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,
568 to_offset == iov_tail(*iov)) {
569 (*iov)->iov_len += copy_bytes;
571 new_iov(iov, iov_cnt, udev);
572 (*iov)->iov_base = (void __user *)to_offset;
573 (*iov)->iov_len = copy_bytes;
576 offset = DATA_BLOCK_SIZE - block_remaining;
578 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;
640 tcmu_flush_dcache_range(from, copy_bytes);
641 memcpy(to + sg->length - sg_remaining, from + offset,
644 sg_remaining -= copy_bytes;
645 block_remaining -= copy_bytes;
647 kunmap_atomic(to - sg->offset);
653 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
655 return DATA_BLOCK_SIZE * (thresh - bitmap_weight(bitmap, thresh));
659 * We can't queue a command until we have space available on the cmd ring *and*
660 * space available on the data area.
662 * Called with ring lock held.
664 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
665 size_t cmd_size, size_t data_needed)
667 struct tcmu_mailbox *mb = udev->mb_addr;
668 uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
670 size_t space, cmd_needed;
673 tcmu_flush_dcache_range(mb, sizeof(*mb));
675 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
678 * If cmd end-of-ring space is too small then we need space for a NOP plus
679 * original cmd - cmds are internally contiguous.
681 if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
682 cmd_needed = cmd_size;
684 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
686 space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
687 if (space < cmd_needed) {
688 pr_debug("no cmd space: %u %u %u\n", cmd_head,
689 udev->cmdr_last_cleaned, udev->cmdr_size);
693 /* try to check and get the data blocks as needed */
694 space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
695 if (space < data_needed) {
696 unsigned long blocks_left = DATA_BLOCK_BITS - udev->dbi_thresh;
699 if (blocks_left < blocks_needed) {
700 pr_debug("no data space: only %lu available, but ask for %zu\n",
701 blocks_left * DATA_BLOCK_SIZE,
706 /* Try to expand the thresh */
707 if (!udev->dbi_thresh) {
708 /* From idle state */
709 uint32_t init_thresh = DATA_BLOCK_INIT_BITS;
711 udev->dbi_thresh = max(blocks_needed, init_thresh);
714 * Grow the data area by max(blocks needed,
715 * dbi_thresh / 2), but limited to the max
716 * DATA_BLOCK_BITS size.
718 grow = max(blocks_needed, udev->dbi_thresh / 2);
719 udev->dbi_thresh += grow;
720 if (udev->dbi_thresh > DATA_BLOCK_BITS)
721 udev->dbi_thresh = DATA_BLOCK_BITS;
725 return tcmu_get_empty_blocks(udev, cmd);
728 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
730 return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
731 sizeof(struct tcmu_cmd_entry));
734 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
735 size_t base_command_size)
737 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
740 command_size = base_command_size +
741 round_up(scsi_command_size(se_cmd->t_task_cdb),
744 WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
749 static sense_reason_t
750 tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
752 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
753 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
754 size_t base_command_size, command_size;
755 struct tcmu_mailbox *mb;
756 struct tcmu_cmd_entry *entry;
761 bool copy_to_data_area;
762 size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
764 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
765 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
768 * Must be a certain minimum size for response sense info, but
769 * also may be larger if the iov array is large.
771 * We prepare as many iovs as possbile for potential uses here,
772 * because it's expensive to tell how many regions are freed in
773 * the bitmap & global data pool, as the size calculated here
774 * will only be used to do the checks.
776 * The size will be recalculated later as actually needed to save
779 base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
780 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
782 mutex_lock(&udev->cmdr_lock);
785 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
786 if ((command_size > (udev->cmdr_size / 2)) ||
787 data_length > udev->data_size) {
788 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
789 "cmd ring/data area\n", command_size, data_length,
790 udev->cmdr_size, udev->data_size);
791 mutex_unlock(&udev->cmdr_lock);
792 return TCM_INVALID_CDB_FIELD;
795 while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
800 * Don't leave commands partially setup because the unmap
801 * thread might need the blocks to make forward progress.
