1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
4 * Copyright (C) 2014 Red Hat, Inc.
5 * Copyright (C) 2015 Arrikto, Inc.
6 * Copyright (C) 2017 Chinamobile, Inc.
9 #include <linux/spinlock.h>
10 #include <linux/module.h>
11 #include <linux/idr.h>
12 #include <linux/kernel.h>
13 #include <linux/timer.h>
14 #include <linux/parser.h>
15 #include <linux/vmalloc.h>
16 #include <linux/uio_driver.h>
17 #include <linux/radix-tree.h>
18 #include <linux/stringify.h>
19 #include <linux/bitops.h>
20 #include <linux/highmem.h>
21 #include <linux/configfs.h>
22 #include <linux/mutex.h>
23 #include <linux/workqueue.h>
24 #include <net/genetlink.h>
25 #include <scsi/scsi_common.h>
26 #include <scsi/scsi_proto.h>
27 #include <target/target_core_base.h>
28 #include <target/target_core_fabric.h>
29 #include <target/target_core_backend.h>
31 #include <linux/target_core_user.h>
38 * Define a shared-memory interface for LIO to pass SCSI commands and
39 * data to userspace for processing. This is to allow backends that
40 * are too complex for in-kernel support to be possible.
42 * It uses the UIO framework to do a lot of the device-creation and
43 * introspection work for us.
45 * See the .h file for how the ring is laid out. Note that while the
46 * command ring is defined, the particulars of the data area are
47 * not. Offset values in the command entry point to other locations
48 * internal to the mmap-ed area. There is separate space outside the
49 * command ring for data buffers. This leaves maximum flexibility for
50 * moving buffer allocations, or even page flipping or other
51 * allocation techniques, without altering the command ring layout.
54 * The user process must be assumed to be malicious. There's no way to
55 * prevent it breaking the command ring protocol if it wants, but in
56 * order to prevent other issues we must only ever read *data* from
57 * the shared memory area, not offsets or sizes. This applies to
58 * command ring entries as well as the mailbox. Extra code needed for
59 * this may have a 'UAM' comment.
62 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
64 /* For cmd area, the size is fixed 8MB */
65 #define CMDR_SIZE (8 * 1024 * 1024)
68 * For data area, the block size is PAGE_SIZE and
69 * the total size is 256K * PAGE_SIZE.
71 #define DATA_BLOCK_SIZE PAGE_SIZE
72 #define DATA_BLOCK_SHIFT PAGE_SHIFT
73 #define DATA_BLOCK_BITS_DEF (256 * 1024)
75 #define TCMU_MBS_TO_BLOCKS(_mbs) (_mbs << (20 - DATA_BLOCK_SHIFT))
76 #define TCMU_BLOCKS_TO_MBS(_blocks) (_blocks >> (20 - DATA_BLOCK_SHIFT))
79 * Default number of global data blocks(512K * PAGE_SIZE)
80 * when the unmap thread will be started.
82 #define TCMU_GLOBAL_MAX_BLOCKS_DEF (512 * 1024)
84 static u8 tcmu_kern_cmd_reply_supported;
85 static u8 tcmu_netlink_blocked;
87 static struct device *tcmu_root_device;
93 #define TCMU_CONFIG_LEN 256
95 static DEFINE_MUTEX(tcmu_nl_cmd_mutex);
96 static LIST_HEAD(tcmu_nl_cmd_list);
101 /* wake up thread waiting for reply */
102 struct completion complete;
103 struct list_head nl_list;
104 struct tcmu_dev *udev;
110 struct list_head node;
113 struct se_device se_dev;
118 #define TCMU_DEV_BIT_OPEN 0
119 #define TCMU_DEV_BIT_BROKEN 1
120 #define TCMU_DEV_BIT_BLOCKED 2
121 #define TCMU_DEV_BIT_TMR_NOTIFY 3
124 struct uio_info uio_info;
128 struct tcmu_mailbox *mb_addr;
131 u32 cmdr_last_cleaned;
132 /* Offset of data area from start of mb */
133 /* Must add data_off and mb_addr to get the address */
139 struct mutex cmdr_lock;
140 struct list_head qfull_queue;
141 struct list_head tmr_queue;
145 unsigned long *data_bitmap;
146 struct radix_tree_root data_blocks;
150 struct timer_list cmd_timer;
151 unsigned int cmd_time_out;
152 struct list_head inflight_queue;
154 struct timer_list qfull_timer;
157 struct list_head timedout_entry;
159 struct tcmu_nl_cmd curr_nl_cmd;
161 char dev_config[TCMU_CONFIG_LEN];
163 int nl_reply_supported;
166 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
168 #define CMDR_OFF sizeof(struct tcmu_mailbox)
171 struct se_cmd *se_cmd;
172 struct tcmu_dev *tcmu_dev;
173 struct list_head queue_entry;
177 /* Can't use se_cmd when cleaning up expired cmds, because if
178 cmd has been completed then accessing se_cmd is off limits */
180 uint32_t dbi_bidi_cnt;
184 uint32_t data_len_bidi;
186 unsigned long deadline;
188 #define TCMU_CMD_BIT_EXPIRED 0
193 struct list_head queue_entry;
196 uint32_t tmr_cmd_cnt;
197 int16_t tmr_cmd_ids[];
201 * To avoid dead lock the mutex lock order should always be:
203 * mutex_lock(&root_udev_mutex);
205 * mutex_lock(&tcmu_dev->cmdr_lock);
206 * mutex_unlock(&tcmu_dev->cmdr_lock);
208 * mutex_unlock(&root_udev_mutex);
210 static DEFINE_MUTEX(root_udev_mutex);
211 static LIST_HEAD(root_udev);
213 static DEFINE_SPINLOCK(timed_out_udevs_lock);
214 static LIST_HEAD(timed_out_udevs);
216 static struct kmem_cache *tcmu_cmd_cache;
218 static atomic_t global_db_count = ATOMIC_INIT(0);
219 static struct delayed_work tcmu_unmap_work;
220 static int tcmu_global_max_blocks = TCMU_GLOBAL_MAX_BLOCKS_DEF;
222 static int tcmu_set_global_max_data_area(const char *str,
223 const struct kernel_param *kp)
225 int ret, max_area_mb;
227 ret = kstrtoint(str, 10, &max_area_mb);
231 if (max_area_mb <= 0) {
232 pr_err("global_max_data_area must be larger than 0.\n");
236 tcmu_global_max_blocks = TCMU_MBS_TO_BLOCKS(max_area_mb);
237 if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
238 schedule_delayed_work(&tcmu_unmap_work, 0);
240 cancel_delayed_work_sync(&tcmu_unmap_work);
245 static int tcmu_get_global_max_data_area(char *buffer,
246 const struct kernel_param *kp)
248 return sprintf(buffer, "%d\n", TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
251 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
252 .set = tcmu_set_global_max_data_area,
253 .get = tcmu_get_global_max_data_area,
256 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
258 MODULE_PARM_DESC(global_max_data_area_mb,
259 "Max MBs allowed to be allocated to all the tcmu device's "
262 static int tcmu_get_block_netlink(char *buffer,
263 const struct kernel_param *kp)
265 return sprintf(buffer, "%s\n", tcmu_netlink_blocked ?
