1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
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.
7  */
8 
9 #include <linux/spinlock.h>
10 #include <linux/module.h>
11 #include <linux/kernel.h>
12 #include <linux/timer.h>
13 #include <linux/parser.h>
14 #include <linux/vmalloc.h>
15 #include <linux/uio_driver.h>
16 #include <linux/xarray.h>
17 #include <linux/stringify.h>
18 #include <linux/bitops.h>
19 #include <linux/highmem.h>
20 #include <linux/configfs.h>
21 #include <linux/mutex.h>
22 #include <linux/workqueue.h>
23 #include <linux/pagemap.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>
30 
31 #include <linux/target_core_user.h>
32 
33 /**
34  * DOC: Userspace I/O
35  * Userspace I/O
36  * -------------
37  *
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.
41  *
42  * It uses the UIO framework to do a lot of the device-creation and
43  * introspection work for us.
44  *
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.
52  *
53  * SECURITY:
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.
60  */
61 
62 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
63 
64 /* For mailbox plus cmd ring, the size is fixed 8MB */
65 #define MB_CMDR_SIZE_DEF (8 * 1024 * 1024)
66 /* Offset of cmd ring is size of mailbox */
67 #define CMDR_OFF ((__u32)sizeof(struct tcmu_mailbox))
68 #define CMDR_SIZE_DEF (MB_CMDR_SIZE_DEF - CMDR_OFF)
69 
70 /*
71  * For data area, the default block size is PAGE_SIZE and
72  * the default total size is 256K * PAGE_SIZE.
73  */
74 #define DATA_PAGES_PER_BLK_DEF 1
75 #define DATA_AREA_PAGES_DEF (256 * 1024)
76 
77 #define TCMU_MBS_TO_PAGES(_mbs) ((size_t)_mbs << (20 - PAGE_SHIFT))
78 #define TCMU_PAGES_TO_MBS(_pages) (_pages >> (20 - PAGE_SHIFT))
79 
80 /*
81  * Default number of global data blocks(512K * PAGE_SIZE)
82  * when the unmap thread will be started.
83  */
84 #define TCMU_GLOBAL_MAX_PAGES_DEF (512 * 1024)
85 
86 static u8 tcmu_kern_cmd_reply_supported;
87 static u8 tcmu_netlink_blocked;
88 
89 static struct device *tcmu_root_device;
90 
91 struct tcmu_hba {
92 	u32 host_id;
93 };
94 
95 #define TCMU_CONFIG_LEN 256
96 
97 static DEFINE_MUTEX(tcmu_nl_cmd_mutex);
98 static LIST_HEAD(tcmu_nl_cmd_list);
99 
100 struct tcmu_dev;
101 
102 struct tcmu_nl_cmd {
103 	/* wake up thread waiting for reply */
104 	struct completion complete;
105 	struct list_head nl_list;
106 	struct tcmu_dev *udev;
107 	int cmd;
108 	int status;
109 };
110 
111 struct tcmu_dev {
112 	struct list_head node;
113 	struct kref kref;
114 
115 	struct se_device se_dev;
116 	struct se_dev_plug se_plug;
117 
118 	char *name;
119 	struct se_hba *hba;
120 
121 #define TCMU_DEV_BIT_OPEN 0
122 #define TCMU_DEV_BIT_BROKEN 1
123 #define TCMU_DEV_BIT_BLOCKED 2
124 #define TCMU_DEV_BIT_TMR_NOTIFY 3
125 #define TCMU_DEV_BIT_PLUGGED 4
126 	unsigned long flags;
127 
128 	struct uio_info uio_info;
129 
130 	struct inode *inode;
131 
132 	uint64_t dev_size;
133 
134 	struct tcmu_mailbox *mb_addr;
135 	void *cmdr;
136 	u32 cmdr_size;
137 	u32 cmdr_last_cleaned;
138 	/* Offset of data area from start of mb */
139 	/* Must add data_off and mb_addr to get the address */
140 	size_t data_off;
141 	int data_area_mb;
142 	uint32_t max_blocks;
143 	size_t mmap_pages;
144 
145 	struct mutex cmdr_lock;
146 	struct list_head qfull_queue;
147 	struct list_head tmr_queue;
148 
149 	uint32_t dbi_max;
150 	uint32_t dbi_thresh;
151 	unsigned long *data_bitmap;
152 	struct xarray data_pages;
153 	uint32_t data_pages_per_blk;
154 	uint32_t data_blk_size;
155 
156 	struct xarray commands;
157 
158 	struct timer_list cmd_timer;
159 	unsigned int cmd_time_out;
160 	struct list_head inflight_queue;
161 
162 	struct timer_list qfull_timer;
163 	int qfull_time_out;
164 
165 	struct list_head timedout_entry;
166 
167 	struct tcmu_nl_cmd curr_nl_cmd;
168 
169 	char dev_config[TCMU_CONFIG_LEN];
170 
171 	int nl_reply_supported;
172 };
173 
174 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
175 
176 struct tcmu_cmd {
177 	struct se_cmd *se_cmd;
178 	struct tcmu_dev *tcmu_dev;
179 	struct list_head queue_entry;
180 
181 	uint16_t cmd_id;
182 
183 	/* Can't use se_cmd when cleaning up expired cmds, because if
184 	   cmd has been completed then accessing se_cmd is off limits */
185 	uint32_t dbi_cnt;
186 	uint32_t dbi_bidi_cnt;
187 	uint32_t dbi_cur;
188 	uint32_t *dbi;
189 
190 	uint32_t data_len_bidi;
191 
192 	unsigned long deadline;
193 
194 #define TCMU_CMD_BIT_EXPIRED 0
195 #define TCMU_CMD_BIT_KEEP_BUF 1
196 	unsigned long flags;
197 };
198 
199 struct tcmu_tmr {
200 	struct list_head queue_entry;
201 
202 	uint8_t tmr_type;
203 	uint32_t tmr_cmd_cnt;
204 	int16_t tmr_cmd_ids[];
205 };
206 
207 /*
208  * To avoid dead lock the mutex lock order should always be:
209  *
210  * mutex_lock(&root_udev_mutex);
211  * ...
212  * mutex_lock(&tcmu_dev->cmdr_lock);
213  * mutex_unlock(&tcmu_dev->cmdr_lock);
214  * ...
215  * mutex_unlock(&root_udev_mutex);
216  */
217 static DEFINE_MUTEX(root_udev_mutex);
218 static LIST_HEAD(root_udev);
219 
220 static DEFINE_SPINLOCK(timed_out_udevs_lock);
221 static LIST_HEAD(timed_out_udevs);
222 
223 static struct kmem_cache *tcmu_cmd_cache;
224 
225 static atomic_t global_page_count = ATOMIC_INIT(0);
226 static struct delayed_work tcmu_unmap_work;
227 static int tcmu_global_max_pages = TCMU_GLOBAL_MAX_PAGES_DEF;
228 
229 static int tcmu_set_global_max_data_area(const char *str,
230 					 const struct kernel_param *kp)
231 {
232 	int ret, max_area_mb;
233 
234 	ret = kstrtoint(str, 10, &max_area_mb);
235 	if (ret)
236 		return -EINVAL;
237 
238 	if (max_area_mb <= 0) {
239 		pr_err("global_max_data_area must be larger than 0.\n");
240 		return -EINVAL;
241 	}
242 
243 	tcmu_global_max_pages = TCMU_MBS_TO_PAGES(max_area_mb);
244 	if (atomic_read(&global_page_count) > tcmu_global_max_pages)
245 		schedule_delayed_work(&tcmu_unmap_work, 0);
246 	else
247 		cancel_delayed_work_sync(&tcmu_unmap_work);
248 
249 	return 0;
250 }
251 
252 static int tcmu_get_global_max_data_area(char *buffer,
253 					 const struct kernel_param *kp)
254 {
255 	return sprintf(buffer, "%d\n", TCMU_PAGES_TO_MBS(tcmu_global_max_pages));
256 }
257 
258 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
259 	.set = tcmu_set_global_max_data_area,
260 	.get = tcmu_get_global_max_data_area,
261 };
262 
263 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
264 		S_IWUSR | S_IRUGO);
265 MODULE_PARM_DESC(global_max_data_area_mb,
266 		 "Max MBs allowed to be allocated to all the tcmu device's "
267 		 "data areas.");
268 
269 static int tcmu_get_block_netlink(char *buffer,
270 				  const struct kernel_param *kp)
271 {
272 	return sprintf(buffer, "%s\n", tcmu_netlink_blocked ?
273 		       "blocked" : "unblocked");
274 }
275 
276 static int tcmu_set_block_netlink(const char *str,
277 				  const struct kernel_param *kp)
278 {
279 	int ret;
280 	u8 val;
281 
282 	ret = kstrtou8(str, 0, &val);
283 	if (ret < 0)
284 		return ret;
285 
286 	if (val > 1) {
287 		pr_err("Invalid block netlink value %u\n", val);
288 		return -EINVAL;
289 	}
290 
291 	tcmu_netlink_blocked = val;
292 	return 0;
293 }
294 
295 static const struct kernel_param_ops tcmu_block_netlink_op = {
296 	.set = tcmu_set_block_netlink,
297 	.get = tcmu_get_block_netlink,
298 };
299 
300 module_param_cb(block_netlink, &tcmu_block_netlink_op, NULL, S_IWUSR | S_IRUGO);
301 MODULE_PARM_DESC(block_netlink, "Block new netlink commands.");
302 
303 static int tcmu_fail_netlink_cmd(struct tcmu_nl_cmd *nl_cmd)
304 {
305 	struct tcmu_dev *udev = nl_cmd->udev;
306 
307 	if (!tcmu_netlink_blocked) {
308 		pr_err("Could not reset device's netlink interface. Netlink is not blocked.\n");
309 		return -EBUSY;
310 	}
311 
312 	if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
313 		pr_debug("Aborting nl cmd %d on %s\n", nl_cmd->cmd, udev->name);
314 		nl_cmd->status = -EINTR;
315 		list_del(&nl_cmd->nl_list);
316 		complete(&nl_cmd->complete);
317 	}
318 	return 0;
319 }
320 
321 static int tcmu_set_reset_netlink(const char *str,
322 				  const struct kernel_param *kp)
323 {
324 	struct tcmu_nl_cmd *nl_cmd, *tmp_cmd;
325 	int ret;
326 	u8 val;
327 
328 	ret = kstrtou8(str, 0, &val);
329 	if (ret < 0)
330 		return ret;
331 
332 	if (val != 1) {
333 		pr_err("Invalid reset netlink value %u\n", val);
334 		return -EINVAL;
335 	}
336 
337 	mutex_lock(&tcmu_nl_cmd_mutex);
338 	list_for_each_entry_safe(nl_cmd, tmp_cmd, &tcmu_nl_cmd_list, nl_list) {
339 		ret = tcmu_fail_netlink_cmd(nl_cmd);
340 		if (ret)
341 			break;
342 	}
343 	mutex_unlock(&tcmu_nl_cmd_mutex);
344 
345 	return ret;
346 }
347 
348 static const struct kernel_param_ops tcmu_reset_netlink_op = {
349 	.set = tcmu_set_reset_netlink,
350 };
351 
352 module_param_cb(reset_netlink, &tcmu_reset_netlink_op, NULL, S_IWUSR);
353 MODULE_PARM_DESC(reset_netlink, "Reset netlink commands.");
354 
355 /* multicast group */
356 enum tcmu_multicast_groups {
357 	TCMU_MCGRP_CONFIG,
358 };
359 
360 static const struct genl_multicast_group tcmu_mcgrps[] = {
361 	[TCMU_MCGRP_CONFIG] = { .name = "config", },
362 };
363 
364 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
365 	[TCMU_ATTR_DEVICE]	= { .type = NLA_STRING },
366 	[TCMU_ATTR_MINOR]	= { .type = NLA_U32 },
367 	[TCMU_ATTR_CMD_STATUS]	= { .type = NLA_S32 },
368 	[TCMU_ATTR_DEVICE_ID]	= { .type = NLA_U32 },
369 	[TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
370 };
371 
372 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
373 {
374 	struct tcmu_dev *udev = NULL;
375 	struct tcmu_nl_cmd *nl_cmd;
376 	int dev_id, rc, ret = 0;
377 
378 	if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
379 	    !info->attrs[TCMU_ATTR_DEVICE_ID]) {
380 		printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
381 		return -EINVAL;
382         }
383 
384 	dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
385 	rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
386 
387 	mutex_lock(&tcmu_nl_cmd_mutex);
388 	list_for_each_entry(nl_cmd, &tcmu_nl_cmd_list, nl_list) {
389 		if (nl_cmd->udev->se_dev.dev_index == dev_id) {
390 			udev = nl_cmd->udev;
391 			break;
392 		}
393 	}
394 
395 	if (!udev) {
396 		pr_err("tcmu nl cmd %u/%d completion could not find device with dev id %u.\n",
397 		       completed_cmd, rc, dev_id);
398 		ret = -ENODEV;
399 		goto unlock;
400 	}
401 	list_del(&nl_cmd->nl_list);
402 
403 	pr_debug("%s genl cmd done got id %d curr %d done %d rc %d stat %d\n",
404 		 udev->name, dev_id, nl_cmd->cmd, completed_cmd, rc,
405 		 nl_cmd->status);
406 
407 	if (nl_cmd->cmd != completed_cmd) {
408 		pr_err("Mismatched commands on %s (Expecting reply for %d. Current %d).\n",
409 		       udev->name, completed_cmd, nl_cmd->cmd);
410 		ret = -EINVAL;
411 		goto unlock;
412 	}
413 
414 	nl_cmd->status = rc;
415 	complete(&nl_cmd->complete);
416 unlock:
417 	mutex_unlock(&tcmu_nl_cmd_mutex);
418 	return ret;
419 }
420 
421 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
422 {
423 	return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
424 }
425 
426 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
427 {
428 	return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
429 }
430 
431 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
432 				       struct genl_info *info)
433 {
434 	return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
435 }
436 
437 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
438 {
439 	if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
440 		tcmu_kern_cmd_reply_supported  =
441 			nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
442 		printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
443 		       tcmu_kern_cmd_reply_supported);
444 	}
445 
446 	return 0;
447 }
448 
449 static const struct genl_small_ops tcmu_genl_ops[] = {
450 	{
451 		.cmd	= TCMU_CMD_SET_FEATURES,
452 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
453 		.flags	= GENL_ADMIN_PERM,
454 		.doit	= tcmu_genl_set_features,
455 	},
456 	{
457 		.cmd	= TCMU_CMD_ADDED_DEVICE_DONE,
458 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
459 		.flags	= GENL_ADMIN_PERM,