803 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
804 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
806 prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
808 pr_debug("sleeping for ring space\n");
809 mutex_unlock(&udev->cmdr_lock);
810 if (udev->cmd_time_out)
811 ret = schedule_timeout(
812 msecs_to_jiffies(udev->cmd_time_out));
814 ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
815 finish_wait(&udev->wait_cmdr, &__wait);
817 pr_warn("tcmu: command timed out\n");
818 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
821 mutex_lock(&udev->cmdr_lock);
823 /* We dropped cmdr_lock, cmd_head is stale */
824 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
827 /* Insert a PAD if end-of-ring space is too small */
828 if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
829 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
831 entry = (void *) mb + CMDR_OFF + cmd_head;
832 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
833 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
834 entry->hdr.cmd_id = 0; /* not used for PAD */
835 entry->hdr.kflags = 0;
836 entry->hdr.uflags = 0;
837 tcmu_flush_dcache_range(entry, sizeof(*entry));
839 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
840 tcmu_flush_dcache_range(mb, sizeof(*mb));
842 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
843 WARN_ON(cmd_head != 0);
846 entry = (void *) mb + CMDR_OFF + cmd_head;
847 memset(entry, 0, command_size);
848 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
849 entry->hdr.cmd_id = tcmu_cmd->cmd_id;
851 /* Handle allocating space from the data area */
852 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
853 iov = &entry->req.iov[0];
855 copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
856 || se_cmd->se_cmd_flags & SCF_BIDI);
857 ret = scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
858 se_cmd->t_data_nents, &iov, &iov_cnt,
861 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
862 mutex_unlock(&udev->cmdr_lock);
864 pr_err("tcmu: alloc and scatter data failed\n");
865 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
867 entry->req.iov_cnt = iov_cnt;
869 /* Handle BIDI commands */
871 if (se_cmd->se_cmd_flags & SCF_BIDI) {
873 ret = scatter_data_area(udev, tcmu_cmd,
874 se_cmd->t_bidi_data_sg,
875 se_cmd->t_bidi_data_nents,
876 &iov, &iov_cnt, false);
878 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
879 mutex_unlock(&udev->cmdr_lock);
881 pr_err("tcmu: alloc and scatter bidi data failed\n");
882 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
885 entry->req.iov_bidi_cnt = iov_cnt;
888 * Recalaulate the command's base size and size according
889 * to the actual needs
891 base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
892 entry->req.iov_bidi_cnt);
893 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
895 tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
897 /* All offsets relative to mb_addr, not start of entry! */
898 cdb_off = CMDR_OFF + cmd_head + base_command_size;
899 memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
900 entry->req.cdb_off = cdb_off;
901 tcmu_flush_dcache_range(entry, sizeof(*entry));
903 UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
904 tcmu_flush_dcache_range(mb, sizeof(*mb));
905 mutex_unlock(&udev->cmdr_lock);
907 /* TODO: only if FLUSH and FUA? */
908 uio_event_notify(&udev->uio_info);
910 if (udev->cmd_time_out)
911 mod_timer(&udev->timeout, round_jiffies_up(jiffies +
912 msecs_to_jiffies(udev->cmd_time_out)));
917 static sense_reason_t
918 tcmu_queue_cmd(struct se_cmd *se_cmd)
920 struct se_device *se_dev = se_cmd->se_dev;
921 struct tcmu_dev *udev = TCMU_DEV(se_dev);
922 struct tcmu_cmd *tcmu_cmd;
925 tcmu_cmd = tcmu_alloc_cmd(se_cmd);
927 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
929 ret = tcmu_queue_cmd_ring(tcmu_cmd);
930 if (ret != TCM_NO_SENSE) {
931 pr_err("TCMU: Could not queue command\n");
932 spin_lock_irq(&udev->commands_lock);
933 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
934 spin_unlock_irq(&udev->commands_lock);
936 tcmu_free_cmd(tcmu_cmd);
942 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
944 struct se_cmd *se_cmd = cmd->se_cmd;
945 struct tcmu_dev *udev = cmd->tcmu_dev;
948 * cmd has been completed already from timeout, just reclaim
949 * data area space and free cmd