266 "blocked" : "unblocked");
269 static int tcmu_set_block_netlink(const char *str,
270 const struct kernel_param *kp)
275 ret = kstrtou8(str, 0, &val);
280 pr_err("Invalid block netlink value %u\n", val);
284 tcmu_netlink_blocked = val;
288 static const struct kernel_param_ops tcmu_block_netlink_op = {
289 .set = tcmu_set_block_netlink,
290 .get = tcmu_get_block_netlink,
293 module_param_cb(block_netlink, &tcmu_block_netlink_op, NULL, S_IWUSR | S_IRUGO);
294 MODULE_PARM_DESC(block_netlink, "Block new netlink commands.");
296 static int tcmu_fail_netlink_cmd(struct tcmu_nl_cmd *nl_cmd)
298 struct tcmu_dev *udev = nl_cmd->udev;
300 if (!tcmu_netlink_blocked) {
301 pr_err("Could not reset device's netlink interface. Netlink is not blocked.\n");
305 if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
306 pr_debug("Aborting nl cmd %d on %s\n", nl_cmd->cmd, udev->name);
307 nl_cmd->status = -EINTR;
308 list_del(&nl_cmd->nl_list);
309 complete(&nl_cmd->complete);
314 static int tcmu_set_reset_netlink(const char *str,
315 const struct kernel_param *kp)
317 struct tcmu_nl_cmd *nl_cmd, *tmp_cmd;
321 ret = kstrtou8(str, 0, &val);
326 pr_err("Invalid reset netlink value %u\n", val);
330 mutex_lock(&tcmu_nl_cmd_mutex);
331 list_for_each_entry_safe(nl_cmd, tmp_cmd, &tcmu_nl_cmd_list, nl_list) {
332 ret = tcmu_fail_netlink_cmd(nl_cmd);
336 mutex_unlock(&tcmu_nl_cmd_mutex);
341 static const struct kernel_param_ops tcmu_reset_netlink_op = {
342 .set = tcmu_set_reset_netlink,
345 module_param_cb(reset_netlink, &tcmu_reset_netlink_op, NULL, S_IWUSR);
346 MODULE_PARM_DESC(reset_netlink, "Reset netlink commands.");
348 /* multicast group */
349 enum tcmu_multicast_groups {
353 static const struct genl_multicast_group tcmu_mcgrps[] = {
354 [TCMU_MCGRP_CONFIG] = { .name = "config", },
357 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
358 [TCMU_ATTR_DEVICE] = { .type = NLA_STRING },
359 [TCMU_ATTR_MINOR] = { .type = NLA_U32 },
360 [TCMU_ATTR_CMD_STATUS] = { .type = NLA_S32 },
361 [TCMU_ATTR_DEVICE_ID] = { .type = NLA_U32 },
362 [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
365 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
367 struct tcmu_dev *udev = NULL;
368 struct tcmu_nl_cmd *nl_cmd;
369 int dev_id, rc, ret = 0;
371 if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
372 !info->attrs[TCMU_ATTR_DEVICE_ID]) {
373 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
377 dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
378 rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
380 mutex_lock(&tcmu_nl_cmd_mutex);
381 list_for_each_entry(nl_cmd, &tcmu_nl_cmd_list, nl_list) {
382 if (nl_cmd->udev->se_dev.dev_index == dev_id) {
389 pr_err("tcmu nl cmd %u/%d completion could not find device with dev id %u.\n",
390 completed_cmd, rc, dev_id);
394 list_del(&nl_cmd->nl_list);
396 pr_debug("%s genl cmd done got id %d curr %d done %d rc %d stat %d\n",
397 udev->name, dev_id, nl_cmd->cmd, completed_cmd, rc,
400 if (nl_cmd->cmd != completed_cmd) {
401 pr_err("Mismatched commands on %s (Expecting reply for %d. Current %d).\n",
402 udev->name, completed_cmd, nl_cmd->cmd);
408 complete(&nl_cmd->complete);
410 mutex_unlock(&tcmu_nl_cmd_mutex);
414 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
416 return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
419 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
421 return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
424 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
425 struct genl_info *info)
427 return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
430 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
432 if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
433 tcmu_kern_cmd_reply_supported =
434 nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
435 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
436 tcmu_kern_cmd_reply_supported);
442 static const struct genl_small_ops tcmu_genl_ops[] = {
444 .cmd = TCMU_CMD_SET_FEATURES,
445 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
446 .flags = GENL_ADMIN_PERM,
447 .doit = tcmu_genl_set_features,
450 .cmd = TCMU_CMD_ADDED_DEVICE_DONE,
451 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
452 .flags = GENL_ADMIN_PERM,
453 .doit = tcmu_genl_add_dev_done,
456 .cmd = TCMU_CMD_REMOVED_DEVICE_DONE,
457 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
458 .flags = GENL_ADMIN_PERM,
459 .doit = tcmu_genl_rm_dev_done,
462 .cmd = TCMU_CMD_RECONFIG_DEVICE_DONE,
463 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
464 .flags = GENL_ADMIN_PERM,
465 .doit = tcmu_genl_reconfig_dev_done,
469 /* Our generic netlink family */
470 static struct genl_family tcmu_genl_family __ro_after_init = {
471 .module = THIS_MODULE,
475 .maxattr = TCMU_ATTR_MAX,
476 .policy = tcmu_attr_policy,
477 .mcgrps = tcmu_mcgrps,
478 .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
480 .small_ops = tcmu_genl_ops,
481 .n_small_ops = ARRAY_SIZE(tcmu_genl_ops),
484 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
485 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
486 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
487 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
489 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
491 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
494 for (i = 0; i < len; i++)
495 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
498 static inline int tcmu_get_empty_block(struct tcmu_dev *udev,
499 struct tcmu_cmd *tcmu_cmd,
500 int prev_dbi, int *iov_cnt)
505 dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
506 if (dbi == udev->dbi_thresh)
509 page = radix_tree_lookup(&udev->data_blocks, dbi);
511 if (atomic_add_return(1, &global_db_count) >
512 tcmu_global_max_blocks)
513 schedule_delayed_work(&tcmu_unmap_work, 0);
515 /* try to get new page from the mm */
516 page = alloc_page(GFP_NOIO);
520 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
525 if (dbi > udev->dbi_max)
528 set_bit(dbi, udev->data_bitmap);
529 tcmu_cmd_set_dbi(tcmu_cmd, dbi);
531 if (dbi != prev_dbi + 1)
538 atomic_dec(&global_db_count);
542 static int tcmu_get_empty_blocks(struct tcmu_dev *udev,
543 struct tcmu_cmd *tcmu_cmd, int dbi_cnt)
545 /* start value of dbi + 1 must not be a valid dbi */
549 for (i = 0; i < dbi_cnt; i++) {
550 dbi = tcmu_get_empty_block(udev, tcmu_cmd, dbi, &iov_cnt);
557 static inline struct page *
558 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
560 return radix_tree_lookup(&udev->data_blocks, dbi);
563 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
565 kfree(tcmu_cmd->dbi);
566 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
569 static inline void tcmu_cmd_set_block_cnts(struct tcmu_cmd *cmd)
572 struct se_cmd *se_cmd = cmd->se_cmd;
574 cmd->dbi_cnt = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
576 if (se_cmd->se_cmd_flags & SCF_BIDI) {
577 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
578 for (i = 0, len = 0; i < se_cmd->t_bidi_data_nents; i++)
579 len += se_cmd->t_bidi_data_sg[i].length;
580 cmd->dbi_bidi_cnt = DIV_ROUND_UP(len, DATA_BLOCK_SIZE);
581 cmd->dbi_cnt += cmd->dbi_bidi_cnt;
582 cmd->data_len_bidi = len;
586 static int new_block_to_iov(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
587 struct iovec **iov, int prev_dbi, int *remain)
589 /* Get the next dbi */
590 int dbi = tcmu_cmd_get_dbi(cmd);
591 /* Do not add more than DATA_BLOCK_SIZE to iov */
592 int len = min_t(int, DATA_BLOCK_SIZE, *remain);
596 * The following code will gather and map the blocks to the same iovec
597 * when the blocks are all next to each other.
599 if (dbi != prev_dbi + 1) {
600 /* dbi is not next to previous dbi, so start new iov */
603 /* write offset relative to mb_addr */
604 (*iov)->iov_base = (void __user *)
605 (udev->data_off + dbi * DATA_BLOCK_SIZE);
607 (*iov)->iov_len += len;
612 static void tcmu_setup_iovs(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
613 struct iovec **iov, int data_length)
615 /* start value of dbi + 1 must not be a valid dbi */
618 /* We prepare the IOVs for DMA_FROM_DEVICE transfer direction */
619 while (data_length > 0)
620 dbi = new_block_to_iov(udev, cmd, iov, dbi, &data_length);
623 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
625 struct se_device *se_dev = se_cmd->se_dev;
626 struct tcmu_dev *udev = TCMU_DEV(se_dev);
627 struct tcmu_cmd *tcmu_cmd;
629 tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_NOIO);
633 INIT_LIST_HEAD(&tcmu_cmd->queue_entry);
634 tcmu_cmd->se_cmd = se_cmd;
635 tcmu_cmd->tcmu_dev = udev;
637 tcmu_cmd_set_block_cnts(tcmu_cmd);
638 tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
640 if (!tcmu_cmd->dbi) {
641 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
648 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
650 unsigned long offset = offset_in_page(vaddr);
651 void *start = vaddr - offset;
653 size = round_up(size+offset, PAGE_SIZE);
656 flush_dcache_page(vmalloc_to_page(start));
663 * Some ring helper functions. We don't assume size is a power of 2 so
664 * we can't use circ_buf.h.
666 static inline size_t spc_used(size_t head, size_t tail, size_t size)
668 int diff = head - tail;
676 static inline size_t spc_free(size_t head, size_t tail, size_t size)
678 /* Keep 1 byte unused or we can't tell full from empty */
679 return (size - spc_used(head, tail, size) - 1);
682 static inline size_t head_to_end(size_t head, size_t size)
687 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
689 static void scatter_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
692 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
693 /* start value of dbi + 1 must not be a valid dbi */
695 int block_remaining = 0;
696 int data_len = se_cmd->data_length;
697 void *from, *to = NULL;
698 size_t copy_bytes, offset;
699 struct scatterlist *sg;
700 struct page *page = NULL;
702 for_each_sg(se_cmd->t_data_sg, sg, se_cmd->t_data_nents, i) {
703 int sg_remaining = sg->length;
704 from = kmap_atomic(sg_page(sg)) + sg->offset;
705 while (sg_remaining > 0) {
706 if (block_remaining == 0) {
708 flush_dcache_page(page);
712 /* get next dbi and add to IOVs */
713 dbi = new_block_to_iov(udev, tcmu_cmd, iov, dbi,
715 page = tcmu_get_block_page(udev, dbi);
716 to = kmap_atomic(page);
717 block_remaining = DATA_BLOCK_SIZE;
720 copy_bytes = min_t(size_t, sg_remaining,
722 offset = DATA_BLOCK_SIZE - block_remaining;
723 memcpy(to + offset, from + sg->length - sg_remaining,
726 sg_remaining -= copy_bytes;
727 block_remaining -= copy_bytes;
729 kunmap_atomic(from - sg->offset);
733 flush_dcache_page(page);
738 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
739 bool bidi, uint32_t read_len)
741 struct se_cmd *se_cmd = cmd->se_cmd;
743 int block_remaining = 0;
744 void *from = NULL, *to;
745 size_t copy_bytes, offset;
746 struct scatterlist *sg, *data_sg;
748 unsigned int data_nents;
752 data_sg = se_cmd->t_data_sg;
753 data_nents = se_cmd->t_data_nents;
756 * For bidi case, the first count blocks are for Data-Out
757 * buffer blocks, and before gathering the Data-In buffer
758 * the Data-Out buffer blocks should be skipped.
