1 /*
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.
6  *
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.
10  *
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
14  * more details.
15  *
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.
19  */
20 
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>
42 
43 #include <linux/target_core_user.h>
44 
45 /*
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.
49  *
50  * It uses the UIO framework to do a lot of the device-creation and
51  * introspection work for us.
52  *
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.
60  *
61  * SECURITY:
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.
68  */
69 
70 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
71 
72 /* For cmd area, the size is fixed 8MB */
73 #define CMDR_SIZE (8 * 1024 * 1024)
74 
75 /*
76  * For data area, the block size is PAGE_SIZE and
77  * the total size is 256K * PAGE_SIZE.
78  */
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
83 
84 /* The total size of the ring is 8M + 256K * PAGE_SIZE */
85 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
86 
87 /* Default maximum of the global data blocks(512K * PAGE_SIZE) */
88 #define TCMU_GLOBAL_MAX_BLOCKS (512 * 1024)
89 
90 static u8 tcmu_kern_cmd_reply_supported;
91 
92 static struct device *tcmu_root_device;
93 
94 struct tcmu_hba {
95 	u32 host_id;
96 };
97 
98 #define TCMU_CONFIG_LEN 256
99 
100 struct tcmu_nl_cmd {
101 	/* wake up thread waiting for reply */
102 	struct completion complete;
103 	int cmd;
104 	int status;
105 };
106 
107 struct tcmu_dev {
108 	struct list_head node;
109 	struct kref kref;
110 	struct se_device se_dev;
111 
112 	char *name;
113 	struct se_hba *hba;
114 
115 #define TCMU_DEV_BIT_OPEN 0
116 #define TCMU_DEV_BIT_BROKEN 1
117 	unsigned long flags;
118 
119 	struct uio_info uio_info;
120 
121 	struct inode *inode;
122 
123 	struct tcmu_mailbox *mb_addr;
124 	size_t dev_size;
125 	u32 cmdr_size;
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 */
129 	size_t data_off;
130 	size_t data_size;
131 
132 	wait_queue_head_t wait_cmdr;
133 	struct mutex cmdr_lock;
134 
135 	bool waiting_global;
136 	uint32_t dbi_max;
137 	uint32_t dbi_thresh;
138 	DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
139 	struct radix_tree_root data_blocks;
140 
141 	struct idr commands;
142 	spinlock_t commands_lock;
143 
144 	struct timer_list timeout;
145 	unsigned int cmd_time_out;
146 
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;
151 
152 	char dev_config[TCMU_CONFIG_LEN];
153 
154 	int nl_reply_supported;
155 };
156 
157 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
158 
159 #define CMDR_OFF sizeof(struct tcmu_mailbox)
160 
161 struct tcmu_cmd {
162 	struct se_cmd *se_cmd;
163 	struct tcmu_dev *tcmu_dev;
164 
165 	uint16_t cmd_id;
166 
167 	/* Can't use se_cmd when cleaning up expired cmds, because if
168 	   cmd has been completed then accessing se_cmd is off limits */
169 	uint32_t dbi_cnt;
170 	uint32_t dbi_cur;
171 	uint32_t *dbi;
172 
173 	unsigned long deadline;
174 
175 #define TCMU_CMD_BIT_EXPIRED 0
176 	unsigned long flags;
177 };
178 
179 static struct task_struct *unmap_thread;
180 static wait_queue_head_t unmap_wait;
181 static DEFINE_MUTEX(root_udev_mutex);
182 static LIST_HEAD(root_udev);
183 
184 static atomic_t global_db_count = ATOMIC_INIT(0);
185 
186 static struct kmem_cache *tcmu_cmd_cache;
187 
188 /* multicast group */
189 enum tcmu_multicast_groups {
190 	TCMU_MCGRP_CONFIG,
191 };
192 
193 static const struct genl_multicast_group tcmu_mcgrps[] = {
194 	[TCMU_MCGRP_CONFIG] = { .name = "config", },
195 };
196 
197 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
198 	[TCMU_ATTR_DEVICE]	= { .type = NLA_STRING },
199 	[TCMU_ATTR_MINOR]	= { .type = NLA_U32 },
200 	[TCMU_ATTR_CMD_STATUS]	= { .type = NLA_S32 },
201 	[TCMU_ATTR_DEVICE_ID]	= { .type = NLA_U32 },
202 	[TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
203 };
204 
205 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
206 {
207 	struct se_device *dev;
208 	struct tcmu_dev *udev;
209 	struct tcmu_nl_cmd *nl_cmd;
210 	int dev_id, rc, ret = 0;
211 	bool is_removed = (completed_cmd == TCMU_CMD_REMOVED_DEVICE);
212 
213 	if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
214 	    !info->attrs[TCMU_ATTR_DEVICE_ID]) {
215 		printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
216                 return -EINVAL;
217         }
218 
219 	dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
220 	rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
221 
222 	dev = target_find_device(dev_id, !is_removed);
223 	if (!dev) {
224 		printk(KERN_ERR "tcmu nl cmd %u/%u completion could not find device with dev id %u.\n",
225 		       completed_cmd, rc, dev_id);
226 		return -ENODEV;
227 	}
228 	udev = TCMU_DEV(dev);
229 
230 	spin_lock(&udev->nl_cmd_lock);
231 	nl_cmd = &udev->curr_nl_cmd;
232 
233 	pr_debug("genl cmd done got id %d curr %d done %d rc %d\n", dev_id,
234 		 nl_cmd->cmd, completed_cmd, rc);
235 
236 	if (nl_cmd->cmd != completed_cmd) {
237 		printk(KERN_ERR "Mismatched commands (Expecting reply for %d. Current %d).\n",
238 		       completed_cmd, nl_cmd->cmd);
239 		ret = -EINVAL;
240 	} else {
241 		nl_cmd->status = rc;
242 	}
243 
244 	spin_unlock(&udev->nl_cmd_lock);
245 	if (!is_removed)
246 		 target_undepend_item(&dev->dev_group.cg_item);
247 	if (!ret)
248 		complete(&nl_cmd->complete);
249 	return ret;
250 }
251 
252 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
253 {
254 	return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
255 }
256 
257 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
258 {
259 	return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
260 }
261 
262 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
263 				       struct genl_info *info)
264 {
265 	return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
266 }
267 
268 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
269 {
270 	if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
271 		tcmu_kern_cmd_reply_supported  =
272 			nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
273 		printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
274 		       tcmu_kern_cmd_reply_supported);
275 	}
276 
277 	return 0;
278 }
279 
280 static const struct genl_ops tcmu_genl_ops[] = {
281 	{
282 		.cmd	= TCMU_CMD_SET_FEATURES,
283 		.flags	= GENL_ADMIN_PERM,
284 		.policy	= tcmu_attr_policy,
285 		.doit	= tcmu_genl_set_features,
286 	},
287 	{
288 		.cmd	= TCMU_CMD_ADDED_DEVICE_DONE,
289 		.flags	= GENL_ADMIN_PERM,
290 		.policy	= tcmu_attr_policy,
291 		.doit	= tcmu_genl_add_dev_done,
292 	},
293 	{
294 		.cmd	= TCMU_CMD_REMOVED_DEVICE_DONE,
295 		.flags	= GENL_ADMIN_PERM,
296 		.policy	= tcmu_attr_policy,
297 		.doit	= tcmu_genl_rm_dev_done,
298 	},
299 	{
300 		.cmd	= TCMU_CMD_RECONFIG_DEVICE_DONE,
301 		.flags	= GENL_ADMIN_PERM,
302 		.policy	= tcmu_attr_policy,
303 		.doit	= tcmu_genl_reconfig_dev_done,
304 	},
305 };
306 
307 /* Our generic netlink family */
308 static struct genl_family tcmu_genl_family __ro_after_init = {
309 	.module = THIS_MODULE,
310 	.hdrsize = 0,
311 	.name = "TCM-USER",
312 	.version = 2,
313 	.maxattr = TCMU_ATTR_MAX,
314 	.mcgrps = tcmu_mcgrps,
315 	.n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
316 	.netnsok = true,
317 	.ops = tcmu_genl_ops,
318 	.n_ops = ARRAY_SIZE(tcmu_genl_ops),
319 };
320 
321 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
322 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