951 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
954 tcmu_cmd_reset_dbi_cur(cmd);
956 if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
957 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
959 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
960 } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
961 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
962 } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
963 /* Get Data-In buffer before clean up */
964 gather_data_area(udev, cmd, true);
965 } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
966 gather_data_area(udev, cmd, false);
967 } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
969 } else if (se_cmd->data_direction != DMA_NONE) {
970 pr_warn("TCMU: data direction was %d!\n",
971 se_cmd->data_direction);
974 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
978 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
982 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
984 struct tcmu_mailbox *mb;
987 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
988 pr_err("ring broken, not handling completions\n");
993 tcmu_flush_dcache_range(mb, sizeof(*mb));
995 while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) {
997 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
998 struct tcmu_cmd *cmd;
1001 * Flush max. up to end of cmd ring since current entry might
1002 * be a padding that is shorter than sizeof(*entry)
1004 size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1006 tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1007 ring_left : sizeof(*entry));
1009 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1010 UPDATE_HEAD(udev->cmdr_last_cleaned,
1011 tcmu_hdr_get_len(entry->hdr.len_op),
1015 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1017 spin_lock(&udev->commands_lock);
1018 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1019 spin_unlock(&udev->commands_lock);
1022 pr_err("cmd_id not found, ring is broken\n");
1023 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1027 tcmu_handle_completion(cmd, entry);
1029 UPDATE_HEAD(udev->cmdr_last_cleaned,
1030 tcmu_hdr_get_len(entry->hdr.len_op),
1036 if (mb->cmd_tail == mb->cmd_head)
1037 del_timer(&udev->timeout); /* no more pending cmds */
1039 wake_up(&udev->wait_cmdr);
1044 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1046 struct tcmu_cmd *cmd = p;
1048 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1051 if (!time_after(jiffies, cmd->deadline))
1054 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1055 target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
1061 static void tcmu_device_timedout(unsigned long data)
1063 struct tcmu_dev *udev = (struct tcmu_dev *)data;
1064 unsigned long flags;
1066 spin_lock_irqsave(&udev->commands_lock, flags);
1067 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
1068 spin_unlock_irqrestore(&udev->commands_lock, flags);
1070 /* Try to wake up the ummap thread */
1071 wake_up(&unmap_wait);
1074 * We don't need to wakeup threads on wait_cmdr since they have their
1079 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1081 struct tcmu_hba *tcmu_hba;
1083 tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1087 tcmu_hba->host_id = host_id;
1088 hba->hba_ptr = tcmu_hba;
1093 static void tcmu_detach_hba(struct se_hba *hba)
1095 kfree(hba->hba_ptr);
1096 hba->hba_ptr = NULL;
1099 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1101 struct tcmu_dev *udev;
1103 udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1106 kref_init(&udev->kref);
1108 udev->name = kstrdup(name, GFP_KERNEL);
1115 udev->cmd_time_out = TCMU_TIME_OUT;
1117 init_waitqueue_head(&udev->wait_cmdr);
1118 mutex_init(&udev->cmdr_lock);
1120 idr_init(&udev->commands);
1121 spin_lock_init(&udev->commands_lock);
1123 setup_timer(&udev->timeout, tcmu_device_timedout,
1124 (unsigned long)udev);
1126 init_waitqueue_head(&udev->nl_cmd_wq);
1127 spin_lock_init(&udev->nl_cmd_lock);
1129 return &udev->se_dev;
1132 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1134 struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
1136 mutex_lock(&tcmu_dev->cmdr_lock);
1137 tcmu_handle_completions(tcmu_dev);
1138 mutex_unlock(&tcmu_dev->cmdr_lock);
1144 * mmap code from uio.c. Copied here because we want to hook mmap()
1145 * and this stuff must come along.