760 count = cmd->dbi_cnt - cmd->dbi_bidi_cnt;
762 data_sg = se_cmd->t_bidi_data_sg;
763 data_nents = se_cmd->t_bidi_data_nents;
766 tcmu_cmd_set_dbi_cur(cmd, count);
768 for_each_sg(data_sg, sg, data_nents, i) {
769 int sg_remaining = sg->length;
770 to = kmap_atomic(sg_page(sg)) + sg->offset;
771 while (sg_remaining > 0 && read_len > 0) {
772 if (block_remaining == 0) {
776 block_remaining = DATA_BLOCK_SIZE;
777 dbi = tcmu_cmd_get_dbi(cmd);
778 page = tcmu_get_block_page(udev, dbi);
779 from = kmap_atomic(page);
780 flush_dcache_page(page);
782 copy_bytes = min_t(size_t, sg_remaining,
784 if (read_len < copy_bytes)
785 copy_bytes = read_len;
786 offset = DATA_BLOCK_SIZE - block_remaining;
787 memcpy(to + sg->length - sg_remaining, from + offset,
790 sg_remaining -= copy_bytes;
791 block_remaining -= copy_bytes;
792 read_len -= copy_bytes;
794 kunmap_atomic(to - sg->offset);
802 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
804 return thresh - bitmap_weight(bitmap, thresh);
808 * We can't queue a command until we have space available on the cmd ring.
810 * Called with ring lock held.
812 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size)
814 struct tcmu_mailbox *mb = udev->mb_addr;
815 size_t space, cmd_needed;
818 tcmu_flush_dcache_range(mb, sizeof(*mb));
820 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
823 * If cmd end-of-ring space is too small then we need space for a NOP plus
824 * original cmd - cmds are internally contiguous.
826 if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
827 cmd_needed = cmd_size;
829 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
831 space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
832 if (space < cmd_needed) {
833 pr_debug("no cmd space: %u %u %u\n", cmd_head,
834 udev->cmdr_last_cleaned, udev->cmdr_size);
841 * We have to allocate data buffers before we can queue a command.
842 * Returns -1 on error (not enough space) or number of needed iovs on success
844 * Called with ring lock held.
846 static int tcmu_alloc_data_space(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
849 int space, iov_cnt = 0, ret = 0;
854 /* try to check and get the data blocks as needed */
855 space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
856 if (space < cmd->dbi_cnt) {
857 unsigned long blocks_left =
858 (udev->max_blocks - udev->dbi_thresh) + space;
860 if (blocks_left < cmd->dbi_cnt) {
861 pr_debug("no data space: only %lu available, but ask for %lu\n",
862 blocks_left * DATA_BLOCK_SIZE,
863 cmd->dbi_cnt * DATA_BLOCK_SIZE);
867 udev->dbi_thresh += cmd->dbi_cnt;
868 if (udev->dbi_thresh > udev->max_blocks)
869 udev->dbi_thresh = udev->max_blocks;
872 iov_cnt = tcmu_get_empty_blocks(udev, cmd,
873 cmd->dbi_cnt - cmd->dbi_bidi_cnt);
877 if (cmd->dbi_bidi_cnt) {
878 ret = tcmu_get_empty_blocks(udev, cmd, cmd->dbi_bidi_cnt);
884 return iov_cnt + ret;
887 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
889 return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
890 sizeof(struct tcmu_cmd_entry));
893 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
894 size_t base_command_size)
896 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
899 command_size = base_command_size +
900 round_up(scsi_command_size(se_cmd->t_task_cdb),
903 WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
908 static void tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
909 struct timer_list *timer)
914 tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
915 if (!timer_pending(timer))
916 mod_timer(timer, tcmu_cmd->deadline);
918 pr_debug("Timeout set up for cmd %p, dev = %s, tmo = %lu\n", tcmu_cmd,
919 tcmu_cmd->tcmu_dev->name, tmo / MSEC_PER_SEC);
922 static int add_to_qfull_queue(struct tcmu_cmd *tcmu_cmd)
924 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
928 * For backwards compat if qfull_time_out is not set use
929 * cmd_time_out and if that's not set use the default time out.
931 if (!udev->qfull_time_out)
933 else if (udev->qfull_time_out > 0)
934 tmo = udev->qfull_time_out;
935 else if (udev->cmd_time_out)
936 tmo = udev->cmd_time_out;
940 tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
942 list_add_tail(&tcmu_cmd->queue_entry, &udev->qfull_queue);
943 pr_debug("adding cmd %p on dev %s to ring space wait queue\n",
944 tcmu_cmd, udev->name);
948 static uint32_t ring_insert_padding(struct tcmu_dev *udev, size_t cmd_size)
950 struct tcmu_cmd_entry_hdr *hdr;
951 struct tcmu_mailbox *mb = udev->mb_addr;
952 uint32_t cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
954 /* Insert a PAD if end-of-ring space is too small */
955 if (head_to_end(cmd_head, udev->cmdr_size) < cmd_size) {
956 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
958 hdr = (void *) mb + CMDR_OFF + cmd_head;
959 tcmu_hdr_set_op(&hdr->len_op, TCMU_OP_PAD);
960 tcmu_hdr_set_len(&hdr->len_op, pad_size);
961 hdr->cmd_id = 0; /* not used for PAD */
964 tcmu_flush_dcache_range(hdr, sizeof(*hdr));
966 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
967 tcmu_flush_dcache_range(mb, sizeof(*mb));
969 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
970 WARN_ON(cmd_head != 0);
977 * queue_cmd_ring - queue cmd to ring or internally
978 * @tcmu_cmd: cmd to queue
979 * @scsi_err: TCM error code if failure (-1) returned.
982 * -1 we cannot queue internally or to the ring.
984 * 1 internally queued to wait for ring memory to free.
986 static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
988 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
989 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
990 size_t base_command_size, command_size;
991 struct tcmu_mailbox *mb = udev->mb_addr;
992 struct tcmu_cmd_entry *entry;
994 int iov_cnt, iov_bidi_cnt, cmd_id;
997 /* size of data buffer needed */
998 size_t data_length = (size_t)tcmu_cmd->dbi_cnt * DATA_BLOCK_SIZE;
1000 *scsi_err = TCM_NO_SENSE;
1002 if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
1003 *scsi_err = TCM_LUN_BUSY;
1007 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1008 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1012 if (!list_empty(&udev->qfull_queue))
1015 if (data_length > udev->data_size) {
1016 pr_warn("TCMU: Request of size %zu is too big for %zu data area\n",
1017 data_length, udev->data_size);
1018 *scsi_err = TCM_INVALID_CDB_FIELD;
1022 iov_cnt = tcmu_alloc_data_space(udev, tcmu_cmd, &iov_bidi_cnt);
1024 goto free_and_queue;
1027 * Must be a certain minimum size for response sense info, but
1028 * also may be larger if the iov array is large.
1030 base_command_size = tcmu_cmd_get_base_cmd_size(iov_cnt);
1031 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
1033 if (command_size > (udev->cmdr_size / 2)) {
1034 pr_warn("TCMU: Request of size %zu is too big for %u cmd ring\n",
1035 command_size, udev->cmdr_size);
1036 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1037 *scsi_err = TCM_INVALID_CDB_FIELD;
1041 if (!is_ring_space_avail(udev, command_size))
1043 * Don't leave commands partially setup because the unmap
1044 * thread might need the blocks to make forward progress.
1046 goto free_and_queue;
1048 cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
1050 pr_err("tcmu: Could not allocate cmd id.\n");
1052 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
1053 *scsi_err = TCM_OUT_OF_RESOURCES;
1056 tcmu_cmd->cmd_id = cmd_id;
1058 pr_debug("allocated cmd id %u for cmd %p dev %s\n", tcmu_cmd->cmd_id,
1059 tcmu_cmd, udev->name);
1061 cmd_head = ring_insert_padding(udev, command_size);
1063 entry = (void *) mb + CMDR_OFF + cmd_head;
1064 memset(entry, 0, command_size);
1065 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
1067 /* prepare iov list and copy data to data area if necessary */
1068 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1069 iov = &entry->req.iov[0];
1071 if (se_cmd->data_direction == DMA_TO_DEVICE ||
1072 se_cmd->se_cmd_flags & SCF_BIDI)
1073 scatter_data_area(udev, tcmu_cmd, &iov);
1075 tcmu_setup_iovs(udev, tcmu_cmd, &iov, se_cmd->data_length);
1077 entry->req.iov_cnt = iov_cnt - iov_bidi_cnt;
1079 /* Handle BIDI commands */
1080 if (se_cmd->se_cmd_flags & SCF_BIDI) {
1082 tcmu_setup_iovs(udev, tcmu_cmd, &iov, tcmu_cmd->data_len_bidi);
1083 entry->req.iov_bidi_cnt = iov_bidi_cnt;
1086 tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out, &udev->cmd_timer);
1088 entry->hdr.cmd_id = tcmu_cmd->cmd_id;
1090 tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1092 /* All offsets relative to mb_addr, not start of entry! */
1093 cdb_off = CMDR_OFF + cmd_head + base_command_size;
1094 memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1095 entry->req.cdb_off = cdb_off;
1096 tcmu_flush_dcache_range(entry, command_size);
1098 UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1099 tcmu_flush_dcache_range(mb, sizeof(*mb));
1101 list_add_tail(&tcmu_cmd->queue_entry, &udev->inflight_queue);
1103 /* TODO: only if FLUSH and FUA? */
1104 uio_event_notify(&udev->uio_info);
1109 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1110 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1113 if (add_to_qfull_queue(tcmu_cmd)) {
1114 *scsi_err = TCM_OUT_OF_RESOURCES;
1122 * queue_tmr_ring - queue tmr info to ring or internally
1123 * @udev: related tcmu_dev
1124 * @tmr: tcmu_tmr containing tmr info to queue
1128 * 1 internally queued to wait for ring memory to free.
