2 * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
3 * Copyright (C) 2014 Red Hat, Inc.
4 * Copyright (C) 2015 Arrikto, Inc.
5 * Copyright (C) 2017 Chinamobile, Inc.
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms and conditions of the GNU General Public License,
9 * version 2, as published by the Free Software Foundation.
11 * This program is distributed in the hope it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along with
17 * this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
21 #include <linux/spinlock.h>
22 #include <linux/module.h>
23 #include <linux/idr.h>
24 #include <linux/kernel.h>
25 #include <linux/timer.h>
26 #include <linux/parser.h>
27 #include <linux/vmalloc.h>
28 #include <linux/uio_driver.h>
29 #include <linux/radix-tree.h>
30 #include <linux/stringify.h>
31 #include <linux/bitops.h>
32 #include <linux/highmem.h>
33 #include <linux/configfs.h>
34 #include <linux/mutex.h>
35 #include <linux/kthread.h>
36 #include <net/genetlink.h>
37 #include <scsi/scsi_common.h>
38 #include <scsi/scsi_proto.h>
39 #include <target/target_core_base.h>
40 #include <target/target_core_fabric.h>
41 #include <target/target_core_backend.h>
43 #include <linux/target_core_user.h>
46 * Define a shared-memory interface for LIO to pass SCSI commands and
47 * data to userspace for processing. This is to allow backends that
48 * are too complex for in-kernel support to be possible.
50 * It uses the UIO framework to do a lot of the device-creation and
51 * introspection work for us.
53 * See the .h file for how the ring is laid out. Note that while the
54 * command ring is defined, the particulars of the data area are
55 * not. Offset values in the command entry point to other locations
56 * internal to the mmap()ed area. There is separate space outside the
57 * command ring for data buffers. This leaves maximum flexibility for
58 * moving buffer allocations, or even page flipping or other
59 * allocation techniques, without altering the command ring layout.
62 * The user process must be assumed to be malicious. There's no way to
63 * prevent it breaking the command ring protocol if it wants, but in
64 * order to prevent other issues we must only ever read *data* from
65 * the shared memory area, not offsets or sizes. This applies to
66 * command ring entries as well as the mailbox. Extra code needed for
67 * this may have a 'UAM' comment.
70 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
72 /* For cmd area, the size is fixed 8MB */
73 #define CMDR_SIZE (8 * 1024 * 1024)
76 * For data area, the block size is PAGE_SIZE and
77 * the total size is 256K * PAGE_SIZE.
79 #define DATA_BLOCK_SIZE PAGE_SIZE
80 #define DATA_BLOCK_BITS (256 * 1024)
81 #define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
82 #define DATA_BLOCK_INIT_BITS 128
84 /* The total size of the ring is 8M + 256K * PAGE_SIZE */
85 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
87 /* Default maximum of the global data blocks(512K * PAGE_SIZE) */
88 #define TCMU_GLOBAL_MAX_BLOCKS (512 * 1024)
90 static u8 tcmu_kern_cmd_reply_supported;
92 static struct device *tcmu_root_device;
98 #define TCMU_CONFIG_LEN 256
101 /* wake up thread waiting for reply */
102 struct completion complete;
108 struct list_head node;
110 struct se_device se_dev;
115 #define TCMU_DEV_BIT_OPEN 0
116 #define TCMU_DEV_BIT_BROKEN 1
119 struct uio_info uio_info;
123 struct tcmu_mailbox *mb_addr;
126 u32 cmdr_last_cleaned;
127 /* Offset of data area from start of mb */
128 /* Must add data_off and mb_addr to get the address */
132 wait_queue_head_t wait_cmdr;
133 struct mutex cmdr_lock;
138 DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
139 struct radix_tree_root data_blocks;
142 spinlock_t commands_lock;
144 struct timer_list timeout;
145 unsigned int cmd_time_out;
147 spinlock_t nl_cmd_lock;
148 struct tcmu_nl_cmd curr_nl_cmd;
149 /* wake up threads waiting on curr_nl_cmd */
150 wait_queue_head_t nl_cmd_wq;
152 char dev_config[TCMU_CONFIG_LEN];
155 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
157 #define CMDR_OFF sizeof(struct tcmu_mailbox)
160 struct se_cmd *se_cmd;
161 struct tcmu_dev *tcmu_dev;
165 /* Can't use se_cmd when cleaning up expired cmds, because if
166 cmd has been completed then accessing se_cmd is off limits */
171 unsigned long deadline;
173 #define TCMU_CMD_BIT_EXPIRED 0
177 static struct task_struct *unmap_thread;
178 static wait_queue_head_t unmap_wait;
179 static DEFINE_MUTEX(root_udev_mutex);
180 static LIST_HEAD(root_udev);
182 static atomic_t global_db_count = ATOMIC_INIT(0);
184 static struct kmem_cache *tcmu_cmd_cache;
186 /* multicast group */
187 enum tcmu_multicast_groups {
191 static const struct genl_multicast_group tcmu_mcgrps[] = {
192 [TCMU_MCGRP_CONFIG] = { .name = "config", },
195 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
196 [TCMU_ATTR_DEVICE] = { .type = NLA_STRING },
197 [TCMU_ATTR_MINOR] = { .type = NLA_U32 },
198 [TCMU_ATTR_CMD_STATUS] = { .type = NLA_S32 },
199 [TCMU_ATTR_DEVICE_ID] = { .type = NLA_U32 },
200 [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
203 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
205 struct se_device *dev;
206 struct tcmu_dev *udev;
207 struct tcmu_nl_cmd *nl_cmd;
208 int dev_id, rc, ret = 0;
209 bool is_removed = (completed_cmd == TCMU_CMD_REMOVED_DEVICE);
211 if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
212 !info->attrs[TCMU_ATTR_DEVICE_ID]) {
213 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
217 dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
218 rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
220 dev = target_find_device(dev_id, !is_removed);