1147 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1149 struct tcmu_dev *udev = vma->vm_private_data;
1150 struct uio_info *info = &udev->uio_info;
1152 if (vma->vm_pgoff < MAX_UIO_MAPS) {
1153 if (info->mem[vma->vm_pgoff].size == 0)
1155 return (int)vma->vm_pgoff;
1160 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1165 mutex_lock(&udev->cmdr_lock);
1166 page = tcmu_get_block_page(udev, dbi);
1169 mutex_unlock(&udev->cmdr_lock);
1174 * Normally it shouldn't be here:
1175 * Only when the userspace has touched the blocks which
1176 * are out of the tcmu_cmd's data iov[], and will return
1179 pr_warn("Block(%u) out of cmd's iov[] has been touched!\n", dbi);
1180 pr_warn("Mostly it will be a bug of userspace, please have a check!\n");
1182 if (dbi >= udev->dbi_thresh) {
1183 /* Extern the udev->dbi_thresh to dbi + 1 */
1184 udev->dbi_thresh = dbi + 1;
1185 udev->dbi_max = dbi;
1188 page = radix_tree_lookup(&udev->data_blocks, dbi);
1190 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1192 mutex_unlock(&udev->cmdr_lock);
1196 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
1198 mutex_unlock(&udev->cmdr_lock);
1204 * Since this case is rare in page fault routine, here we
1205 * will allow the global_db_count >= TCMU_GLOBAL_MAX_BLOCKS
1206 * to reduce possible page fault call trace.
1208 atomic_inc(&global_db_count);
1210 mutex_unlock(&udev->cmdr_lock);
1215 static int tcmu_vma_fault(struct vm_fault *vmf)
1217 struct tcmu_dev *udev = vmf->vma->vm_private_data;
1218 struct uio_info *info = &udev->uio_info;
1220 unsigned long offset;
1223 int mi = tcmu_find_mem_index(vmf->vma);
1225 return VM_FAULT_SIGBUS;
1228 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1231 offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1233 if (offset < udev->data_off) {
1234 /* For the vmalloc()ed cmd area pages */
1235 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1236 page = vmalloc_to_page(addr);
1241 /* For the dynamically growing data area pages */
1242 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1243 page = tcmu_try_get_block_page(udev, dbi);
1245 return VM_FAULT_NOPAGE;
1252 static const struct vm_operations_struct tcmu_vm_ops = {
1253 .fault = tcmu_vma_fault,
1256 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1258 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1260 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1261 vma->vm_ops = &tcmu_vm_ops;
1263 vma->vm_private_data = udev;
1265 /* Ensure the mmap is exactly the right size */
1266 if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
1272 static int tcmu_open(struct uio_info *info, struct inode *inode)
1274 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1276 /* O_EXCL not supported for char devs, so fake it? */
1277 if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1280 udev->inode = inode;
1281 kref_get(&udev->kref);
1288 static void tcmu_dev_call_rcu(struct rcu_head *p)
1290 struct se_device *dev = container_of(p, struct se_device, rcu_head);
1291 struct tcmu_dev *udev = TCMU_DEV(dev);
1293 kfree(udev->uio_info.name);
1298 static void tcmu_dev_kref_release(struct kref *kref)
1300 struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1301 struct se_device *dev = &udev->se_dev;
1303 call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1306 static int tcmu_release(struct uio_info *info, struct inode *inode)
1308 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1310 clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1312 pr_debug("close\n");
1313 /* release ref from open */
1314 kref_put(&udev->kref, tcmu_dev_kref_release);
1318 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1320 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1322 if (!tcmu_kern_cmd_reply_supported)
1325 spin_lock(&udev->nl_cmd_lock);
1327 if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1328 spin_unlock(&udev->nl_cmd_lock);
1329 pr_debug("sleeping for open nl cmd\n");
1330 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1334 memset(nl_cmd, 0, sizeof(*nl_cmd));
1336 init_completion(&nl_cmd->complete);