1131 queue_tmr_ring(struct tcmu_dev *udev, struct tcmu_tmr *tmr)
1133 struct tcmu_tmr_entry *entry;
1136 struct tcmu_mailbox *mb = udev->mb_addr;
1139 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
1142 id_list_sz = sizeof(tmr->tmr_cmd_ids[0]) * tmr->tmr_cmd_cnt;
1143 cmd_size = round_up(sizeof(*entry) + id_list_sz, TCMU_OP_ALIGN_SIZE);
1145 if (!list_empty(&udev->tmr_queue) ||
1146 !is_ring_space_avail(udev, cmd_size)) {
1147 list_add_tail(&tmr->queue_entry, &udev->tmr_queue);
1148 pr_debug("adding tmr %p on dev %s to TMR ring space wait queue\n",
1153 cmd_head = ring_insert_padding(udev, cmd_size);
1155 entry = (void *)mb + CMDR_OFF + cmd_head;
1156 memset(entry, 0, cmd_size);
1157 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_TMR);
1158 tcmu_hdr_set_len(&entry->hdr.len_op, cmd_size);
1159 entry->tmr_type = tmr->tmr_type;
1160 entry->cmd_cnt = tmr->tmr_cmd_cnt;
1161 memcpy(&entry->cmd_ids[0], &tmr->tmr_cmd_ids[0], id_list_sz);
1162 tcmu_flush_dcache_range(entry, cmd_size);
1164 UPDATE_HEAD(mb->cmd_head, cmd_size, udev->cmdr_size);
1165 tcmu_flush_dcache_range(mb, sizeof(*mb));
1167 uio_event_notify(&udev->uio_info);
1175 static sense_reason_t
1176 tcmu_queue_cmd(struct se_cmd *se_cmd)
1178 struct se_device *se_dev = se_cmd->se_dev;
1179 struct tcmu_dev *udev = TCMU_DEV(se_dev);
1180 struct tcmu_cmd *tcmu_cmd;
1181 sense_reason_t scsi_ret = TCM_CHECK_CONDITION_ABORT_CMD;
1184 tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1186 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1188 mutex_lock(&udev->cmdr_lock);
1189 if (!(se_cmd->transport_state & CMD_T_ABORTED))
1190 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1192 tcmu_free_cmd(tcmu_cmd);
1194 se_cmd->priv = tcmu_cmd;
1195 mutex_unlock(&udev->cmdr_lock);
1199 static void tcmu_set_next_deadline(struct list_head *queue,
1200 struct timer_list *timer)
1202 struct tcmu_cmd *cmd;
1204 if (!list_empty(queue)) {
1205 cmd = list_first_entry(queue, struct tcmu_cmd, queue_entry);
1206 mod_timer(timer, cmd->deadline);
1212 tcmu_tmr_type(enum tcm_tmreq_table tmf)
1215 case TMR_ABORT_TASK: return TCMU_TMR_ABORT_TASK;
1216 case TMR_ABORT_TASK_SET: return TCMU_TMR_ABORT_TASK_SET;
1217 case TMR_CLEAR_ACA: return TCMU_TMR_CLEAR_ACA;
1218 case TMR_CLEAR_TASK_SET: return TCMU_TMR_CLEAR_TASK_SET;
1219 case TMR_LUN_RESET: return TCMU_TMR_LUN_RESET;
1220 case TMR_TARGET_WARM_RESET: return TCMU_TMR_TARGET_WARM_RESET;
1221 case TMR_TARGET_COLD_RESET: return TCMU_TMR_TARGET_COLD_RESET;
1222 case TMR_LUN_RESET_PRO: return TCMU_TMR_LUN_RESET_PRO;
1223 default: return TCMU_TMR_UNKNOWN;
1228 tcmu_tmr_notify(struct se_device *se_dev, enum tcm_tmreq_table tmf,
1229 struct list_head *cmd_list)
1231 int i = 0, cmd_cnt = 0;
1232 bool unqueued = false;
1233 uint16_t *cmd_ids = NULL;
1234 struct tcmu_cmd *cmd;
1235 struct se_cmd *se_cmd;
1236 struct tcmu_tmr *tmr;
1237 struct tcmu_dev *udev = TCMU_DEV(se_dev);
1239 mutex_lock(&udev->cmdr_lock);
1241 /* First we check for aborted commands in qfull_queue */
1242 list_for_each_entry(se_cmd, cmd_list, state_list) {
1247 /* Commands on qfull queue have no id yet */
1252 pr_debug("Removing aborted command %p from queue on dev %s.\n",
1255 list_del_init(&cmd->queue_entry);
1257 se_cmd->priv = NULL;
1258 target_complete_cmd(se_cmd, SAM_STAT_TASK_ABORTED);
1262 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1264 if (!test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags))
1267 pr_debug("TMR event %d on dev %s, aborted cmds %d, afflicted cmd_ids %d\n",
1268 tcmu_tmr_type(tmf), udev->name, i, cmd_cnt);
1270 tmr = kmalloc(sizeof(*tmr) + cmd_cnt * sizeof(*cmd_ids), GFP_KERNEL);
1274 tmr->tmr_type = tcmu_tmr_type(tmf);
1275 tmr->tmr_cmd_cnt = cmd_cnt;
1279 list_for_each_entry(se_cmd, cmd_list, state_list) {
1284 tmr->tmr_cmd_ids[cmd_cnt++] = cmd->cmd_id;
1288 queue_tmr_ring(udev, tmr);
1291 mutex_unlock(&udev->cmdr_lock);
1294 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1296 struct se_cmd *se_cmd = cmd->se_cmd;
1297 struct tcmu_dev *udev = cmd->tcmu_dev;
1298 bool read_len_valid = false;
1302 * cmd has been completed already from timeout, just reclaim
1303 * data area space and free cmd
1305 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1306 WARN_ON_ONCE(se_cmd);
1310 list_del_init(&cmd->queue_entry);
1312 tcmu_cmd_reset_dbi_cur(cmd);
1314 if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1315 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1317 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1321 read_len = se_cmd->data_length;
1322 if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1323 (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1324 read_len_valid = true;
1325 if (entry->rsp.read_len < read_len)
1326 read_len = entry->rsp.read_len;
1329 if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1330 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1331 if (!read_len_valid )
1334 se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1336 if (se_cmd->se_cmd_flags & SCF_BIDI) {
1337 /* Get Data-In buffer before clean up */
1338 gather_data_area(udev, cmd, true, read_len);
1339 } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1340 gather_data_area(udev, cmd, false, read_len);
1341 } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1343 } else if (se_cmd->data_direction != DMA_NONE) {
1344 pr_warn("TCMU: data direction was %d!\n",
1345 se_cmd->data_direction);
1349 se_cmd->priv = NULL;
1350 if (read_len_valid) {
1351 pr_debug("read_len = %d\n", read_len);
1352 target_complete_cmd_with_length(cmd->se_cmd,
1353 entry->rsp.scsi_status, read_len);
1355 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1358 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1362 static int tcmu_run_tmr_queue(struct tcmu_dev *udev)
1364 struct tcmu_tmr *tmr, *tmp;
1367 if (list_empty(&udev->tmr_queue))
1370 pr_debug("running %s's tmr queue\n", udev->name);
1372 list_splice_init(&udev->tmr_queue, &tmrs);
1374 list_for_each_entry_safe(tmr, tmp, &tmrs, queue_entry) {
1375 list_del_init(&tmr->queue_entry);
1377 pr_debug("removing tmr %p on dev %s from queue\n",
1380 if (queue_tmr_ring(udev, tmr)) {
1381 pr_debug("ran out of space during tmr queue run\n");
1383 * tmr was requeued, so just put all tmrs back in
1386 list_splice_tail(&tmrs, &udev->tmr_queue);
1394 static bool tcmu_handle_completions(struct tcmu_dev *udev)
1396 struct tcmu_mailbox *mb;
1397 struct tcmu_cmd *cmd;
1398 bool free_space = false;
1400 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1401 pr_err("ring broken, not handling completions\n");
1406 tcmu_flush_dcache_range(mb, sizeof(*mb));
1408 while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1410 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1413 * Flush max. up to end of cmd ring since current entry might
1414 * be a padding that is shorter than sizeof(*entry)
1416 size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1418 tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1419 ring_left : sizeof(*entry));
1423 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD ||
1424 tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_TMR) {
1425 UPDATE_HEAD(udev->cmdr_last_cleaned,
1426 tcmu_hdr_get_len(entry->hdr.len_op),
1430 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1432 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1434 pr_err("cmd_id %u not found, ring is broken\n",
1436 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1440 tcmu_handle_completion(cmd, entry);
1442 UPDATE_HEAD(udev->cmdr_last_cleaned,
1443 tcmu_hdr_get_len(entry->hdr.len_op),
1447 free_space = tcmu_run_tmr_queue(udev);
1449 if (atomic_read(&global_db_count) > tcmu_global_max_blocks &&
1450 idr_is_empty(&udev->commands) && list_empty(&udev->qfull_queue)) {
1452 * Allocated blocks exceeded global block limit, currently no
1453 * more pending or waiting commands so try to reclaim blocks.
