (test_and_set_bit(TCMU_DEV_BIT_OPEN
, &udev
->flags
))
779 static int tcmu_release(struct uio_info
*info
, struct inode
*inode
)
781 struct tcmu_dev
*udev
= container_of(info
, struct tcmu_dev
, uio_info
);
783 clear_bit(TCMU_DEV_BIT_OPEN
, &udev
->flags
);
790 static int tcmu_netlink_event(enum tcmu_genl_cmd cmd
, const char *name
, int minor
)
796 skb
= genlmsg_new(NLMSG_GOODSIZE
, GFP_KERNEL
);
800 msg_header
= genlmsg_put(skb
, 0, 0, &tcmu_genl_family
, 0, cmd
);
804 ret
= nla_put_string(skb
, TCMU_ATTR_DEVICE
, name
);
808 ret
= nla_put_u32(skb
, TCMU_ATTR_MINOR
, minor
);
812 genlmsg_end(skb
, msg_header
);
814 ret
= genlmsg_multicast(&tcmu_genl_family
, skb
, 0,
815 TCMU_MCGRP_CONFIG
, GFP_KERNEL
);
817 /* We don't care if no one is listening */
827 static int tcmu_configure_device(struct se_device
*dev
)
829 struct tcmu_dev
*udev
= TCMU_DEV(dev
);
830 struct tcmu_hba
*hba
= udev
->hba
->hba_ptr
;
831 struct uio_info
*info
;
832 struct tcmu_mailbox
*mb
;
838 info
= &udev
->uio_info
;
840 size
= snprintf(NULL
, 0, "tcm-user/%u/%s/%s", hba
->host_id
, udev
->name
,
842 size
+= 1; /* for \0 */
843 str
= kmalloc(size
, GFP_KERNEL
);
847 used
= snprintf(str
, size
, "tcm-user/%u/%s", hba
->host_id
, udev
->name
);
849 if (udev
->dev_config
[0])
850 snprintf(str
+ used
, size
- used
, "/%s", udev
->dev_config
);
854 udev
->mb_addr
= vzalloc(TCMU_RING_SIZE
);
855 if (!udev
->mb_addr
) {
860 /* mailbox fits in first part of CMDR space */
861 udev
->cmdr_size
= CMDR_SIZE
- CMDR_OFF
;
862 udev
->data_off
= CMDR_SIZE
;
863 udev
->data_size
= TCMU_RING_SIZE
- CMDR_SIZE
;
866 mb
->version
= TCMU_MAILBOX_VERSION
;
867 mb
->cmdr_off
= CMDR_OFF
;
868 mb
->cmdr_size
= udev
->cmdr_size
;
870 WARN_ON(!PAGE_ALIGNED(udev
->data_off
));
871 WARN_ON(udev
->data_size
% PAGE_SIZE
);
873 info
->version
= xstr(TCMU_MAILBOX_VERSION
);
875 info
->mem
[0].name
= "tcm-user command & data buffer";
876 info
->mem
[0].addr
= (phys_addr_t
) udev
->mb_addr
;
877 info
->mem
[0].size
= TCMU_RING_SIZE
;
878 info
->mem
[0].memtype
= UIO_MEM_VIRTUAL
;
880 info
->irqcontrol
= tcmu_irqcontrol
;
881 info
->irq
= UIO_IRQ_CUSTOM
;
883 info
->mmap
= tcmu_mmap
;
884 info
->open
= tcmu_open
;
885 info
->release
= tcmu_release
;
887 ret
= uio_register_device(tcmu_root_device
, info
);
891 /* Other attributes can be configured in userspace */
892 dev
->dev_attrib
.hw_block_size
= 512;
893 dev
->dev_attrib
.hw_max_sectors
= 128;
894 dev
->dev_attrib
.hw_queue_depth
= 128;
896 ret
= tcmu_netlink_event(TCMU_CMD_ADDED_DEVICE
, udev
->uio_info
.name
,
897 udev
->uio_info
.uio_dev
->minor
);
904 uio_unregister_device(&udev
->uio_info
);
906 vfree(udev
->mb_addr
);
913 static int tcmu_check_pending_cmd(int id
, void *p
, void *data
)
915 struct tcmu_cmd
*cmd
= p
;
917 if (test_bit(TCMU_CMD_BIT_EXPIRED
, &cmd
->flags
))
922 static void tcmu_free_device(struct se_device
*dev
)
924 struct tcmu_dev
*udev
= TCMU_DEV(dev
);
927 del_timer_sync(&udev
->timeout
);
929 vfree(udev
->mb_addr
);
931 /* Upper layer should drain all requests before calling this */
932 spin_lock_irq(&udev
->commands_lock
);
933 i
= idr_for_each(&udev
->commands
, tcmu_check_pending_cmd
, NULL
);
934 idr_destroy(&udev
->commands
);
935 spin_unlock_irq(&udev
->commands_lock
);
938 /* Device was configured */
939 if (udev
->uio_info
.uio_dev
) {
940 tcmu_netlink_event(TCMU_CMD_REMOVED_DEVICE