1338 spin_unlock(&udev->nl_cmd_lock);
1341 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1343 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1345 DEFINE_WAIT(__wait);
1347 if (!tcmu_kern_cmd_reply_supported)
1350 pr_debug("sleeping for nl reply\n");
1351 wait_for_completion(&nl_cmd->complete);
1353 spin_lock(&udev->nl_cmd_lock);
1354 nl_cmd->cmd = TCMU_CMD_UNSPEC;
1355 ret = nl_cmd->status;
1357 spin_unlock(&udev->nl_cmd_lock);
1359 wake_up_all(&udev->nl_cmd_wq);
1364 static int tcmu_netlink_event(struct tcmu_dev *udev, enum tcmu_genl_cmd cmd,
1365 int reconfig_attr, const void *reconfig_data)
1367 struct sk_buff *skb;
1371 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1375 msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1379 ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1383 ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1387 ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1391 if (cmd == TCMU_CMD_RECONFIG_DEVICE) {
1392 switch (reconfig_attr) {
1393 case TCMU_ATTR_DEV_CFG:
1394 ret = nla_put_string(skb, reconfig_attr, reconfig_data);
1396 case TCMU_ATTR_DEV_SIZE:
1397 ret = nla_put_u64_64bit(skb, reconfig_attr,
1398 *((u64 *)reconfig_data),
1401 case TCMU_ATTR_WRITECACHE:
1402 ret = nla_put_u8(skb, reconfig_attr,
1403 *((u8 *)reconfig_data));
1413 genlmsg_end(skb, msg_header);
1415 tcmu_init_genl_cmd_reply(udev, cmd);
1417 ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1418 TCMU_MCGRP_CONFIG, GFP_KERNEL);
1419 /* We don't care if no one is listening */
1423 ret = tcmu_wait_genl_cmd_reply(udev);
1431 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1433 struct tcmu_hba *hba = udev->hba->hba_ptr;
1434 struct uio_info *info;
1438 info = &udev->uio_info;
1439 size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1441 size += 1; /* for \0 */
1442 str = kmalloc(size, GFP_KERNEL);
1446 used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1447 if (udev->dev_config[0])
1448 snprintf(str + used, size - used, "/%s", udev->dev_config);
1450 /* If the old string exists, free it */
1457 static int tcmu_configure_device(struct se_device *dev)
1459 struct tcmu_dev *udev = TCMU_DEV(dev);
1460 struct uio_info *info;
1461 struct tcmu_mailbox *mb;
1464 ret = tcmu_update_uio_info(udev);
1468 info = &udev->uio_info;
1470 udev->mb_addr = vzalloc(CMDR_SIZE);
1471 if (!udev->mb_addr) {
1476 /* mailbox fits in first part of CMDR space */
1477 udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1478 udev->data_off = CMDR_SIZE;
1479 udev->data_size = DATA_SIZE;
1480 udev->dbi_thresh = 0; /* Default in Idle state */
1481 udev->waiting_global = false;
1483 /* Initialise the mailbox of the ring buffer */
1485 mb->version = TCMU_MAILBOX_VERSION;
1486 mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1487 mb->cmdr_off = CMDR_OFF;
1488 mb->cmdr_size = udev->cmdr_size;
1490 WARN_ON(!PAGE_ALIGNED(udev->data_off));
1491 WARN_ON(udev->data_size % PAGE_SIZE);
1492 WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1494 INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1496 info->version = __stringify(TCMU_MAILBOX_VERSION);
1498 info->mem[0].name = "tcm-user command & data buffer";
1499 info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1500 info->mem[0].size = TCMU_RING_SIZE;
1501 info->mem[0].memtype = UIO_MEM_NONE;
1503 info->irqcontrol = tcmu_irqcontrol;
1504 info->irq = UIO_IRQ_CUSTOM;
1506 info->mmap = tcmu_mmap;
1507 info->open = tcmu_open;
1508 info->release = tcmu_release;
1510 ret = uio_register_device(tcmu_root_device, info);
1514 /* User can set hw_block_size before enable the device */
1515 if (dev->dev_attrib.hw_block_size == 0)
1516 dev->dev_attrib.hw_block_size = 512;
1517 /* Other attributes can be configured in userspace */
1518 if (!dev->dev_attrib.hw_max_sectors)
1519 dev->dev_attrib.hw_max_sectors = 128;
1520 if (!dev->dev_attrib.emulate_write_cache)
1521 dev->dev_attrib.emulate_write_cache = 0;
1522 dev->dev_attrib.hw_queue_depth = 128;
1525 * Get a ref incase userspace does a close on the uio device before
1526 * LIO has initiated tcmu_free_device.