1455 schedule_delayed_work(&tcmu_unmap_work, 0);
1457 if (udev->cmd_time_out)
1458 tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
1463 static void tcmu_check_expired_ring_cmd(struct tcmu_cmd *cmd)
1465 struct se_cmd *se_cmd;
1467 if (!time_after_eq(jiffies, cmd->deadline))
1470 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1471 list_del_init(&cmd->queue_entry);
1472 se_cmd = cmd->se_cmd;
1473 se_cmd->priv = NULL;
1476 pr_debug("Timing out inflight cmd %u on dev %s.\n",
1477 cmd->cmd_id, cmd->tcmu_dev->name);
1479 target_complete_cmd(se_cmd, SAM_STAT_CHECK_CONDITION);
1482 static void tcmu_check_expired_queue_cmd(struct tcmu_cmd *cmd)
1484 struct se_cmd *se_cmd;
1486 if (!time_after_eq(jiffies, cmd->deadline))
1489 pr_debug("Timing out queued cmd %p on dev %s.\n",
1490 cmd, cmd->tcmu_dev->name);
1492 list_del_init(&cmd->queue_entry);
1493 se_cmd = cmd->se_cmd;
1496 se_cmd->priv = NULL;
1497 target_complete_cmd(se_cmd, SAM_STAT_TASK_SET_FULL);
1500 static void tcmu_device_timedout(struct tcmu_dev *udev)
1502 spin_lock(&timed_out_udevs_lock);
1503 if (list_empty(&udev->timedout_entry))
1504 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1505 spin_unlock(&timed_out_udevs_lock);
1507 schedule_delayed_work(&tcmu_unmap_work, 0);
1510 static void tcmu_cmd_timedout(struct timer_list *t)
1512 struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1514 pr_debug("%s cmd timeout has expired\n", udev->name);
1515 tcmu_device_timedout(udev);
1518 static void tcmu_qfull_timedout(struct timer_list *t)
1520 struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1522 pr_debug("%s qfull timeout has expired\n", udev->name);
1523 tcmu_device_timedout(udev);
1526 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1528 struct tcmu_hba *tcmu_hba;
1530 tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1534 tcmu_hba->host_id = host_id;
1535 hba->hba_ptr = tcmu_hba;
1540 static void tcmu_detach_hba(struct se_hba *hba)
1542 kfree(hba->hba_ptr);
1543 hba->hba_ptr = NULL;
1546 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1548 struct tcmu_dev *udev;
1550 udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1553 kref_init(&udev->kref);
1555 udev->name = kstrdup(name, GFP_KERNEL);
1562 udev->cmd_time_out = TCMU_TIME_OUT;
1563 udev->qfull_time_out = -1;
1565 udev->max_blocks = DATA_BLOCK_BITS_DEF;
1566 mutex_init(&udev->cmdr_lock);
1568 INIT_LIST_HEAD(&udev->node);
1569 INIT_LIST_HEAD(&udev->timedout_entry);
1570 INIT_LIST_HEAD(&udev->qfull_queue);
1571 INIT_LIST_HEAD(&udev->tmr_queue);
1572 INIT_LIST_HEAD(&udev->inflight_queue);
1573 idr_init(&udev->commands);
1575 timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1576 timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1578 INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1580 return &udev->se_dev;
1583 static void run_qfull_queue(struct tcmu_dev *udev, bool fail)
1585 struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1587 sense_reason_t scsi_ret;
1590 if (list_empty(&udev->qfull_queue))
1593 pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1595 list_splice_init(&udev->qfull_queue, &cmds);
1597 list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, queue_entry) {
1598 list_del_init(&tcmu_cmd->queue_entry);
1600 pr_debug("removing cmd %p on dev %s from queue\n",
1601 tcmu_cmd, udev->name);
1605 * We were not able to even start the command, so
1606 * fail with busy to allow a retry in case runner
1607 * was only temporarily down. If the device is being
1608 * removed then LIO core will do the right thing and
1611 tcmu_cmd->se_cmd->priv = NULL;
1612 target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1613 tcmu_free_cmd(tcmu_cmd);
1617 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1619 pr_debug("cmd %p on dev %s failed with %u\n",
1620 tcmu_cmd, udev->name, scsi_ret);
1622 * Ignore scsi_ret for now. target_complete_cmd
1625 tcmu_cmd->se_cmd->priv = NULL;
1626 target_complete_cmd(tcmu_cmd->se_cmd,
1627 SAM_STAT_CHECK_CONDITION);
1628 tcmu_free_cmd(tcmu_cmd);
1629 } else if (ret > 0) {
1630 pr_debug("ran out of space during cmdr queue run\n");
1632 * cmd was requeued, so just put all cmds back in
1635 list_splice_tail(&cmds, &udev->qfull_queue);
1640 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1643 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1645 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1647 mutex_lock(&udev->cmdr_lock);
1648 if (tcmu_handle_completions(udev))
1649 run_qfull_queue(udev, false);
1650 mutex_unlock(&udev->cmdr_lock);
1656 * mmap code from uio.c. Copied here because we want to hook mmap()
1657 * and this stuff must come along.
1659 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1661 struct tcmu_dev *udev = vma->vm_private_data;
1662 struct uio_info *info = &udev->uio_info;
1664 if (vma->vm_pgoff < MAX_UIO_MAPS) {
1665 if (info->mem[vma->vm_pgoff].size == 0)
1667 return (int)vma->vm_pgoff;
1672 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1676 mutex_lock(&udev->cmdr_lock);
1677 page = tcmu_get_block_page(udev, dbi);
1680 mutex_unlock(&udev->cmdr_lock);
1685 * Userspace messed up and passed in a address not in the
1686 * data iov passed to it.
1688 pr_err("Invalid addr to data block mapping (dbi %u) on device %s\n",
1691 mutex_unlock(&udev->cmdr_lock);
1696 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1698 struct tcmu_dev *udev = vmf->vma->vm_private_data;
1699 struct uio_info *info = &udev->uio_info;
1701 unsigned long offset;
1704 int mi = tcmu_find_mem_index(vmf->vma);
1706 return VM_FAULT_SIGBUS;
1709 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1712 offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1714 if (offset < udev->data_off) {
1715 /* For the vmalloc()ed cmd area pages */
1716 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1717 page = vmalloc_to_page(addr);
1722 /* For the dynamically growing data area pages */
1723 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1724 page = tcmu_try_get_block_page(udev, dbi);
1726 return VM_FAULT_SIGBUS;
1733 static const struct vm_operations_struct tcmu_vm_ops = {
1734 .fault = tcmu_vma_fault,
1737 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1739 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1741 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1742 vma->vm_ops = &tcmu_vm_ops;
1744 vma->vm_private_data = udev;
1746 /* Ensure the mmap is exactly the right size */
1747 if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
1753 static int tcmu_open(struct uio_info *info, struct inode *inode)
1755 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1757 /* O_EXCL not supported for char devs, so fake it? */
1758 if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1761 udev->inode = inode;
1762 kref_get(&udev->kref);
1769 static void tcmu_dev_call_rcu(struct rcu_head *p)
1771 struct se_device *dev = container_of(p, struct se_device, rcu_head);
1772 struct tcmu_dev *udev = TCMU_DEV(dev);
1774 kfree(udev->uio_info.name);
1779 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1781 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1782 kmem_cache_free(tcmu_cmd_cache, cmd);
1788 static void tcmu_blocks_release(struct radix_tree_root *blocks,
1794 for (i = start; i < end; i++) {
1795 page = radix_tree_delete(blocks, i);
1798 atomic_dec(&global_db_count);
1803 static void tcmu_remove_all_queued_tmr(struct tcmu_dev *udev)
1805 struct tcmu_tmr *tmr, *tmp;
1807 list_for_each_entry_safe(tmr, tmp, &udev->tmr_queue, queue_entry) {
1808 list_del_init(&tmr->queue_entry);
1813 static void tcmu_dev_kref_release(struct kref *kref)
1815 struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1816 struct se_device *dev = &udev->se_dev;
1817 struct tcmu_cmd *cmd;
1818 bool all_expired = true;
1821 vfree(udev->mb_addr);
1822 udev->mb_addr = NULL;
1824 spin_lock_bh(&timed_out_udevs_lock);
1825 if (!list_empty(&udev->timedout_entry))
1826 list_del(&udev->timedout_entry);
1827 spin_unlock_bh(&timed_out_udevs_lock);
1829 /* Upper layer should drain all requests before calling this */
1830 mutex_lock(&udev->cmdr_lock);
1831 idr_for_each_entry(&udev->commands, cmd, i) {
1832 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1833 all_expired = false;
1835 /* There can be left over TMR cmds. Remove them. */
1836 tcmu_remove_all_queued_tmr(udev);
1837 if (!list_empty(&udev->qfull_queue))
1838 all_expired = false;
1839 idr_destroy(&udev->commands);
1840 WARN_ON(!all_expired);
1842 tcmu_blocks_release(&udev->data_blocks, 0, udev->dbi_max + 1);
1843 bitmap_free(udev->data_bitmap);