222 printk(KERN_ERR "tcmu nl cmd %u/%u completion could not find device with dev id %u.\n",
223 completed_cmd, rc, dev_id);
226 udev = TCMU_DEV(dev);
228 spin_lock(&udev->nl_cmd_lock);
229 nl_cmd = &udev->curr_nl_cmd;
231 pr_debug("genl cmd done got id %d curr %d done %d rc %d\n", dev_id,
232 nl_cmd->cmd, completed_cmd, rc);
234 if (nl_cmd->cmd != completed_cmd) {
235 printk(KERN_ERR "Mismatched commands (Expecting reply for %d. Current %d).\n",
236 completed_cmd, nl_cmd->cmd);
242 spin_unlock(&udev->nl_cmd_lock);
244 target_undepend_item(&dev->dev_group.cg_item);
246 complete(&nl_cmd->complete);
250 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
252 return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
255 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
257 return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
260 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
261 struct genl_info *info)
263 return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
266 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
268 if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
269 tcmu_kern_cmd_reply_supported =
270 nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
271 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
272 tcmu_kern_cmd_reply_supported);
278 static const struct genl_ops tcmu_genl_ops[] = {
280 .cmd = TCMU_CMD_SET_FEATURES,
281 .flags = GENL_ADMIN_PERM,
282 .policy = tcmu_attr_policy,
283 .doit = tcmu_genl_set_features,
286 .cmd = TCMU_CMD_ADDED_DEVICE_DONE,
287 .flags = GENL_ADMIN_PERM,
288 .policy = tcmu_attr_policy,
289 .doit = tcmu_genl_add_dev_done,
292 .cmd = TCMU_CMD_REMOVED_DEVICE_DONE,
293 .flags = GENL_ADMIN_PERM,
294 .policy = tcmu_attr_policy,
295 .doit = tcmu_genl_rm_dev_done,
298 .cmd = TCMU_CMD_RECONFIG_DEVICE_DONE,
299 .flags = GENL_ADMIN_PERM,
300 .policy = tcmu_attr_policy,
301 .doit = tcmu_genl_reconfig_dev_done,
305 /* Our generic netlink family */
306 static struct genl_family tcmu_genl_family __ro_after_init = {
307 .module = THIS_MODULE,
311 .maxattr = TCMU_ATTR_MAX,
312 .mcgrps = tcmu_mcgrps,
313 .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
315 .ops = tcmu_genl_ops,
316 .n_ops = ARRAY_SIZE(tcmu_genl_ops),
319 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
320 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
321 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
322 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
324 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
326 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
329 for (i = 0; i < len; i++)
330 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
333 static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
334 struct tcmu_cmd *tcmu_cmd)
339 dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
340 if (dbi == udev->dbi_thresh)
343 page = radix_tree_lookup(&udev->data_blocks, dbi);
345 if (atomic_add_return(1, &global_db_count) >
346 TCMU_GLOBAL_MAX_BLOCKS) {
347 atomic_dec(&global_db_count);
351 /* try to get new page from the mm */
352 page = alloc_page(GFP_KERNEL);
356 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
361 if (dbi > udev->dbi_max)
364 set_bit(dbi, udev->data_bitmap);
365 tcmu_cmd_set_dbi(tcmu_cmd, dbi);
371 atomic_dec(&global_db_count);
375 static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
376 struct tcmu_cmd *tcmu_cmd)
380 udev->waiting_global = false;
382 for (i = tcmu_cmd->dbi_cur; i < tcmu_cmd->dbi_cnt; i++) {
383 if (!tcmu_get_empty_block(udev, tcmu_cmd))
389 udev->waiting_global = true;
390 /* Try to wake up the unmap thread */
391 wake_up(&unmap_wait);
395 static inline struct page *
396 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
398 return radix_tree_lookup(&udev->data_blocks, dbi);
401 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
403 kfree(tcmu_cmd->dbi);
404 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
407 static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
409 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
410 size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
412 if (se_cmd->se_cmd_flags & SCF_BIDI) {
413 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
414 data_length += round_up(se_cmd->t_bidi_data_sg->length,
421 static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
423 size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
425 return data_length / DATA_BLOCK_SIZE;
428 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
430 struct se_device *se_dev = se_cmd->se_dev;
431 struct tcmu_dev *udev = TCMU_DEV(se_dev);
432 struct tcmu_cmd *tcmu_cmd;
435 tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
439 tcmu_cmd->se_cmd = se_cmd;
440 tcmu_cmd->tcmu_dev = udev;
441 if (udev->cmd_time_out)
442 tcmu_cmd->deadline = jiffies +
443 msecs_to_jiffies(udev->cmd_time_out);
445 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
446 tcmu_cmd->dbi_cnt = tcmu_cmd_get_block_cnt(tcmu_cmd);
447 tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
449 if (!tcmu_cmd->dbi) {
450 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
454 idr_preload(GFP_KERNEL);
455 spin_lock_irq(&udev->commands_lock);
456 cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 0,
457 USHRT_MAX, GFP_NOWAIT);
458 spin_unlock_irq(&udev->commands_lock);
462 tcmu_free_cmd(tcmu_cmd);
465 tcmu_cmd->cmd_id = cmd_id;
470 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
472 unsigned long offset = offset_in_page(vaddr);
474 size = round_up(size+offset, PAGE_SIZE);
478 flush_dcache_page(virt_to_page(vaddr));
484 * Some ring helper functions. We don't assume size is a power of 2 so
485 * we can't use circ_buf.h.
