xref: /linux/drivers/target/target_core_iblock.c (revision 34dc1baba215b826e454b8d19e4f24adbeb7d00d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*******************************************************************************
3  * Filename:  target_core_iblock.c
4  *
5  * This file contains the Storage Engine  <-> Linux BlockIO transport
6  * specific functions.
7  *
8  * (c) Copyright 2003-2013 Datera, Inc.
9  *
10  * Nicholas A. Bellinger <nab@kernel.org>
11  *
12  ******************************************************************************/
13 
14 #include <linux/string.h>
15 #include <linux/parser.h>
16 #include <linux/timer.h>
17 #include <linux/fs.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-integrity.h>
20 #include <linux/slab.h>
21 #include <linux/spinlock.h>
22 #include <linux/bio.h>
23 #include <linux/file.h>
24 #include <linux/module.h>
25 #include <linux/scatterlist.h>
26 #include <linux/pr.h>
27 #include <scsi/scsi_proto.h>
28 #include <scsi/scsi_common.h>
29 #include <asm/unaligned.h>
30 
31 #include <target/target_core_base.h>
32 #include <target/target_core_backend.h>
33 
34 #include "target_core_iblock.h"
35 #include "target_core_pr.h"
36 
37 #define IBLOCK_MAX_BIO_PER_TASK	 32	/* max # of bios to submit at a time */
38 #define IBLOCK_BIO_POOL_SIZE	128
39 
40 static inline struct iblock_dev *IBLOCK_DEV(struct se_device *dev)
41 {
42 	return container_of(dev, struct iblock_dev, dev);
43 }
44 
45 
46 static int iblock_attach_hba(struct se_hba *hba, u32 host_id)
47 {
48 	pr_debug("CORE_HBA[%d] - TCM iBlock HBA Driver %s on"
49 		" Generic Target Core Stack %s\n", hba->hba_id,
50 		IBLOCK_VERSION, TARGET_CORE_VERSION);
51 	return 0;
52 }
53 
54 static void iblock_detach_hba(struct se_hba *hba)
55 {
56 }
57 
58 static struct se_device *iblock_alloc_device(struct se_hba *hba, const char *name)
59 {
60 	struct iblock_dev *ib_dev = NULL;
61 
62 	ib_dev = kzalloc(sizeof(struct iblock_dev), GFP_KERNEL);
63 	if (!ib_dev) {
64 		pr_err("Unable to allocate struct iblock_dev\n");
65 		return NULL;
66 	}
67 
68 	ib_dev->ibd_plug = kcalloc(nr_cpu_ids, sizeof(*ib_dev->ibd_plug),
69 				   GFP_KERNEL);
70 	if (!ib_dev->ibd_plug)
71 		goto free_dev;
72 
73 	pr_debug( "IBLOCK: Allocated ib_dev for %s\n", name);
74 
75 	return &ib_dev->dev;
76 
77 free_dev:
78 	kfree(ib_dev);
79 	return NULL;
80 }
81 
82 static bool iblock_configure_unmap(struct se_device *dev)
83 {
84 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
85 
86 	return target_configure_unmap_from_queue(&dev->dev_attrib,
87 						 ib_dev->ibd_bd);
88 }
89 
90 static int iblock_configure_device(struct se_device *dev)
91 {
92 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
93 	struct request_queue *q;
94 	struct block_device *bd = NULL;
95 	struct blk_integrity *bi;
96 	blk_mode_t mode = BLK_OPEN_READ;
97 	unsigned int max_write_zeroes_sectors;
98 	int ret;
99 
100 	if (!(ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)) {
101 		pr_err("Missing udev_path= parameters for IBLOCK\n");
102 		return -EINVAL;
103 	}
104 
105 	ret = bioset_init(&ib_dev->ibd_bio_set, IBLOCK_BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
106 	if (ret) {
107 		pr_err("IBLOCK: Unable to create bioset\n");
108 		goto out;
109 	}
110 
111 	pr_debug( "IBLOCK: Claiming struct block_device: %s\n",
112 			ib_dev->ibd_udev_path);
113 
114 	if (!ib_dev->ibd_readonly)
115 		mode |= BLK_OPEN_WRITE;
116 	else
117 		dev->dev_flags |= DF_READ_ONLY;
118 
119 	bd = blkdev_get_by_path(ib_dev->ibd_udev_path, mode, ib_dev, NULL);
120 	if (IS_ERR(bd)) {
121 		ret = PTR_ERR(bd);
122 		goto out_free_bioset;
123 	}
124 	ib_dev->ibd_bd = bd;
125 
126 	q = bdev_get_queue(bd);
127 
128 	dev->dev_attrib.hw_block_size = bdev_logical_block_size(bd);
129 	dev->dev_attrib.hw_max_sectors = mult_frac(queue_max_hw_sectors(q),
130 			SECTOR_SIZE,
131 			dev->dev_attrib.hw_block_size);
132 	dev->dev_attrib.hw_queue_depth = q->nr_requests;
133 
134 	/*
135 	 * Enable write same emulation for IBLOCK and use 0xFFFF as
136 	 * the smaller WRITE_SAME(10) only has a two-byte block count.