1844 mutex_unlock(&udev->cmdr_lock);
1846 call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1849 static int tcmu_release(struct uio_info *info, struct inode *inode)
1851 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1853 clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1855 pr_debug("close\n");
1856 /* release ref from open */
1857 kref_put(&udev->kref, tcmu_dev_kref_release);
1861 static int tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1863 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1865 if (!tcmu_kern_cmd_reply_supported)
1868 if (udev->nl_reply_supported <= 0)
1871 mutex_lock(&tcmu_nl_cmd_mutex);
1873 if (tcmu_netlink_blocked) {
1874 mutex_unlock(&tcmu_nl_cmd_mutex);
1875 pr_warn("Failing nl cmd %d on %s. Interface is blocked.\n", cmd,
1880 if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1881 mutex_unlock(&tcmu_nl_cmd_mutex);
1882 pr_warn("netlink cmd %d already executing on %s\n",
1883 nl_cmd->cmd, udev->name);
1887 memset(nl_cmd, 0, sizeof(*nl_cmd));
1889 nl_cmd->udev = udev;
1890 init_completion(&nl_cmd->complete);
1891 INIT_LIST_HEAD(&nl_cmd->nl_list);
1893 list_add_tail(&nl_cmd->nl_list, &tcmu_nl_cmd_list);
1895 mutex_unlock(&tcmu_nl_cmd_mutex);
1899 static void tcmu_destroy_genl_cmd_reply(struct tcmu_dev *udev)
1901 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1903 if (!tcmu_kern_cmd_reply_supported)
1906 if (udev->nl_reply_supported <= 0)
1909 mutex_lock(&tcmu_nl_cmd_mutex);
1911 list_del(&nl_cmd->nl_list);
1912 memset(nl_cmd, 0, sizeof(*nl_cmd));
1914 mutex_unlock(&tcmu_nl_cmd_mutex);
1917 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1919 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1922 if (!tcmu_kern_cmd_reply_supported)
1925 if (udev->nl_reply_supported <= 0)
1928 pr_debug("sleeping for nl reply\n");
1929 wait_for_completion(&nl_cmd->complete);
1931 mutex_lock(&tcmu_nl_cmd_mutex);
1932 nl_cmd->cmd = TCMU_CMD_UNSPEC;
1933 ret = nl_cmd->status;
1934 mutex_unlock(&tcmu_nl_cmd_mutex);
1939 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
1940 enum tcmu_genl_cmd cmd,
1941 struct sk_buff **buf, void **hdr)
1943 struct sk_buff *skb;
1947 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1951 msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1955 ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1959 ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1963 ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1976 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
1977 enum tcmu_genl_cmd cmd,
1978 struct sk_buff *skb, void *msg_header)
1982 genlmsg_end(skb, msg_header);
1984 ret = tcmu_init_genl_cmd_reply(udev, cmd);
1990 ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1991 TCMU_MCGRP_CONFIG, GFP_KERNEL);
1993 /* Wait during an add as the listener may not be up yet */
1995 (ret == -ESRCH && cmd == TCMU_CMD_ADDED_DEVICE))
1996 return tcmu_wait_genl_cmd_reply(udev);
1998 tcmu_destroy_genl_cmd_reply(udev);
2003 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
2005 struct sk_buff *skb = NULL;
2006 void *msg_header = NULL;
2009 ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
2013 return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, skb,
2017 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
2019 struct sk_buff *skb = NULL;
2020 void *msg_header = NULL;
2023 ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
2027 return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
2031 static int tcmu_update_uio_info(struct tcmu_dev *udev)
2033 struct tcmu_hba *hba = udev->hba->hba_ptr;
2034 struct uio_info *info;
2037 info = &udev->uio_info;
2039 if (udev->dev_config[0])
2040 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s/%s", hba->host_id,
2041 udev->name, udev->dev_config);
2043 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s", hba->host_id,
2048 /* If the old string exists, free it */
2055 static int tcmu_configure_device(struct se_device *dev)
2057 struct tcmu_dev *udev = TCMU_DEV(dev);
2058 struct uio_info *info;
2059 struct tcmu_mailbox *mb;
2062 ret = tcmu_update_uio_info(udev);
2066 info = &udev->uio_info;
2068 mutex_lock(&udev->cmdr_lock);
2069 udev->data_bitmap = bitmap_zalloc(udev->max_blocks, GFP_KERNEL);
2070 mutex_unlock(&udev->cmdr_lock);
2071 if (!udev->data_bitmap) {
2073 goto err_bitmap_alloc;
2076 udev->mb_addr = vzalloc(CMDR_SIZE);
2077 if (!udev->mb_addr) {
2082 /* mailbox fits in first part of CMDR space */
2083 udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
2084 udev->data_off = CMDR_SIZE;
2085 udev->data_size = udev->max_blocks * DATA_BLOCK_SIZE;
2086 udev->dbi_thresh = 0; /* Default in Idle state */
2088 /* Initialise the mailbox of the ring buffer */
2090 mb->version = TCMU_MAILBOX_VERSION;
2091 mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC |
2092 TCMU_MAILBOX_FLAG_CAP_READ_LEN |
2093 TCMU_MAILBOX_FLAG_CAP_TMR;
2094 mb->cmdr_off = CMDR_OFF;
2095 mb->cmdr_size = udev->cmdr_size;
2097 WARN_ON(!PAGE_ALIGNED(udev->data_off));
2098 WARN_ON(udev->data_size % PAGE_SIZE);
2099 WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
2101 info->version = __stringify(TCMU_MAILBOX_VERSION);
2103 info->mem[0].name = "tcm-user command & data buffer";
2104 info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
2105 info->mem[0].size = udev->ring_size = udev->data_size + CMDR_SIZE;
2106 info->mem[0].memtype = UIO_MEM_NONE;
2108 info->irqcontrol = tcmu_irqcontrol;
2109 info->irq = UIO_IRQ_CUSTOM;
2111 info->mmap = tcmu_mmap;
2112 info->open = tcmu_open;
2113 info->release = tcmu_release;
2115 ret = uio_register_device(tcmu_root_device, info);
2119 /* User can set hw_block_size before enable the device */
2120 if (dev->dev_attrib.hw_block_size == 0)
2121 dev->dev_attrib.hw_block_size = 512;
2122 /* Other attributes can be configured in userspace */
2123 if (!dev->dev_attrib.hw_max_sectors)
2124 dev->dev_attrib.hw_max_sectors = 128;
2125 if (!dev->dev_attrib.emulate_write_cache)
2126 dev->dev_attrib.emulate_write_cache = 0;
2127 dev->dev_attrib.hw_queue_depth = 128;
2129 /* If user didn't explicitly disable netlink reply support, use
2130 * module scope setting.
2132 if (udev->nl_reply_supported >= 0)
2133 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
2136 * Get a ref incase userspace does a close on the uio device before
2137 * LIO has initiated tcmu_free_device.
2139 kref_get(&udev->kref);
2141 ret = tcmu_send_dev_add_event(udev);
2145 mutex_lock(&root_udev_mutex);
2146 list_add(&udev->node, &root_udev);
2147 mutex_unlock(&root_udev_mutex);
2152 kref_put(&udev->kref, tcmu_dev_kref_release);
2153 uio_unregister_device(&udev->uio_info);
2155 vfree(udev->mb_addr);
2156 udev->mb_addr = NULL;
2158 bitmap_free(udev->data_bitmap);
2159 udev->data_bitmap = NULL;
2167 static void tcmu_free_device(struct se_device *dev)
2169 struct tcmu_dev *udev = TCMU_DEV(dev);
2171 /* release ref from init */
2172 kref_put(&udev->kref, tcmu_dev_kref_release);
2175 static void tcmu_destroy_device(struct se_device *dev)
2177 struct tcmu_dev *udev = TCMU_DEV(dev);
2179 del_timer_sync(&udev->cmd_timer);
2180 del_timer_sync(&udev->qfull_timer);
2182 mutex_lock(&root_udev_mutex);
2183 list_del(&udev->node);
2184 mutex_unlock(&root_udev_mutex);
2186 tcmu_send_dev_remove_event(udev);
2188 uio_unregister_device(&udev->uio_info);
2190 /* release ref from configure */
2191 kref_put(&udev->kref, tcmu_dev_kref_release);
2194 static void tcmu_unblock_dev(struct tcmu_dev *udev)
2196 mutex_lock(&udev->cmdr_lock);
2197 clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
2198 mutex_unlock(&udev->cmdr_lock);
2201 static void tcmu_block_dev(struct tcmu_dev *udev)
2203 mutex_lock(&udev->cmdr_lock);
2205 if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2208 /* complete IO that has executed successfully */
2209 tcmu_handle_completions(udev);
2210 /* fail IO waiting to be queued */
2211 run_qfull_queue(udev, true);
2214 mutex_unlock(&udev->cmdr_lock);
2217 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
2219 struct tcmu_mailbox *mb;
2220 struct tcmu_cmd *cmd;
2223 mutex_lock(&udev->cmdr_lock);
2225 idr_for_each_entry(&udev->commands, cmd, i) {
2226 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
2227 cmd->cmd_id, udev->name,
2228 test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
2230 idr_remove(&udev->commands, i);
2231 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
2232 WARN_ON(!cmd->se_cmd);
2233 list_del_init(&cmd->queue_entry);
2234 cmd->se_cmd->priv = NULL;
2235 if (err_level == 1) {
2237 * Userspace was not able to start the
2238 * command or it is retryable.