1528 kref_get(&udev->kref);
1530 ret = tcmu_netlink_event(udev, TCMU_CMD_ADDED_DEVICE, 0, NULL);
1534 mutex_lock(&root_udev_mutex);
1535 list_add(&udev->node, &root_udev);
1536 mutex_unlock(&root_udev_mutex);
1541 kref_put(&udev->kref, tcmu_dev_kref_release);
1542 uio_unregister_device(&udev->uio_info);
1544 vfree(udev->mb_addr);
1552 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1554 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1555 kmem_cache_free(tcmu_cmd_cache, cmd);
1561 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1563 return udev->uio_info.uio_dev ? true : false;
1566 static void tcmu_blocks_release(struct tcmu_dev *udev)
1571 /* Try to release all block pages */
1572 mutex_lock(&udev->cmdr_lock);
1573 for (i = 0; i <= udev->dbi_max; i++) {
1574 page = radix_tree_delete(&udev->data_blocks, i);
1577 atomic_dec(&global_db_count);
1580 mutex_unlock(&udev->cmdr_lock);
1583 static void tcmu_free_device(struct se_device *dev)
1585 struct tcmu_dev *udev = TCMU_DEV(dev);
1587 /* release ref from init */
1588 kref_put(&udev->kref, tcmu_dev_kref_release);
1591 static void tcmu_destroy_device(struct se_device *dev)
1593 struct tcmu_dev *udev = TCMU_DEV(dev);
1594 struct tcmu_cmd *cmd;
1595 bool all_expired = true;
1598 del_timer_sync(&udev->timeout);
1600 mutex_lock(&root_udev_mutex);
1601 list_del(&udev->node);
1602 mutex_unlock(&root_udev_mutex);
1604 vfree(udev->mb_addr);
1606 /* Upper layer should drain all requests before calling this */
1607 spin_lock_irq(&udev->commands_lock);
1608 idr_for_each_entry(&udev->commands, cmd, i) {
1609 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1610 all_expired = false;
1612 idr_destroy(&udev->commands);
1613 spin_unlock_irq(&udev->commands_lock);
1614 WARN_ON(!all_expired);
1616 tcmu_blocks_release(udev);
1618 tcmu_netlink_event(udev, TCMU_CMD_REMOVED_DEVICE, 0, NULL);
1620 uio_unregister_device(&udev->uio_info);
1622 /* release ref from configure */
1623 kref_put(&udev->kref, tcmu_dev_kref_release);
1627 Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1631 static match_table_t tokens = {
1632 {Opt_dev_config, "dev_config=%s"},
1633 {Opt_dev_size, "dev_size=%u"},
1634 {Opt_hw_block_size, "hw_block_size=%u"},
1635 {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1639 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1641 unsigned long tmp_ul;
1645 arg_p = match_strdup(arg);
1649 ret = kstrtoul(arg_p, 0, &tmp_ul);
1652 pr_err("kstrtoul() failed for dev attrib\n");
1656 pr_err("dev attrib must be nonzero\n");
1659 *dev_attrib = tmp_ul;
1663 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1664 const char *page, ssize_t count)
1666 struct tcmu_dev *udev = TCMU_DEV(dev);
1667 char *orig, *ptr, *opts, *arg_p;
1668 substring_t args[MAX_OPT_ARGS];
1671 opts = kstrdup(page, GFP_KERNEL);
1677 while ((ptr = strsep(&opts, ",\n")) != NULL) {
1681 token = match_token(ptr, tokens, args);
1683 case Opt_dev_config:
1684 if (match_strlcpy(udev->dev_config, &args[0],
1685 TCMU_CONFIG_LEN) == 0) {
1689 pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1692 arg_p = match_strdup(&args[0]);
1697 ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1700 pr_err("kstrtoul() failed for dev_size=\n");
1702 case Opt_hw_block_size:
1703 ret = tcmu_set_dev_attrib(&args[0],
1704 &(dev->dev_attrib.hw_block_size));
1706 case Opt_hw_max_sectors:
1707 ret = tcmu_set_dev_attrib(&args[0],
1708 &(dev->dev_attrib.hw_max_sectors));
1719 return (!ret) ? count : ret;
1722 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
1724 struct tcmu_dev *udev = TCMU_DEV(dev);
1727 bl = sprintf(b + bl, "Config: %s ",
1728 udev->dev_config[0] ? udev->dev_config : "NULL");
1729 bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size);
1734 static sector_t tcmu_get_blocks(struct se_device *dev)
1736 struct tcmu_dev *udev = TCMU_DEV(dev);
1738 return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1739 dev->dev_attrib.block_size);
1742 static sense_reason_t
1743 tcmu_parse_cdb(struct se_cmd *cmd)
1745 return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
1748 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
1750 struct se_dev_attrib *da = container_of(to_config_group(item),
1751 struct se_dev_attrib, da_group);
1752 struct tcmu_dev *udev = container_of(da->da_dev,
1753 struct tcmu_dev, se_dev);
1755 return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
1758 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
1761 struct se_dev_attrib *da = container_of(to_config_group(item),
1762 struct se_dev_attrib, da_group);
1763 struct tcmu_dev *udev = container_of(da->da_dev,
1764 struct tcmu_dev, se_dev);