137 	 */
138 	max_write_zeroes_sectors = bdev_write_zeroes_sectors(bd);
139 	if (max_write_zeroes_sectors)
140 		dev->dev_attrib.max_write_same_len = max_write_zeroes_sectors;
141 	else
142 		dev->dev_attrib.max_write_same_len = 0xFFFF;
143 
144 	if (bdev_nonrot(bd))
145 		dev->dev_attrib.is_nonrot = 1;
146 
147 	bi = bdev_get_integrity(bd);
148 	if (bi) {
149 		struct bio_set *bs = &ib_dev->ibd_bio_set;
150 
151 		if (!strcmp(bi->profile->name, "T10-DIF-TYPE3-IP") ||
152 		    !strcmp(bi->profile->name, "T10-DIF-TYPE1-IP")) {
153 			pr_err("IBLOCK export of blk_integrity: %s not"
154 			       " supported\n", bi->profile->name);
155 			ret = -ENOSYS;
156 			goto out_blkdev_put;
157 		}
158 
159 		if (!strcmp(bi->profile->name, "T10-DIF-TYPE3-CRC")) {
160 			dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE3_PROT;
161 		} else if (!strcmp(bi->profile->name, "T10-DIF-TYPE1-CRC")) {
162 			dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE1_PROT;
163 		}
164 
165 		if (dev->dev_attrib.pi_prot_type) {
166 			if (bioset_integrity_create(bs, IBLOCK_BIO_POOL_SIZE) < 0) {
167 				pr_err("Unable to allocate bioset for PI\n");
168 				ret = -ENOMEM;
169 				goto out_blkdev_put;
170 			}
171 			pr_debug("IBLOCK setup BIP bs->bio_integrity_pool: %p\n",
172 				 &bs->bio_integrity_pool);
173 		}
174 		dev->dev_attrib.hw_pi_prot_type = dev->dev_attrib.pi_prot_type;
175 	}
176 
177 	return 0;
178 
179 out_blkdev_put:
180 	blkdev_put(ib_dev->ibd_bd, ib_dev);
181 out_free_bioset:
182 	bioset_exit(&ib_dev->ibd_bio_set);
183 out:
184 	return ret;
185 }
186 
187 static void iblock_dev_call_rcu(struct rcu_head *p)
188 {
189 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
190 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
191 
192 	kfree(ib_dev->ibd_plug);
193 	kfree(ib_dev);
194 }
195 
196 static void iblock_free_device(struct se_device *dev)
197 {
198 	call_rcu(&dev->rcu_head, iblock_dev_call_rcu);
199 }
200 
201 static void iblock_destroy_device(struct se_device *dev)
202 {
203 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
204 
205 	if (ib_dev->ibd_bd != NULL)
206 		blkdev_put(ib_dev->ibd_bd, ib_dev);
207 	bioset_exit(&ib_dev->ibd_bio_set);
208 }
209 
210 static struct se_dev_plug *iblock_plug_device(struct se_device *se_dev)
211 {
212 	struct iblock_dev *ib_dev = IBLOCK_DEV(se_dev);
213 	struct iblock_dev_plug *ib_dev_plug;
214 
215 	/*
216 	 * Each se_device has a per cpu work this can be run from. We
217 	 * shouldn't have multiple threads on the same cpu calling this
218 	 * at the same time.