323 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
324 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
325 
326 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
327 {
328 	struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
329 	uint32_t i;
330 
331 	for (i = 0; i < len; i++)
332 		clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
333 }
334 
335 static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
336 					struct tcmu_cmd *tcmu_cmd)
337 {
338 	struct page *page;
339 	int ret, dbi;
340 
341 	dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
342 	if (dbi == udev->dbi_thresh)
343 		return false;
344 
345 	page = radix_tree_lookup(&udev->data_blocks, dbi);
346 	if (!page) {
347 		if (atomic_add_return(1, &global_db_count) >
348 					TCMU_GLOBAL_MAX_BLOCKS) {
349 			atomic_dec(&global_db_count);
350 			return false;
351 		}
352 
353 		/* try to get new page from the mm */
354 		page = alloc_page(GFP_KERNEL);
355 		if (!page)
356 			goto err_alloc;
357 
358 		ret = radix_tree_insert(&udev->data_blocks, dbi, page);
359 		if (ret)
360 			goto err_insert;
361 	}
362 
363 	if (dbi > udev->dbi_max)
364 		udev->dbi_max = dbi;
365 
366 	set_bit(dbi, udev->data_bitmap);
367 	tcmu_cmd_set_dbi(tcmu_cmd, dbi);
368 
369 	return true;
370 err_insert:
371 	__free_page(page);
372 err_alloc:
373 	atomic_dec(&global_db_count);
374 	return false;
375 }
376 
377 static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
378 				  struct tcmu_cmd *tcmu_cmd)
379 {
380 	int i;
381 
382 	udev->waiting_global = false;
383 
384 	for (i = tcmu_cmd->dbi_cur; i < tcmu_cmd->dbi_cnt; i++) {
385 		if (!tcmu_get_empty_block(udev, tcmu_cmd))
386 			goto err;
387 	}
388 	return true;
389 
390 err:
391 	udev->waiting_global = true;
392 	/* Try to wake up the unmap thread */
393 	wake_up(&unmap_wait);
394 	return false;
395 }
396 
397 static inline struct page *
398 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
399 {
400 	return radix_tree_lookup(&udev->data_blocks, dbi);
401 }
402 
403 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
404 {
405 	kfree(tcmu_cmd->dbi);
406 	kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
407 }
408 
409 static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
410 {
411 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
412 	size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
413 
414 	if (se_cmd->se_cmd_flags & SCF_BIDI) {
415 		BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
416 		data_length += round_up(se_cmd->t_bidi_data_sg->length,
417 				DATA_BLOCK_SIZE);
418 	}
419 
420 	return data_length;
421 }
422 
423 static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
424 {
425 	size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
426 
427 	return data_length / DATA_BLOCK_SIZE;
428 }
429 
430 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
431 {
432 	struct se_device *se_dev = se_cmd->se_dev;
433 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
434 	struct tcmu_cmd *tcmu_cmd;
435 
436 	tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
437 	if (!tcmu_cmd)
438 		return NULL;
439 
440 	tcmu_cmd->se_cmd = se_cmd;
441 	tcmu_cmd->tcmu_dev = udev;
442 
443 	tcmu_cmd_reset_dbi_cur(tcmu_cmd);
444 	tcmu_cmd->dbi_cnt = tcmu_cmd_get_block_cnt(tcmu_cmd);
445 	tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
446 				GFP_KERNEL);
447 	if (!tcmu_cmd->dbi) {
448 		kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
449 		return NULL;
450 	}
451 
452 	return tcmu_cmd;
453 }
454 
455 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
456 {
457 	unsigned long offset = offset_in_page(vaddr);
458 
459 	size = round_up(size+offset, PAGE_SIZE);
460 	vaddr -= offset;
461 
462 	while (size) {
463 		flush_dcache_page(virt_to_page(vaddr));
464 		size -= PAGE_SIZE;
465 	}
466 }
467 
468 /*
469  * Some ring helper functions. We don't assume size is a power of 2 so
470  * we can't use circ_buf.h.
471  */
472 static inline size_t spc_used(size_t head, size_t tail, size_t size)
473 {
474 	int diff = head - tail;
475 
476 	if (diff >= 0)
477 		return diff;
478 	else
479 		return size + diff;
480 }
481 
482 static inline size_t spc_free(size_t head, size_t tail, size_t size)
483 {
484 	/* Keep 1 byte unused or we can't tell full from empty */
485 	return (size - spc_used(head, tail, size) - 1);
486 }
487 
488 static inline size_t head_to_end(size_t head, size_t size)
489 {
490 	return size - head;
491 }
492 
493 static inline void new_iov(struct iovec **iov, int *iov_cnt,
494 			   struct tcmu_dev *udev)
495 {
496 	struct iovec *iovec;
497 
498 	if (*iov_cnt != 0)
499 		(*iov)++;
500 	(*iov_cnt)++;
501 
502 	iovec = *iov;
503 	memset(iovec, 0, sizeof(struct iovec));
504 }
505 
506 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
507 
508 /* offset is relative to mb_addr */
509 static inline size_t get_block_offset_user(struct tcmu_dev *dev,
510 		int dbi, int remaining)
511 {
512 	return dev->data_off + dbi * DATA_BLOCK_SIZE +
513 		DATA_BLOCK_SIZE - remaining;
514 }
515 
516 static inline size_t iov_tail(struct iovec *iov)
517 {
518 	return (size_t)iov->iov_base + iov->iov_len;
519 }
520 
521 static int scatter_data_area(struct tcmu_dev *udev,
522 	struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
523 	unsigned int data_nents, struct iovec **iov,
524 	int *iov_cnt, bool copy_data)
525 {
526 	int i, dbi;
527 	int block_remaining = 0;
528 	void *from, *to = NULL;
529 	size_t copy_bytes, to_offset, offset;
530 	struct scatterlist *sg;
531 	struct page *page;
532 
533 	for_each_sg(data_sg, sg, data_nents, i) {
534 		int sg_remaining = sg->length;
535 		from = kmap_atomic(sg_page(sg)) + sg->offset;
536 		while (sg_remaining > 0) {
537 			if (block_remaining == 0) {
538 				if (to)
539 					kunmap_atomic(to);
540 
541 				block_remaining = DATA_BLOCK_SIZE;
542 				dbi = tcmu_cmd_get_dbi(tcmu_cmd);
543 				page = tcmu_get_block_page(udev, dbi);
544 				to = kmap_atomic(page);
545 			}
546 
547 			copy_bytes = min_t(size_t, sg_remaining,
548 					block_remaining);
549 			to_offset = get_block_offset_user(udev, dbi,
550 					block_remaining);
551 
552 			if (*iov_cnt != 0 &&
553 			    to_offset == iov_tail(*iov)) {
554 				(*iov)->iov_len += copy_bytes;
555 			} else {
556 				new_iov(iov, iov_cnt, udev);
557 				(*iov)->iov_base = (void __user *)to_offset;
558 				(*iov)->iov_len = copy_bytes;
559 			}
560 			if (copy_data) {
561 				offset = DATA_BLOCK_SIZE - block_remaining;
562 				memcpy(to + offset,
563 				       from + sg->length - sg_remaining,
564 				       copy_bytes);
565 				tcmu_flush_dcache_range(to, copy_bytes);
566 			}
567 			sg_remaining -= copy_bytes;
568 			block_remaining -= copy_bytes;
569 		}
570 		kunmap_atomic(from - sg->offset);
571 	}
572 	if (to)
573 		kunmap_atomic(to);
574 
575 	return 0;
576 }
577 
578 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
579 			     bool bidi)
580 {
581 	struct se_cmd *se_cmd = cmd->se_cmd;
582 	int i, dbi;
583 	int block_remaining = 0;
584 	void *from = NULL, *to;
585 	size_t copy_bytes, offset;
586 	struct scatterlist *sg, *data_sg;
587 	struct page *page;
588 	unsigned int data_nents;
589 	uint32_t count = 0;
590 
591 	if (!bidi) {
592 		data_sg = se_cmd->t_data_sg;
593 		data_nents = se_cmd->t_data_nents;
594 	} else {
595 
596 		/*
597 		 * For bidi case, the first count blocks are for Data-Out
598 		 * buffer blocks, and before gathering the Data-In buffer
599 		 * the Data-Out buffer blocks should be discarded.