460 		.doit	= tcmu_genl_add_dev_done,
461 	},
462 	{
463 		.cmd	= TCMU_CMD_REMOVED_DEVICE_DONE,
464 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
465 		.flags	= GENL_ADMIN_PERM,
466 		.doit	= tcmu_genl_rm_dev_done,
467 	},
468 	{
469 		.cmd	= TCMU_CMD_RECONFIG_DEVICE_DONE,
470 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
471 		.flags	= GENL_ADMIN_PERM,
472 		.doit	= tcmu_genl_reconfig_dev_done,
473 	},
474 };
475 
476 /* Our generic netlink family */
477 static struct genl_family tcmu_genl_family __ro_after_init = {
478 	.module = THIS_MODULE,
479 	.hdrsize = 0,
480 	.name = "TCM-USER",
481 	.version = 2,
482 	.maxattr = TCMU_ATTR_MAX,
483 	.policy = tcmu_attr_policy,
484 	.mcgrps = tcmu_mcgrps,
485 	.n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
486 	.netnsok = true,
487 	.small_ops = tcmu_genl_ops,
488 	.n_small_ops = ARRAY_SIZE(tcmu_genl_ops),
489 	.resv_start_op = TCMU_CMD_SET_FEATURES + 1,
490 };
491 
492 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
493 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
494 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
495 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
496 
497 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
498 {
499 	struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
500 	uint32_t i;
501 
502 	for (i = 0; i < len; i++)
503 		clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
504 }
505 
506 static inline int tcmu_get_empty_block(struct tcmu_dev *udev,
507 				       struct tcmu_cmd *tcmu_cmd,
508 				       int prev_dbi, int length, int *iov_cnt)
509 {
510 	XA_STATE(xas, &udev->data_pages, 0);
511 	struct page *page;
512 	int i, cnt, dbi, dpi;
513 	int page_cnt = DIV_ROUND_UP(length, PAGE_SIZE);
514 
515 	dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
516 	if (dbi == udev->dbi_thresh)
517 		return -1;
518 
519 	dpi = dbi * udev->data_pages_per_blk;
520 	/* Count the number of already allocated pages */
521 	xas_set(&xas, dpi);
522 	rcu_read_lock();
523 	for (cnt = 0; xas_next(&xas) && cnt < page_cnt;)
524 		cnt++;
525 	rcu_read_unlock();
526 
527 	for (i = cnt; i < page_cnt; i++) {
528 		/* try to get new zeroed page from the mm */
529 		page = alloc_page(GFP_NOIO | __GFP_ZERO);
530 		if (!page)
531 			break;
532 
533 		if (xa_store(&udev->data_pages, dpi + i, page, GFP_NOIO)) {
534 			__free_page(page);
535 			break;
536 		}
537 	}
538 	if (atomic_add_return(i - cnt, &global_page_count) >
539 			      tcmu_global_max_pages)
540 		schedule_delayed_work(&tcmu_unmap_work, 0);
541 
542 	if (i && dbi > udev->dbi_max)
543 		udev->dbi_max = dbi;
544 
545 	set_bit(dbi, udev->data_bitmap);
546 	tcmu_cmd_set_dbi(tcmu_cmd, dbi);
547 
548 	if (dbi != prev_dbi + 1)
549 		*iov_cnt += 1;
550 
551 	return i == page_cnt ? dbi : -1;
552 }
553 
554 static int tcmu_get_empty_blocks(struct tcmu_dev *udev,
555 				 struct tcmu_cmd *tcmu_cmd, int length)
556 {
557 	/* start value of dbi + 1 must not be a valid dbi */
558 	int dbi = -2;
559 	int blk_data_len, iov_cnt = 0;
560 	uint32_t blk_size = udev->data_blk_size;
561 
562 	for (; length > 0; length -= blk_size) {
563 		blk_data_len = min_t(uint32_t, length, blk_size);
564 		dbi = tcmu_get_empty_block(udev, tcmu_cmd, dbi, blk_data_len,
565 					   &iov_cnt);
566 		if (dbi < 0)
567 			return -1;
568 	}
569 	return iov_cnt;
570 }
571 
572 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
573 {
574 	kfree(tcmu_cmd->dbi);
575 	kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
576 }
577 
578 static inline void tcmu_cmd_set_block_cnts(struct tcmu_cmd *cmd)
579 {
580 	int i, len;
581 	struct se_cmd *se_cmd = cmd->se_cmd;
582 	uint32_t blk_size = cmd->tcmu_dev->data_blk_size;
583 
584 	cmd->dbi_cnt = DIV_ROUND_UP(se_cmd->data_length, blk_size);
585 
586 	if (se_cmd->se_cmd_flags & SCF_BIDI) {
587 		BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
588 		for (i = 0, len = 0; i < se_cmd->t_bidi_data_nents; i++)
589 			len += se_cmd->t_bidi_data_sg[i].length;
590 		cmd->dbi_bidi_cnt = DIV_ROUND_UP(len, blk_size);
591 		cmd->dbi_cnt += cmd->dbi_bidi_cnt;
592 		cmd->data_len_bidi = len;
593 	}
594 }
595 
596 static int new_block_to_iov(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
597 			    struct iovec **iov, int prev_dbi, int len)
598 {
599 	/* Get the next dbi */
600 	int dbi = tcmu_cmd_get_dbi(cmd);
601 
602 	/* Do not add more than udev->data_blk_size to iov */
603 	len = min_t(int,  len, udev->data_blk_size);
604 
605 	/*
606 	 * The following code will gather and map the blocks to the same iovec
607 	 * when the blocks are all next to each other.
608 	 */
609 	if (dbi != prev_dbi + 1) {
610 		/* dbi is not next to previous dbi, so start new iov */
611 		if (prev_dbi >= 0)
612 			(*iov)++;
613 		/* write offset relative to mb_addr */
614 		(*iov)->iov_base = (void __user *)
615 				   (udev->data_off + dbi * udev->data_blk_size);
616 	}
617 	(*iov)->iov_len += len;
618 
619 	return dbi;
620 }
621 
622 static void tcmu_setup_iovs(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
623 			    struct iovec **iov, int data_length)
624 {
625 	/* start value of dbi + 1 must not be a valid dbi */
626 	int dbi = -2;
627 
628 	/* We prepare the IOVs for DMA_FROM_DEVICE transfer direction */
629 	for (; data_length > 0; data_length -= udev->data_blk_size)
630 		dbi = new_block_to_iov(udev, cmd, iov, dbi, data_length);
631 }
632 
633 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
634 {
635 	struct se_device *se_dev = se_cmd->se_dev;
636 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
637 	struct tcmu_cmd *tcmu_cmd;
638 
639 	tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_NOIO);
640 	if (!tcmu_cmd)
641 		return NULL;
642 
643 	INIT_LIST_HEAD(&tcmu_cmd->queue_entry);
644 	tcmu_cmd->se_cmd = se_cmd;
645 	tcmu_cmd->tcmu_dev = udev;
646 
647 	tcmu_cmd_set_block_cnts(tcmu_cmd);
648 	tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
649 				GFP_NOIO);
650 	if (!tcmu_cmd->dbi) {
651 		kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
652 		return NULL;
653 	}
654 
655 	return tcmu_cmd;
656 }
657 
658 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
659 {
660 	unsigned long offset = offset_in_page(vaddr);
661 	void *start = vaddr - offset;
662 
663 	size = round_up(size+offset, PAGE_SIZE);
664 
665 	while (size) {
666 		flush_dcache_page(vmalloc_to_page(start));
667 		start += PAGE_SIZE;
668 		size -= PAGE_SIZE;
669 	}
670 }
671 
672 /*
673  * Some ring helper functions. We don't assume size is a power of 2 so
674  * we can't use circ_buf.h.
675  */
676 static inline size_t spc_used(size_t head, size_t tail, size_t size)
677 {
678 	int diff = head - tail;
679 
680 	if (diff >= 0)
681 		return diff;
682 	else
683 		return size + diff;
684 }
685 
686 static inline size_t spc_free(size_t head, size_t tail, size_t size)
687 {
688 	/* Keep 1 byte unused or we can't tell full from empty */
689 	return (size - spc_used(head, tail, size) - 1);
690 }
691 
692 static inline size_t head_to_end(size_t head, size_t size)
693 {
694 	return size - head;
695 }
696 
697 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
698 
699 #define TCMU_SG_TO_DATA_AREA 1
700 #define TCMU_DATA_AREA_TO_SG 2
701 
702 static inline void tcmu_copy_data(struct tcmu_dev *udev,
703 				  struct tcmu_cmd *tcmu_cmd, uint32_t direction,
704 				  struct scatterlist *sg, unsigned int sg_nents,
705 				  struct iovec **iov, size_t data_len)
706 {
707 	/* start value of dbi + 1 must not be a valid dbi */
708 	int dbi = -2;
709 	size_t page_remaining, cp_len;
710 	int page_cnt, page_inx, dpi;
711 	struct sg_mapping_iter sg_iter;
712 	unsigned int sg_flags;
713 	struct page *page;
714 	void *data_page_start, *data_addr;
715 
716 	if (direction == TCMU_SG_TO_DATA_AREA)
717 		sg_flags = SG_MITER_ATOMIC | SG_MITER_FROM_SG;
718 	else
719 		sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
720 	sg_miter_start(&sg_iter, sg, sg_nents, sg_flags);
721 
722 	while (data_len) {
723 		if (direction == TCMU_SG_TO_DATA_AREA)
724 			dbi = new_block_to_iov(udev, tcmu_cmd, iov, dbi,
725 					       data_len);
726 		else
727 			dbi = tcmu_cmd_get_dbi(tcmu_cmd);
728 
729 		page_cnt = DIV_ROUND_UP(data_len, PAGE_SIZE);
730 		if (page_cnt > udev->data_pages_per_blk)
731 			page_cnt = udev->data_pages_per_blk;
732 
733 		dpi = dbi * udev->data_pages_per_blk;
734 		for (page_inx = 0; page_inx < page_cnt && data_len;
735 		     page_inx++, dpi++) {
736 			page = xa_load(&udev->data_pages, dpi);
737 
738 			if (direction == TCMU_DATA_AREA_TO_SG)
739 				flush_dcache_page(page);
740 			data_page_start = kmap_atomic(page);
741 			page_remaining = PAGE_SIZE;
742 
743 			while (page_remaining && data_len) {
744 				if (!sg_miter_next(&sg_iter)) {
745 					/* set length to 0 to abort outer loop */
746 					data_len = 0;
747 					pr_debug("%s: aborting data copy due to exhausted sg_list\n",
748 						 __func__);
749 					break;
750 				}
751 				cp_len = min3(sg_iter.length, page_remaining,
752 					      data_len);
753 
754 				data_addr = data_page_start +
755 					    PAGE_SIZE - page_remaining;
756 				if (direction == TCMU_SG_TO_DATA_AREA)
757 					memcpy(data_addr, sg_iter.addr, cp_len);
758 				else
759 					memcpy(sg_iter.addr, data_addr, cp_len);
760 
761 				data_len -= cp_len;
762 				page_remaining -= cp_len;
763 				sg_iter.consumed = cp_len;
764 			}
765 			sg_miter_stop(&sg_iter);
766 
767 			kunmap_atomic(data_page_start);
768 			if (direction == TCMU_SG_TO_DATA_AREA)
769 				flush_dcache_page(page);
770 		}
771 	}
772 }
773 
774 static void scatter_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
775 			      struct iovec **iov)
776 {
777 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
778 
779 	tcmu_copy_data(udev, tcmu_cmd, TCMU_SG_TO_DATA_AREA, se_cmd->t_data_sg,
780 		       se_cmd->t_data_nents, iov, se_cmd->data_length);
781 }
782 
783 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
784 			     bool bidi, uint32_t read_len)
785 {
786 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
787 	struct scatterlist *data_sg;
788 	unsigned int data_nents;
789 
790 	if (!bidi) {
791 		data_sg = se_cmd->t_data_sg;
792 		data_nents = se_cmd->t_data_nents;
793 	} else {
794 		/*
795 		 * For bidi case, the first count blocks are for Data-Out
796 		 * buffer blocks, and before gathering the Data-In buffer
797 		 * the Data-Out buffer blocks should be skipped.
798 		 */
799 		tcmu_cmd_set_dbi_cur(tcmu_cmd,
800 				     tcmu_cmd->dbi_cnt - tcmu_cmd->dbi_bidi_cnt);
801 
802 		data_sg = se_cmd->t_bidi_data_sg;
803 		data_nents = se_cmd->t_bidi_data_nents;
804 	}
805 
806 	tcmu_copy_data(udev, tcmu_cmd, TCMU_DATA_AREA_TO_SG, data_sg,
807 		       data_nents, NULL, read_len);
808 }
809 
810 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
811 {
812 	return thresh - bitmap_weight(bitmap, thresh);
813 }
814 
815 /*
816  * We can't queue a command until we have space available on the cmd ring.
817  *
818  * Called with ring lock held.
819  */
820 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size)
821 {
822 	struct tcmu_mailbox *mb = udev->mb_addr;
823 	size_t space, cmd_needed;
824 	u32 cmd_head;
825 
826 	tcmu_flush_dcache_range(mb, sizeof(*mb));
827 
828 	cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
829 
830 	/*
831 	 * If cmd end-of-ring space is too small then we need space for a NOP plus
832 	 * original cmd - cmds are internally contiguous.
833 	 */
834 	if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
835 		cmd_needed = cmd_size;
836 	else
837 		cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
838 
839 	space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
840 	if (space < cmd_needed) {
841 		pr_debug("no cmd space: %u %u %u\n", cmd_head,
842 		       udev->cmdr_last_cleaned, udev->cmdr_size);
843 		return false;
844 	}
845 	return true;
846 }
847 
848 /*
849  * We have to allocate data buffers before we can queue a command.
850  * Returns -1 on error (not enough space) or number of needed iovs on success
851  *
852  * Called with ring lock held.