1768 if (da->da_dev->export_count) {
1769 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
1773 ret = kstrtou32(page, 0, &val);
1777 udev->cmd_time_out = val * MSEC_PER_SEC;
1780 CONFIGFS_ATTR(tcmu_, cmd_time_out);
1782 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
1784 struct se_dev_attrib *da = container_of(to_config_group(item),
1785 struct se_dev_attrib, da_group);
1786 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1788 return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
1791 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
1794 struct se_dev_attrib *da = container_of(to_config_group(item),
1795 struct se_dev_attrib, da_group);
1796 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1800 if (!len || len > TCMU_CONFIG_LEN - 1)
1803 /* Check if device has been configured before */
1804 if (tcmu_dev_configured(udev)) {
1805 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1806 TCMU_ATTR_DEV_CFG, page);
1808 pr_err("Unable to reconfigure device\n");
1811 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1813 ret = tcmu_update_uio_info(udev);
1818 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1822 CONFIGFS_ATTR(tcmu_, dev_config);
1824 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
1826 struct se_dev_attrib *da = container_of(to_config_group(item),
1827 struct se_dev_attrib, da_group);
1828 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1830 return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
1833 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
1836 struct se_dev_attrib *da = container_of(to_config_group(item),
1837 struct se_dev_attrib, da_group);
1838 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1842 ret = kstrtou64(page, 0, &val);
1846 /* Check if device has been configured before */
1847 if (tcmu_dev_configured(udev)) {
1848 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1849 TCMU_ATTR_DEV_SIZE, &val);
1851 pr_err("Unable to reconfigure device\n");
1855 udev->dev_size = val;
1858 CONFIGFS_ATTR(tcmu_, dev_size);
1860 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
1863 struct se_dev_attrib *da = container_of(to_config_group(item),
1864 struct se_dev_attrib, da_group);
1866 return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
1869 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
1870 const char *page, size_t count)
1872 struct se_dev_attrib *da = container_of(to_config_group(item),
1873 struct se_dev_attrib, da_group);
1874 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1878 ret = kstrtou8(page, 0, &val);
1882 /* Check if device has been configured before */
1883 if (tcmu_dev_configured(udev)) {
1884 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1885 TCMU_ATTR_WRITECACHE, &val);
1887 pr_err("Unable to reconfigure device\n");
1892 da->emulate_write_cache = val;
1895 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
1897 static struct configfs_attribute *tcmu_attrib_attrs[] = {
1898 &tcmu_attr_cmd_time_out,
1899 &tcmu_attr_dev_config,
1900 &tcmu_attr_dev_size,
1901 &tcmu_attr_emulate_write_cache,
1905 static struct configfs_attribute **tcmu_attrs;
1907 static struct target_backend_ops tcmu_ops = {
1909 .owner = THIS_MODULE,
1910 .transport_flags = TRANSPORT_FLAG_PASSTHROUGH,
1911 .attach_hba = tcmu_attach_hba,
1912 .detach_hba = tcmu_detach_hba,
1913 .alloc_device = tcmu_alloc_device,
1914 .configure_device = tcmu_configure_device,
1915 .destroy_device = tcmu_destroy_device,
1916 .free_device = tcmu_free_device,
1917 .parse_cdb = tcmu_parse_cdb,
1918 .set_configfs_dev_params = tcmu_set_configfs_dev_params,
1919 .show_configfs_dev_params = tcmu_show_configfs_dev_params,
1920 .get_device_type = sbc_get_device_type,
1921 .get_blocks = tcmu_get_blocks,
1922 .tb_dev_attrib_attrs = NULL,
1925 static int unmap_thread_fn(void *data)
1927 struct tcmu_dev *udev;
1929 uint32_t start, end, block;
1933 while (!kthread_should_stop()) {
1934 DEFINE_WAIT(__wait);
1936 prepare_to_wait(&unmap_wait, &__wait, TASK_INTERRUPTIBLE);
1938 finish_wait(&unmap_wait, &__wait);
1940 if (kthread_should_stop())
1943 mutex_lock(&root_udev_mutex);
1944 list_for_each_entry(udev, &root_udev, node) {
1945 mutex_lock(&udev->cmdr_lock);
1947 /* Try to complete the finished commands first */
1948 tcmu_handle_completions(udev);
1950 /* Skip the udevs waiting the global pool or in idle */
1951 if (udev->waiting_global || !udev->dbi_thresh) {
1952 mutex_unlock(&udev->cmdr_lock);
1956 end = udev->dbi_max + 1;
1957 block = find_last_bit(udev->data_bitmap, end);
1958 if (block == udev->dbi_max) {
1960 * The last bit is dbi_max, so there is
1961 * no need to shrink any blocks.