, udev
->uio_info
.name
,
941 udev
->uio_info
.uio_dev
->minor
);
943 uio_unregister_device(&udev
->uio_info
);
944 kfree(udev
->uio_info
.name
);
952 Opt_dev_config
, Opt_dev_size
, Opt_err
, Opt_pass_level
,
955 static match_table_t tokens
= {
956 {Opt_dev_config
, "dev_config=%s"},
957 {Opt_dev_size
, "dev_size=%u"},
958 {Opt_pass_level
, "pass_level=%u"},
962 static ssize_t
tcmu_set_configfs_dev_params(struct se_device
*dev
,
963 const char *page
, ssize_t count
)
965 struct tcmu_dev
*udev
= TCMU_DEV(dev
);
966 char *orig
, *ptr
, *opts
, *arg_p
;
967 substring_t args
[MAX_OPT_ARGS
];
971 opts
= kstrdup(page
, GFP_KERNEL
);
977 while ((ptr
= strsep(&opts
, ",\n")) != NULL
) {
981 token
= match_token(ptr
, tokens
, args
);
984 if (match_strlcpy(udev
->dev_config
, &args
[0],
985 TCMU_CONFIG_LEN
) == 0) {
989 pr_debug("TCMU: Referencing Path: %s\n", udev
->dev_config
);
992 arg_p
= match_strdup(&args
[0]);
997 ret
= kstrtoul(arg_p
, 0, (unsigned long *) &udev
->dev_size
);
1000 pr_err("kstrtoul() failed for dev_size=\n");
1002 case Opt_pass_level
:
1003 match_int(args
, &arg
);
1004 if (arg
>= TCMU_PASS_INVALID
) {
1005 pr_warn("TCMU: Invalid pass_level: %d\n", arg
);
1009 pr_debug("TCMU: Setting pass_level to %d\n", arg
);
1010 udev
->pass_level
= arg
;
1018 return (!ret
) ? count
: ret
;
1021 static ssize_t
tcmu_show_configfs_dev_params(struct se_device
*dev
, char *b
)
1023 struct tcmu_dev
*udev
= TCMU_DEV(dev
);
1026 bl
= sprintf(b
+ bl
, "Config: %s ",
1027 udev
->dev_config
[0] ? udev
->dev_config
: "NULL");
1028 bl
+= sprintf(b
+ bl
, "Size: %zu PassLevel: %u\n",
1029 udev
->dev_size
, udev
->pass_level
);
1034 static sector_t
tcmu_get_blocks(struct se_device
*dev
)
1036 struct tcmu_dev
*udev
= TCMU_DEV(dev
);
1038 return div_u64(udev
->dev_size
- dev
->dev_attrib
.block_size
,
1039 dev
->dev_attrib
.block_size
);
1042 static sense_reason_t
1043 tcmu_execute_rw(struct se_cmd
*se_cmd
, struct scatterlist
*sgl
, u32 sgl_nents
,
1044 enum dma_data_direction data_direction
)
1048 ret
= tcmu_queue_cmd(se_cmd
);
1051 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
;
1053 return TCM_NO_SENSE
;
1056 static sense_reason_t
1057 tcmu_pass_op(struct se_cmd
*se_cmd
)
1059 int ret
= tcmu_queue_cmd(se_cmd
);
1062 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
;
1064 return TCM_NO_SENSE
;
1067 static struct sbc_ops tcmu_sbc_ops
= {
1068 .execute_rw
= tcmu_execute_rw
,
1069 .execute_sync_cache
= tcmu_pass_op
,
1070 .execute_write_same
= tcmu_pass_op
,
1071 .execute_write_same_unmap
= tcmu_pass_op
,
1072 .execute_unmap
= tcmu_pass_op
,
1075 static sense_reason_t
1076 tcmu_parse_cdb(struct se_cmd
*cmd
)
1078 unsigned char *cdb
= cmd
->t_task_cdb
;
1079 struct tcmu_dev
*udev
= TCMU_DEV(cmd
->se_dev
);
1082 switch (udev
->pass_level
) {
1084 /* We're just like pscsi, then */
1086 * For REPORT LUNS we always need to emulate the response, for everything
1091 cmd
->execute_cmd
= spc_emulate_report_luns
;
1102 cmd
->se_cmd_flags
|= SCF_SCSI_DATA_CDB
;
1105 cmd
->execute_cmd
= tcmu_pass_op
;
1110 ret
= sbc_parse_cdb(cmd
, &tcmu_sbc_ops
);
1113 pr_err("Unknown tcm-user pass level %d\n", udev
->pass_level
);
1114 ret
= TCM_CHECK_CONDITION_ABORT_CMD
;
1120 DEF_TB_DEFAULT_ATTRIBS(tcmu
);
1122 static struct configfs_attribute
*tcmu_backend_dev_attrs
[] = {
1123 &tcmu_dev_attrib_emulate_model_alias
.attr
,