853  */
854 static int tcmu_alloc_data_space(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
855 				  int *iov_bidi_cnt)
856 {
857 	int space, iov_cnt = 0, ret = 0;
858 
859 	if (!cmd->dbi_cnt)
860 		goto wr_iov_cnts;
861 
862 	/* try to check and get the data blocks as needed */
863 	space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
864 	if (space < cmd->dbi_cnt) {
865 		unsigned long blocks_left =
866 				(udev->max_blocks - udev->dbi_thresh) + space;
867 
868 		if (blocks_left < cmd->dbi_cnt) {
869 			pr_debug("no data space: only %lu available, but ask for %u\n",
870 					blocks_left * udev->data_blk_size,
871 					cmd->dbi_cnt * udev->data_blk_size);
872 			return -1;
873 		}
874 
875 		udev->dbi_thresh += cmd->dbi_cnt;
876 		if (udev->dbi_thresh > udev->max_blocks)
877 			udev->dbi_thresh = udev->max_blocks;
878 	}
879 
880 	iov_cnt = tcmu_get_empty_blocks(udev, cmd, cmd->se_cmd->data_length);
881 	if (iov_cnt < 0)
882 		return -1;
883 
884 	if (cmd->dbi_bidi_cnt) {
885 		ret = tcmu_get_empty_blocks(udev, cmd, cmd->data_len_bidi);
886 		if (ret < 0)
887 			return -1;
888 	}
889 wr_iov_cnts:
890 	*iov_bidi_cnt = ret;
891 	return iov_cnt + ret;
892 }
893 
894 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
895 {
896 	return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
897 			sizeof(struct tcmu_cmd_entry));
898 }
899 
900 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
901 					   size_t base_command_size)
902 {
903 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
904 	size_t command_size;
905 
906 	command_size = base_command_size +
907 		round_up(scsi_command_size(se_cmd->t_task_cdb),
908 				TCMU_OP_ALIGN_SIZE);
909 
910 	WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
911 
912 	return command_size;
913 }
914 
915 static void tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
916 				 struct timer_list *timer)
917 {
918 	if (!tmo)
919 		return;
920 
921 	tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
922 	if (!timer_pending(timer))
923 		mod_timer(timer, tcmu_cmd->deadline);
924 
925 	pr_debug("Timeout set up for cmd %p, dev = %s, tmo = %lu\n", tcmu_cmd,
926 		 tcmu_cmd->tcmu_dev->name, tmo / MSEC_PER_SEC);
927 }
928 
929 static int add_to_qfull_queue(struct tcmu_cmd *tcmu_cmd)
930 {
931 	struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
932 	unsigned int tmo;
933 
934 	/*
935 	 * For backwards compat if qfull_time_out is not set use
936 	 * cmd_time_out and if that's not set use the default time out.
937 	 */
938 	if (!udev->qfull_time_out)
939 		return -ETIMEDOUT;
940 	else if (udev->qfull_time_out > 0)
941 		tmo = udev->qfull_time_out;
942 	else if (udev->cmd_time_out)
943 		tmo = udev->cmd_time_out;
944 	else
945 		tmo = TCMU_TIME_OUT;
946 
947 	tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
948 
949 	list_add_tail(&tcmu_cmd->queue_entry, &udev->qfull_queue);
950 	pr_debug("adding cmd %p on dev %s to ring space wait queue\n",
951 		 tcmu_cmd, udev->name);
952 	return 0;
953 }
954 
955 static uint32_t ring_insert_padding(struct tcmu_dev *udev, size_t cmd_size)
956 {
957 	struct tcmu_cmd_entry_hdr *hdr;
958 	struct tcmu_mailbox *mb = udev->mb_addr;
959 	uint32_t cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
960 
961 	/* Insert a PAD if end-of-ring space is too small */
962 	if (head_to_end(cmd_head, udev->cmdr_size) < cmd_size) {
963 		size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
964 
965 		hdr = udev->cmdr + cmd_head;
966 		tcmu_hdr_set_op(&hdr->len_op, TCMU_OP_PAD);
967 		tcmu_hdr_set_len(&hdr->len_op, pad_size);
968 		hdr->cmd_id = 0; /* not used for PAD */
969 		hdr->kflags = 0;
970 		hdr->uflags = 0;
971 		tcmu_flush_dcache_range(hdr, sizeof(*hdr));
972 
973 		UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
974 		tcmu_flush_dcache_range(mb, sizeof(*mb));
975 
976 		cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
977 		WARN_ON(cmd_head != 0);
978 	}
979 
980 	return cmd_head;
981 }
982 
983 static void tcmu_unplug_device(struct se_dev_plug *se_plug)
984 {
985 	struct se_device *se_dev = se_plug->se_dev;
986 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
987 
988 	clear_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags);
989 	uio_event_notify(&udev->uio_info);
990 }
991 
992 static struct se_dev_plug *tcmu_plug_device(struct se_device *se_dev)
993 {
994 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
995 
996 	if (!test_and_set_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags))
997 		return &udev->se_plug;
998 
999 	return NULL;
1000 }
1001 
1002 /**
1003  * queue_cmd_ring - queue cmd to ring or internally
1004  * @tcmu_cmd: cmd to queue
1005  * @scsi_err: TCM error code if failure (-1) returned.
1006  *
1007  * Returns:
1008  * -1 we cannot queue internally or to the ring.
1009  *  0 success
1010  *  1 internally queued to wait for ring memory to free.
1011  */
1012 static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
1013 {
1014 	struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
1015 	struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
1016 	size_t base_command_size, command_size;
1017 	struct tcmu_mailbox *mb = udev->mb_addr;
1018 	struct tcmu_cmd_entry *entry;
1019 	struct iovec *iov;
1020 	int iov_cnt, iov_bidi_cnt;
1021 	uint32_t cmd_id, cmd_head;
1022 	uint64_t cdb_off;
1023 	uint32_t blk_size = udev->data_blk_size;
1024 	/* size of data buffer needed */
1025 	size_t data_length = (size_t)tcmu_cmd->dbi_cnt * blk_size;
1026 
1027 	*scsi_err = TCM_NO_SENSE;
1028 
1029 	if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
1030 		*scsi_err = TCM_LUN_BUSY;
1031 		return -1;
1032 	}
1033 
1034 	if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1035 		*scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1036 		return -1;
1037 	}
1038 
1039 	if (!list_empty(&udev->qfull_queue))
1040 		goto queue;
1041 
1042 	if (data_length > (size_t)udev->max_blocks * blk_size) {
1043 		pr_warn("TCMU: Request of size %zu is too big for %zu data area\n",
1044 			data_length, (size_t)udev->max_blocks * blk_size);
1045 		*scsi_err = TCM_INVALID_CDB_FIELD;
1046 		return -1;
1047 	}
1048 
1049 	iov_cnt = tcmu_alloc_data_space(udev, tcmu_cmd, &iov_bidi_cnt);
1050 	if (iov_cnt < 0)
1051 		goto free_and_queue;
1052 
1053 	/*
1054 	 * Must be a certain minimum size for response sense info, but
1055 	 * also may be larger if the iov array is large.
1056 	 */
1057 	base_command_size = tcmu_cmd_get_base_cmd_size(iov_cnt);
1058 	command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
1059 
1060 	if (command_size > (udev->cmdr_size / 2)) {
1061 		pr_warn("TCMU: Request of size %zu is too big for %u cmd ring\n",
1062 			command_size, udev->cmdr_size);
1063 		tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1064 		*scsi_err = TCM_INVALID_CDB_FIELD;
1065 		return -1;
1066 	}
1067 
1068 	if (!is_ring_space_avail(udev, command_size))
1069 		/*
1070 		 * Don't leave commands partially setup because the unmap
1071 		 * thread might need the blocks to make forward progress.
1072 		 */
1073 		goto free_and_queue;
1074 
1075 	if (xa_alloc(&udev->commands, &cmd_id, tcmu_cmd, XA_LIMIT(1, 0xffff),
1076 		     GFP_NOWAIT) < 0) {
1077 		pr_err("tcmu: Could not allocate cmd id.\n");
1078 
1079 		tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
1080 		*scsi_err = TCM_OUT_OF_RESOURCES;
1081 		return -1;
1082 	}
1083 	tcmu_cmd->cmd_id = cmd_id;
1084 
1085 	pr_debug("allocated cmd id %u for cmd %p dev %s\n", tcmu_cmd->cmd_id,
1086 		 tcmu_cmd, udev->name);
1087 
1088 	cmd_head = ring_insert_padding(udev, command_size);
1089 
1090 	entry = udev->cmdr + cmd_head;
1091 	memset(entry, 0, command_size);
1092 	tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
1093 
1094 	/* prepare iov list and copy data to data area if necessary */
1095 	tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1096 	iov = &entry->req.iov[0];
1097 
1098 	if (se_cmd->data_direction == DMA_TO_DEVICE ||
1099 	    se_cmd->se_cmd_flags & SCF_BIDI)
1100 		scatter_data_area(udev, tcmu_cmd, &iov);
1101 	else
1102 		tcmu_setup_iovs(udev, tcmu_cmd, &iov, se_cmd->data_length);
1103 
1104 	entry->req.iov_cnt = iov_cnt - iov_bidi_cnt;
1105 
1106 	/* Handle BIDI commands */
1107 	if (se_cmd->se_cmd_flags & SCF_BIDI) {
1108 		iov++;
1109 		tcmu_setup_iovs(udev, tcmu_cmd, &iov, tcmu_cmd->data_len_bidi);
1110 		entry->req.iov_bidi_cnt = iov_bidi_cnt;
1111 	}
1112 
1113 	tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out, &udev->cmd_timer);
1114 
1115 	entry->hdr.cmd_id = tcmu_cmd->cmd_id;
1116 
1117 	tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1118 
1119 	/* All offsets relative to mb_addr, not start of entry! */
1120 	cdb_off = CMDR_OFF + cmd_head + base_command_size;
1121 	memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1122 	entry->req.cdb_off = cdb_off;
1123 	tcmu_flush_dcache_range(entry, command_size);
1124 
1125 	UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1126 	tcmu_flush_dcache_range(mb, sizeof(*mb));
1127 
1128 	list_add_tail(&tcmu_cmd->queue_entry, &udev->inflight_queue);
1129 
1130 	if (!test_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags))
1131 		uio_event_notify(&udev->uio_info);
1132 
1133 	return 0;
1134 
1135 free_and_queue:
1136 	tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1137 	tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1138 
1139 queue:
1140 	if (add_to_qfull_queue(tcmu_cmd)) {
1141 		*scsi_err = TCM_OUT_OF_RESOURCES;
1142 		return -1;
1143 	}
1144 
1145 	return 1;
1146 }
1147 
1148 /**
1149  * queue_tmr_ring - queue tmr info to ring or internally
1150  * @udev: related tcmu_dev
1151  * @tmr: tcmu_tmr containing tmr info to queue
1152  *
1153  * Returns:
1154  *  0 success
1155  *  1 internally queued to wait for ring memory to free.
1156  */
1157 static int
1158 queue_tmr_ring(struct tcmu_dev *udev, struct tcmu_tmr *tmr)
1159 {
1160 	struct tcmu_tmr_entry *entry;
1161 	int cmd_size;
1162 	int id_list_sz;
1163 	struct tcmu_mailbox *mb = udev->mb_addr;
1164 	uint32_t cmd_head;
1165 
1166 	if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
1167 		goto out_free;
1168 
1169 	id_list_sz = sizeof(tmr->tmr_cmd_ids[0]) * tmr->tmr_cmd_cnt;
1170 	cmd_size = round_up(sizeof(*entry) + id_list_sz, TCMU_OP_ALIGN_SIZE);
1171 
1172 	if (!list_empty(&udev->tmr_queue) ||
1173 	    !is_ring_space_avail(udev, cmd_size)) {
1174 		list_add_tail(&tmr->queue_entry, &udev->tmr_queue);
1175 		pr_debug("adding tmr %p on dev %s to TMR ring space wait queue\n",
1176 			 tmr, udev->name);
1177 		return 1;
1178 	}
1179 
1180 	cmd_head = ring_insert_padding(udev, cmd_size);
1181 
1182 	entry = udev->cmdr + cmd_head;
1183 	memset(entry, 0, cmd_size);
1184 	tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_TMR);
1185 	tcmu_hdr_set_len(&entry->hdr.len_op, cmd_size);
1186 	entry->tmr_type = tmr->tmr_type;
1187 	entry->cmd_cnt = tmr->tmr_cmd_cnt;
1188 	memcpy(&entry->cmd_ids[0], &tmr->tmr_cmd_ids[0], id_list_sz);
1189 	tcmu_flush_dcache_range(entry, cmd_size);
1190 
1191 	UPDATE_HEAD(mb->cmd_head, cmd_size, udev->cmdr_size);
1192 	tcmu_flush_dcache_range(mb, sizeof(*mb));
1193 
1194 	uio_event_notify(&udev->uio_info);
1195 
1196 out_free:
1197 	kfree(tmr);
1198 
1199 	return 0;
1200 }
1201 
1202 static sense_reason_t
1203 tcmu_queue_cmd(struct se_cmd *se_cmd)
1204 {
1205 	struct se_device *se_dev = se_cmd->se_dev;
1206 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
1207 	struct tcmu_cmd *tcmu_cmd;
1208 	sense_reason_t scsi_ret = TCM_CHECK_CONDITION_ABORT_CMD;
1209 	int ret = -1;
1210 
1211 	tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1212 	if (!tcmu_cmd)
1213 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1214 
1215 	mutex_lock(&udev->cmdr_lock);
1216 	if (!(se_cmd->transport_state & CMD_T_ABORTED))
1217 		ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1218 	if (ret < 0)
1219 		tcmu_free_cmd(tcmu_cmd);
1220 	else
1221 		se_cmd->priv = tcmu_cmd;
1222 	mutex_unlock(&udev->cmdr_lock);
1223 	return scsi_ret;
1224 }
1225 
1226 static void tcmu_set_next_deadline(struct list_head *queue,
1227 				   struct timer_list *timer)
1228 {
1229 	struct tcmu_cmd *cmd;
1230 
1231 	if (!list_empty(queue)) {
1232 		cmd = list_first_entry(queue, struct tcmu_cmd, queue_entry);
1233 		mod_timer(timer, cmd->deadline);
1234 	} else
1235 		del_timer(timer);
1236 }
1237 
1238 static int
1239 tcmu_tmr_type(enum tcm_tmreq_table tmf)
1240 {
1241 	switch (tmf) {
1242 	case TMR_ABORT_TASK:		return TCMU_TMR_ABORT_TASK;
1243 	case TMR_ABORT_TASK_SET:	return TCMU_TMR_ABORT_TASK_SET;
1244 	case TMR_CLEAR_ACA:		return TCMU_TMR_CLEAR_ACA;
1245 	case TMR_CLEAR_TASK_SET:	return TCMU_TMR_CLEAR_TASK_SET;
1246 	case TMR_LUN_RESET:		return TCMU_TMR_LUN_RESET;
1247 	case TMR_TARGET_WARM_RESET:	return TCMU_TMR_TARGET_WARM_RESET;
1248 	case TMR_TARGET_COLD_RESET:	return TCMU_TMR_TARGET_COLD_RESET;
1249 	case TMR_LUN_RESET_PRO:		return TCMU_TMR_LUN_RESET_PRO;
1250 	default:			return TCMU_TMR_UNKNOWN;
1251 	}
1252 }
1253 
1254 static void
1255 tcmu_tmr_notify(struct se_device *se_dev, enum tcm_tmreq_table tmf,
1256 		struct list_head *cmd_list)
1257 {
1258 	int i = 0, cmd_cnt = 0;
1259 	bool unqueued = false;
1260 	struct tcmu_cmd *cmd;
1261 	struct se_cmd *se_cmd;
1262 	struct tcmu_tmr *tmr;