487 static inline size_t spc_used(size_t head, size_t tail, size_t size)
489 int diff = head - tail;
497 static inline size_t spc_free(size_t head, size_t tail, size_t size)
499 /* Keep 1 byte unused or we can't tell full from empty */
500 return (size - spc_used(head, tail, size) - 1);
503 static inline size_t head_to_end(size_t head, size_t size)
508 static inline void new_iov(struct iovec **iov, int *iov_cnt,
509 struct tcmu_dev *udev)
518 memset(iovec, 0, sizeof(struct iovec));
521 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
523 /* offset is relative to mb_addr */
524 static inline size_t get_block_offset_user(struct tcmu_dev *dev,
525 int dbi, int remaining)
527 return dev->data_off + dbi * DATA_BLOCK_SIZE +
528 DATA_BLOCK_SIZE - remaining;
531 static inline size_t iov_tail(struct iovec *iov)
533 return (size_t)iov->iov_base + iov->iov_len;
536 static int scatter_data_area(struct tcmu_dev *udev,
537 struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
538 unsigned int data_nents, struct iovec **iov,
539 int *iov_cnt, bool copy_data)
542 int block_remaining = 0;
543 void *from, *to = NULL;
544 size_t copy_bytes, to_offset, offset;
545 struct scatterlist *sg;
548 for_each_sg(data_sg, sg, data_nents, i) {
549 int sg_remaining = sg->length;
550 from = kmap_atomic(sg_page(sg)) + sg->offset;
551 while (sg_remaining > 0) {
552 if (block_remaining == 0) {
556 block_remaining = DATA_BLOCK_SIZE;
557 dbi = tcmu_cmd_get_dbi(tcmu_cmd);
558 page = tcmu_get_block_page(udev, dbi);
559 to = kmap_atomic(page);
562 copy_bytes = min_t(size_t, sg_remaining,
564 to_offset = get_block_offset_user(udev, dbi,
566 offset = DATA_BLOCK_SIZE - block_remaining;
570 to_offset == iov_tail(*iov)) {
571 (*iov)->iov_len += copy_bytes;
573 new_iov(iov, iov_cnt, udev);
574 (*iov)->iov_base = (void __user *)to_offset;
575 (*iov)->iov_len = copy_bytes;
578 memcpy(to, from + sg->length - sg_remaining,
580 tcmu_flush_dcache_range(to, copy_bytes);
582 sg_remaining -= copy_bytes;
583 block_remaining -= copy_bytes;
585 kunmap_atomic(from - sg->offset);
593 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
596 struct se_cmd *se_cmd = cmd->se_cmd;
598 int block_remaining = 0;
599 void *from = NULL, *to;
600 size_t copy_bytes, offset;
601 struct scatterlist *sg, *data_sg;
603 unsigned int data_nents;
607 data_sg = se_cmd->t_data_sg;
608 data_nents = se_cmd->t_data_nents;
612 * For bidi case, the first count blocks are for Data-Out
613 * buffer blocks, and before gathering the Data-In buffer
614 * the Data-Out buffer blocks should be discarded.
616 count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
618 data_sg = se_cmd->t_bidi_data_sg;
619 data_nents = se_cmd->t_bidi_data_nents;
622 tcmu_cmd_set_dbi_cur(cmd, count);
624 for_each_sg(data_sg, sg, data_nents, i) {
625 int sg_remaining = sg->length;
626 to = kmap_atomic(sg_page(sg)) + sg->offset;
627 while (sg_remaining > 0) {
628 if (block_remaining == 0) {
632 block_remaining = DATA_BLOCK_SIZE;
633 dbi = tcmu_cmd_get_dbi(cmd);
634 page = tcmu_get_block_page(udev, dbi);
635 from = kmap_atomic(page);
637 copy_bytes = min_t(size_t, sg_remaining,
639 offset = DATA_BLOCK_SIZE - block_remaining;
641 tcmu_flush_dcache_range(from, copy_bytes);
642 memcpy(to + sg->length - sg_remaining, from,
645 sg_remaining -= copy_bytes;
646 block_remaining -= copy_bytes;
648 kunmap_atomic(to - sg->offset);
654 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
656 return DATA_BLOCK_SIZE * (thresh - bitmap_weight(bitmap, thresh));
660 * We can't queue a command until we have space available on the cmd ring *and*
661 * space available on the data area.
663 * Called with ring lock held.
665 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
666 size_t cmd_size, size_t data_needed)
668 struct tcmu_mailbox *mb = udev->mb_addr;
669 uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
671 size_t space, cmd_needed;
674 tcmu_flush_dcache_range(mb, sizeof(*mb));
676 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
679 * If cmd end-of-ring space is too small then we need space for a NOP plus
680 * original cmd - cmds are internally contiguous.
682 if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
683 cmd_needed = cmd_size;
685 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
687 space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
688 if (space < cmd_needed) {
689 pr_debug("no cmd space: %u %u %u\n", cmd_head,
690 udev->cmdr_last_cleaned, udev->cmdr_size);
694 /* try to check and get the data blocks as needed */
695 space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
696 if (space < data_needed) {
697 unsigned long blocks_left = DATA_BLOCK_BITS - udev->dbi_thresh;
700 if (blocks_left < blocks_needed) {
701 pr_debug("no data space: only %lu available, but ask for %zu\n",
702 blocks_left * DATA_BLOCK_SIZE,
707 /* Try to expand the thresh */
708 if (!udev->dbi_thresh) {
709 /* From idle state */
710 uint32_t init_thresh = DATA_BLOCK_INIT_BITS;
712 udev->dbi_thresh = max(blocks_needed, init_thresh);
715 * Grow the data area by max(blocks needed,
716 * dbi_thresh / 2), but limited to the max
717 * DATA_BLOCK_BITS size.
719 grow = max(blocks_needed, udev->dbi_thresh / 2);
720 udev->dbi_thresh += grow;
721 if (udev->dbi_thresh > DATA_BLOCK_BITS)
722 udev->dbi_thresh = DATA_BLOCK_BITS;
726 return tcmu_get_empty_blocks(udev, cmd);
729 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
731 return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
732 sizeof(struct tcmu_cmd_entry));
735 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
736 size_t base_command_size)
738 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
741 command_size = base_command_size +
742 round_up(scsi_command_size(se_cmd->t_task_cdb),
745 WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
750 static sense_reason_t
751 tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
753 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
754 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
755 size_t base_command_size, command_size;
756 struct tcmu_mailbox *mb;
757 struct tcmu_cmd_entry *entry;
762 bool copy_to_data_area;
763 size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
765 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
766 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
769 * Must be a certain minimum size for response sense info, but
770 * also may be larger if the iov array is large.