600 		 */
601 		count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
602 
603 		data_sg = se_cmd->t_bidi_data_sg;
604 		data_nents = se_cmd->t_bidi_data_nents;
605 	}
606 
607 	tcmu_cmd_set_dbi_cur(cmd, count);
608 
609 	for_each_sg(data_sg, sg, data_nents, i) {
610 		int sg_remaining = sg->length;
611 		to = kmap_atomic(sg_page(sg)) + sg->offset;
612 		while (sg_remaining > 0) {
613 			if (block_remaining == 0) {
614 				if (from)
615 					kunmap_atomic(from);
616 
617 				block_remaining = DATA_BLOCK_SIZE;
618 				dbi = tcmu_cmd_get_dbi(cmd);
619 				page = tcmu_get_block_page(udev, dbi);
620 				from = kmap_atomic(page);
621 			}
622 			copy_bytes = min_t(size_t, sg_remaining,
623 					block_remaining);
624 			offset = DATA_BLOCK_SIZE - block_remaining;
625 			tcmu_flush_dcache_range(from, copy_bytes);
626 			memcpy(to + sg->length - sg_remaining, from + offset,
627 					copy_bytes);
628 
629 			sg_remaining -= copy_bytes;
630 			block_remaining -= copy_bytes;
631 		}
632 		kunmap_atomic(to - sg->offset);
633 	}
634 	if (from)
635 		kunmap_atomic(from);
636 }
637 
638 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
639 {
640 	return DATA_BLOCK_SIZE * (thresh - bitmap_weight(bitmap, thresh));
641 }
642 
643 /*
644  * We can't queue a command until we have space available on the cmd ring *and*
645  * space available on the data area.
646  *
647  * Called with ring lock held.
648  */
649 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
650 		size_t cmd_size, size_t data_needed)
651 {
652 	struct tcmu_mailbox *mb = udev->mb_addr;
653 	uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
654 				/ DATA_BLOCK_SIZE;
655 	size_t space, cmd_needed;
656 	u32 cmd_head;
657 
658 	tcmu_flush_dcache_range(mb, sizeof(*mb));
659 
660 	cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
661 
662 	/*
663 	 * If cmd end-of-ring space is too small then we need space for a NOP plus
664 	 * original cmd - cmds are internally contiguous.
665 	 */
666 	if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
667 		cmd_needed = cmd_size;
668 	else
669 		cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
670 
671 	space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
672 	if (space < cmd_needed) {
673 		pr_debug("no cmd space: %u %u %u\n", cmd_head,
674 		       udev->cmdr_last_cleaned, udev->cmdr_size);
675 		return false;
676 	}
677 
678 	/* try to check and get the data blocks as needed */
679 	space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
680 	if (space < data_needed) {
681 		unsigned long blocks_left = DATA_BLOCK_BITS - udev->dbi_thresh;
682 		unsigned long grow;
683 
684 		if (blocks_left < blocks_needed) {
685 			pr_debug("no data space: only %lu available, but ask for %zu\n",
686 					blocks_left * DATA_BLOCK_SIZE,
687 					data_needed);
688 			return false;
689 		}
690 
691 		/* Try to expand the thresh */
692 		if (!udev->dbi_thresh) {
693 			/* From idle state */
694 			uint32_t init_thresh = DATA_BLOCK_INIT_BITS;
695 
696 			udev->dbi_thresh = max(blocks_needed, init_thresh);
697 		} else {
698 			/*
699 			 * Grow the data area by max(blocks needed,
700 			 * dbi_thresh / 2), but limited to the max
701 			 * DATA_BLOCK_BITS size.
702 			 */
703 			grow = max(blocks_needed, udev->dbi_thresh / 2);
704 			udev->dbi_thresh += grow;
705 			if (udev->dbi_thresh > DATA_BLOCK_BITS)
706 				udev->dbi_thresh = DATA_BLOCK_BITS;
707 		}
708 	}
709 
710 	return tcmu_get_empty_blocks(udev, cmd);
711 }
712 
713 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
714 {
715 	return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
716 			sizeof(struct tcmu_cmd_entry));
717 }
718 
719 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
720 					   size_t base_command_size)
721 {
722 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
723 	size_t command_size;
724 
725 	command_size = base_command_size +
726 		round_up(scsi_command_size(se_cmd->t_task_cdb),
727 				TCMU_OP_ALIGN_SIZE);
728 
729 	WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
730 
731 	return command_size;
732 }
733 
734 static int tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd)
735 {
736 	struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
737 	unsigned long tmo = udev->cmd_time_out;
738 	int cmd_id;
739 
740 	if (tcmu_cmd->cmd_id)
741 		return 0;
742 
743 	cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
744 	if (cmd_id < 0) {
745 		pr_err("tcmu: Could not allocate cmd id.\n");
746 		return cmd_id;
747 	}
748 	tcmu_cmd->cmd_id = cmd_id;
749 
750 	if (!tmo)
751 		return 0;
752 
753 	tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
754 	mod_timer(&udev->timeout, tcmu_cmd->deadline);
755 	return 0;
756 }
757 
758 static sense_reason_t
759 tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
760 {
761 	struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
762 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
763 	size_t base_command_size, command_size;
764 	struct tcmu_mailbox *mb;
765 	struct tcmu_cmd_entry *entry;
766 	struct iovec *iov;
767 	int iov_cnt, ret;
768 	uint32_t cmd_head;
769 	uint64_t cdb_off;
770 	bool copy_to_data_area;
771 	size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
772 
773 	if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
774 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
775 
776 	/*
777 	 * Must be a certain minimum size for response sense info, but
778 	 * also may be larger if the iov array is large.
779 	 *
780 	 * We prepare as many iovs as possbile for potential uses here,
781 	 * because it's expensive to tell how many regions are freed in
782 	 * the bitmap & global data pool, as the size calculated here
783 	 * will only be used to do the checks.
784 	 *
785 	 * The size will be recalculated later as actually needed to save
786 	 * cmd area memories.
787 	 */
788 	base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
789 	command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
790 
791 	mutex_lock(&udev->cmdr_lock);
792 
793 	mb = udev->mb_addr;
794 	cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
795 	if ((command_size > (udev->cmdr_size / 2)) ||
796 	    data_length > udev->data_size) {
797 		pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
798 			"cmd ring/data area\n", command_size, data_length,
799 			udev->cmdr_size, udev->data_size);
800 		mutex_unlock(&udev->cmdr_lock);
801 		return TCM_INVALID_CDB_FIELD;
802 	}
803 
804 	while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
805 		int ret;
806 		DEFINE_WAIT(__wait);
807 
808 		prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
809 
810 		pr_debug("sleeping for ring space\n");
811 		mutex_unlock(&udev->cmdr_lock);
812 		if (udev->cmd_time_out)
813 			ret = schedule_timeout(
814 					msecs_to_jiffies(udev->cmd_time_out));
815 		else
816 			ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
817 		finish_wait(&udev->wait_cmdr, &__wait);
818 		if (!ret) {
819 			pr_warn("tcmu: command timed out\n");
820 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
821 		}
822 
823 		mutex_lock(&udev->cmdr_lock);
824 
825 		/* We dropped cmdr_lock, cmd_head is stale */
826 		cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
827 	}
828 
829 	/* Insert a PAD if end-of-ring space is too small */
830 	if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
831 		size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
832 
833 		entry = (void *) mb + CMDR_OFF + cmd_head;
834 		tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
835 		tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
836 		entry->hdr.cmd_id = 0; /* not used for PAD */
837 		entry->hdr.kflags = 0;
838 		entry->hdr.uflags = 0;
839 		tcmu_flush_dcache_range(entry, sizeof(*entry));
840 
841 		UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
842 		tcmu_flush_dcache_range(mb, sizeof(*mb));
843 
844 		cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
845 		WARN_ON(cmd_head != 0);
846 	}
847 
848 	entry = (void *) mb + CMDR_OFF + cmd_head;
849 	memset(entry, 0, command_size);
850 	tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
851 
852 	/* Handle allocating space from the data area */
853 	tcmu_cmd_reset_dbi_cur(tcmu_cmd);
854 	iov = &entry->req.iov[0];
855 	iov_cnt = 0;
856 	copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
857 		|| se_cmd->se_cmd_flags & SCF_BIDI);