1263 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
1264 
1265 	mutex_lock(&udev->cmdr_lock);
1266 
1267 	/* First we check for aborted commands in qfull_queue */
1268 	list_for_each_entry(se_cmd, cmd_list, state_list) {
1269 		i++;
1270 		if (!se_cmd->priv)
1271 			continue;
1272 		cmd = se_cmd->priv;
1273 		/* Commands on qfull queue have no id yet */
1274 		if (cmd->cmd_id) {
1275 			cmd_cnt++;
1276 			continue;
1277 		}
1278 		pr_debug("Removing aborted command %p from queue on dev %s.\n",
1279 			 cmd, udev->name);
1280 
1281 		list_del_init(&cmd->queue_entry);
1282 		tcmu_free_cmd(cmd);
1283 		se_cmd->priv = NULL;
1284 		target_complete_cmd(se_cmd, SAM_STAT_TASK_ABORTED);
1285 		unqueued = true;
1286 	}
1287 	if (unqueued)
1288 		tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1289 
1290 	if (!test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags))
1291 		goto unlock;
1292 
1293 	pr_debug("TMR event %d on dev %s, aborted cmds %d, afflicted cmd_ids %d\n",
1294 		 tcmu_tmr_type(tmf), udev->name, i, cmd_cnt);
1295 
1296 	tmr = kmalloc(struct_size(tmr, tmr_cmd_ids, cmd_cnt), GFP_NOIO);
1297 	if (!tmr)
1298 		goto unlock;
1299 
1300 	tmr->tmr_type = tcmu_tmr_type(tmf);
1301 	tmr->tmr_cmd_cnt = cmd_cnt;
1302 
1303 	if (cmd_cnt != 0) {
1304 		cmd_cnt = 0;
1305 		list_for_each_entry(se_cmd, cmd_list, state_list) {
1306 			if (!se_cmd->priv)
1307 				continue;
1308 			cmd = se_cmd->priv;
1309 			if (cmd->cmd_id)
1310 				tmr->tmr_cmd_ids[cmd_cnt++] = cmd->cmd_id;
1311 		}
1312 	}
1313 
1314 	queue_tmr_ring(udev, tmr);
1315 
1316 unlock:
1317 	mutex_unlock(&udev->cmdr_lock);
1318 }
1319 
1320 static bool tcmu_handle_completion(struct tcmu_cmd *cmd,
1321 				   struct tcmu_cmd_entry *entry, bool keep_buf)
1322 {
1323 	struct se_cmd *se_cmd = cmd->se_cmd;
1324 	struct tcmu_dev *udev = cmd->tcmu_dev;
1325 	bool read_len_valid = false;
1326 	bool ret = true;
1327 	uint32_t read_len;
1328 
1329 	/*
1330 	 * cmd has been completed already from timeout, just reclaim
1331 	 * data area space and free cmd
1332 	 */
1333 	if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1334 		WARN_ON_ONCE(se_cmd);
1335 		goto out;
1336 	}
1337 	if (test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
1338 		pr_err("cmd_id %u already completed with KEEP_BUF, ring is broken\n",
1339 		       entry->hdr.cmd_id);
1340 		set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1341 		ret = false;
1342 		goto out;
1343 	}
1344 
1345 	list_del_init(&cmd->queue_entry);
1346 
1347 	tcmu_cmd_reset_dbi_cur(cmd);
1348 
1349 	if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1350 		pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1351 			cmd->se_cmd);
1352 		entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1353 		goto done;
1354 	}
1355 
1356 	read_len = se_cmd->data_length;
1357 	if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1358 	    (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1359 		read_len_valid = true;
1360 		if (entry->rsp.read_len < read_len)
1361 			read_len = entry->rsp.read_len;
1362 	}
1363 
1364 	if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1365 		transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1366 		if (!read_len_valid )
1367 			goto done;
1368 		else
1369 			se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1370 	}
1371 	if (se_cmd->se_cmd_flags & SCF_BIDI) {
1372 		/* Get Data-In buffer before clean up */
1373 		gather_data_area(udev, cmd, true, read_len);
1374 	} else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1375 		gather_data_area(udev, cmd, false, read_len);
1376 	} else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1377 		/* TODO: */
1378 	} else if (se_cmd->data_direction != DMA_NONE) {
1379 		pr_warn("TCMU: data direction was %d!\n",
1380 			se_cmd->data_direction);
1381 	}
1382 
1383 done:
1384 	se_cmd->priv = NULL;
1385 	if (read_len_valid) {
1386 		pr_debug("read_len = %d\n", read_len);
1387 		target_complete_cmd_with_length(cmd->se_cmd,
1388 					entry->rsp.scsi_status, read_len);
1389 	} else
1390 		target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1391 
1392 out:
1393 	if (!keep_buf) {
1394 		tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1395 		tcmu_free_cmd(cmd);
1396 	} else {
1397 		/*
1398 		 * Keep this command after completion, since userspace still
1399 		 * needs the data buffer. Mark it with TCMU_CMD_BIT_KEEP_BUF
1400 		 * and reset potential TCMU_CMD_BIT_EXPIRED, so we don't accept
1401 		 * a second completion later.
1402 		 * Userspace can free the buffer later by writing the cmd_id
1403 		 * to new action attribute free_kept_buf.
1404 		 */
1405 		clear_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1406 		set_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags);
1407 	}
1408 	return ret;
1409 }
1410 
1411 static int tcmu_run_tmr_queue(struct tcmu_dev *udev)
1412 {
1413 	struct tcmu_tmr *tmr, *tmp;
1414 	LIST_HEAD(tmrs);
1415 
1416 	if (list_empty(&udev->tmr_queue))
1417 		return 1;
1418 
1419 	pr_debug("running %s's tmr queue\n", udev->name);
1420 
1421 	list_splice_init(&udev->tmr_queue, &tmrs);
1422 
1423 	list_for_each_entry_safe(tmr, tmp, &tmrs, queue_entry) {
1424 		list_del_init(&tmr->queue_entry);
1425 
1426 		pr_debug("removing tmr %p on dev %s from queue\n",
1427 			 tmr, udev->name);
1428 
1429 		if (queue_tmr_ring(udev, tmr)) {
1430 			pr_debug("ran out of space during tmr queue run\n");
1431 			/*
1432 			 * tmr was requeued, so just put all tmrs back in
1433 			 * the queue
1434 			 */
1435 			list_splice_tail(&tmrs, &udev->tmr_queue);
1436 			return 0;
1437 		}
1438 	}
1439 
1440 	return 1;
1441 }
1442 
1443 static bool tcmu_handle_completions(struct tcmu_dev *udev)
1444 {
1445 	struct tcmu_mailbox *mb;
1446 	struct tcmu_cmd *cmd;
1447 	bool free_space = false;
1448 
1449 	if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1450 		pr_err("ring broken, not handling completions\n");
1451 		return false;
1452 	}
1453 
1454 	mb = udev->mb_addr;
1455 	tcmu_flush_dcache_range(mb, sizeof(*mb));
1456 
1457 	while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1458 
1459 		struct tcmu_cmd_entry *entry = udev->cmdr + udev->cmdr_last_cleaned;
1460 		bool keep_buf;
1461 
1462 		/*
1463 		 * Flush max. up to end of cmd ring since current entry might
1464 		 * be a padding that is shorter than sizeof(*entry)
1465 		 */
1466 		size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1467 					       udev->cmdr_size);
1468 		tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1469 					ring_left : sizeof(*entry));
1470 
1471 		free_space = true;
1472 
1473 		if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD ||
1474 		    tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_TMR) {
1475 			UPDATE_HEAD(udev->cmdr_last_cleaned,
1476 				    tcmu_hdr_get_len(entry->hdr.len_op),
1477 				    udev->cmdr_size);
1478 			continue;
1479 		}
1480 		WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1481 
1482 		keep_buf = !!(entry->hdr.uflags & TCMU_UFLAG_KEEP_BUF);
1483 		if (keep_buf)
1484 			cmd = xa_load(&udev->commands, entry->hdr.cmd_id);
1485 		else
1486 			cmd = xa_erase(&udev->commands, entry->hdr.cmd_id);
1487 		if (!cmd) {
1488 			pr_err("cmd_id %u not found, ring is broken\n",
1489 			       entry->hdr.cmd_id);
1490 			set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1491 			return false;
1492 		}
1493 
1494 		if (!tcmu_handle_completion(cmd, entry, keep_buf))
1495 			break;
1496 
1497 		UPDATE_HEAD(udev->cmdr_last_cleaned,
1498 			    tcmu_hdr_get_len(entry->hdr.len_op),
1499 			    udev->cmdr_size);
1500 	}
1501 	if (free_space)
1502 		free_space = tcmu_run_tmr_queue(udev);
1503 
1504 	if (atomic_read(&global_page_count) > tcmu_global_max_pages &&
1505 	    xa_empty(&udev->commands) && list_empty(&udev->qfull_queue)) {
1506 		/*
1507 		 * Allocated blocks exceeded global block limit, currently no
1508 		 * more pending or waiting commands so try to reclaim blocks.
1509 		 */
1510 		schedule_delayed_work(&tcmu_unmap_work, 0);
1511 	}
1512 	if (udev->cmd_time_out)
1513 		tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
1514 
1515 	return free_space;
1516 }
1517 
1518 static void tcmu_check_expired_ring_cmd(struct tcmu_cmd *cmd)
1519 {
1520 	struct se_cmd *se_cmd;
1521 
1522 	if (!time_after_eq(jiffies, cmd->deadline))
1523 		return;
1524 
1525 	set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1526 	list_del_init(&cmd->queue_entry);
1527 	se_cmd = cmd->se_cmd;
1528 	se_cmd->priv = NULL;
1529 	cmd->se_cmd = NULL;
1530 
1531 	pr_debug("Timing out inflight cmd %u on dev %s.\n",
1532 		 cmd->cmd_id, cmd->tcmu_dev->name);
1533 
1534 	target_complete_cmd(se_cmd, SAM_STAT_CHECK_CONDITION);
1535 }
1536 
1537 static void tcmu_check_expired_queue_cmd(struct tcmu_cmd *cmd)
1538 {
1539 	struct se_cmd *se_cmd;
1540 
1541 	if (!time_after_eq(jiffies, cmd->deadline))
1542 		return;
1543 
1544 	pr_debug("Timing out queued cmd %p on dev %s.\n",
1545 		  cmd, cmd->tcmu_dev->name);
1546 
1547 	list_del_init(&cmd->queue_entry);
1548 	se_cmd = cmd->se_cmd;
1549 	tcmu_free_cmd(cmd);
1550 
1551 	se_cmd->priv = NULL;
1552 	target_complete_cmd(se_cmd, SAM_STAT_TASK_SET_FULL);
1553 }
1554 
1555 static void tcmu_device_timedout(struct tcmu_dev *udev)
1556 {
1557 	spin_lock(&timed_out_udevs_lock);
1558 	if (list_empty(&udev->timedout_entry))
1559 		list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1560 	spin_unlock(&timed_out_udevs_lock);
1561 
1562 	schedule_delayed_work(&tcmu_unmap_work, 0);
1563 }
1564 
1565 static void tcmu_cmd_timedout(struct timer_list *t)
1566 {
1567 	struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1568 
1569 	pr_debug("%s cmd timeout has expired\n", udev->name);
1570 	tcmu_device_timedout(udev);
1571 }
1572 
1573 static void tcmu_qfull_timedout(struct timer_list *t)
1574 {
1575 	struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1576 
1577 	pr_debug("%s qfull timeout has expired\n", udev->name);
1578 	tcmu_device_timedout(udev);
1579 }
1580 
1581 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1582 {
1583 	struct tcmu_hba *tcmu_hba;
1584 
1585 	tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1586 	if (!tcmu_hba)
1587 		return -ENOMEM;
1588 
1589 	tcmu_hba->host_id = host_id;
1590 	hba->hba_ptr = tcmu_hba;
1591 
1592 	return 0;
1593 }
1594 
1595 static void tcmu_detach_hba(struct se_hba *hba)
1596 {
1597 	kfree(hba->hba_ptr);
1598 	hba->hba_ptr = NULL;
1599 }
1600 
1601 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1602 {
1603 	struct tcmu_dev *udev;
1604 
1605 	udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1606 	if (!udev)
1607 		return NULL;
1608 	kref_init(&udev->kref);
1609 
1610 	udev->name = kstrdup(name, GFP_KERNEL);
1611 	if (!udev->name) {
1612 		kfree(udev);
1613 		return NULL;
1614 	}
1615 
1616 	udev->hba = hba;
1617 	udev->cmd_time_out = TCMU_TIME_OUT;
1618 	udev->qfull_time_out = -1;
1619 
1620 	udev->data_pages_per_blk = DATA_PAGES_PER_BLK_DEF;
1621 	udev->max_blocks = DATA_AREA_PAGES_DEF / udev->data_pages_per_blk;
1622 	udev->cmdr_size = CMDR_SIZE_DEF;
1623 	udev->data_area_mb = TCMU_PAGES_TO_MBS(DATA_AREA_PAGES_DEF);
1624 
1625 	mutex_init(&udev->cmdr_lock);
1626 
1627 	INIT_LIST_HEAD(&udev->node);
1628 	INIT_LIST_HEAD(&udev->timedout_entry);
1629 	INIT_LIST_HEAD(&udev->qfull_queue);
1630 	INIT_LIST_HEAD(&udev->tmr_queue);
1631 	INIT_LIST_HEAD(&udev->inflight_queue);
1632 	xa_init_flags(&udev->commands, XA_FLAGS_ALLOC1);
1633 
1634 	timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1635 	timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1636 
1637 	xa_init(&udev->data_pages);
1638 
1639 	return &udev->se_dev;
1640 }
1641 
1642 static void tcmu_dev_call_rcu(struct rcu_head *p)
1643 {
1644 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
1645 	struct tcmu_dev *udev = TCMU_DEV(dev);
1646 
1647 	kfree(udev->uio_info.name);
1648 	kfree(udev->name);
1649 	kfree(udev);
1650 }
1651 
1652 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1653 {
1654 	if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags) ||
1655 	    test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
1656 		kmem_cache_free(tcmu_cmd_cache, cmd);
1657 		return 0;
1658 	}
1659 	return -EINVAL;
1660 }
1661 
1662 static u32 tcmu_blocks_release(struct tcmu_dev *udev, unsigned long first,
1663 				unsigned long last)
1664 {
1665 	struct page *page;
1666 	unsigned long dpi;
1667 	u32 pages_freed = 0;
1668 
1669 	first = first * udev->data_pages_per_blk;
1670 	last = (last + 1) * udev->data_pages_per_blk - 1;
1671 	xa_for_each_range(&udev->data_pages, dpi, page, first, last) {
1672 		xa_erase(&udev->data_pages, dpi);
1673 		/*
1674 		 * While reaching here there may be page faults occurring on
1675 		 * the to-be-released pages. A race condition may occur if
1676 		 * unmap_mapping_range() is called before page faults on these
1677 		 * pages have completed; a valid but stale map is created.
1678 		 *
1679 		 * If another command subsequently runs and needs to extend
1680 		 * dbi_thresh, it may reuse the slot corresponding to the
1681 		 * previous page in data_bitmap. Though we will allocate a new
1682 		 * page for the slot in data_area, no page fault will happen
1683 		 * because we have a valid map. Therefore the command's data
1684 		 * will be lost.