772 * We prepare as many iovs as possbile for potential uses here,
773 * because it's expensive to tell how many regions are freed in
774 * the bitmap & global data pool, as the size calculated here
775 * will only be used to do the checks.
777 * The size will be recalculated later as actually needed to save
780 base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
781 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
783 mutex_lock(&udev->cmdr_lock);
786 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
787 if ((command_size > (udev->cmdr_size / 2)) ||
788 data_length > udev->data_size) {
789 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
790 "cmd ring/data area\n", command_size, data_length,
791 udev->cmdr_size, udev->data_size);
792 mutex_unlock(&udev->cmdr_lock);
793 return TCM_INVALID_CDB_FIELD;
796 while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
800 prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
802 pr_debug("sleeping for ring space\n");
803 mutex_unlock(&udev->cmdr_lock);
804 if (udev->cmd_time_out)
805 ret = schedule_timeout(
806 msecs_to_jiffies(udev->cmd_time_out));
808 ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
809 finish_wait(&udev->wait_cmdr, &__wait);
811 pr_warn("tcmu: command timed out\n");
812 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
815 mutex_lock(&udev->cmdr_lock);
817 /* We dropped cmdr_lock, cmd_head is stale */
818 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
821 /* Insert a PAD if end-of-ring space is too small */
822 if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
823 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
825 entry = (void *) mb + CMDR_OFF + cmd_head;
826 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
827 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
828 entry->hdr.cmd_id = 0; /* not used for PAD */
829 entry->hdr.kflags = 0;
830 entry->hdr.uflags = 0;
831 tcmu_flush_dcache_range(entry, sizeof(*entry));
833 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
834 tcmu_flush_dcache_range(mb, sizeof(*mb));
836 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
837 WARN_ON(cmd_head != 0);
840 entry = (void *) mb + CMDR_OFF + cmd_head;
841 memset(entry, 0, command_size);
842 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
843 entry->hdr.cmd_id = tcmu_cmd->cmd_id;
845 /* Handle allocating space from the data area */
846 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
847 iov = &entry->req.iov[0];
849 copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
850 || se_cmd->se_cmd_flags & SCF_BIDI);
851 ret = scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
852 se_cmd->t_data_nents, &iov, &iov_cnt,
855 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
856 mutex_unlock(&udev->cmdr_lock);
858 pr_err("tcmu: alloc and scatter data failed\n");
859 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
861 entry->req.iov_cnt = iov_cnt;
863 /* Handle BIDI commands */
865 if (se_cmd->se_cmd_flags & SCF_BIDI) {
867 ret = scatter_data_area(udev, tcmu_cmd,
868 se_cmd->t_bidi_data_sg,
869 se_cmd->t_bidi_data_nents,
870 &iov, &iov_cnt, false);
872 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
873 mutex_unlock(&udev->cmdr_lock);
875 pr_err("tcmu: alloc and scatter bidi data failed\n");
876 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
879 entry->req.iov_bidi_cnt = iov_cnt;
882 * Recalaulate the command's base size and size according
883 * to the actual needs
885 base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
886 entry->req.iov_bidi_cnt);
887 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
889 tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
891 /* All offsets relative to mb_addr, not start of entry! */
892 cdb_off = CMDR_OFF + cmd_head + base_command_size;
893 memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
894 entry->req.cdb_off = cdb_off;
895 tcmu_flush_dcache_range(entry, sizeof(*entry));
897 UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
898 tcmu_flush_dcache_range(mb, sizeof(*mb));
899 mutex_unlock(&udev->cmdr_lock);
901 /* TODO: only if FLUSH and FUA? */
902 uio_event_notify(&udev->uio_info);
904 if (udev->cmd_time_out)
905 mod_timer(&udev->timeout, round_jiffies_up(jiffies +
906 msecs_to_jiffies(udev->cmd_time_out)));
911 static sense_reason_t
912 tcmu_queue_cmd(struct se_cmd *se_cmd)
914 struct se_device *se_dev = se_cmd->se_dev;
915 struct tcmu_dev *udev = TCMU_DEV(se_dev);
916 struct tcmu_cmd *tcmu_cmd;
919 tcmu_cmd = tcmu_alloc_cmd(se_cmd);
921 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
923 ret = tcmu_queue_cmd_ring(tcmu_cmd);
924 if (ret != TCM_NO_SENSE) {
925 pr_err("TCMU: Could not queue command\n");
926 spin_lock_irq(&udev->commands_lock);
927 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
928 spin_unlock_irq(&udev->commands_lock);
930 tcmu_free_cmd(tcmu_cmd);
936 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
938 struct se_cmd *se_cmd = cmd->se_cmd;
939 struct tcmu_dev *udev = cmd->tcmu_dev;
942 * cmd has been completed already from timeout, just reclaim
943 * data area space and free cmd
945 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
948 tcmu_cmd_reset_dbi_cur(cmd);
950 if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
951 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
953 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
954 } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
955 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
956 } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
957 /* Get Data-In buffer before clean up */
958 gather_data_area(udev, cmd, true);
959 } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
960 gather_data_area(udev, cmd, false);
961 } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