858 	ret = scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
859 				se_cmd->t_data_nents, &iov, &iov_cnt,
860 				copy_to_data_area);
861 	if (ret) {
862 		tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
863 		mutex_unlock(&udev->cmdr_lock);
864 
865 		pr_err("tcmu: alloc and scatter data failed\n");
866 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
867 	}
868 	entry->req.iov_cnt = iov_cnt;
869 
870 	/* Handle BIDI commands */
871 	iov_cnt = 0;
872 	if (se_cmd->se_cmd_flags & SCF_BIDI) {
873 		iov++;
874 		ret = scatter_data_area(udev, tcmu_cmd,
875 					se_cmd->t_bidi_data_sg,
876 					se_cmd->t_bidi_data_nents,
877 					&iov, &iov_cnt, false);
878 		if (ret) {
879 			tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
880 			mutex_unlock(&udev->cmdr_lock);
881 
882 			pr_err("tcmu: alloc and scatter bidi data failed\n");
883 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
884 		}
885 	}
886 	entry->req.iov_bidi_cnt = iov_cnt;
887 
888 	ret = tcmu_setup_cmd_timer(tcmu_cmd);
889 	if (ret) {
890 		tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
891 		mutex_unlock(&udev->cmdr_lock);
892 		return TCM_OUT_OF_RESOURCES;
893 	}
894 	entry->hdr.cmd_id = tcmu_cmd->cmd_id;
895 
896 	/*
897 	 * Recalaulate the command's base size and size according
898 	 * to the actual needs
899 	 */
900 	base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
901 						       entry->req.iov_bidi_cnt);
902 	command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
903 
904 	tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
905 
906 	/* All offsets relative to mb_addr, not start of entry! */
907 	cdb_off = CMDR_OFF + cmd_head + base_command_size;
908 	memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
909 	entry->req.cdb_off = cdb_off;
910 	tcmu_flush_dcache_range(entry, sizeof(*entry));
911 
912 	UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
913 	tcmu_flush_dcache_range(mb, sizeof(*mb));
914 	mutex_unlock(&udev->cmdr_lock);
915 
916 	/* TODO: only if FLUSH and FUA? */
917 	uio_event_notify(&udev->uio_info);
918 
919 	if (udev->cmd_time_out)
920 		mod_timer(&udev->timeout, round_jiffies_up(jiffies +
921 			  msecs_to_jiffies(udev->cmd_time_out)));
922 
923 	return TCM_NO_SENSE;
924 }
925 
926 static sense_reason_t
927 tcmu_queue_cmd(struct se_cmd *se_cmd)
928 {
929 	struct tcmu_cmd *tcmu_cmd;
930 	sense_reason_t ret;
931 
932 	tcmu_cmd = tcmu_alloc_cmd(se_cmd);
933 	if (!tcmu_cmd)
934 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
935 
936 	ret = tcmu_queue_cmd_ring(tcmu_cmd);
937 	if (ret != TCM_NO_SENSE) {
938 		pr_err("TCMU: Could not queue command\n");
939 
940 		tcmu_free_cmd(tcmu_cmd);
941 	}
942 
943 	return ret;
944 }
945 
946 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
947 {
948 	struct se_cmd *se_cmd = cmd->se_cmd;
949 	struct tcmu_dev *udev = cmd->tcmu_dev;
950 
951 	/*
952 	 * cmd has been completed already from timeout, just reclaim
953 	 * data area space and free cmd
954 	 */
955 	if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
956 		goto out;
957 
958 	tcmu_cmd_reset_dbi_cur(cmd);
959 
960 	if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
961 		pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
962 			cmd->se_cmd);
963 		entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
964 	} else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
965 		transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
966 	} else if (se_cmd->se_cmd_flags & SCF_BIDI) {
967 		/* Get Data-In buffer before clean up */
968 		gather_data_area(udev, cmd, true);
969 	} else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
970 		gather_data_area(udev, cmd, false);
971 	} else if (se_cmd->data_direction == DMA_TO_DEVICE) {
972 		/* TODO: */
973 	} else if (se_cmd->data_direction != DMA_NONE) {
974 		pr_warn("TCMU: data direction was %d!\n",
975 			se_cmd->data_direction);
976 	}
977 
978 	target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
979 
980 out:
981 	cmd->se_cmd = NULL;
982 	tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
983 	tcmu_free_cmd(cmd);
984 }
985 
986 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
987 {
988 	struct tcmu_mailbox *mb;
989 	int handled = 0;
990 
991 	if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
992 		pr_err("ring broken, not handling completions\n");
993 		return 0;
994 	}
995 
996 	mb = udev->mb_addr;
997 	tcmu_flush_dcache_range(mb, sizeof(*mb));
998 
999 	while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1000 
1001 		struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1002 		struct tcmu_cmd *cmd;
1003 
1004 		tcmu_flush_dcache_range(entry, sizeof(*entry));
1005 
1006 		if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1007 			UPDATE_HEAD(udev->cmdr_last_cleaned,
1008 				    tcmu_hdr_get_len(entry->hdr.len_op),
1009 				    udev->cmdr_size);
1010 			continue;
1011 		}
1012 		WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1013 
1014 		spin_lock(&udev->commands_lock);
1015 		cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1016 		spin_unlock(&udev->commands_lock);
1017 
1018 		if (!cmd) {
1019 			pr_err("cmd_id not found, ring is broken\n");
1020 			set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1021 			break;
1022 		}
1023 
1024 		tcmu_handle_completion(cmd, entry);
1025 
1026 		UPDATE_HEAD(udev->cmdr_last_cleaned,
1027 			    tcmu_hdr_get_len(entry->hdr.len_op),
1028 			    udev->cmdr_size);
1029 
1030 		handled++;
1031 	}
1032 
1033 	if (mb->cmd_tail == mb->cmd_head)
1034 		del_timer(&udev->timeout); /* no more pending cmds */
1035 
1036 	wake_up(&udev->wait_cmdr);
1037 
1038 	return handled;
1039 }
1040 
1041 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1042 {
1043 	struct tcmu_cmd *cmd = p;
1044 
1045 	if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1046 		return 0;
1047 
1048 	if (!time_after(jiffies, cmd->deadline))
1049 		return 0;
1050 
1051 	set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1052 	target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
1053 	cmd->se_cmd = NULL;
1054 
1055 	return 0;
1056 }
1057 
1058 static void tcmu_device_timedout(struct timer_list *t)
1059 {
1060 	struct tcmu_dev *udev = from_timer(udev, t, timeout);
1061 	unsigned long flags;
1062 
1063 	spin_lock_irqsave(&udev->commands_lock, flags);
1064 	idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
1065 	spin_unlock_irqrestore(&udev->commands_lock, flags);
1066 
1067 	/* Try to wake up the ummap thread */
1068 	wake_up(&unmap_wait);
1069 
1070 	/*
1071 	 * We don't need to wakeup threads on wait_cmdr since they have their
1072 	 * own timeout.
1073 	 */
1074 }
1075 
1076 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1077 {
1078 	struct tcmu_hba *tcmu_hba;
1079 
1080 	tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1081 	if (!tcmu_hba)
1082 		return -ENOMEM;
1083 
1084 	tcmu_hba->host_id = host_id;
1085 	hba->hba_ptr = tcmu_hba;
1086 
1087 	return 0;
1088 }
1089 
1090 static void tcmu_detach_hba(struct se_hba *hba)
1091 {
1092 	kfree(hba->hba_ptr);
1093 	hba->hba_ptr = NULL;
1094 }
1095 
1096 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1097 {
1098 	struct tcmu_dev *udev;
1099 
1100 	udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1101 	if (!udev)
1102 		return NULL;
1103 	kref_init(&udev->kref);
1104 
1105 	udev->name = kstrdup(name, GFP_KERNEL);
1106 	if (!udev->name) {
1107 		kfree(udev);
1108 		return NULL;
1109 	}
1110 
1111 	udev->hba = hba;
1112 	udev->cmd_time_out = TCMU_TIME_OUT;
1113 
1114 	init_waitqueue_head(&udev->wait_cmdr);
1115 	mutex_init(&udev->cmdr_lock);
1116 
1117 	idr_init(&udev->commands);
1118 	spin_lock_init(&udev->commands_lock);
1119 
1120 	timer_setup(&udev->timeout, tcmu_device_timedout, 0);
1121 
1122 	init_waitqueue_head(&udev->nl_cmd_wq);
1123 	spin_lock_init(&udev->nl_cmd_lock);
1124 
1125 	INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1126 
1127 	return &udev->se_dev;
1128 }
1129 
1130 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1131 {
1132 	struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
1133 
1134 	mutex_lock(&tcmu_dev->cmdr_lock);
1135 	tcmu_handle_completions(tcmu_dev);
1136 	mutex_unlock(&tcmu_dev->cmdr_lock);
1137 
1138 	return 0;
1139 }
1140 
1141 /*
1142  * mmap code from uio.c. Copied here because we want to hook mmap()
1143  * and this stuff must come along.