1685 		 *
1686 		 * We lock and unlock pages that are to be released to ensure
1687 		 * all page faults have completed. This way
1688 		 * unmap_mapping_range() can ensure stale maps are cleanly
1689 		 * removed.
1690 		 */
1691 		lock_page(page);
1692 		unlock_page(page);
1693 		__free_page(page);
1694 		pages_freed++;
1695 	}
1696 
1697 	atomic_sub(pages_freed, &global_page_count);
1698 
1699 	return pages_freed;
1700 }
1701 
1702 static void tcmu_remove_all_queued_tmr(struct tcmu_dev *udev)
1703 {
1704 	struct tcmu_tmr *tmr, *tmp;
1705 
1706 	list_for_each_entry_safe(tmr, tmp, &udev->tmr_queue, queue_entry) {
1707 		list_del_init(&tmr->queue_entry);
1708 		kfree(tmr);
1709 	}
1710 }
1711 
1712 static void tcmu_dev_kref_release(struct kref *kref)
1713 {
1714 	struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1715 	struct se_device *dev = &udev->se_dev;
1716 	struct tcmu_cmd *cmd;
1717 	bool all_expired = true;
1718 	unsigned long i;
1719 
1720 	vfree(udev->mb_addr);
1721 	udev->mb_addr = NULL;
1722 
1723 	spin_lock_bh(&timed_out_udevs_lock);
1724 	if (!list_empty(&udev->timedout_entry))
1725 		list_del(&udev->timedout_entry);
1726 	spin_unlock_bh(&timed_out_udevs_lock);
1727 
1728 	/* Upper layer should drain all requests before calling this */
1729 	mutex_lock(&udev->cmdr_lock);
1730 	xa_for_each(&udev->commands, i, cmd) {
1731 		if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1732 			all_expired = false;
1733 	}
1734 	/* There can be left over TMR cmds. Remove them. */
1735 	tcmu_remove_all_queued_tmr(udev);
1736 	if (!list_empty(&udev->qfull_queue))
1737 		all_expired = false;
1738 	xa_destroy(&udev->commands);
1739 	WARN_ON(!all_expired);
1740 
1741 	tcmu_blocks_release(udev, 0, udev->dbi_max);
1742 	bitmap_free(udev->data_bitmap);
1743 	mutex_unlock(&udev->cmdr_lock);
1744 
1745 	pr_debug("dev_kref_release\n");
1746 
1747 	call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1748 }
1749 
1750 static void run_qfull_queue(struct tcmu_dev *udev, bool fail)
1751 {
1752 	struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1753 	LIST_HEAD(cmds);
1754 	sense_reason_t scsi_ret;
1755 	int ret;
1756 
1757 	if (list_empty(&udev->qfull_queue))
1758 		return;
1759 
1760 	pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1761 
1762 	list_splice_init(&udev->qfull_queue, &cmds);
1763 
1764 	list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, queue_entry) {
1765 		list_del_init(&tcmu_cmd->queue_entry);
1766 
1767 		pr_debug("removing cmd %p on dev %s from queue\n",
1768 			 tcmu_cmd, udev->name);
1769 
1770 		if (fail) {
1771 			/*
1772 			 * We were not able to even start the command, so
1773 			 * fail with busy to allow a retry in case runner
1774 			 * was only temporarily down. If the device is being
1775 			 * removed then LIO core will do the right thing and
1776 			 * fail the retry.
1777 			 */
1778 			tcmu_cmd->se_cmd->priv = NULL;
1779 			target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1780 			tcmu_free_cmd(tcmu_cmd);
1781 			continue;
1782 		}
1783 
1784 		ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1785 		if (ret < 0) {
1786 			pr_debug("cmd %p on dev %s failed with %u\n",
1787 				 tcmu_cmd, udev->name, scsi_ret);
1788 			/*
1789 			 * Ignore scsi_ret for now. target_complete_cmd
1790 			 * drops it.
1791 			 */
1792 			tcmu_cmd->se_cmd->priv = NULL;
1793 			target_complete_cmd(tcmu_cmd->se_cmd,
1794 					    SAM_STAT_CHECK_CONDITION);
1795 			tcmu_free_cmd(tcmu_cmd);
1796 		} else if (ret > 0) {
1797 			pr_debug("ran out of space during cmdr queue run\n");
1798 			/*
1799 			 * cmd was requeued, so just put all cmds back in
1800 			 * the queue
1801 			 */
1802 			list_splice_tail(&cmds, &udev->qfull_queue);
1803 			break;
1804 		}
1805 	}
1806 
1807 	tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1808 }
1809 
1810 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1811 {
1812 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1813 
1814 	mutex_lock(&udev->cmdr_lock);
1815 	if (tcmu_handle_completions(udev))
1816 		run_qfull_queue(udev, false);
1817 	mutex_unlock(&udev->cmdr_lock);
1818 
1819 	return 0;
1820 }
1821 
1822 /*
1823  * mmap code from uio.c. Copied here because we want to hook mmap()
1824  * and this stuff must come along.
1825  */
1826 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1827 {
1828 	struct tcmu_dev *udev = vma->vm_private_data;
1829 	struct uio_info *info = &udev->uio_info;
1830 
1831 	if (vma->vm_pgoff < MAX_UIO_MAPS) {
1832 		if (info->mem[vma->vm_pgoff].size == 0)
1833 			return -1;
1834 		return (int)vma->vm_pgoff;
1835 	}
1836 	return -1;
1837 }
1838 
1839 static struct page *tcmu_try_get_data_page(struct tcmu_dev *udev, uint32_t dpi)
1840 {
1841 	struct page *page;
1842 
1843 	mutex_lock(&udev->cmdr_lock);
1844 	page = xa_load(&udev->data_pages, dpi);
1845 	if (likely(page)) {
1846 		get_page(page);
1847 		lock_page(page);
1848 		mutex_unlock(&udev->cmdr_lock);
1849 		return page;
1850 	}
1851 
1852 	/*
1853 	 * Userspace messed up and passed in a address not in the
1854 	 * data iov passed to it.
1855 	 */
1856 	pr_err("Invalid addr to data page mapping (dpi %u) on device %s\n",
1857 	       dpi, udev->name);
1858 	mutex_unlock(&udev->cmdr_lock);
1859 
1860 	return NULL;
1861 }
1862 
1863 static void tcmu_vma_open(struct vm_area_struct *vma)
1864 {
1865 	struct tcmu_dev *udev = vma->vm_private_data;
1866 
1867 	pr_debug("vma_open\n");
1868 
1869 	kref_get(&udev->kref);
1870 }
1871 
1872 static void tcmu_vma_close(struct vm_area_struct *vma)
1873 {
1874 	struct tcmu_dev *udev = vma->vm_private_data;
1875 
1876 	pr_debug("vma_close\n");
1877 
1878 	/* release ref from tcmu_vma_open */
1879 	kref_put(&udev->kref, tcmu_dev_kref_release);
1880 }
1881 
1882 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1883 {
1884 	struct tcmu_dev *udev = vmf->vma->vm_private_data;
1885 	struct uio_info *info = &udev->uio_info;
1886 	struct page *page;
1887 	unsigned long offset;
1888 	void *addr;
1889 	vm_fault_t ret = 0;
1890 
1891 	int mi = tcmu_find_mem_index(vmf->vma);
1892 	if (mi < 0)
1893 		return VM_FAULT_SIGBUS;
1894 
1895 	/*
1896 	 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1897 	 * to use mem[N].
1898 	 */
1899 	offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1900 
1901 	if (offset < udev->data_off) {
1902 		/* For the vmalloc()ed cmd area pages */
1903 		addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1904 		page = vmalloc_to_page(addr);
1905 		get_page(page);
1906 	} else {
1907 		uint32_t dpi;
1908 
1909 		/* For the dynamically growing data area pages */
1910 		dpi = (offset - udev->data_off) / PAGE_SIZE;
1911 		page = tcmu_try_get_data_page(udev, dpi);
1912 		if (!page)
1913 			return VM_FAULT_SIGBUS;
1914 		ret = VM_FAULT_LOCKED;
1915 	}
1916 
1917 	vmf->page = page;
1918 	return ret;
1919 }
1920 
1921 static const struct vm_operations_struct tcmu_vm_ops = {
1922 	.open = tcmu_vma_open,
1923 	.close = tcmu_vma_close,
1924 	.fault = tcmu_vma_fault,
1925 };
1926 
1927 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1928 {
1929 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1930 
1931 	vm_flags_set(vma, VM_DONTEXPAND | VM_DONTDUMP);
1932 	vma->vm_ops = &tcmu_vm_ops;
1933 
1934 	vma->vm_private_data = udev;
1935 
1936 	/* Ensure the mmap is exactly the right size */
1937 	if (vma_pages(vma) != udev->mmap_pages)
1938 		return -EINVAL;
1939 
1940 	tcmu_vma_open(vma);
1941 
1942 	return 0;
1943 }
1944 
1945 static int tcmu_open(struct uio_info *info, struct inode *inode)
1946 {
1947 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1948 
1949 	/* O_EXCL not supported for char devs, so fake it? */
1950 	if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1951 		return -EBUSY;
1952 
1953 	udev->inode = inode;
1954 
1955 	pr_debug("open\n");
1956 
1957 	return 0;
1958 }
1959 
1960 static int tcmu_release(struct uio_info *info, struct inode *inode)
1961 {
1962 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1963 	struct tcmu_cmd *cmd;
1964 	unsigned long i;
1965 	bool freed = false;
1966 
1967 	mutex_lock(&udev->cmdr_lock);
1968 
1969 	xa_for_each(&udev->commands, i, cmd) {
1970 		/* Cmds with KEEP_BUF set are no longer on the ring, but
1971 		 * userspace still holds the data buffer. If userspace closes
1972 		 * we implicitly free these cmds and buffers, since after new
1973 		 * open the (new ?) userspace cannot find the cmd in the ring
1974 		 * and thus never will release the buffer by writing cmd_id to
1975 		 * free_kept_buf action attribute.
1976 		 */
1977 		if (!test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags))
1978 			continue;
1979 		pr_debug("removing KEEP_BUF cmd %u on dev %s from ring\n",
1980 			 cmd->cmd_id, udev->name);
1981 		freed = true;
1982 
1983 		xa_erase(&udev->commands, i);
1984 		tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1985 		tcmu_free_cmd(cmd);
1986 	}
1987 	/*
1988 	 * We only freed data space, not ring space. Therefore we dont call
1989 	 * run_tmr_queue, but call run_qfull_queue if tmr_list is empty.