2240 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
2243 target_complete_cmd(cmd->se_cmd,
2244 SAM_STAT_CHECK_CONDITION);
2247 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
2252 tcmu_flush_dcache_range(mb, sizeof(*mb));
2253 pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
2254 mb->cmd_tail, mb->cmd_head);
2256 udev->cmdr_last_cleaned = 0;
2259 tcmu_flush_dcache_range(mb, sizeof(*mb));
2260 clear_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
2262 del_timer(&udev->cmd_timer);
2265 * ring is empty and qfull queue never contains aborted commands.
2266 * So TMRs in tmr queue do not contain relevant cmd_ids.
2267 * After a ring reset userspace should do a fresh start, so
2268 * even LUN RESET message is no longer relevant.
2269 * Therefore remove all TMRs from qfull queue
2271 tcmu_remove_all_queued_tmr(udev);
2273 run_qfull_queue(udev, false);
2275 mutex_unlock(&udev->cmdr_lock);
2279 Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
2280 Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_err,
2283 static match_table_t tokens = {
2284 {Opt_dev_config, "dev_config=%s"},
2285 {Opt_dev_size, "dev_size=%s"},
2286 {Opt_hw_block_size, "hw_block_size=%d"},
2287 {Opt_hw_max_sectors, "hw_max_sectors=%d"},
2288 {Opt_nl_reply_supported, "nl_reply_supported=%d"},
2289 {Opt_max_data_area_mb, "max_data_area_mb=%d"},
2293 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
2297 ret = match_int(arg, &val);
2299 pr_err("match_int() failed for dev attrib. Error %d.\n",
2305 pr_err("Invalid dev attrib value %d. Must be greater than zero.\n",
2313 static int tcmu_set_max_blocks_param(struct tcmu_dev *udev, substring_t *arg)
2317 ret = match_int(arg, &val);
2319 pr_err("match_int() failed for max_data_area_mb=. Error %d.\n",
2325 pr_err("Invalid max_data_area %d.\n", val);
2329 mutex_lock(&udev->cmdr_lock);
2330 if (udev->data_bitmap) {
2331 pr_err("Cannot set max_data_area_mb after it has been enabled.\n");
2336 udev->max_blocks = TCMU_MBS_TO_BLOCKS(val);
2337 if (udev->max_blocks > tcmu_global_max_blocks) {
2338 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2339 val, TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
2340 udev->max_blocks = tcmu_global_max_blocks;
2344 mutex_unlock(&udev->cmdr_lock);
2348 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
2349 const char *page, ssize_t count)
2351 struct tcmu_dev *udev = TCMU_DEV(dev);
2352 char *orig, *ptr, *opts;
2353 substring_t args[MAX_OPT_ARGS];
2356 opts = kstrdup(page, GFP_KERNEL);
2362 while ((ptr = strsep(&opts, ",\n")) != NULL) {
2366 token = match_token(ptr, tokens, args);
2368 case Opt_dev_config:
2369 if (match_strlcpy(udev->dev_config, &args[0],
2370 TCMU_CONFIG_LEN) == 0) {
2374 pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2377 ret = match_u64(&args[0], &udev->dev_size);
2379 pr_err("match_u64() failed for dev_size=. Error %d.\n",
2382 case Opt_hw_block_size:
2383 ret = tcmu_set_dev_attrib(&args[0],
2384 &(dev->dev_attrib.hw_block_size));
2386 case Opt_hw_max_sectors:
2387 ret = tcmu_set_dev_attrib(&args[0],
2388 &(dev->dev_attrib.hw_max_sectors));
2390 case Opt_nl_reply_supported:
2391 ret = match_int(&args[0], &udev->nl_reply_supported);
2393 pr_err("match_int() failed for nl_reply_supported=. Error %d.\n",
2396 case Opt_max_data_area_mb:
2397 ret = tcmu_set_max_blocks_param(udev, &args[0]);
2408 return (!ret) ? count : ret;
2411 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2413 struct tcmu_dev *udev = TCMU_DEV(dev);
2416 bl = sprintf(b + bl, "Config: %s ",
2417 udev->dev_config[0] ? udev->dev_config : "NULL");
2418 bl += sprintf(b + bl, "Size: %llu ", udev->dev_size);
2419 bl += sprintf(b + bl, "MaxDataAreaMB: %u\n",
2420 TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2425 static sector_t tcmu_get_blocks(struct se_device *dev)
2427 struct tcmu_dev *udev = TCMU_DEV(dev);
2429 return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2430 dev->dev_attrib.block_size);
2433 static sense_reason_t
2434 tcmu_parse_cdb(struct se_cmd *cmd)
2436 return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2439 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2441 struct se_dev_attrib *da = container_of(to_config_group(item),
2442 struct se_dev_attrib, da_group);
2443 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2445 return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2448 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2451 struct se_dev_attrib *da = container_of(to_config_group(item),
2452 struct se_dev_attrib, da_group);
2453 struct tcmu_dev *udev = container_of(da->da_dev,
2454 struct tcmu_dev, se_dev);
2458 if (da->da_dev->export_count) {
2459 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2463 ret = kstrtou32(page, 0, &val);
2467 udev->cmd_time_out = val * MSEC_PER_SEC;
2470 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2472 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2474 struct se_dev_attrib *da = container_of(to_config_group(item),
2475 struct se_dev_attrib, da_group);
2476 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2478 return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2479 udev->qfull_time_out :
2480 udev->qfull_time_out / MSEC_PER_SEC);
2483 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2484 const char *page, size_t count)
2486 struct se_dev_attrib *da = container_of(to_config_group(item),
2487 struct se_dev_attrib, da_group);
2488 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2492 ret = kstrtos32(page, 0, &val);
2497 udev->qfull_time_out = val * MSEC_PER_SEC;
2498 } else if (val == -1) {
2499 udev->qfull_time_out = val;
2501 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2506 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2508 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2510 struct se_dev_attrib *da = container_of(to_config_group(item),
2511 struct se_dev_attrib, da_group);
2512 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2514 return snprintf(page, PAGE_SIZE, "%u\n",
2515 TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2517 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2519 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2521 struct se_dev_attrib *da = container_of(to_config_group(item),
2522 struct se_dev_attrib, da_group);
2523 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2525 return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2528 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2529 const char *reconfig_data)
2531 struct sk_buff *skb = NULL;
2532 void *msg_header = NULL;
2535 ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2539 ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2544 return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2549 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2552 struct se_dev_attrib *da = container_of(to_config_group(item),
2553 struct se_dev_attrib, da_group);
2554 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2558 if (!len || len > TCMU_CONFIG_LEN - 1)
2561 /* Check if device has been configured before */
2562 if (target_dev_configured(&udev->se_dev)) {
2563 ret = tcmu_send_dev_config_event(udev, page);
2565 pr_err("Unable to reconfigure device\n");
2568 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2570 ret = tcmu_update_uio_info(udev);
2575 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2579 CONFIGFS_ATTR(tcmu_, dev_config);
2581 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2583 struct se_dev_attrib *da = container_of(to_config_group(item),
2584 struct se_dev_attrib, da_group);
2585 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2587 return snprintf(page, PAGE_SIZE, "%llu\n", udev->dev_size);
2590 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2592 struct sk_buff *skb = NULL;
2593 void *msg_header = NULL;
2596 ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2600 ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2601 size, TCMU_ATTR_PAD);
2606 return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2610 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2613 struct se_dev_attrib *da = container_of(to_config_group(item),
2614 struct se_dev_attrib, da_group);
2615 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2619 ret = kstrtou64(page, 0, &val);
2623 /* Check if device has been configured before */
2624 if (target_dev_configured(&udev->se_dev)) {
2625 ret = tcmu_send_dev_size_event(udev, val);
2627 pr_err("Unable to reconfigure device\n");
2631 udev->dev_size = val;
2634 CONFIGFS_ATTR(tcmu_, dev_size);
2636 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2639 struct se_dev_attrib *da = container_of(to_config_group(item),
2640 struct se_dev_attrib, da_group);
2641 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2643 return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2646 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2647 const char *page, size_t count)
2649 struct se_dev_attrib *da = container_of(to_config_group(item),
2650 struct se_dev_attrib, da_group);
2651 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2655 ret = kstrtos8(page, 0, &val);
2659 udev->nl_reply_supported = val;
2662 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2664 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2667 struct se_dev_attrib *da = container_of(to_config_group(item),