1144  */
1145 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1146 {
1147 	struct tcmu_dev *udev = vma->vm_private_data;
1148 	struct uio_info *info = &udev->uio_info;
1149 
1150 	if (vma->vm_pgoff < MAX_UIO_MAPS) {
1151 		if (info->mem[vma->vm_pgoff].size == 0)
1152 			return -1;
1153 		return (int)vma->vm_pgoff;
1154 	}
1155 	return -1;
1156 }
1157 
1158 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1159 {
1160 	struct page *page;
1161 	int ret;
1162 
1163 	mutex_lock(&udev->cmdr_lock);
1164 	page = tcmu_get_block_page(udev, dbi);
1165 	if (likely(page)) {
1166 		mutex_unlock(&udev->cmdr_lock);
1167 		return page;
1168 	}
1169 
1170 	/*
1171 	 * Normally it shouldn't be here:
1172 	 * Only when the userspace has touched the blocks which
1173 	 * are out of the tcmu_cmd's data iov[], and will return
1174 	 * one zeroed page.
1175 	 */
1176 	pr_warn("Block(%u) out of cmd's iov[] has been touched!\n", dbi);
1177 	pr_warn("Mostly it will be a bug of userspace, please have a check!\n");
1178 
1179 	if (dbi >= udev->dbi_thresh) {
1180 		/* Extern the udev->dbi_thresh to dbi + 1 */
1181 		udev->dbi_thresh = dbi + 1;
1182 		udev->dbi_max = dbi;
1183 	}
1184 
1185 	page = radix_tree_lookup(&udev->data_blocks, dbi);
1186 	if (!page) {
1187 		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1188 		if (!page) {
1189 			mutex_unlock(&udev->cmdr_lock);
1190 			return NULL;
1191 		}
1192 
1193 		ret = radix_tree_insert(&udev->data_blocks, dbi, page);
1194 		if (ret) {
1195 			mutex_unlock(&udev->cmdr_lock);
1196 			__free_page(page);
1197 			return NULL;
1198 		}
1199 
1200 		/*
1201 		 * Since this case is rare in page fault routine, here we
1202 		 * will allow the global_db_count >= TCMU_GLOBAL_MAX_BLOCKS
1203 		 * to reduce possible page fault call trace.
1204 		 */
1205 		atomic_inc(&global_db_count);
1206 	}
1207 	mutex_unlock(&udev->cmdr_lock);
1208 
1209 	return page;
1210 }
1211 
1212 static int tcmu_vma_fault(struct vm_fault *vmf)
1213 {
1214 	struct tcmu_dev *udev = vmf->vma->vm_private_data;
1215 	struct uio_info *info = &udev->uio_info;
1216 	struct page *page;
1217 	unsigned long offset;
1218 	void *addr;
1219 
1220 	int mi = tcmu_find_mem_index(vmf->vma);
1221 	if (mi < 0)
1222 		return VM_FAULT_SIGBUS;
1223 
1224 	/*
1225 	 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1226 	 * to use mem[N].
1227 	 */
1228 	offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1229 
1230 	if (offset < udev->data_off) {
1231 		/* For the vmalloc()ed cmd area pages */
1232 		addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1233 		page = vmalloc_to_page(addr);
1234 	} else {
1235 		uint32_t dbi;
1236 
1237 		/* For the dynamically growing data area pages */
1238 		dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1239 		page = tcmu_try_get_block_page(udev, dbi);
1240 		if (!page)
1241 			return VM_FAULT_NOPAGE;
1242 	}
1243 
1244 	get_page(page);
1245 	vmf->page = page;
1246 	return 0;
1247 }
1248 
1249 static const struct vm_operations_struct tcmu_vm_ops = {
1250 	.fault = tcmu_vma_fault,
1251 };
1252 
1253 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1254 {
1255 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1256 
1257 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1258 	vma->vm_ops = &tcmu_vm_ops;
1259 
1260 	vma->vm_private_data = udev;
1261 
1262 	/* Ensure the mmap is exactly the right size */
1263 	if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
1264 		return -EINVAL;
1265 
1266 	return 0;
1267 }
1268 
1269 static int tcmu_open(struct uio_info *info, struct inode *inode)
1270 {
1271 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1272 
1273 	/* O_EXCL not supported for char devs, so fake it? */
1274 	if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1275 		return -EBUSY;
1276 
1277 	udev->inode = inode;
1278 	kref_get(&udev->kref);
1279 
1280 	pr_debug("open\n");
1281 
1282 	return 0;
1283 }
1284 
1285 static void tcmu_dev_call_rcu(struct rcu_head *p)
1286 {
1287 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
1288 	struct tcmu_dev *udev = TCMU_DEV(dev);
1289 
1290 	kfree(udev->uio_info.name);
1291 	kfree(udev->name);
1292 	kfree(udev);
1293 }
1294 
1295 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1296 {
1297 	if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1298 		kmem_cache_free(tcmu_cmd_cache, cmd);
1299 		return 0;
1300 	}
1301 	return -EINVAL;
1302 }
1303 
1304 static void tcmu_blocks_release(struct tcmu_dev *udev)
1305 {
1306 	int i;
1307 	struct page *page;
1308 
1309 	/* Try to release all block pages */
1310 	mutex_lock(&udev->cmdr_lock);
1311 	for (i = 0; i <= udev->dbi_max; i++) {
1312 		page = radix_tree_delete(&udev->data_blocks, i);
1313 		if (page) {
1314 			__free_page(page);
1315 			atomic_dec(&global_db_count);
1316 		}
1317 	}
1318 	mutex_unlock(&udev->cmdr_lock);
1319 }
1320 
1321 static void tcmu_dev_kref_release(struct kref *kref)
1322 {
1323 	struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1324 	struct se_device *dev = &udev->se_dev;
1325 	struct tcmu_cmd *cmd;
1326 	bool all_expired = true;
1327 	int i;
1328 
1329 	vfree(udev->mb_addr);
1330 	udev->mb_addr = NULL;
1331 
1332 	/* Upper layer should drain all requests before calling this */
1333 	spin_lock_irq(&udev->commands_lock);
1334 	idr_for_each_entry(&udev->commands, cmd, i) {
1335 		if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1336 			all_expired = false;
1337 	}
1338 	idr_destroy(&udev->commands);
1339 	spin_unlock_irq(&udev->commands_lock);
1340 	WARN_ON(!all_expired);
1341 
1342 	tcmu_blocks_release(udev);
1343 
1344 	call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1345 }
1346 
1347 static int tcmu_release(struct uio_info *info, struct inode *inode)
1348 {
1349 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1350 
1351 	clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1352 
1353 	pr_debug("close\n");
1354 	/* release ref from open */
1355 	kref_put(&udev->kref, tcmu_dev_kref_release);
1356 	return 0;
1357 }
1358 
1359 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1360 {
1361 	struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1362 
1363 	if (!tcmu_kern_cmd_reply_supported)