1990 	 */
1991 	if (freed && list_empty(&udev->tmr_queue))
1992 		run_qfull_queue(udev, false);
1993 
1994 	mutex_unlock(&udev->cmdr_lock);
1995 
1996 	clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1997 
1998 	pr_debug("close\n");
1999 
2000 	return 0;
2001 }
2002 
2003 static int tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
2004 {
2005 	struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
2006 
2007 	if (!tcmu_kern_cmd_reply_supported)
2008 		return 0;
2009 
2010 	if (udev->nl_reply_supported <= 0)
2011 		return 0;
2012 
2013 	mutex_lock(&tcmu_nl_cmd_mutex);
2014 
2015 	if (tcmu_netlink_blocked) {
2016 		mutex_unlock(&tcmu_nl_cmd_mutex);
2017 		pr_warn("Failing nl cmd %d on %s. Interface is blocked.\n", cmd,
2018 			udev->name);
2019 		return -EAGAIN;
2020 	}
2021 
2022 	if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
2023 		mutex_unlock(&tcmu_nl_cmd_mutex);
2024 		pr_warn("netlink cmd %d already executing on %s\n",
2025 			 nl_cmd->cmd, udev->name);
2026 		return -EBUSY;
2027 	}
2028 
2029 	memset(nl_cmd, 0, sizeof(*nl_cmd));
2030 	nl_cmd->cmd = cmd;
2031 	nl_cmd->udev = udev;
2032 	init_completion(&nl_cmd->complete);
2033 	INIT_LIST_HEAD(&nl_cmd->nl_list);
2034 
2035 	list_add_tail(&nl_cmd->nl_list, &tcmu_nl_cmd_list);
2036 
2037 	mutex_unlock(&tcmu_nl_cmd_mutex);
2038 	return 0;
2039 }
2040 
2041 static void tcmu_destroy_genl_cmd_reply(struct tcmu_dev *udev)
2042 {
2043 	struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
2044 
2045 	if (!tcmu_kern_cmd_reply_supported)
2046 		return;
2047 
2048 	if (udev->nl_reply_supported <= 0)
2049 		return;
2050 
2051 	mutex_lock(&tcmu_nl_cmd_mutex);
2052 
2053 	list_del(&nl_cmd->nl_list);
2054 	memset(nl_cmd, 0, sizeof(*nl_cmd));
2055 
2056 	mutex_unlock(&tcmu_nl_cmd_mutex);
2057 }
2058 
2059 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
2060 {
2061 	struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
2062 	int ret;
2063 
2064 	if (!tcmu_kern_cmd_reply_supported)
2065 		return 0;
2066 
2067 	if (udev->nl_reply_supported <= 0)
2068 		return 0;
2069 
2070 	pr_debug("sleeping for nl reply\n");
2071 	wait_for_completion(&nl_cmd->complete);
2072 
2073 	mutex_lock(&tcmu_nl_cmd_mutex);
2074 	nl_cmd->cmd = TCMU_CMD_UNSPEC;
2075 	ret = nl_cmd->status;
2076 	mutex_unlock(&tcmu_nl_cmd_mutex);
2077 
2078 	return ret;
2079 }
2080 
2081 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
2082 				   enum tcmu_genl_cmd cmd,
2083 				   struct sk_buff **buf, void **hdr)
2084 {
2085 	struct sk_buff *skb;
2086 	void *msg_header;
2087 	int ret = -ENOMEM;
2088 
2089 	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
2090 	if (!skb)
2091 		return ret;
2092 
2093 	msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
2094 	if (!msg_header)
2095 		goto free_skb;
2096 
2097 	ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
2098 	if (ret < 0)
2099 		goto free_skb;
2100 
2101 	ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
2102 	if (ret < 0)
2103 		goto free_skb;
2104 
2105 	ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
2106 	if (ret < 0)
2107 		goto free_skb;
2108 
2109 	*buf = skb;
2110 	*hdr = msg_header;
2111 	return ret;
2112 
2113 free_skb:
2114 	nlmsg_free(skb);
2115 	return ret;
2116 }
2117 
2118 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
2119 				   enum tcmu_genl_cmd cmd,
2120 				   struct sk_buff *skb, void *msg_header)
2121 {
2122 	int ret;
2123 
2124 	genlmsg_end(skb, msg_header);
2125 
2126 	ret = tcmu_init_genl_cmd_reply(udev, cmd);
2127 	if (ret) {
2128 		nlmsg_free(skb);
2129 		return ret;
2130 	}
2131 
2132 	ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
2133 				      TCMU_MCGRP_CONFIG, GFP_KERNEL);
2134 
2135 	/* Wait during an add as the listener may not be up yet */
2136 	if (ret == 0 ||
2137 	   (ret == -ESRCH && cmd == TCMU_CMD_ADDED_DEVICE))
2138 		return tcmu_wait_genl_cmd_reply(udev);
2139 	else
2140 		tcmu_destroy_genl_cmd_reply(udev);
2141 
2142 	return ret;
2143 }
2144 
2145 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
2146 {
2147 	struct sk_buff *skb = NULL;
2148 	void *msg_header = NULL;
2149 	int ret = 0;
2150 
2151 	ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
2152 				      &msg_header);
2153 	if (ret < 0)
2154 		return ret;
2155 	return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, skb,
2156 				       msg_header);
2157 }
2158 
2159 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
2160 {
2161 	struct sk_buff *skb = NULL;
2162 	void *msg_header = NULL;
2163 	int ret = 0;
2164 
2165 	ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
2166 				      &skb, &msg_header);
2167 	if (ret < 0)
2168 		return ret;
2169 	return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
2170 				       skb, msg_header);
2171 }
2172 
2173 static int tcmu_update_uio_info(struct tcmu_dev *udev)
2174 {
2175 	struct tcmu_hba *hba = udev->hba->hba_ptr;
2176 	struct uio_info *info;
2177 	char *str;
2178 
2179 	info = &udev->uio_info;
2180 
2181 	if (udev->dev_config[0])
2182 		str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s/%s", hba->host_id,
2183 				udev->name, udev->dev_config);
2184 	else
2185 		str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s", hba->host_id,
2186 				udev->name);
2187 	if (!str)
2188 		return -ENOMEM;
2189 
2190 	/* If the old string exists, free it */
2191 	kfree(info->name);
2192 	info->name = str;
2193 
2194 	return 0;
2195 }
2196 
2197 static int tcmu_configure_device(struct se_device *dev)
2198 {
2199 	struct tcmu_dev *udev = TCMU_DEV(dev);
2200 	struct uio_info *info;
2201 	struct tcmu_mailbox *mb;
2202 	size_t data_size;
2203 	int ret = 0;
2204 
2205 	ret = tcmu_update_uio_info(udev);
2206 	if (ret)
2207 		return ret;
2208 
2209 	info = &udev->uio_info;
2210 
2211 	mutex_lock(&udev->cmdr_lock);
2212 	udev->data_bitmap = bitmap_zalloc(udev->max_blocks, GFP_KERNEL);
2213 	mutex_unlock(&udev->cmdr_lock);
2214 	if (!udev->data_bitmap) {
2215 		ret = -ENOMEM;
2216 		goto err_bitmap_alloc;
2217 	}
2218 
2219 	mb = vzalloc(udev->cmdr_size + CMDR_OFF);
2220 	if (!mb) {
2221 		ret = -ENOMEM;
2222 		goto err_vzalloc;
2223 	}
2224 
2225 	/* mailbox fits in first part of CMDR space */
2226 	udev->mb_addr = mb;
2227 	udev->cmdr = (void *)mb + CMDR_OFF;
2228 	udev->data_off = udev->cmdr_size + CMDR_OFF;
2229 	data_size = TCMU_MBS_TO_PAGES(udev->data_area_mb) << PAGE_SHIFT;
2230 	udev->mmap_pages = (data_size + udev->cmdr_size + CMDR_OFF) >> PAGE_SHIFT;
2231 	udev->data_blk_size = udev->data_pages_per_blk * PAGE_SIZE;
2232 	udev->dbi_thresh = 0; /* Default in Idle state */
2233 
2234 	/* Initialise the mailbox of the ring buffer */
2235 	mb->version = TCMU_MAILBOX_VERSION;
2236 	mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC |
2237 		    TCMU_MAILBOX_FLAG_CAP_READ_LEN |
2238 		    TCMU_MAILBOX_FLAG_CAP_TMR |
2239 		    TCMU_MAILBOX_FLAG_CAP_KEEP_BUF;
2240 	mb->cmdr_off = CMDR_OFF;
2241 	mb->cmdr_size = udev->cmdr_size;
2242 
2243 	WARN_ON(!PAGE_ALIGNED(udev->data_off));
2244 	WARN_ON(data_size % PAGE_SIZE);
2245 
2246 	info->version = __stringify(TCMU_MAILBOX_VERSION);
2247 
2248 	info->mem[0].name = "tcm-user command & data buffer";
2249 	info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
2250 	info->mem[0].size = data_size + udev->cmdr_size + CMDR_OFF;
2251 	info->mem[0].memtype = UIO_MEM_NONE;
2252 
2253 	info->irqcontrol = tcmu_irqcontrol;
2254 	info->irq = UIO_IRQ_CUSTOM;
2255 
2256 	info->mmap = tcmu_mmap;
2257 	info->open = tcmu_open;
2258 	info->release = tcmu_release;
2259 
2260 	ret = uio_register_device(tcmu_root_device, info);
2261 	if (ret)
2262 		goto err_register;
2263 
2264 	/* User can set hw_block_size before enable the device */
2265 	if (dev->dev_attrib.hw_block_size == 0)
2266 		dev->dev_attrib.hw_block_size = 512;
2267 	/* Other attributes can be configured in userspace */
2268 	if (!dev->dev_attrib.hw_max_sectors)
2269 		dev->dev_attrib.hw_max_sectors = 128;
2270 	if (!dev->dev_attrib.emulate_write_cache)
2271 		dev->dev_attrib.emulate_write_cache = 0;
2272 	dev->dev_attrib.hw_queue_depth = 128;
2273 
2274 	/* If user didn't explicitly disable netlink reply support, use
2275 	 * module scope setting.
2276 	 */
2277 	if (udev->nl_reply_supported >= 0)
2278 		udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
2279 
2280 	/*
2281 	 * Get a ref incase userspace does a close on the uio device before
2282 	 * LIO has initiated tcmu_free_device.
2283 	 */
2284 	kref_get(&udev->kref);
2285 
2286 	ret = tcmu_send_dev_add_event(udev);
2287 	if (ret)
2288 		goto err_netlink;
2289 
2290 	mutex_lock(&root_udev_mutex);
2291 	list_add(&udev->node, &root_udev);
2292 	mutex_unlock(&root_udev_mutex);
2293 
2294 	return 0;
2295 
2296 err_netlink:
2297 	kref_put(&udev->kref, tcmu_dev_kref_release);
2298 	uio_unregister_device(&udev->uio_info);
2299 err_register:
2300 	vfree(udev->mb_addr);
2301 	udev->mb_addr = NULL;
2302 err_vzalloc:
2303 	bitmap_free(udev->data_bitmap);
2304 	udev->data_bitmap = NULL;
2305 err_bitmap_alloc:
2306 	kfree(info->name);
2307 	info->name = NULL;
2308 
2309 	return ret;
2310 }
2311 
2312 static void tcmu_free_device(struct se_device *dev)
2313 {
2314 	struct tcmu_dev *udev = TCMU_DEV(dev);
2315 
2316 	/* release ref from init */
2317 	kref_put(&udev->kref, tcmu_dev_kref_release);
2318 }
2319 
2320 static void tcmu_destroy_device(struct se_device *dev)
2321 {
2322 	struct tcmu_dev *udev = TCMU_DEV(dev);
2323 
2324 	del_timer_sync(&udev->cmd_timer);
2325 	del_timer_sync(&udev->qfull_timer);
2326 
2327 	mutex_lock(&root_udev_mutex);
2328 	list_del(&udev->node);
2329 	mutex_unlock(&root_udev_mutex);
2330 
2331 	tcmu_send_dev_remove_event(udev);
2332 
2333 	uio_unregister_device(&udev->uio_info);
2334 
2335 	/* release ref from configure */
2336 	kref_put(&udev->kref, tcmu_dev_kref_release);
2337 }
2338 
2339 static void tcmu_unblock_dev(struct tcmu_dev *udev)
2340 {
2341 	mutex_lock(&udev->cmdr_lock);
2342 	clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
2343 	mutex_unlock(&udev->cmdr_lock);
2344 }
2345 
2346 static void tcmu_block_dev(struct tcmu_dev *udev)
2347 {
2348 	mutex_lock(&udev->cmdr_lock);
2349 
2350 	if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2351 		goto unlock;
2352 
2353 	/* complete IO that has executed successfully */
2354 	tcmu_handle_completions(udev);
2355 	/* fail IO waiting to be queued */
2356 	run_qfull_queue(udev, true);
2357 
2358 unlock:
2359 	mutex_unlock(&udev->cmdr_lock);
2360 }
2361 
2362 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
2363 {
2364 	struct tcmu_mailbox *mb;
2365 	struct tcmu_cmd *cmd;
2366 	unsigned long i;
2367 
2368 	mutex_lock(&udev->cmdr_lock);
2369 
2370 	xa_for_each(&udev->commands, i, cmd) {
2371 		pr_debug("removing cmd %u on dev %s from ring %s\n",
2372 			 cmd->cmd_id, udev->name,
2373 			 test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags) ?
2374 			 "(is expired)" :
2375 			 (test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags) ?
2376 			 "(is keep buffer)" : ""));
2377 
2378 		xa_erase(&udev->commands, i);
2379 		if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags) &&
2380 		    !test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
2381 			WARN_ON(!cmd->se_cmd);
2382 			list_del_init(&cmd->queue_entry);
2383 			cmd->se_cmd->priv = NULL;
2384 			if (err_level == 1) {
2385 				/*
2386 				 * Userspace was not able to start the
2387 				 * command or it is retryable.
2388 				 */
2389 				target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
2390 			} else {
2391 				/* hard failure */
2392 				target_complete_cmd(cmd->se_cmd,
2393 						    SAM_STAT_CHECK_CONDITION);
2394 			}
2395 		}
2396 		tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
2397 		tcmu_free_cmd(cmd);
2398 	}
2399 
2400 	mb = udev->mb_addr;
2401 	tcmu_flush_dcache_range(mb, sizeof(*mb));
2402 	pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
2403 		 mb->cmd_tail, mb->cmd_head);
2404 
2405 	udev->cmdr_last_cleaned = 0;
2406 	mb->cmd_tail = 0;
2407 	mb->cmd_head = 0;
2408 	tcmu_flush_dcache_range(mb, sizeof(*mb));
2409 	clear_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
2410 
2411 	del_timer(&udev->cmd_timer);
2412 
2413 	/*
2414 	 * ring is empty and qfull queue never contains aborted commands.
2415 	 * So TMRs in tmr queue do not contain relevant cmd_ids.
2416 	 * After a ring reset userspace should do a fresh start, so
2417 	 * even LUN RESET message is no longer relevant.