963 } else if (se_cmd->data_direction != DMA_NONE) {
964 pr_warn("TCMU: data direction was %d!\n",
965 se_cmd->data_direction);
968 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
972 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
976 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
978 struct tcmu_mailbox *mb;
981 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
982 pr_err("ring broken, not handling completions\n");
987 tcmu_flush_dcache_range(mb, sizeof(*mb));
989 while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) {
991 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
992 struct tcmu_cmd *cmd;
994 tcmu_flush_dcache_range(entry, sizeof(*entry));
996 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
997 UPDATE_HEAD(udev->cmdr_last_cleaned,
998 tcmu_hdr_get_len(entry->hdr.len_op),
1002 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1004 spin_lock(&udev->commands_lock);
1005 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1006 spin_unlock(&udev->commands_lock);
1009 pr_err("cmd_id not found, ring is broken\n");
1010 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1014 tcmu_handle_completion(cmd, entry);
1016 UPDATE_HEAD(udev->cmdr_last_cleaned,
1017 tcmu_hdr_get_len(entry->hdr.len_op),
1023 if (mb->cmd_tail == mb->cmd_head)
1024 del_timer(&udev->timeout); /* no more pending cmds */
1026 wake_up(&udev->wait_cmdr);
1031 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1033 struct tcmu_cmd *cmd = p;
1035 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1038 if (!time_after(jiffies, cmd->deadline))
1041 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1042 target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
1048 static void tcmu_device_timedout(unsigned long data)
1050 struct tcmu_dev *udev = (struct tcmu_dev *)data;
1051 unsigned long flags;
1053 spin_lock_irqsave(&udev->commands_lock, flags);
1054 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
1055 spin_unlock_irqrestore(&udev->commands_lock, flags);
1057 /* Try to wake up the ummap thread */
1058 wake_up(&unmap_wait);
1061 * We don't need to wakeup threads on wait_cmdr since they have their
1066 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1068 struct tcmu_hba *tcmu_hba;
1070 tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1074 tcmu_hba->host_id = host_id;
1075 hba->hba_ptr = tcmu_hba;
1080 static void tcmu_detach_hba(struct se_hba *hba)
1082 kfree(hba->hba_ptr);
1083 hba->hba_ptr = NULL;
1086 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1088 struct tcmu_dev *udev;
1090 udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1093 kref_init(&udev->kref);
1095 udev->name = kstrdup(name, GFP_KERNEL);
1102 udev->cmd_time_out = TCMU_TIME_OUT;
1104 init_waitqueue_head(&udev->wait_cmdr);
1105 mutex_init(&udev->cmdr_lock);
1107 idr_init(&udev->commands);
1108 spin_lock_init(&udev->commands_lock);
1110 setup_timer(&udev->timeout, tcmu_device_timedout,
1111 (unsigned long)udev);
1113 init_waitqueue_head(&udev->nl_cmd_wq);
1114 spin_lock_init(&udev->nl_cmd_lock);
1116 return &udev->se_dev;
1119 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1121 struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
1123 mutex_lock(&tcmu_dev->cmdr_lock);
1124 tcmu_handle_completions(tcmu_dev);
1125 mutex_unlock(&tcmu_dev->cmdr_lock);
1131 * mmap code from uio.c. Copied here because we want to hook mmap()
1132 * and this stuff must come along.
1134 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1136 struct tcmu_dev *udev = vma->vm_private_data;
1137 struct uio_info *info = &udev->uio_info;
1139 if (vma->vm_pgoff < MAX_UIO_MAPS) {
1140 if (info->mem[vma->vm_pgoff].size == 0)
1142 return (int)vma->vm_pgoff;
1147 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1152 mutex_lock(&udev->cmdr_lock);
1153 page = tcmu_get_block_page(udev, dbi);
1155 mutex_unlock(&udev->cmdr_lock);
1160 * Normally it shouldn't be here:
1161 * Only when the userspace has touched the blocks which
1162 * are out of the tcmu_cmd's data iov[], and will return
1165 pr_warn("Block(%u) out of cmd's iov[] has been touched!\n", dbi);
1166 pr_warn("Mostly it will be a bug of userspace, please have a check!\n");
1168 if (dbi >= udev->dbi_thresh) {
1169 /* Extern the udev->dbi_thresh to dbi + 1 */
1170 udev->dbi_thresh = dbi + 1;
1171 udev->dbi_max = dbi;
1174 page = radix_tree_lookup(&udev->data_blocks, dbi);
1176 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1178 mutex_unlock(&udev->cmdr_lock);
1182 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
1184 mutex_unlock(&udev->cmdr_lock);
1190 * Since this case is rare in page fault routine, here we
1191 * will allow the global_db_count >= TCMU_GLOBAL_MAX_BLOCKS
1192 * to reduce possible page fault call trace.
1194 atomic_inc(&global_db_count);
1196 mutex_unlock(&udev->cmdr_lock);
1201 static int tcmu_vma_fault(struct vm_fault *vmf)
1203 struct tcmu_dev *udev = vmf->vma->vm_private_data;
1204 struct uio_info *info = &udev->uio_info;
1206 unsigned long offset;
1209 int mi = tcmu_find_mem_index(vmf->vma);
1211 return VM_FAULT_SIGBUS;
1214 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1217 offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1219 if (offset < udev->data_off) {
1220 /* For the vmalloc()ed cmd area pages */
1221 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1222 page = vmalloc_to_page(addr);
1226 /* For the dynamically growing data area pages */
1227 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1228 page = tcmu_try_get_block_page(udev, dbi);
1230 return VM_FAULT_NOPAGE;
1238 static const struct vm_operations_struct tcmu_vm_ops = {
1239 .fault = tcmu_vma_fault,
1242 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1244 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1246 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1247 vma->vm_ops = &tcmu_vm_ops;