1963 mutex_unlock(&udev->cmdr_lock);
1965 } else if (block == end) {
1966 /* The current udev will goto idle state */
1967 udev->dbi_thresh = start = 0;
1970 udev->dbi_thresh = start = block + 1;
1971 udev->dbi_max = block;
1974 /* Here will truncate the data area from off */
1975 off = udev->data_off + start * DATA_BLOCK_SIZE;
1976 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
1978 /* Release the block pages */
1979 for (i = start; i < end; i++) {
1980 page = radix_tree_delete(&udev->data_blocks, i);
1983 atomic_dec(&global_db_count);
1986 mutex_unlock(&udev->cmdr_lock);
1990 * Try to wake up the udevs who are waiting
1991 * for the global data pool.
1993 list_for_each_entry(udev, &root_udev, node) {
1994 if (udev->waiting_global)
1995 wake_up(&udev->wait_cmdr);
1997 mutex_unlock(&root_udev_mutex);
2003 static int __init tcmu_module_init(void)
2005 int ret, i, k, len = 0;
2007 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2009 tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2010 sizeof(struct tcmu_cmd),
2011 __alignof__(struct tcmu_cmd),
2013 if (!tcmu_cmd_cache)
2016 tcmu_root_device = root_device_register("tcm_user");
2017 if (IS_ERR(tcmu_root_device)) {
2018 ret = PTR_ERR(tcmu_root_device);
2019 goto out_free_cache;
2022 ret = genl_register_family(&tcmu_genl_family);
2024 goto out_unreg_device;
2027 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2028 len += sizeof(struct configfs_attribute *);
2030 for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2031 len += sizeof(struct configfs_attribute *);
2033 len += sizeof(struct configfs_attribute *);
2035 tcmu_attrs = kzalloc(len, GFP_KERNEL);
2038 goto out_unreg_genl;
2041 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2042 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2044 for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2045 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2048 tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2050 ret = transport_backend_register(&tcmu_ops);
2054 init_waitqueue_head(&unmap_wait);
2055 unmap_thread = kthread_run(unmap_thread_fn, NULL, "tcmu_unmap");
2056 if (IS_ERR(unmap_thread)) {
2057 ret = PTR_ERR(unmap_thread);
2058 goto out_unreg_transport;
2063 out_unreg_transport:
2064 target_backend_unregister(&tcmu_ops);
2068 genl_unregister_family(&tcmu_genl_family);
2070 root_device_unregister(tcmu_root_device);
2072 kmem_cache_destroy(tcmu_cmd_cache);
2077 static void __exit tcmu_module_exit(void)
2079 kthread_stop(unmap_thread);
2080 target_backend_unregister(&tcmu_ops);
2082 genl_unregister_family(&tcmu_genl_family);
2083 root_device_unregister(tcmu_root_device);
2084 kmem_cache_destroy(tcmu_cmd_cache);
2087 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2088 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2089 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2090 MODULE_LICENSE("GPL");
2092 module_init(tcmu_module_init);
2093 module_exit(tcmu_module_exit);