1364 		return;
1365 
1366 	if (udev->nl_reply_supported <= 0)
1367 		return;
1368 
1369 relock:
1370 	spin_lock(&udev->nl_cmd_lock);
1371 
1372 	if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1373 		spin_unlock(&udev->nl_cmd_lock);
1374 		pr_debug("sleeping for open nl cmd\n");
1375 		wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1376 		goto relock;
1377 	}
1378 
1379 	memset(nl_cmd, 0, sizeof(*nl_cmd));
1380 	nl_cmd->cmd = cmd;
1381 	init_completion(&nl_cmd->complete);
1382 
1383 	spin_unlock(&udev->nl_cmd_lock);
1384 }
1385 
1386 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1387 {
1388 	struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1389 	int ret;
1390 	DEFINE_WAIT(__wait);
1391 
1392 	if (!tcmu_kern_cmd_reply_supported)
1393 		return 0;
1394 
1395 	if (udev->nl_reply_supported <= 0)
1396 		return 0;
1397 
1398 	pr_debug("sleeping for nl reply\n");
1399 	wait_for_completion(&nl_cmd->complete);
1400 
1401 	spin_lock(&udev->nl_cmd_lock);
1402 	nl_cmd->cmd = TCMU_CMD_UNSPEC;
1403 	ret = nl_cmd->status;
1404 	nl_cmd->status = 0;
1405 	spin_unlock(&udev->nl_cmd_lock);
1406 
1407 	wake_up_all(&udev->nl_cmd_wq);
1408 
1409 	return ret;;
1410 }
1411 
1412 static int tcmu_netlink_event(struct tcmu_dev *udev, enum tcmu_genl_cmd cmd,
1413 			      int reconfig_attr, const void *reconfig_data)
1414 {
1415 	struct sk_buff *skb;
1416 	void *msg_header;
1417 	int ret = -ENOMEM;
1418 
1419 	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1420 	if (!skb)
1421 		return ret;
1422 
1423 	msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1424 	if (!msg_header)
1425 		goto free_skb;
1426 
1427 	ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1428 	if (ret < 0)
1429 		goto free_skb;
1430 
1431 	ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1432 	if (ret < 0)
1433 		goto free_skb;
1434 
1435 	ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1436 	if (ret < 0)
1437 		goto free_skb;
1438 
1439 	if (cmd == TCMU_CMD_RECONFIG_DEVICE) {
1440 		switch (reconfig_attr) {
1441 		case TCMU_ATTR_DEV_CFG:
1442 			ret = nla_put_string(skb, reconfig_attr, reconfig_data);
1443 			break;
1444 		case TCMU_ATTR_DEV_SIZE:
1445 			ret = nla_put_u64_64bit(skb, reconfig_attr,
1446 						*((u64 *)reconfig_data),
1447 						TCMU_ATTR_PAD);
1448 			break;
1449 		case TCMU_ATTR_WRITECACHE:
1450 			ret = nla_put_u8(skb, reconfig_attr,
1451 					  *((u8 *)reconfig_data));
1452 			break;
1453 		default:
1454 			BUG();
1455 		}
1456 
1457 		if (ret < 0)
1458 			goto free_skb;
1459 	}
1460 
1461 	genlmsg_end(skb, msg_header);
1462 
1463 	tcmu_init_genl_cmd_reply(udev, cmd);
1464 
1465 	ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1466 				TCMU_MCGRP_CONFIG, GFP_KERNEL);
1467 	/* We don't care if no one is listening */
1468 	if (ret == -ESRCH)
1469 		ret = 0;
1470 	if (!ret)
1471 		ret = tcmu_wait_genl_cmd_reply(udev);
1472 
1473 	return ret;
1474 free_skb:
1475 	nlmsg_free(skb);
1476 	return ret;
1477 }
1478 
1479 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1480 {
1481 	struct tcmu_hba *hba = udev->hba->hba_ptr;
1482 	struct uio_info *info;
1483 	size_t size, used;
1484 	char *str;
1485 
1486 	info = &udev->uio_info;
1487 	size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1488 			udev->dev_config);
1489 	size += 1; /* for \0 */
1490 	str = kmalloc(size, GFP_KERNEL);
1491 	if (!str)
1492 		return -ENOMEM;
1493 
1494 	used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1495 	if (udev->dev_config[0])
1496 		snprintf(str + used, size - used, "/%s", udev->dev_config);
1497 
1498 	/* If the old string exists, free it */
1499 	kfree(info->name);
1500 	info->name = str;
1501 
1502 	return 0;
1503 }
1504 
1505 static int tcmu_configure_device(struct se_device *dev)
1506 {
1507 	struct tcmu_dev *udev = TCMU_DEV(dev);
1508 	struct uio_info *info;
1509 	struct tcmu_mailbox *mb;
1510 	int ret = 0;
1511 
1512 	ret = tcmu_update_uio_info(udev);
1513 	if (ret)
1514 		return ret;
1515 
1516 	info = &udev->uio_info;
1517 
1518 	udev->mb_addr = vzalloc(CMDR_SIZE);
1519 	if (!udev->mb_addr) {
1520 		ret = -ENOMEM;
1521 		goto err_vzalloc;
1522 	}
1523 
1524 	/* mailbox fits in first part of CMDR space */
1525 	udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1526 	udev->data_off = CMDR_SIZE;
1527 	udev->data_size = DATA_SIZE;
1528 	udev->dbi_thresh = 0; /* Default in Idle state */
1529 	udev->waiting_global = false;
1530 
1531 	/* Initialise the mailbox of the ring buffer */
1532 	mb = udev->mb_addr;
1533 	mb->version = TCMU_MAILBOX_VERSION;
1534 	mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1535 	mb->cmdr_off = CMDR_OFF;
1536 	mb->cmdr_size = udev->cmdr_size;
1537 
1538 	WARN_ON(!PAGE_ALIGNED(udev->data_off));
1539 	WARN_ON(udev->data_size % PAGE_SIZE);
1540 	WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1541 
1542 	info->version = __stringify(TCMU_MAILBOX_VERSION);
1543 
1544 	info->mem[0].name = "tcm-user command & data buffer";
1545 	info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1546 	info->mem[0].size = TCMU_RING_SIZE;
1547 	info->mem[0].memtype = UIO_MEM_NONE;
1548 
1549 	info->irqcontrol = tcmu_irqcontrol;
1550 	info->irq = UIO_IRQ_CUSTOM;
1551 
1552 	info->mmap = tcmu_mmap;
1553 	info->open = tcmu_open;
1554 	info->release = tcmu_release;
1555 
1556 	ret = uio_register_device(tcmu_root_device, info);
1557 	if (ret)
1558 		goto err_register;
1559 
1560 	/* User can set hw_block_size before enable the device */
1561 	if (dev->dev_attrib.hw_block_size == 0)
1562 		dev->dev_attrib.hw_block_size = 512;
1563 	/* Other attributes can be configured in userspace */
1564 	if (!dev->dev_attrib.hw_max_sectors)
1565 		dev->dev_attrib.hw_max_sectors = 128;
1566 	if (!dev->dev_attrib.emulate_write_cache)
1567 		dev->dev_attrib.emulate_write_cache = 0;
1568 	dev->dev_attrib.hw_queue_depth = 128;
1569 
1570 	/* If user didn't explicitly disable netlink reply support, use
1571 	 * module scope setting.
1572 	 */
1573 	if (udev->nl_reply_supported >= 0)
1574 		udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
1575 
1576 	/*
1577 	 * Get a ref incase userspace does a close on the uio device before
1578 	 * LIO has initiated tcmu_free_device.