2668 struct se_dev_attrib, da_group);
2670 return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2673 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2675 struct sk_buff *skb = NULL;
2676 void *msg_header = NULL;
2679 ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2683 ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2688 return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2692 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2693 const char *page, size_t count)
2695 struct se_dev_attrib *da = container_of(to_config_group(item),
2696 struct se_dev_attrib, da_group);
2697 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2701 ret = kstrtou8(page, 0, &val);
2705 /* Check if device has been configured before */
2706 if (target_dev_configured(&udev->se_dev)) {
2707 ret = tcmu_send_emulate_write_cache(udev, val);
2709 pr_err("Unable to reconfigure device\n");
2714 da->emulate_write_cache = val;
2717 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2719 static ssize_t tcmu_tmr_notification_show(struct config_item *item, char *page)
2721 struct se_dev_attrib *da = container_of(to_config_group(item),
2722 struct se_dev_attrib, da_group);
2723 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2725 return snprintf(page, PAGE_SIZE, "%i\n",
2726 test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags));
2729 static ssize_t tcmu_tmr_notification_store(struct config_item *item,
2730 const char *page, size_t count)
2732 struct se_dev_attrib *da = container_of(to_config_group(item),
2733 struct se_dev_attrib, da_group);
2734 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2738 ret = kstrtou8(page, 0, &val);
2745 set_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2747 clear_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2750 CONFIGFS_ATTR(tcmu_, tmr_notification);
2752 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2754 struct se_device *se_dev = container_of(to_config_group(item),
2757 struct tcmu_dev *udev = TCMU_DEV(se_dev);
2759 if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2760 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2762 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2765 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2768 struct se_device *se_dev = container_of(to_config_group(item),
2771 struct tcmu_dev *udev = TCMU_DEV(se_dev);
2775 if (!target_dev_configured(&udev->se_dev)) {
2776 pr_err("Device is not configured.\n");
2780 ret = kstrtou8(page, 0, &val);
2785 pr_err("Invalid block value %d\n", val);
2790 tcmu_unblock_dev(udev);
2792 tcmu_block_dev(udev);
2795 CONFIGFS_ATTR(tcmu_, block_dev);
2797 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2800 struct se_device *se_dev = container_of(to_config_group(item),
2803 struct tcmu_dev *udev = TCMU_DEV(se_dev);
2807 if (!target_dev_configured(&udev->se_dev)) {
2808 pr_err("Device is not configured.\n");
2812 ret = kstrtou8(page, 0, &val);
2816 if (val != 1 && val != 2) {
2817 pr_err("Invalid reset ring value %d\n", val);
2821 tcmu_reset_ring(udev, val);
2824 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2826 static struct configfs_attribute *tcmu_attrib_attrs[] = {
2827 &tcmu_attr_cmd_time_out,
2828 &tcmu_attr_qfull_time_out,
2829 &tcmu_attr_max_data_area_mb,
2830 &tcmu_attr_dev_config,
2831 &tcmu_attr_dev_size,
2832 &tcmu_attr_emulate_write_cache,
2833 &tcmu_attr_tmr_notification,
2834 &tcmu_attr_nl_reply_supported,
2838 static struct configfs_attribute **tcmu_attrs;
2840 static struct configfs_attribute *tcmu_action_attrs[] = {
2841 &tcmu_attr_block_dev,
2842 &tcmu_attr_reset_ring,
2846 static struct target_backend_ops tcmu_ops = {
2848 .owner = THIS_MODULE,
2849 .transport_flags_default = TRANSPORT_FLAG_PASSTHROUGH,
2850 .transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR |
2851 TRANSPORT_FLAG_PASSTHROUGH_ALUA,
2852 .attach_hba = tcmu_attach_hba,
2853 .detach_hba = tcmu_detach_hba,
2854 .alloc_device = tcmu_alloc_device,
2855 .configure_device = tcmu_configure_device,
2856 .destroy_device = tcmu_destroy_device,
2857 .free_device = tcmu_free_device,
2858 .parse_cdb = tcmu_parse_cdb,
2859 .tmr_notify = tcmu_tmr_notify,
2860 .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2861 .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2862 .get_device_type = sbc_get_device_type,
2863 .get_blocks = tcmu_get_blocks,
2864 .tb_dev_action_attrs = tcmu_action_attrs,
2867 static void find_free_blocks(void)
2869 struct tcmu_dev *udev;
2871 u32 start, end, block, total_freed = 0;
2873 if (atomic_read(&global_db_count) <= tcmu_global_max_blocks)
2876 mutex_lock(&root_udev_mutex);
2877 list_for_each_entry(udev, &root_udev, node) {
2878 mutex_lock(&udev->cmdr_lock);
2880 if (!target_dev_configured(&udev->se_dev)) {
2881 mutex_unlock(&udev->cmdr_lock);
2885 /* Try to complete the finished commands first */
2886 if (tcmu_handle_completions(udev))
2887 run_qfull_queue(udev, false);
2889 /* Skip the udevs in idle */
2890 if (!udev->dbi_thresh) {
2891 mutex_unlock(&udev->cmdr_lock);
2895 end = udev->dbi_max + 1;
2896 block = find_last_bit(udev->data_bitmap, end);
2897 if (block == udev->dbi_max) {
2899 * The last bit is dbi_max, so it is not possible
2900 * reclaim any blocks.
2902 mutex_unlock(&udev->cmdr_lock);
2904 } else if (block == end) {
2905 /* The current udev will goto idle state */
2906 udev->dbi_thresh = start = 0;
2909 udev->dbi_thresh = start = block + 1;
2910 udev->dbi_max = block;
2913 /* Here will truncate the data area from off */
2914 off = udev->data_off + start * DATA_BLOCK_SIZE;
2915 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2917 /* Release the block pages */
2918 tcmu_blocks_release(&udev->data_blocks, start, end);
2919 mutex_unlock(&udev->cmdr_lock);
2921 total_freed += end - start;
2922 pr_debug("Freed %u blocks (total %u) from %s.\n", end - start,
2923 total_freed, udev->name);
2925 mutex_unlock(&root_udev_mutex);
2927 if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
2928 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
2931 static void check_timedout_devices(void)
2933 struct tcmu_dev *udev, *tmp_dev;
2934 struct tcmu_cmd *cmd, *tmp_cmd;
2937 spin_lock_bh(&timed_out_udevs_lock);
2938 list_splice_init(&timed_out_udevs, &devs);
2940 list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
2941 list_del_init(&udev->timedout_entry);
2942 spin_unlock_bh(&timed_out_udevs_lock);
2944 mutex_lock(&udev->cmdr_lock);
2947 * If cmd_time_out is disabled but qfull is set deadline
2948 * will only reflect the qfull timeout. Ignore it.
2950 if (udev->cmd_time_out) {
2951 list_for_each_entry_safe(cmd, tmp_cmd,
2952 &udev->inflight_queue,
2954 tcmu_check_expired_ring_cmd(cmd);
2956 tcmu_set_next_deadline(&udev->inflight_queue,
2959 list_for_each_entry_safe(cmd, tmp_cmd, &udev->qfull_queue,
2961 tcmu_check_expired_queue_cmd(cmd);
2963 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
2965 mutex_unlock(&udev->cmdr_lock);
2967 spin_lock_bh(&timed_out_udevs_lock);
2970 spin_unlock_bh(&timed_out_udevs_lock);
2973 static void tcmu_unmap_work_fn(struct work_struct *work)
2975 check_timedout_devices();
2979 static int __init tcmu_module_init(void)
2981 int ret, i, k, len = 0;
2983 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2985 INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
2987 tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2988 sizeof(struct tcmu_cmd),
2989 __alignof__(struct tcmu_cmd),
2991 if (!tcmu_cmd_cache)
2994 tcmu_root_device = root_device_register("tcm_user");
2995 if (IS_ERR(tcmu_root_device)) {
2996 ret = PTR_ERR(tcmu_root_device);
2997 goto out_free_cache;
3000 ret = genl_register_family(&tcmu_genl_family);
3002 goto out_unreg_device;
3005 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3006 len += sizeof(struct configfs_attribute *);
3007 for (i = 0; passthrough_pr_attrib_attrs[i] != NULL; i++)
3008 len += sizeof(struct configfs_attribute *);
3009 for (i = 0; tcmu_attrib_attrs[i] != NULL; i++)
3010 len += sizeof(struct configfs_attribute *);
3011 len += sizeof(struct configfs_attribute *);
3013 tcmu_attrs = kzalloc(len, GFP_KERNEL);
3016 goto out_unreg_genl;
3019 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3020 tcmu_attrs[i] = passthrough_attrib_attrs[i];
3021 for (k = 0; passthrough_pr_attrib_attrs[k] != NULL; k++)
3022 tcmu_attrs[i++] = passthrough_pr_attrib_attrs[k];
3023 for (k = 0; tcmu_attrib_attrs[k] != NULL; k++)
3024 tcmu_attrs[i++] = tcmu_attrib_attrs[k];
3025 tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
3027 ret = transport_backend_register(&tcmu_ops);
3036 genl_unregister_family(&tcmu_genl_family);
3038 root_device_unregister(tcmu_root_device);
3040 kmem_cache_destroy(tcmu_cmd_cache);
3045 static void __exit tcmu_module_exit(void)
3047 cancel_delayed_work_sync(&tcmu_unmap_work);
3048 target_backend_unregister(&tcmu_ops);
3050 genl_unregister_family(&tcmu_genl_family);
3051 root_device_unregister(tcmu_root_device);
3052 kmem_cache_destroy(tcmu_cmd_cache);
3055 MODULE_DESCRIPTION("TCM USER subsystem plugin");
3056 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
3057 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
3058 MODULE_LICENSE("GPL");
3060 module_init(tcmu_module_init);
3061 module_exit(tcmu_module_exit);