2418 	 * Therefore remove all TMRs from qfull queue
2419 	 */
2420 	tcmu_remove_all_queued_tmr(udev);
2421 
2422 	run_qfull_queue(udev, false);
2423 
2424 	mutex_unlock(&udev->cmdr_lock);
2425 }
2426 
2427 enum {
2428 	Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
2429 	Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_data_pages_per_blk,
2430 	Opt_cmd_ring_size_mb, Opt_err,
2431 };
2432 
2433 static match_table_t tokens = {
2434 	{Opt_dev_config, "dev_config=%s"},
2435 	{Opt_dev_size, "dev_size=%s"},
2436 	{Opt_hw_block_size, "hw_block_size=%d"},
2437 	{Opt_hw_max_sectors, "hw_max_sectors=%d"},
2438 	{Opt_nl_reply_supported, "nl_reply_supported=%d"},
2439 	{Opt_max_data_area_mb, "max_data_area_mb=%d"},
2440 	{Opt_data_pages_per_blk, "data_pages_per_blk=%d"},
2441 	{Opt_cmd_ring_size_mb, "cmd_ring_size_mb=%d"},
2442 	{Opt_err, NULL}
2443 };
2444 
2445 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
2446 {
2447 	int val, ret;
2448 
2449 	ret = match_int(arg, &val);
2450 	if (ret < 0) {
2451 		pr_err("match_int() failed for dev attrib. Error %d.\n",
2452 		       ret);
2453 		return ret;
2454 	}
2455 
2456 	if (val <= 0) {
2457 		pr_err("Invalid dev attrib value %d. Must be greater than zero.\n",
2458 		       val);
2459 		return -EINVAL;
2460 	}
2461 	*dev_attrib = val;
2462 	return 0;
2463 }
2464 
2465 static int tcmu_set_max_blocks_param(struct tcmu_dev *udev, substring_t *arg)
2466 {
2467 	int val, ret;
2468 	uint32_t pages_per_blk = udev->data_pages_per_blk;
2469 
2470 	ret = match_int(arg, &val);
2471 	if (ret < 0) {
2472 		pr_err("match_int() failed for max_data_area_mb=. Error %d.\n",
2473 		       ret);
2474 		return ret;
2475 	}
2476 	if (val <= 0) {
2477 		pr_err("Invalid max_data_area %d.\n", val);
2478 		return -EINVAL;
2479 	}
2480 	if (val > TCMU_PAGES_TO_MBS(tcmu_global_max_pages)) {
2481 		pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2482 		       val, TCMU_PAGES_TO_MBS(tcmu_global_max_pages));
2483 		val = TCMU_PAGES_TO_MBS(tcmu_global_max_pages);
2484 	}
2485 	if (TCMU_MBS_TO_PAGES(val) < pages_per_blk) {
2486 		pr_err("Invalid max_data_area %d (%zu pages): smaller than data_pages_per_blk (%u pages).\n",
2487 		       val, TCMU_MBS_TO_PAGES(val), pages_per_blk);
2488 		return -EINVAL;
2489 	}
2490 
2491 	mutex_lock(&udev->cmdr_lock);
2492 	if (udev->data_bitmap) {
2493 		pr_err("Cannot set max_data_area_mb after it has been enabled.\n");
2494 		ret = -EINVAL;
2495 		goto unlock;
2496 	}
2497 
2498 	udev->data_area_mb = val;
2499 	udev->max_blocks = TCMU_MBS_TO_PAGES(val) / pages_per_blk;
2500 
2501 unlock:
2502 	mutex_unlock(&udev->cmdr_lock);
2503 	return ret;
2504 }
2505 
2506 static int tcmu_set_data_pages_per_blk(struct tcmu_dev *udev, substring_t *arg)
2507 {
2508 	int val, ret;
2509 
2510 	ret = match_int(arg, &val);
2511 	if (ret < 0) {
2512 		pr_err("match_int() failed for data_pages_per_blk=. Error %d.\n",
2513 		       ret);
2514 		return ret;
2515 	}
2516 
2517 	if (val > TCMU_MBS_TO_PAGES(udev->data_area_mb)) {
2518 		pr_err("Invalid data_pages_per_blk %d: greater than max_data_area_mb %d -> %zd pages).\n",
2519 		       val, udev->data_area_mb,
2520 		       TCMU_MBS_TO_PAGES(udev->data_area_mb));
2521 		return -EINVAL;
2522 	}
2523 
2524 	mutex_lock(&udev->cmdr_lock);
2525 	if (udev->data_bitmap) {
2526 		pr_err("Cannot set data_pages_per_blk after it has been enabled.\n");
2527 		ret = -EINVAL;
2528 		goto unlock;
2529 	}
2530 
2531 	udev->data_pages_per_blk = val;
2532 	udev->max_blocks = TCMU_MBS_TO_PAGES(udev->data_area_mb) / val;
2533 
2534 unlock:
2535 	mutex_unlock(&udev->cmdr_lock);
2536 	return ret;
2537 }
2538 
2539 static int tcmu_set_cmd_ring_size(struct tcmu_dev *udev, substring_t *arg)
2540 {
2541 	int val, ret;
2542 
2543 	ret = match_int(arg, &val);
2544 	if (ret < 0) {
2545 		pr_err("match_int() failed for cmd_ring_size_mb=. Error %d.\n",
2546 		       ret);
2547 		return ret;
2548 	}
2549 
2550 	if (val <= 0) {
2551 		pr_err("Invalid cmd_ring_size_mb %d.\n", val);
2552 		return -EINVAL;
2553 	}
2554 
2555 	mutex_lock(&udev->cmdr_lock);
2556 	if (udev->data_bitmap) {
2557 		pr_err("Cannot set cmd_ring_size_mb after it has been enabled.\n");
2558 		ret = -EINVAL;
2559 		goto unlock;
2560 	}
2561 
2562 	udev->cmdr_size = (val << 20) - CMDR_OFF;
2563 	if (val > (MB_CMDR_SIZE_DEF >> 20)) {
2564 		pr_err("%d is too large. Adjusting cmd_ring_size_mb to global limit of %u\n",
2565 		       val, (MB_CMDR_SIZE_DEF >> 20));
2566 		udev->cmdr_size = CMDR_SIZE_DEF;
2567 	}
2568 
2569 unlock:
2570 	mutex_unlock(&udev->cmdr_lock);
2571 	return ret;
2572 }
2573 
2574 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
2575 		const char *page, ssize_t count)
2576 {
2577 	struct tcmu_dev *udev = TCMU_DEV(dev);
2578 	char *orig, *ptr, *opts;
2579 	substring_t args[MAX_OPT_ARGS];
2580 	int ret = 0, token;
2581 
2582 	opts = kstrdup(page, GFP_KERNEL);
2583 	if (!opts)
2584 		return -ENOMEM;
2585 
2586 	orig = opts;
2587 
2588 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
2589 		if (!*ptr)
2590 			continue;
2591 
2592 		token = match_token(ptr, tokens, args);
2593 		switch (token) {
2594 		case Opt_dev_config:
2595 			if (match_strlcpy(udev->dev_config, &args[0],
2596 					  TCMU_CONFIG_LEN) == 0) {
2597 				ret = -EINVAL;
2598 				break;
2599 			}
2600 			pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2601 			break;
2602 		case Opt_dev_size:
2603 			ret = match_u64(&args[0], &udev->dev_size);
2604 			if (ret < 0)
2605 				pr_err("match_u64() failed for dev_size=. Error %d.\n",
2606 				       ret);
2607 			break;
2608 		case Opt_hw_block_size:
2609 			ret = tcmu_set_dev_attrib(&args[0],
2610 					&(dev->dev_attrib.hw_block_size));
2611 			break;
2612 		case Opt_hw_max_sectors:
2613 			ret = tcmu_set_dev_attrib(&args[0],
2614 					&(dev->dev_attrib.hw_max_sectors));
2615 			break;
2616 		case Opt_nl_reply_supported:
2617 			ret = match_int(&args[0], &udev->nl_reply_supported);
2618 			if (ret < 0)
2619 				pr_err("match_int() failed for nl_reply_supported=. Error %d.\n",
2620 				       ret);
2621 			break;
2622 		case Opt_max_data_area_mb:
2623 			ret = tcmu_set_max_blocks_param(udev, &args[0]);
2624 			break;
2625 		case Opt_data_pages_per_blk:
2626 			ret = tcmu_set_data_pages_per_blk(udev, &args[0]);
2627 			break;
2628 		case Opt_cmd_ring_size_mb:
2629 			ret = tcmu_set_cmd_ring_size(udev, &args[0]);
2630 			break;
2631 		default:
2632 			break;
2633 		}
2634 
2635 		if (ret)
2636 			break;
2637 	}
2638 
2639 	kfree(orig);
2640 	return (!ret) ? count : ret;
2641 }
2642 
2643 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2644 {
2645 	struct tcmu_dev *udev = TCMU_DEV(dev);
2646 	ssize_t bl = 0;
2647 
2648 	bl = sprintf(b + bl, "Config: %s ",
2649 		     udev->dev_config[0] ? udev->dev_config : "NULL");
2650 	bl += sprintf(b + bl, "Size: %llu ", udev->dev_size);
2651 	bl += sprintf(b + bl, "MaxDataAreaMB: %u ", udev->data_area_mb);
2652 	bl += sprintf(b + bl, "DataPagesPerBlk: %u ", udev->data_pages_per_blk);
2653 	bl += sprintf(b + bl, "CmdRingSizeMB: %u\n",
2654 		      (udev->cmdr_size + CMDR_OFF) >> 20);
2655 
2656 	return bl;
2657 }
2658 
2659 static sector_t tcmu_get_blocks(struct se_device *dev)
2660 {
2661 	struct tcmu_dev *udev = TCMU_DEV(dev);
2662 
2663 	return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2664 		       dev->dev_attrib.block_size);
2665 }
2666 
2667 static sense_reason_t
2668 tcmu_parse_cdb(struct se_cmd *cmd)
2669 {
2670 	return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2671 }
2672 
2673 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2674 {
2675 	struct se_dev_attrib *da = container_of(to_config_group(item),
2676 					struct se_dev_attrib, da_group);
2677 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2678 
2679 	return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2680 }
2681 
2682 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2683 				       size_t count)
2684 {
2685 	struct se_dev_attrib *da = container_of(to_config_group(item),
2686 					struct se_dev_attrib, da_group);
2687 	struct tcmu_dev *udev = container_of(da->da_dev,
2688 					struct tcmu_dev, se_dev);
2689 	u32 val;
2690 	int ret;
2691 
2692 	if (da->da_dev->export_count) {
2693 		pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2694 		return -EINVAL;
2695 	}
2696 
2697 	ret = kstrtou32(page, 0, &val);
2698 	if (ret < 0)
2699 		return ret;
2700 
2701 	udev->cmd_time_out = val * MSEC_PER_SEC;
2702 	return count;
2703 }
2704 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2705 
2706 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2707 {
2708 	struct se_dev_attrib *da = container_of(to_config_group(item),
2709 						struct se_dev_attrib, da_group);
2710 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2711 
2712 	return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2713 			udev->qfull_time_out :
2714 			udev->qfull_time_out / MSEC_PER_SEC);
2715 }
2716 
2717 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2718 					 const char *page, size_t count)
2719 {
2720 	struct se_dev_attrib *da = container_of(to_config_group(item),
2721 					struct se_dev_attrib, da_group);
2722 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2723 	s32 val;
2724 	int ret;
2725 
2726 	ret = kstrtos32(page, 0, &val);
2727 	if (ret < 0)
2728 		return ret;
2729 
2730 	if (val >= 0) {
2731 		udev->qfull_time_out = val * MSEC_PER_SEC;
2732 	} else if (val == -1) {
2733 		udev->qfull_time_out = val;
2734 	} else {
2735 		printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2736 		return -EINVAL;
2737 	}
2738 	return count;
2739 }
2740 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2741 
2742 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2743 {
2744 	struct se_dev_attrib *da = container_of(to_config_group(item),
2745 						struct se_dev_attrib, da_group);
2746 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2747 
2748 	return snprintf(page, PAGE_SIZE, "%u\n", udev->data_area_mb);
2749 }
2750 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2751 
2752 static ssize_t tcmu_data_pages_per_blk_show(struct config_item *item,
2753 					    char *page)
2754 {
2755 	struct se_dev_attrib *da = container_of(to_config_group(item),
2756 						struct se_dev_attrib, da_group);
2757 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2758 
2759 	return snprintf(page, PAGE_SIZE, "%u\n", udev->data_pages_per_blk);
2760 }
2761 CONFIGFS_ATTR_RO(tcmu_, data_pages_per_blk);
2762 
2763 static ssize_t tcmu_cmd_ring_size_mb_show(struct config_item *item, char *page)
2764 {
2765 	struct se_dev_attrib *da = container_of(to_config_group(item),
2766 						struct se_dev_attrib, da_group);
2767 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2768 
2769 	return snprintf(page, PAGE_SIZE, "%u\n",
2770 			(udev->cmdr_size + CMDR_OFF) >> 20);
2771 }
2772 CONFIGFS_ATTR_RO(tcmu_, cmd_ring_size_mb);
2773 
2774 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2775 {
2776 	struct se_dev_attrib *da = container_of(to_config_group(item),
2777 						struct se_dev_attrib, da_group);
2778 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2779 
2780 	return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2781 }
2782 
2783 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2784 				      const char *reconfig_data)
2785 {
2786 	struct sk_buff *skb = NULL;
2787 	void *msg_header = NULL;
2788 	int ret = 0;
2789 
2790 	ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2791 				      &skb, &msg_header);
2792 	if (ret < 0)
2793 		return ret;
2794 	ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2795 	if (ret < 0) {
2796 		nlmsg_free(skb);
2797 		return ret;
2798 	}
2799 	return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2800 				       skb, msg_header);
2801 }
2802 
2803 
2804 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2805 				     size_t count)
2806 {
2807 	struct se_dev_attrib *da = container_of(to_config_group(item),
2808 						struct se_dev_attrib, da_group);
2809 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2810 	int ret, len;
2811 
2812 	len = strlen(page);
2813 	if (!len || len > TCMU_CONFIG_LEN - 1)
2814 		return -EINVAL;
2815 
2816 	/* Check if device has been configured before */
2817 	if (target_dev_configured(&udev->se_dev)) {
2818 		ret = tcmu_send_dev_config_event(udev, page);
2819 		if (ret) {
2820 			pr_err("Unable to reconfigure device\n");
2821 			return ret;
2822 		}
2823 		strscpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2824 
2825 		ret = tcmu_update_uio_info(udev);
2826 		if (ret)
2827 			return ret;
2828 		return count;
2829 	}
2830 	strscpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2831 
2832 	return count;
2833 }
2834 CONFIGFS_ATTR(tcmu_, dev_config);
2835 
2836 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2837 {
2838 	struct se_dev_attrib *da = container_of(to_config_group(item),
2839 						struct se_dev_attrib, da_group);
2840 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2841 
2842 	return snprintf(page, PAGE_SIZE, "%llu\n", udev->dev_size);
2843 }
2844 
2845 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2846 {
2847 	struct sk_buff *skb = NULL;
2848 	void *msg_header = NULL;
2849 	int ret = 0;
2850 
2851 	ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2852 				      &skb, &msg_header);
2853 	if (ret < 0)
2854 		return ret;
2855 	ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2856 				size, TCMU_ATTR_PAD);
2857 	if (ret < 0) {
2858 		nlmsg_free(skb);
2859 		return ret;
2860 	}
2861 	return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2862 				       skb, msg_header);
2863 }
2864 
2865 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2866 				   size_t count)
2867 {
2868 	struct se_dev_attrib *da = container_of(to_config_group(item),
2869 						struct se_dev_attrib, da_group);
2870 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2871 	u64 val;
2872 	int ret;
2873 
2874 	ret = kstrtou64(page, 0, &val);
2875 	if (ret < 0)
2876 		return ret;
2877 
2878 	/* Check if device has been configured before */
2879 	if (target_dev_configured(&udev->se_dev)) {
2880 		ret = tcmu_send_dev_size_event(udev, val);
2881 		if (ret) {
2882 			pr_err("Unable to reconfigure device\n");
2883 			return ret;
2884 		}
2885 	}
2886 	udev->dev_size = val;
2887 	return count;
2888 }
2889 CONFIGFS_ATTR(tcmu_, dev_size);
2890 
2891 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2892 		char *page)
2893 {
2894 	struct se_dev_attrib *da = container_of(to_config_group(item),
2895 						struct se_dev_attrib, da_group);
2896 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2897 
2898 	return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2899 }
2900 
2901 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2902 		const char *page, size_t count)
2903 {
2904 	struct se_dev_attrib *da = container_of(to_config_group(item),
2905 						struct se_dev_attrib, da_group);
2906 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2907 	s8 val;
2908 	int ret;
2909 
2910 	ret = kstrtos8(page, 0, &val);
2911 	if (ret < 0)
2912 		return ret;
2913 
2914 	udev->nl_reply_supported = val;
2915 	return count;
2916 }
2917 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2918 
2919 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2920 					     char *page)
2921 {
2922 	struct se_dev_attrib *da = container_of(to_config_group(item),
2923 					struct se_dev_attrib, da_group);
2924 
2925 	return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2926 }
2927 
2928 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2929 {
2930 	struct sk_buff *skb = NULL;
2931 	void *msg_header = NULL;
2932 	int ret = 0;
2933 
2934 	ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2935 				      &skb, &msg_header);
2936 	if (ret < 0)
2937 		return ret;
2938 	ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2939 	if (ret < 0) {
2940 		nlmsg_free(skb);
2941 		return ret;
2942 	}
2943 	return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2944 				       skb, msg_header);
2945 }
2946 
2947 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2948 					      const char *page, size_t count)
2949 {
2950 	struct se_dev_attrib *da = container_of(to_config_group(item),
2951 					struct se_dev_attrib, da_group);
2952 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2953 	u8 val;
2954 	int ret;
2955 
2956 	ret = kstrtou8(page, 0, &val);
2957 	if (ret < 0)
2958 		return ret;
2959 
2960 	/* Check if device has been configured before */
2961 	if (target_dev_configured(&udev->se_dev)) {
2962 		ret = tcmu_send_emulate_write_cache(udev, val);
2963 		if (ret) {
2964 			pr_err("Unable to reconfigure device\n");
2965 			return ret;
2966 		}
2967 	}
2968 
2969 	da->emulate_write_cache = val;
2970 	return count;
2971 }
2972 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2973 
2974 static ssize_t tcmu_tmr_notification_show(struct config_item *item, char *page)
2975 {
2976 	struct se_dev_attrib *da = container_of(to_config_group(item),
2977 					struct se_dev_attrib, da_group);
2978 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2979 
2980 	return snprintf(page, PAGE_SIZE, "%i\n",
2981 			test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags));
2982 }
2983 
2984 static ssize_t tcmu_tmr_notification_store(struct config_item *item,
2985 					   const char *page, size_t count)
2986 {
2987 	struct se_dev_attrib *da = container_of(to_config_group(item),
2988 					struct se_dev_attrib, da_group);
2989 	struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2990 	u8 val;
2991 	int ret;
2992 
2993 	ret = kstrtou8(page, 0, &val);
2994 	if (ret < 0)
2995 		return ret;
2996 	if (val > 1)
2997 		return -EINVAL;
2998 
2999 	if (val)
3000 		set_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
3001 	else
3002 		clear_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
3003 	return count;
3004 }
3005 CONFIGFS_ATTR(tcmu_, tmr_notification);
3006 
3007 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
3008 {
3009 	struct se_device *se_dev = container_of(to_config_group(item),
3010 						struct se_device,
3011 						dev_action_group);
3012 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
3013 
3014 	if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
3015 		return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
3016 	else
3017 		return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
3018 }
3019 
3020 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
3021 				    size_t count)
3022 {
3023 	struct se_device *se_dev = container_of(to_config_group(item),
3024 						struct se_device,
3025 						dev_action_group);
3026 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
3027 	u8 val;
3028 	int ret;
3029 
3030 	if (!target_dev_configured(&udev->se_dev)) {
3031 		pr_err("Device is not configured.\n");
3032 		return -EINVAL;
3033 	}
3034 
3035 	ret = kstrtou8(page, 0, &val);
3036 	if (ret < 0)
3037 		return ret;
3038 
3039 	if (val > 1) {
3040 		pr_err("Invalid block value %d\n", val);
3041 		return -EINVAL;
3042 	}
3043 
3044 	if (!val)
3045 		tcmu_unblock_dev(udev);
3046 	else
3047 		tcmu_block_dev(udev);
3048 	return count;
3049 }
3050 CONFIGFS_ATTR(tcmu_, block_dev);
3051 
3052 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
3053 				     size_t count)
3054 {
3055 	struct se_device *se_dev = container_of(to_config_group(item),
3056 						struct se_device,
3057 						dev_action_group);
3058 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
3059 	u8 val;
3060 	int ret;
3061 
3062 	if (!target_dev_configured(&udev->se_dev)) {
3063 		pr_err("Device is not configured.\n");
3064 		return -EINVAL;
3065 	}
3066 
3067 	ret = kstrtou8(page, 0, &val);
3068 	if (ret < 0)
3069 		return ret;
3070 
3071 	if (val != 1 && val != 2) {
3072 		pr_err("Invalid reset ring value %d\n", val);
3073 		return -EINVAL;
3074 	}
3075 
3076 	tcmu_reset_ring(udev, val);
3077 	return count;
3078 }
3079 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
3080 
3081 static ssize_t tcmu_free_kept_buf_store(struct config_item *item, const char *page,
3082 					size_t count)
3083 {
3084 	struct se_device *se_dev = container_of(to_config_group(item),
3085 						struct se_device,
3086 						dev_action_group);
3087 	struct tcmu_dev *udev = TCMU_DEV(se_dev);
3088 	struct tcmu_cmd *cmd;
3089 	u16 cmd_id;
3090 	int ret;
3091 
3092 	if (!target_dev_configured(&udev->se_dev)) {
3093 		pr_err("Device is not configured.\n");
3094 		return -EINVAL;
3095 	}
3096 
3097 	ret = kstrtou16(page, 0, &cmd_id);
3098 	if (ret < 0)
3099 		return ret;
3100 
3101 	mutex_lock(&udev->cmdr_lock);
3102 
3103 	{
3104 		XA_STATE(xas, &udev->commands, cmd_id);
3105 
3106 		xas_lock(&xas);
3107 		cmd = xas_load(&xas);
3108 		if (!cmd) {
3109 			pr_err("free_kept_buf: cmd_id %d not found\n", cmd_id);
3110 			count = -EINVAL;
3111 			xas_unlock(&xas);
3112 			goto out_unlock;
3113 		}
3114 		if (!test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
3115 			pr_err("free_kept_buf: cmd_id %d was not completed with KEEP_BUF\n",
3116 			       cmd_id);
3117 			count = -EINVAL;
3118 			xas_unlock(&xas);
3119 			goto out_unlock;
3120 		}
3121 		xas_store(&xas, NULL);
3122 		xas_unlock(&xas);
3123 	}
3124 
3125 	tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
3126 	tcmu_free_cmd(cmd);
3127 	/*
3128 	 * We only freed data space, not ring space. Therefore we dont call
3129 	 * run_tmr_queue, but call run_qfull_queue if tmr_list is empty.