1579 	 */
1580 	kref_get(&udev->kref);
1581 
1582 	ret = tcmu_netlink_event(udev, TCMU_CMD_ADDED_DEVICE, 0, NULL);
1583 	if (ret)
1584 		goto err_netlink;
1585 
1586 	mutex_lock(&root_udev_mutex);
1587 	list_add(&udev->node, &root_udev);
1588 	mutex_unlock(&root_udev_mutex);
1589 
1590 	return 0;
1591 
1592 err_netlink:
1593 	kref_put(&udev->kref, tcmu_dev_kref_release);
1594 	uio_unregister_device(&udev->uio_info);
1595 err_register:
1596 	vfree(udev->mb_addr);
1597 	udev->mb_addr = NULL;
1598 err_vzalloc:
1599 	kfree(info->name);
1600 	info->name = NULL;
1601 
1602 	return ret;
1603 }
1604 
1605 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1606 {
1607 	return udev->uio_info.uio_dev ? true : false;
1608 }
1609 
1610 static void tcmu_free_device(struct se_device *dev)
1611 {
1612 	struct tcmu_dev *udev = TCMU_DEV(dev);
1613 
1614 	/* release ref from init */
1615 	kref_put(&udev->kref, tcmu_dev_kref_release);
1616 }
1617 
1618 static void tcmu_destroy_device(struct se_device *dev)
1619 {
1620 	struct tcmu_dev *udev = TCMU_DEV(dev);
1621 
1622 	del_timer_sync(&udev->timeout);
1623 
1624 	mutex_lock(&root_udev_mutex);
1625 	list_del(&udev->node);
1626 	mutex_unlock(&root_udev_mutex);
1627 
1628 	tcmu_netlink_event(udev, TCMU_CMD_REMOVED_DEVICE, 0, NULL);
1629 
1630 	uio_unregister_device(&udev->uio_info);
1631 
1632 	/* release ref from configure */
1633 	kref_put(&udev->kref, tcmu_dev_kref_release);
1634 }
1635 
1636 enum {
1637 	Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1638 	Opt_nl_reply_supported, Opt_err,
1639 };
1640 
1641 static match_table_t tokens = {
1642 	{Opt_dev_config, "dev_config=%s"},
1643 	{Opt_dev_size, "dev_size=%u"},
1644 	{Opt_hw_block_size, "hw_block_size=%u"},
1645 	{Opt_hw_max_sectors, "hw_max_sectors=%u"},
1646 	{Opt_nl_reply_supported, "nl_reply_supported=%d"},
1647 	{Opt_err, NULL}
1648 };
1649 
1650 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1651 {
1652 	unsigned long tmp_ul;
1653 	char *arg_p;
1654 	int ret;
1655 
1656 	arg_p = match_strdup(arg);
1657 	if (!arg_p)
1658 		return -ENOMEM;
1659 
1660 	ret = kstrtoul(arg_p, 0, &tmp_ul);
1661 	kfree(arg_p);
1662 	if (ret < 0) {
1663 		pr_err("kstrtoul() failed for dev attrib\n");
1664 		return ret;
1665 	}
1666 	if (!tmp_ul) {
1667 		pr_err("dev attrib must be nonzero\n");
1668 		return -EINVAL;
1669 	}
1670 	*dev_attrib = tmp_ul;
1671 	return 0;
1672 }
1673 
1674 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1675 		const char *page, ssize_t count)
1676 {
1677 	struct tcmu_dev *udev = TCMU_DEV(dev);
1678 	char *orig, *ptr, *opts, *arg_p;
1679 	substring_t args[MAX_OPT_ARGS];
1680 	int ret = 0, token;
1681 
1682 	opts = kstrdup(page, GFP_KERNEL);
1683 	if (!opts)
1684 		return -ENOMEM;
1685 
1686 	orig = opts;
1687 
1688 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
1689 		if (!*ptr)
1690 			continue;
1691 
1692 		token = match_token(ptr, tokens, args);
1693 		switch (token) {
1694 		case Opt_dev_config:
1695 			if (match_strlcpy(udev->dev_config, &args[0],
1696 					  TCMU_CONFIG_LEN) == 0) {
1697 				ret = -EINVAL;
1698 				break;
1699 			}
1700 			pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1701 			break;
1702 		case Opt_dev_size:
1703 			arg_p = match_strdup(&args[0]);
1704 			if (!arg_p) {
1705 				ret = -ENOMEM;
1706 				break;
1707 			}
1708 			ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1709 			kfree(arg_p);
1710 			if (ret < 0)
1711 				pr_err("kstrtoul() failed for dev_size=\n");
1712 			break;
1713 		case Opt_hw_block_size:
1714 			ret = tcmu_set_dev_attrib(&args[0],
1715 					&(dev->dev_attrib.hw_block_size));
1716 			break;
1717 		case Opt_hw_max_sectors:
1718 			ret = tcmu_set_dev_attrib(&args[0],
1719 					&(dev->dev_attrib.hw_max_sectors));
1720 			break;
1721 		case Opt_nl_reply_supported:
1722 			arg_p = match_strdup(&args[0]);
1723 			if (!arg_p) {
1724 				ret = -ENOMEM;
1725 				break;
1726 			}
1727 			ret = kstrtoint(arg_p, 0, &udev->nl_reply_supported);
1728 			kfree(arg_p);
1729 			if (ret < 0)
1730 				pr_err("kstrtoint() failed for nl_reply_supported=\n");
1731 			break;
1732 		default:
1733 			break;
1734 		}
1735 
1736 		if (ret)
1737 			break;
1738 	}
1739 
1740 	kfree(orig);
1741 	return (!ret) ? count : ret;
1742 }
1743 
1744 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
1745 {
1746 	struct tcmu_dev *udev = TCMU_DEV(dev);
1747 	ssize_t bl = 0;
1748 
1749 	bl = sprintf(b + bl, "Config: %s ",
1750 		     udev->dev_config[0] ? udev->dev_config : "NULL");
1751 	bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size);
1752 
1753 	return bl;
1754 }
1755 
1756 static sector_t tcmu_get_blocks(struct se_device *dev)
1757 {
1758 	struct tcmu_dev *udev = TCMU_DEV(dev);
1759 
1760 	return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1761 		       dev->dev_attrib.block_size);
1762 }
1763 
1764 static sense_reason_t
1765 tcmu_parse_cdb(struct se_cmd *cmd)
1766 {
1767 	return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
1768 }
1769 
1770 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
1771 {
1772 	struct se_dev_attrib *da = container_of(to_config_group(item),
1773 					struct se_dev_attrib, da_group);
1774 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1775 
1776 	return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
1777 }
1778 
1779 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
1780 				       size_t count)
1781 {
1782 	struct se_dev_attrib *da = container_of(to_config_group(item),
1783 					struct se_dev_attrib, da_group);
1784 	struct tcmu_dev *udev = container_of(da->da_dev,
1785 					struct tcmu_dev, se_dev);
1786 	u32 val;
1787 	int ret;
1788 
1789 	if (da->da_dev->export_count) {
1790 		pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
1791 		return -EINVAL;
1792 	}
1793 
1794 	ret = kstrtou32(page, 0, &val);
1795 	if (ret < 0)
1796 		return ret;
1797 
1798 	udev->cmd_time_out = val * MSEC_PER_SEC;
1799 	return count;
1800 }
1801 CONFIGFS_ATTR(tcmu_, cmd_time_out);
1802 
1803 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
1804 {
1805 	struct se_dev_attrib *da = container_of(to_config_group(item),
1806 						struct se_dev_attrib, da_group);
1807 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1808 
1809 	return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
1810 }
1811 
1812 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
1813 				     size_t count)
1814 {
1815 	struct se_dev_attrib *da = container_of(to_config_group(item),
1816 						struct se_dev_attrib, da_group);
1817 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1818 	int ret, len;
1819 
1820 	len = strlen(page);
1821 	if (!len || len > TCMU_CONFIG_LEN - 1)
1822 		return -EINVAL;
1823 
1824 	/* Check if device has been configured before */
1825 	if (tcmu_dev_configured(udev)) {
1826 		ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1827 					 TCMU_ATTR_DEV_CFG, page);
1828 		if (ret) {
1829 			pr_err("Unable to reconfigure device\n");
1830 			return ret;
1831 		}
1832 		strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1833 
1834 		ret = tcmu_update_uio_info(udev);
1835 		if (ret)
1836 			return ret;
1837 		return count;
1838 	}
1839 	strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1840 
1841 	return count;
1842 }
1843 CONFIGFS_ATTR(tcmu_, dev_config);
1844 
1845 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
1846 {
1847 	struct se_dev_attrib *da = container_of(to_config_group(item),
1848 						struct se_dev_attrib, da_group);
1849 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1850 
1851 	return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
1852 }
1853 
1854 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
1855 				   size_t count)
1856 {
1857 	struct se_dev_attrib *da = container_of(to_config_group(item),
1858 						struct se_dev_attrib, da_group);
1859 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1860 	u64 val;
1861 	int ret;
1862 
1863 	ret = kstrtou64(page, 0, &val);
1864 	if (ret < 0)
1865 		return ret;
1866 
1867 	/* Check if device has been configured before */
1868 	if (tcmu_dev_configured(udev)) {
1869 		ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1870 					 TCMU_ATTR_DEV_SIZE, &val);