1249 vma->vm_private_data = udev;
1251 /* Ensure the mmap is exactly the right size */
1252 if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
1258 static int tcmu_open(struct uio_info *info, struct inode *inode)
1260 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1262 /* O_EXCL not supported for char devs, so fake it? */
1263 if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1266 udev->inode = inode;
1267 kref_get(&udev->kref);
1274 static void tcmu_dev_call_rcu(struct rcu_head *p)
1276 struct se_device *dev = container_of(p, struct se_device, rcu_head);
1277 struct tcmu_dev *udev = TCMU_DEV(dev);
1279 kfree(udev->uio_info.name);
1284 static void tcmu_dev_kref_release(struct kref *kref)
1286 struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1287 struct se_device *dev = &udev->se_dev;
1289 call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1292 static int tcmu_release(struct uio_info *info, struct inode *inode)
1294 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1296 clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1298 pr_debug("close\n");
1299 /* release ref from open */
1300 kref_put(&udev->kref, tcmu_dev_kref_release);
1304 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1306 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1308 if (!tcmu_kern_cmd_reply_supported)
1311 spin_lock(&udev->nl_cmd_lock);
1313 if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1314 spin_unlock(&udev->nl_cmd_lock);
1315 pr_debug("sleeping for open nl cmd\n");
1316 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1320 memset(nl_cmd, 0, sizeof(*nl_cmd));
1322 init_completion(&nl_cmd->complete);
1324 spin_unlock(&udev->nl_cmd_lock);
1327 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1329 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1331 DEFINE_WAIT(__wait);
1333 if (!tcmu_kern_cmd_reply_supported)
1336 pr_debug("sleeping for nl reply\n");
1337 wait_for_completion(&nl_cmd->complete);
1339 spin_lock(&udev->nl_cmd_lock);
1340 nl_cmd->cmd = TCMU_CMD_UNSPEC;
1341 ret = nl_cmd->status;
1343 spin_unlock(&udev->nl_cmd_lock);
1345 wake_up_all(&udev->nl_cmd_wq);
1350 static int tcmu_netlink_event(struct tcmu_dev *udev, enum tcmu_genl_cmd cmd,
1351 int reconfig_attr, const void *reconfig_data)
1353 struct sk_buff *skb;
1357 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1361 msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1365 ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1369 ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1373 ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1377 if (cmd == TCMU_CMD_RECONFIG_DEVICE) {
1378 switch (reconfig_attr) {
1379 case TCMU_ATTR_DEV_CFG:
1380 ret = nla_put_string(skb, reconfig_attr, reconfig_data);
1382 case TCMU_ATTR_DEV_SIZE:
1383 ret = nla_put_u64_64bit(skb, reconfig_attr,
1384 *((u64 *)reconfig_data),
1387 case TCMU_ATTR_WRITECACHE:
1388 ret = nla_put_u8(skb, reconfig_attr,
1389 *((u8 *)reconfig_data));
1399 genlmsg_end(skb, msg_header);
1401 tcmu_init_genl_cmd_reply(udev, cmd);
1403 ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1404 TCMU_MCGRP_CONFIG, GFP_KERNEL);
1405 /* We don't care if no one is listening */
1409 ret = tcmu_wait_genl_cmd_reply(udev);
1417 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1419 struct tcmu_hba *hba = udev->hba->hba_ptr;
1420 struct uio_info *info;
1424 info = &udev->uio_info;
1425 size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1427 size += 1; /* for \0 */
1428 str = kmalloc(size, GFP_KERNEL);
1432 used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1433 if (udev->dev_config[0])
1434 snprintf(str + used, size - used, "/%s", udev->dev_config);
1441 static int tcmu_configure_device(struct se_device *dev)
1443 struct tcmu_dev *udev = TCMU_DEV(dev);
1444 struct uio_info *info;
1445 struct tcmu_mailbox *mb;
1448 ret = tcmu_update_uio_info(udev);
1452 info = &udev->uio_info;
1454 udev->mb_addr = vzalloc(CMDR_SIZE);
1455 if (!udev->mb_addr) {
1460 /* mailbox fits in first part of CMDR space */
1461 udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1462 udev->data_off = CMDR_SIZE;
1463 udev->data_size = DATA_SIZE;
1464 udev->dbi_thresh = 0; /* Default in Idle state */
1465 udev->waiting_global = false;
1467 /* Initialise the mailbox of the ring buffer */
1469 mb->version = TCMU_MAILBOX_VERSION;
1470 mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1471 mb->cmdr_off = CMDR_OFF;
1472 mb->cmdr_size = udev->cmdr_size;
1474 WARN_ON(!PAGE_ALIGNED(udev->data_off));
1475 WARN_ON(udev->data_size % PAGE_SIZE);
1476 WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1478 INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1480 info->version = __stringify(TCMU_MAILBOX_VERSION);
1482 info->mem[0].name = "tcm-user command & data buffer";
1483 info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1484 info->mem[0].size = TCMU_RING_SIZE;
1485 info->mem[0].memtype = UIO_MEM_NONE;
1487 info->irqcontrol = tcmu_irqcontrol;
1488 info->irq = UIO_IRQ_CUSTOM;
1490 info->mmap = tcmu_mmap;
1491 info->open = tcmu_open;
1492 info->release = tcmu_release;
1494 ret = uio_register_device(tcmu_root_device, info);
1498 /* User can set hw_block_size before enable the device */
1499 if (dev->dev_attrib.hw_block_size == 0)
1500 dev->dev_attrib.hw_block_size = 512;
1501 /* Other attributes can be configured in userspace */
1502 if (!dev->dev_attrib.hw_max_sectors)
1503 dev->dev_attrib.hw_max_sectors = 128;
1504 if (!dev->dev_attrib.emulate_write_cache)
1505 dev->dev_attrib.emulate_write_cache = 0;
1506 dev->dev_attrib.hw_queue_depth = 128;
1509 * Get a ref incase userspace does a close on the uio device before
1510 * LIO has initiated tcmu_free_device.