3130 	 */
3131 	if (list_empty(&udev->tmr_queue))
3132 		run_qfull_queue(udev, false);
3133 
3134 out_unlock:
3135 	mutex_unlock(&udev->cmdr_lock);
3136 	return count;
3137 }
3138 CONFIGFS_ATTR_WO(tcmu_, free_kept_buf);
3139 
3140 static struct configfs_attribute *tcmu_attrib_attrs[] = {
3141 	&tcmu_attr_cmd_time_out,
3142 	&tcmu_attr_qfull_time_out,
3143 	&tcmu_attr_max_data_area_mb,
3144 	&tcmu_attr_data_pages_per_blk,
3145 	&tcmu_attr_cmd_ring_size_mb,
3146 	&tcmu_attr_dev_config,
3147 	&tcmu_attr_dev_size,
3148 	&tcmu_attr_emulate_write_cache,
3149 	&tcmu_attr_tmr_notification,
3150 	&tcmu_attr_nl_reply_supported,
3151 	NULL,
3152 };
3153 
3154 static struct configfs_attribute **tcmu_attrs;
3155 
3156 static struct configfs_attribute *tcmu_action_attrs[] = {
3157 	&tcmu_attr_block_dev,
3158 	&tcmu_attr_reset_ring,
3159 	&tcmu_attr_free_kept_buf,
3160 	NULL,
3161 };
3162 
3163 static struct target_backend_ops tcmu_ops = {
3164 	.name			= "user",
3165 	.owner			= THIS_MODULE,
3166 	.transport_flags_default = TRANSPORT_FLAG_PASSTHROUGH,
3167 	.transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR |
3168 				      TRANSPORT_FLAG_PASSTHROUGH_ALUA,
3169 	.attach_hba		= tcmu_attach_hba,
3170 	.detach_hba		= tcmu_detach_hba,
3171 	.alloc_device		= tcmu_alloc_device,
3172 	.configure_device	= tcmu_configure_device,
3173 	.destroy_device		= tcmu_destroy_device,
3174 	.free_device		= tcmu_free_device,
3175 	.unplug_device		= tcmu_unplug_device,
3176 	.plug_device		= tcmu_plug_device,
3177 	.parse_cdb		= tcmu_parse_cdb,
3178 	.tmr_notify		= tcmu_tmr_notify,
3179 	.set_configfs_dev_params = tcmu_set_configfs_dev_params,
3180 	.show_configfs_dev_params = tcmu_show_configfs_dev_params,
3181 	.get_device_type	= sbc_get_device_type,
3182 	.get_blocks		= tcmu_get_blocks,
3183 	.tb_dev_action_attrs	= tcmu_action_attrs,
3184 };
3185 
3186 static void find_free_blocks(void)
3187 {
3188 	struct tcmu_dev *udev;
3189 	loff_t off;
3190 	u32 pages_freed, total_pages_freed = 0;
3191 	u32 start, end, block, total_blocks_freed = 0;
3192 
3193 	if (atomic_read(&global_page_count) <= tcmu_global_max_pages)
3194 		return;
3195 
3196 	mutex_lock(&root_udev_mutex);
3197 	list_for_each_entry(udev, &root_udev, node) {
3198 		mutex_lock(&udev->cmdr_lock);
3199 
3200 		if (!target_dev_configured(&udev->se_dev)) {
3201 			mutex_unlock(&udev->cmdr_lock);
3202 			continue;
3203 		}
3204 
3205 		/* Try to complete the finished commands first */
3206 		if (tcmu_handle_completions(udev))
3207 			run_qfull_queue(udev, false);
3208 
3209 		/* Skip the udevs in idle */
3210 		if (!udev->dbi_thresh) {
3211 			mutex_unlock(&udev->cmdr_lock);
3212 			continue;
3213 		}
3214 
3215 		end = udev->dbi_max + 1;
3216 		block = find_last_bit(udev->data_bitmap, end);
3217 		if (block == udev->dbi_max) {
3218 			/*
3219 			 * The last bit is dbi_max, so it is not possible
3220 			 * reclaim any blocks.
3221 			 */
3222 			mutex_unlock(&udev->cmdr_lock);
3223 			continue;
3224 		} else if (block == end) {
3225 			/* The current udev will goto idle state */
3226 			udev->dbi_thresh = start = 0;
3227 			udev->dbi_max = 0;
3228 		} else {
3229 			udev->dbi_thresh = start = block + 1;
3230 			udev->dbi_max = block;
3231 		}
3232 
3233 		/*
3234 		 * Release the block pages.
3235 		 *
3236 		 * Also note that since tcmu_vma_fault() gets an extra page
3237 		 * refcount, tcmu_blocks_release() won't free pages if pages
3238 		 * are mapped. This means it is safe to call
3239 		 * tcmu_blocks_release() before unmap_mapping_range() which
3240 		 * drops the refcount of any pages it unmaps and thus releases
3241 		 * them.
3242 		 */
3243 		pages_freed = tcmu_blocks_release(udev, start, end - 1);
3244 
3245 		/* Here will truncate the data area from off */
3246 		off = udev->data_off + (loff_t)start * udev->data_blk_size;
3247 		unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
3248 
3249 		mutex_unlock(&udev->cmdr_lock);
3250 
3251 		total_pages_freed += pages_freed;
3252 		total_blocks_freed += end - start;
3253 		pr_debug("Freed %u pages (total %u) from %u blocks (total %u) from %s.\n",
3254 			 pages_freed, total_pages_freed, end - start,
3255 			 total_blocks_freed, udev->name);
3256 	}
3257 	mutex_unlock(&root_udev_mutex);
3258 
3259 	if (atomic_read(&global_page_count) > tcmu_global_max_pages)
3260 		schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
3261 }
3262 
3263 static void check_timedout_devices(void)
3264 {
3265 	struct tcmu_dev *udev, *tmp_dev;
3266 	struct tcmu_cmd *cmd, *tmp_cmd;
3267 	LIST_HEAD(devs);
3268 
3269 	spin_lock_bh(&timed_out_udevs_lock);
3270 	list_splice_init(&timed_out_udevs, &devs);
3271 
3272 	list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
3273 		list_del_init(&udev->timedout_entry);
3274 		spin_unlock_bh(&timed_out_udevs_lock);
3275 
3276 		mutex_lock(&udev->cmdr_lock);
3277 
3278 		/*
3279 		 * If cmd_time_out is disabled but qfull is set deadline
3280 		 * will only reflect the qfull timeout. Ignore it.
3281 		 */
3282 		if (udev->cmd_time_out) {
3283 			list_for_each_entry_safe(cmd, tmp_cmd,
3284 						 &udev->inflight_queue,
3285 						 queue_entry) {
3286 				tcmu_check_expired_ring_cmd(cmd);
3287 			}
3288 			tcmu_set_next_deadline(&udev->inflight_queue,
3289 					       &udev->cmd_timer);
3290 		}
3291 		list_for_each_entry_safe(cmd, tmp_cmd, &udev->qfull_queue,
3292 					 queue_entry) {
3293 			tcmu_check_expired_queue_cmd(cmd);
3294 		}
3295 		tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
3296 
3297 		mutex_unlock(&udev->cmdr_lock);
3298 
3299 		spin_lock_bh(&timed_out_udevs_lock);
3300 	}
3301 
3302 	spin_unlock_bh(&timed_out_udevs_lock);
3303 }
3304 
3305 static void tcmu_unmap_work_fn(struct work_struct *work)
3306 {
3307 	check_timedout_devices();
3308 	find_free_blocks();
3309 }
3310 
3311 static int __init tcmu_module_init(void)
3312 {
3313 	int ret, i, k, len = 0;
3314 
3315 	BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
3316 
3317 	INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
3318 
3319 	tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
3320 				sizeof(struct tcmu_cmd),
3321 				__alignof__(struct tcmu_cmd),
3322 				0, NULL);
3323 	if (!tcmu_cmd_cache)
3324 		return -ENOMEM;
3325 
3326 	tcmu_root_device = root_device_register("tcm_user");
3327 	if (IS_ERR(tcmu_root_device)) {
3328 		ret = PTR_ERR(tcmu_root_device);
3329 		goto out_free_cache;
3330 	}
3331 
3332 	ret = genl_register_family(&tcmu_genl_family);
3333 	if (ret < 0) {
3334 		goto out_unreg_device;
3335 	}
3336 
3337 	for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3338 		len += sizeof(struct configfs_attribute *);
3339 	for (i = 0; passthrough_pr_attrib_attrs[i] != NULL; i++)
3340 		len += sizeof(struct configfs_attribute *);
3341 	for (i = 0; tcmu_attrib_attrs[i] != NULL; i++)
3342 		len += sizeof(struct configfs_attribute *);
3343 	len += sizeof(struct configfs_attribute *);
3344 
3345 	tcmu_attrs = kzalloc(len, GFP_KERNEL);
3346 	if (!tcmu_attrs) {
3347 		ret = -ENOMEM;
3348 		goto out_unreg_genl;
3349 	}
3350 
3351 	for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3352 		tcmu_attrs[i] = passthrough_attrib_attrs[i];
3353 	for (k = 0; passthrough_pr_attrib_attrs[k] != NULL; k++)
3354 		tcmu_attrs[i++] = passthrough_pr_attrib_attrs[k];
3355 	for (k = 0; tcmu_attrib_attrs[k] != NULL; k++)
3356 		tcmu_attrs[i++] = tcmu_attrib_attrs[k];
3357 	tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
3358 
3359 	ret = transport_backend_register(&tcmu_ops);
3360 	if (ret)
3361 		goto out_attrs;
3362 
3363 	return 0;
3364 
3365 out_attrs:
3366 	kfree(tcmu_attrs);
3367 out_unreg_genl:
3368 	genl_unregister_family(&tcmu_genl_family);
3369 out_unreg_device:
3370 	root_device_unregister(tcmu_root_device);
3371 out_free_cache:
3372 	kmem_cache_destroy(tcmu_cmd_cache);
3373 
3374 	return ret;
3375 }
3376 
3377 static void __exit tcmu_module_exit(void)
3378 {
3379 	cancel_delayed_work_sync(&tcmu_unmap_work);
3380 	target_backend_unregister(&tcmu_ops);
3381 	kfree(tcmu_attrs);
3382 	genl_unregister_family(&tcmu_genl_family);
3383 	root_device_unregister(tcmu_root_device);
3384 	kmem_cache_destroy(tcmu_cmd_cache);
3385 }
3386 
3387 MODULE_DESCRIPTION("TCM USER subsystem plugin");
3388 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
3389 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
3390 MODULE_LICENSE("GPL");
3391 
3392 module_init(tcmu_module_init);
3393 module_exit(tcmu_module_exit);
3394