1871 		if (ret) {
1872 			pr_err("Unable to reconfigure device\n");
1873 			return ret;
1874 		}
1875 	}
1876 	udev->dev_size = val;
1877 	return count;
1878 }
1879 CONFIGFS_ATTR(tcmu_, dev_size);
1880 
1881 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
1882 		char *page)
1883 {
1884 	struct se_dev_attrib *da = container_of(to_config_group(item),
1885 						struct se_dev_attrib, da_group);
1886 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1887 
1888 	return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
1889 }
1890 
1891 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
1892 		const char *page, size_t count)
1893 {
1894 	struct se_dev_attrib *da = container_of(to_config_group(item),
1895 						struct se_dev_attrib, da_group);
1896 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1897 	s8 val;
1898 	int ret;
1899 
1900 	ret = kstrtos8(page, 0, &val);
1901 	if (ret < 0)
1902 		return ret;
1903 
1904 	udev->nl_reply_supported = val;
1905 	return count;
1906 }
1907 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
1908 
1909 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
1910 					     char *page)
1911 {
1912 	struct se_dev_attrib *da = container_of(to_config_group(item),
1913 					struct se_dev_attrib, da_group);
1914 
1915 	return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
1916 }
1917 
1918 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
1919 					      const char *page, size_t count)
1920 {
1921 	struct se_dev_attrib *da = container_of(to_config_group(item),
1922 					struct se_dev_attrib, da_group);
1923 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1924 	u8 val;
1925 	int ret;
1926 
1927 	ret = kstrtou8(page, 0, &val);
1928 	if (ret < 0)
1929 		return ret;
1930 
1931 	/* Check if device has been configured before */
1932 	if (tcmu_dev_configured(udev)) {
1933 		ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1934 					 TCMU_ATTR_WRITECACHE, &val);
1935 		if (ret) {
1936 			pr_err("Unable to reconfigure device\n");
1937 			return ret;
1938 		}
1939 	}
1940 
1941 	da->emulate_write_cache = val;
1942 	return count;
1943 }
1944 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
1945 
1946 static struct configfs_attribute *tcmu_attrib_attrs[] = {
1947 	&tcmu_attr_cmd_time_out,
1948 	&tcmu_attr_dev_config,
1949 	&tcmu_attr_dev_size,
1950 	&tcmu_attr_emulate_write_cache,
1951 	&tcmu_attr_nl_reply_supported,
1952 	NULL,
1953 };
1954 
1955 static struct configfs_attribute **tcmu_attrs;
1956 
1957 static struct target_backend_ops tcmu_ops = {
1958 	.name			= "user",
1959 	.owner			= THIS_MODULE,
1960 	.transport_flags	= TRANSPORT_FLAG_PASSTHROUGH,
1961 	.attach_hba		= tcmu_attach_hba,
1962 	.detach_hba		= tcmu_detach_hba,
1963 	.alloc_device		= tcmu_alloc_device,
1964 	.configure_device	= tcmu_configure_device,
1965 	.destroy_device		= tcmu_destroy_device,
1966 	.free_device		= tcmu_free_device,
1967 	.parse_cdb		= tcmu_parse_cdb,
1968 	.set_configfs_dev_params = tcmu_set_configfs_dev_params,
1969 	.show_configfs_dev_params = tcmu_show_configfs_dev_params,
1970 	.get_device_type	= sbc_get_device_type,
1971 	.get_blocks		= tcmu_get_blocks,
1972 	.tb_dev_attrib_attrs	= NULL,
1973 };
1974 
1975 static int unmap_thread_fn(void *data)
1976 {
1977 	struct tcmu_dev *udev;
1978 	loff_t off;
1979 	uint32_t start, end, block;
1980 	struct page *page;
1981 	int i;
1982 
1983 	while (!kthread_should_stop()) {
1984 		DEFINE_WAIT(__wait);
1985 
1986 		prepare_to_wait(&unmap_wait, &__wait, TASK_INTERRUPTIBLE);
1987 		schedule();
1988 		finish_wait(&unmap_wait, &__wait);
1989 
1990 		if (kthread_should_stop())
1991 			break;
1992 
1993 		mutex_lock(&root_udev_mutex);
1994 		list_for_each_entry(udev, &root_udev, node) {
1995 			mutex_lock(&udev->cmdr_lock);
1996 
1997 			/* Try to complete the finished commands first */
1998 			tcmu_handle_completions(udev);
1999 
2000 			/* Skip the udevs waiting the global pool or in idle */
2001 			if (udev->waiting_global || !udev->dbi_thresh) {
2002 				mutex_unlock(&udev->cmdr_lock);
2003 				continue;
2004 			}
2005 
2006 			end = udev->dbi_max + 1;
2007 			block = find_last_bit(udev->data_bitmap, end);
2008 			if (block == udev->dbi_max) {
2009 				/*
2010 				 * The last bit is dbi_max, so there is
2011 				 * no need to shrink any blocks.
2012 				 */
2013 				mutex_unlock(&udev->cmdr_lock);
2014 				continue;
2015 			} else if (block == end) {
2016 				/* The current udev will goto idle state */
2017 				udev->dbi_thresh = start = 0;
2018 				udev->dbi_max = 0;
2019 			} else {
2020 				udev->dbi_thresh = start = block + 1;
2021 				udev->dbi_max = block;
2022 			}
2023 
2024 			/* Here will truncate the data area from off */
2025 			off = udev->data_off + start * DATA_BLOCK_SIZE;
2026 			unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2027 
2028 			/* Release the block pages */
2029 			for (i = start; i < end; i++) {
2030 				page = radix_tree_delete(&udev->data_blocks, i);
2031 				if (page) {
2032 					__free_page(page);
2033 					atomic_dec(&global_db_count);
2034 				}
2035 			}
2036 			mutex_unlock(&udev->cmdr_lock);
2037 		}
2038 
2039 		/*
2040 		 * Try to wake up the udevs who are waiting
2041 		 * for the global data pool.
2042 		 */
2043 		list_for_each_entry(udev, &root_udev, node) {
2044 			if (udev->waiting_global)
2045 				wake_up(&udev->wait_cmdr);
2046 		}
2047 		mutex_unlock(&root_udev_mutex);
2048 	}
2049 
2050 	return 0;
2051 }
2052 
2053 static int __init tcmu_module_init(void)
2054 {
2055 	int ret, i, k, len = 0;
2056 
2057 	BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2058 
2059 	tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2060 				sizeof(struct tcmu_cmd),
2061 				__alignof__(struct tcmu_cmd),
2062 				0, NULL);
2063 	if (!tcmu_cmd_cache)
2064 		return -ENOMEM;
2065 
2066 	tcmu_root_device = root_device_register("tcm_user");
2067 	if (IS_ERR(tcmu_root_device)) {
2068 		ret = PTR_ERR(tcmu_root_device);
2069 		goto out_free_cache;
2070 	}
2071 
2072 	ret = genl_register_family(&tcmu_genl_family);
2073 	if (ret < 0) {
2074 		goto out_unreg_device;
2075 	}
2076 
2077 	for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2078 		len += sizeof(struct configfs_attribute *);
2079 	}
2080 	for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2081 		len += sizeof(struct configfs_attribute *);
2082 	}
2083 	len += sizeof(struct configfs_attribute *);
2084 
2085 	tcmu_attrs = kzalloc(len, GFP_KERNEL);
2086 	if (!tcmu_attrs) {
2087 		ret = -ENOMEM;
2088 		goto out_unreg_genl;
2089 	}
2090 
2091 	for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2092 		tcmu_attrs[i] = passthrough_attrib_attrs[i];
2093 	}
2094 	for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2095 		tcmu_attrs[i] = tcmu_attrib_attrs[k];
2096 		i++;
2097 	}
2098 	tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2099 
2100 	ret = transport_backend_register(&tcmu_ops);
2101 	if (ret)
2102 		goto out_attrs;
2103 
2104 	init_waitqueue_head(&unmap_wait);
2105 	unmap_thread = kthread_run(unmap_thread_fn, NULL, "tcmu_unmap");
2106 	if (IS_ERR(unmap_thread)) {
2107 		ret = PTR_ERR(unmap_thread);
2108 		goto out_unreg_transport;
2109 	}
2110 
2111 	return 0;
2112 
2113 out_unreg_transport:
2114 	target_backend_unregister(&tcmu_ops);
2115 out_attrs:
2116 	kfree(tcmu_attrs);
2117 out_unreg_genl:
2118 	genl_unregister_family(&tcmu_genl_family);
2119 out_unreg_device:
2120 	root_device_unregister(tcmu_root_device);
2121 out_free_cache:
2122 	kmem_cache_destroy(tcmu_cmd_cache);
2123 
2124 	return ret;
2125 }
2126 
2127 static void __exit tcmu_module_exit(void)
2128 {
2129 	kthread_stop(unmap_thread);
2130 	target_backend_unregister(&tcmu_ops);
2131 	kfree(tcmu_attrs);
2132 	genl_unregister_family(&tcmu_genl_family);
2133 	root_device_unregister(tcmu_root_device);
2134 	kmem_cache_destroy(tcmu_cmd_cache);
2135 }
2136 
2137 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2138 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2139 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2140 MODULE_LICENSE("GPL");
2141 
2142 module_init(tcmu_module_init);
2143 module_exit(tcmu_module_exit);
2144