1512 kref_get(&udev->kref);
1514 ret = tcmu_netlink_event(udev, TCMU_CMD_ADDED_DEVICE, 0, NULL);
1518 mutex_lock(&root_udev_mutex);
1519 list_add(&udev->node, &root_udev);
1520 mutex_unlock(&root_udev_mutex);
1525 kref_put(&udev->kref, tcmu_dev_kref_release);
1526 uio_unregister_device(&udev->uio_info);
1528 vfree(udev->mb_addr);
1536 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1538 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1539 kmem_cache_free(tcmu_cmd_cache, cmd);
1545 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1547 return udev->uio_info.uio_dev ? true : false;
1550 static void tcmu_blocks_release(struct tcmu_dev *udev)
1555 /* Try to release all block pages */
1556 mutex_lock(&udev->cmdr_lock);
1557 for (i = 0; i <= udev->dbi_max; i++) {
1558 page = radix_tree_delete(&udev->data_blocks, i);
1561 atomic_dec(&global_db_count);
1564 mutex_unlock(&udev->cmdr_lock);
1567 static void tcmu_free_device(struct se_device *dev)
1569 struct tcmu_dev *udev = TCMU_DEV(dev);
1571 /* release ref from init */
1572 kref_put(&udev->kref, tcmu_dev_kref_release);
1575 static void tcmu_destroy_device(struct se_device *dev)
1577 struct tcmu_dev *udev = TCMU_DEV(dev);
1578 struct tcmu_cmd *cmd;
1579 bool all_expired = true;
1582 del_timer_sync(&udev->timeout);
1584 mutex_lock(&root_udev_mutex);
1585 list_del(&udev->node);
1586 mutex_unlock(&root_udev_mutex);
1588 vfree(udev->mb_addr);
1590 /* Upper layer should drain all requests before calling this */
1591 spin_lock_irq(&udev->commands_lock);
1592 idr_for_each_entry(&udev->commands, cmd, i) {
1593 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1594 all_expired = false;
1596 idr_destroy(&udev->commands);
1597 spin_unlock_irq(&udev->commands_lock);
1598 WARN_ON(!all_expired);
1600 tcmu_blocks_release(udev);
1602 tcmu_netlink_event(udev, TCMU_CMD_REMOVED_DEVICE, 0, NULL);
1604 uio_unregister_device(&udev->uio_info);
1606 /* release ref from configure */
1607 kref_put(&udev->kref, tcmu_dev_kref_release);
1611 Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1615 static match_table_t tokens = {
1616 {Opt_dev_config, "dev_config=%s"},
1617 {Opt_dev_size, "dev_size=%u"},
1618 {Opt_hw_block_size, "hw_block_size=%u"},
1619 {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1623 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1625 unsigned long tmp_ul;
1629 arg_p = match_strdup(arg);
1633 ret = kstrtoul(arg_p, 0, &tmp_ul);
1636 pr_err("kstrtoul() failed for dev attrib\n");
1640 pr_err("dev attrib must be nonzero\n");
1643 *dev_attrib = tmp_ul;
1647 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1648 const char *page, ssize_t count)
1650 struct tcmu_dev *udev = TCMU_DEV(dev);
1651 char *orig, *ptr, *opts, *arg_p;
1652 substring_t args[MAX_OPT_ARGS];
1655 opts = kstrdup(page, GFP_KERNEL);
1661 while ((ptr = strsep(&opts, ",\n")) != NULL) {
1665 token = match_token(ptr, tokens, args);
1667 case Opt_dev_config:
1668 if (match_strlcpy(udev->dev_config, &args[0],
1669 TCMU_CONFIG_LEN) == 0) {
1673 pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1676 arg_p = match_strdup(&args[0]);
1681 ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1684 pr_err("kstrtoul() failed for dev_size=\n");
1686 case Opt_hw_block_size:
1687 ret = tcmu_set_dev_attrib(&args[0],
1688 &(dev->dev_attrib.hw_block_size));
1690 case Opt_hw_max_sectors:
1691 ret = tcmu_set_dev_attrib(&args[0],
1692 &(dev->dev_attrib.hw_max_sectors));
1703 return (!ret) ? count : ret;
1706 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
1708 struct tcmu_dev *udev = TCMU_DEV(dev);
1711 bl = sprintf(b + bl, "Config: %s ",
1712 udev->dev_config[0] ? udev->dev_config : "NULL");
1713 bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size);
1718 static sector_t tcmu_get_blocks(struct se_device *dev)
1720 struct tcmu_dev *udev = TCMU_DEV(dev);
1722 return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1723 dev->dev_attrib.block_size);
1726 static sense_reason_t
1727 tcmu_parse_cdb(struct se_cmd *cmd)
1729 return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
1732 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
1734 struct se_dev_attrib *da = container_of(to_config_group(item),
1735 struct se_dev_attrib, da_group);
1736 struct tcmu_dev *udev = container_of(da->da_dev,
1737 struct tcmu_dev, se_dev);
1739 return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
1742 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
1745 struct se_dev_attrib *da = container_of(to_config_group(item),
1746 struct se_dev_attrib, da_group);
1747 struct tcmu_dev *udev = container_of(da->da_dev,
1748 struct tcmu_dev, se_dev);
1752 if (da->da_dev->export_count) {
1753 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
1757 ret = kstrtou32(page, 0, &val);
1761 udev->cmd_time_out = val * MSEC_PER_SEC;
1764 CONFIGFS_ATTR(tcmu_, cmd_time_out);
1766 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
1768 struct se_dev_attrib *da = container_of(to_config_group(item),
1769 struct se_dev_attrib, da_group);
1770 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1772 return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
1775 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
1778 struct se_dev_attrib *da = container_of(to_config_group(item),
1779 struct se_dev_attrib, da_group);
1780 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1784 if (!len || len > TCMU_CONFIG_LEN - 1)
1787 /* Check if device has been configured before */
1788 if (tcmu_dev_configured(udev)) {
1789 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1790 TCMU_ATTR_DEV_CFG, page);
1792 pr_err("Unable to reconfigure device\n");
1795 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1797 ret = tcmu_update_uio_info(udev);
1802 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1806 CONFIGFS_ATTR(tcmu_, dev_config);
1808 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
1810 struct se_dev_attrib *da = container_of(to_config_group(item),
1811 struct se_dev_attrib, da_group);
1812 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1814 return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