1124 &tcmu_dev_attrib_emulate_dpo
.attr
,
1125 &tcmu_dev_attrib_emulate_fua_write
.attr
,
1126 &tcmu_dev_attrib_emulate_fua_read
.attr
,
1127 &tcmu_dev_attrib_emulate_write_cache
.attr
,
1128 &tcmu_dev_attrib_emulate_ua_intlck_ctrl
.attr
,
1129 &tcmu_dev_attrib_emulate_tas
.attr
,
1130 &tcmu_dev_attrib_emulate_tpu
.attr
,
1131 &tcmu_dev_attrib_emulate_tpws
.attr
,
1132 &tcmu_dev_attrib_emulate_caw
.attr
,
1133 &tcmu_dev_attrib_emulate_3pc
.attr
,
1134 &tcmu_dev_attrib_pi_prot_type
.attr
,
1135 &tcmu_dev_attrib_hw_pi_prot_type
.attr
,
1136 &tcmu_dev_attrib_pi_prot_format
.attr
,
1137 &tcmu_dev_attrib_enforce_pr_isids
.attr
,
1138 &tcmu_dev_attrib_is_nonrot
.attr
,
1139 &tcmu_dev_attrib_emulate_rest_reord
.attr
,
1140 &tcmu_dev_attrib_force_pr_aptpl
.attr
,
1141 &tcmu_dev_attrib_hw_block_size
.attr
,
1142 &tcmu_dev_attrib_block_size
.attr
,
1143 &tcmu_dev_attrib_hw_max_sectors
.attr
,
1144 &tcmu_dev_attrib_optimal_sectors
.attr
,
1145 &tcmu_dev_attrib_hw_queue_depth
.attr
,
1146 &tcmu_dev_attrib_queue_depth
.attr
,
1147 &tcmu_dev_attrib_max_unmap_lba_count
.attr
,
1148 &tcmu_dev_attrib_max_unmap_block_desc_count
.attr
,
1149 &tcmu_dev_attrib_unmap_granularity
.attr
,
1150 &tcmu_dev_attrib_unmap_granularity_alignment
.attr
,
1151 &tcmu_dev_attrib_max_write_same_len
.attr
,
1155 static struct se_subsystem_api tcmu_template
= {
1157 .inquiry_prod
= "USER",
1158 .inquiry_rev
= TCMU_VERSION
,
1159 .owner
= THIS_MODULE
,
1160 .transport_type
= TRANSPORT_PLUGIN_VHBA_PDEV
,
1161 .attach_hba
= tcmu_attach_hba
,
1162 .detach_hba
= tcmu_detach_hba
,
1163 .alloc_device
= tcmu_alloc_device
,
1164 .configure_device
= tcmu_configure_device
,
1165 .free_device
= tcmu_free_device
,
1166 .parse_cdb
= tcmu_parse_cdb
,
1167 .set_configfs_dev_params
= tcmu_set_configfs_dev_params
,
1168 .show_configfs_dev_params
= tcmu_show_configfs_dev_params
,
1169 .get_device_type
= sbc_get_device_type
,
1170 .get_blocks
= tcmu_get_blocks
,
1173 static int __init
tcmu_module_init(void)
1175 struct target_backend_cits
*tbc
= &tcmu_template
.tb_cits
;
1178 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry
) % TCMU_OP_ALIGN_SIZE
) != 0);
1180 tcmu_cmd_cache
= kmem_cache_create("tcmu_cmd_cache",
1181 sizeof(struct tcmu_cmd
),
1182 __alignof__(struct tcmu_cmd
),
1184 if (!tcmu_cmd_cache
)
1187 tcmu_root_device
= root_device_register("tcm_user");
1188 if (IS_ERR(tcmu_root_device
)) {
1189 ret
= PTR_ERR(tcmu_root_device
);
1190 goto out_free_cache
;
1193 ret
= genl_register_family(&tcmu_genl_family
);
1195 goto out_unreg_device
;
1198 target_core_setup_sub_cits(&tcmu_template
);
1199 tbc
->tb_dev_attrib_cit
.ct_attrs
= tcmu_backend_dev_attrs
;
1201 ret
= transport_subsystem_register(&tcmu_template
);
1203 goto out_unreg_genl
;
1208 genl_unregister_family(&tcmu_genl_family
);
1210 root_device_unregister(tcmu_root_device
);
1212 kmem_cache_destroy(tcmu_cmd_cache
);
1217 static void __exit
tcmu_module_exit(void)
1219 transport_subsystem_release(&tcmu_template
);
1220 genl_unregister_family(&tcmu_genl_family
);
1221 root_device_unregister(tcmu_root_device
);
1222 kmem_cache_destroy(tcmu_cmd_cache
);
1225 MODULE_DESCRIPTION("TCM USER subsystem plugin");
1226 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
1227 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
1228 MODULE_LICENSE("GPL");
1230 module_init(tcmu_module_init
);
1231 module_exit(tcmu_module_exit
);