1817 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
1820 struct se_dev_attrib *da = container_of(to_config_group(item),
1821 struct se_dev_attrib, da_group);
1822 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1826 ret = kstrtou64(page, 0, &val);
1830 /* Check if device has been configured before */
1831 if (tcmu_dev_configured(udev)) {
1832 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1833 TCMU_ATTR_DEV_SIZE, &val);
1835 pr_err("Unable to reconfigure device\n");
1839 udev->dev_size = val;
1842 CONFIGFS_ATTR(tcmu_, dev_size);
1844 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
1847 struct se_dev_attrib *da = container_of(to_config_group(item),
1848 struct se_dev_attrib, da_group);
1850 return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
1853 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
1854 const char *page, size_t count)
1856 struct se_dev_attrib *da = container_of(to_config_group(item),
1857 struct se_dev_attrib, da_group);
1858 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1862 ret = kstrtou8(page, 0, &val);
1866 /* Check if device has been configured before */
1867 if (tcmu_dev_configured(udev)) {
1868 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1869 TCMU_ATTR_WRITECACHE, &val);
1871 pr_err("Unable to reconfigure device\n");
1876 da->emulate_write_cache = val;
1879 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
1881 static struct configfs_attribute *tcmu_attrib_attrs[] = {
1882 &tcmu_attr_cmd_time_out,
1883 &tcmu_attr_dev_config,
1884 &tcmu_attr_dev_size,
1885 &tcmu_attr_emulate_write_cache,
1889 static struct configfs_attribute **tcmu_attrs;
1891 static struct target_backend_ops tcmu_ops = {
1893 .owner = THIS_MODULE,
1894 .transport_flags = TRANSPORT_FLAG_PASSTHROUGH,
1895 .attach_hba = tcmu_attach_hba,
1896 .detach_hba = tcmu_detach_hba,
1897 .alloc_device = tcmu_alloc_device,
1898 .configure_device = tcmu_configure_device,
1899 .destroy_device = tcmu_destroy_device,
1900 .free_device = tcmu_free_device,
1901 .parse_cdb = tcmu_parse_cdb,
1902 .set_configfs_dev_params = tcmu_set_configfs_dev_params,
1903 .show_configfs_dev_params = tcmu_show_configfs_dev_params,
1904 .get_device_type = sbc_get_device_type,
1905 .get_blocks = tcmu_get_blocks,
1906 .tb_dev_attrib_attrs = NULL,
1909 static int unmap_thread_fn(void *data)
1911 struct tcmu_dev *udev;
1913 uint32_t start, end, block;
1917 while (!kthread_should_stop()) {
1918 DEFINE_WAIT(__wait);
1920 prepare_to_wait(&unmap_wait, &__wait, TASK_INTERRUPTIBLE);
1922 finish_wait(&unmap_wait, &__wait);
1924 if (kthread_should_stop())
1927 mutex_lock(&root_udev_mutex);
1928 list_for_each_entry(udev, &root_udev, node) {
1929 mutex_lock(&udev->cmdr_lock);
1931 /* Try to complete the finished commands first */
1932 tcmu_handle_completions(udev);
1934 /* Skip the udevs waiting the global pool or in idle */
1935 if (udev->waiting_global || !udev->dbi_thresh) {
1936 mutex_unlock(&udev->cmdr_lock);
1940 end = udev->dbi_max + 1;
1941 block = find_last_bit(udev->data_bitmap, end);
1942 if (block == udev->dbi_max) {
1944 * The last bit is dbi_max, so there is
1945 * no need to shrink any blocks.
1947 mutex_unlock(&udev->cmdr_lock);
1949 } else if (block == end) {
1950 /* The current udev will goto idle state */
1951 udev->dbi_thresh = start = 0;
1954 udev->dbi_thresh = start = block + 1;
1955 udev->dbi_max = block;
1958 /* Here will truncate the data area from off */
1959 off = udev->data_off + start * DATA_BLOCK_SIZE;
1960 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
1962 /* Release the block pages */
1963 for (i = start; i < end; i++) {
1964 page = radix_tree_delete(&udev->data_blocks, i);
1967 atomic_dec(&global_db_count);
1970 mutex_unlock(&udev->cmdr_lock);
1974 * Try to wake up the udevs who are waiting
1975 * for the global data pool.
1977 list_for_each_entry(udev, &root_udev, node) {
1978 if (udev->waiting_global)
1979 wake_up(&udev->wait_cmdr);
1981 mutex_unlock(&root_udev_mutex);
1987 static int __init tcmu_module_init(void)
1989 int ret, i, k, len = 0;
1991 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
1993 tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
1994 sizeof(struct tcmu_cmd),
1995 __alignof__(struct tcmu_cmd),
1997 if (!tcmu_cmd_cache)
2000 tcmu_root_device = root_device_register("tcm_user");
2001 if (IS_ERR(tcmu_root_device)) {
2002 ret = PTR_ERR(tcmu_root_device);
2003 goto out_free_cache;
2006 ret = genl_register_family(&tcmu_genl_family);
2008 goto out_unreg_device;
2011 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2012 len += sizeof(struct configfs_attribute *);
2014 for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2015 len += sizeof(struct configfs_attribute *);
2017 len += sizeof(struct configfs_attribute *);
2019 tcmu_attrs = kzalloc(len, GFP_KERNEL);
2022 goto out_unreg_genl;
2025 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2026 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2028 for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2029 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2032 tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2034 ret = transport_backend_register(&tcmu_ops);
2038 init_waitqueue_head(&unmap_wait);
2039 unmap_thread = kthread_run(unmap_thread_fn, NULL, "tcmu_unmap");
2040 if (IS_ERR(unmap_thread)) {
2041 ret = PTR_ERR(unmap_thread);
2042 goto out_unreg_transport;
2047 out_unreg_transport:
2048 target_backend_unregister(&tcmu_ops);
2052 genl_unregister_family(&tcmu_genl_family);
2054 root_device_unregister(tcmu_root_device);
2056 kmem_cache_destroy(tcmu_cmd_cache);
2061 static void __exit tcmu_module_exit(void)
2063 kthread_stop(unmap_thread);
2064 target_backend_unregister(&tcmu_ops);
2066 genl_unregister_family(&tcmu_genl_family);
2067 root_device_unregister(tcmu_root_device);
2068 kmem_cache_destroy(tcmu_cmd_cache);
2071 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2072 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2073 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2074 MODULE_LICENSE("GPL");
2076 module_init(tcmu_module_init);
2077 module_exit(tcmu_module_exit);