219 	 */
220 	ib_dev_plug = &ib_dev->ibd_plug[raw_smp_processor_id()];
221 	if (test_and_set_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags))
222 		return NULL;
223 
224 	blk_start_plug(&ib_dev_plug->blk_plug);
225 	return &ib_dev_plug->se_plug;
226 }
227 
228 static void iblock_unplug_device(struct se_dev_plug *se_plug)
229 {
230 	struct iblock_dev_plug *ib_dev_plug = container_of(se_plug,
231 					struct iblock_dev_plug, se_plug);
232 
233 	blk_finish_plug(&ib_dev_plug->blk_plug);
234 	clear_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags);
235 }
236 
237 static sector_t iblock_get_blocks(struct se_device *dev)
238 {
239 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
240 	u32 block_size = bdev_logical_block_size(ib_dev->ibd_bd);
241 	unsigned long long blocks_long =
242 		div_u64(bdev_nr_bytes(ib_dev->ibd_bd), block_size) - 1;
243 
244 	if (block_size == dev->dev_attrib.block_size)
245 		return blocks_long;
246 
247 	switch (block_size) {
248 	case 4096:
249 		switch (dev->dev_attrib.block_size) {
250 		case 2048:
251 			blocks_long <<= 1;
252 			break;
253 		case 1024:
254 			blocks_long <<= 2;
255 			break;
256 		case 512:
257 			blocks_long <<= 3;
258 			break;
259 		default:
260 			break;
261 		}
262 		break;
263 	case 2048:
264 		switch (dev->dev_attrib.block_size) {
265 		case 4096:
266 			blocks_long >>= 1;
267 			break;
268 		case 1024:
269 			blocks_long <<= 1;
270 			break;
271 		case 512:
272 			blocks_long <<= 2;
273 			break;
274 		default:
275 			break;
276 		}
277 		break;
278 	case 1024:
279 		switch (dev->dev_attrib.block_size) {
280 		case 4096:
281 			blocks_long >>= 2;
282 			break;
283 		case 2048:
284 			blocks_long >>= 1;
285 			break;
286 		case 512:
287 			blocks_long <<= 1;
288 			break;
289 		default:
290 			break;
291 		}
292 		break;
293 	case 512:
294 		switch (dev->dev_attrib.block_size) {
295 		case 4096:
296 			blocks_long >>= 3;
297 			break;
298 		case 2048:
299 			blocks_long >>= 2;
300 			break;
301 		case 1024:
302 			blocks_long >>= 1;
303 			break;
304 		default:
305 			break;
306 		}
307 		break;
308 	default:
309 		break;
310 	}
311 
312 	return blocks_long;
313 }
314 
315 static void iblock_complete_cmd(struct se_cmd *cmd, blk_status_t blk_status)
316 {
317 	struct iblock_req *ibr = cmd->priv;
318 	u8 status;
319 
320 	if (!refcount_dec_and_test(&ibr->pending))
321 		return;
322 
323 	if (blk_status == BLK_STS_RESV_CONFLICT)
324 		status = SAM_STAT_RESERVATION_CONFLICT;
325 	else if (atomic_read(&ibr->ib_bio_err_cnt))
326 		status = SAM_STAT_CHECK_CONDITION;
327 	else
328 		status = SAM_STAT_GOOD;
329 
330 	target_complete_cmd(cmd, status);
331 	kfree(ibr);
332 }
333 
334 static void iblock_bio_done(struct bio *bio)
335 {
336 	struct se_cmd *cmd = bio->bi_private;
337 	struct iblock_req *ibr = cmd->priv;
338 	blk_status_t blk_status = bio->bi_status;
339 
340 	if (bio->bi_status) {
341 		pr_err("bio error: %p,  err: %d\n", bio, bio->bi_status);
342 		/*
343 		 * Bump the ib_bio_err_cnt and release bio.
344 		 */
345 		atomic_inc(&ibr->ib_bio_err_cnt);
346 		smp_mb__after_atomic();
347 	}
348 
349 	bio_put(bio);
350 
351 	iblock_complete_cmd(cmd, blk_status);
352 }
353 
354 static struct bio *iblock_get_bio(struct se_cmd *cmd, sector_t lba, u32 sg_num,
355 				  blk_opf_t opf)
356 {
357 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
358 	struct bio *bio;
359 
360 	/*
361 	 * Only allocate as many vector entries as the bio code allows us to,
362 	 * we'll loop later on until we have handled the whole request.
363 	 */
364 	bio = bio_alloc_bioset(ib_dev->ibd_bd, bio_max_segs(sg_num), opf,
365 			       GFP_NOIO, &ib_dev->ibd_bio_set);
366 	if (!bio) {
367 		pr_err("Unable to allocate memory for bio\n");
368 		return NULL;
369 	}
370 
371 	bio->bi_private = cmd;
372 	bio->bi_end_io = &iblock_bio_done;
373 	bio->bi_iter.bi_sector = lba;
374 
375 	return bio;
376 }
377 
378 static void iblock_submit_bios(struct bio_list *list)
379 {
380 	struct blk_plug plug;
381 	struct bio *bio;
382 	/*
383 	 * The block layer handles nested plugs, so just plug/unplug to handle
384 	 * fabric drivers that didn't support batching and multi bio cmds.
385 	 */
386 	blk_start_plug(&plug);
387 	while ((bio = bio_list_pop(list)))
388 		submit_bio(bio);
389 	blk_finish_plug(&plug);
390 }
391 
392 static void iblock_end_io_flush(struct bio *bio)
393 {
394 	struct se_cmd *cmd = bio->bi_private;
395 
396 	if (bio->bi_status)
397 		pr_err("IBLOCK: cache flush failed: %d\n", bio->bi_status);
398 
399 	if (cmd) {
400 		if (bio->bi_status)
401 			target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
402 		else
403 			target_complete_cmd(cmd, SAM_STAT_GOOD);
404 	}
405 
406 	bio_put(bio);
407 }
408 
409 /*
410  * Implement SYCHRONIZE CACHE.  Note that we can't handle lba ranges and must
411  * always flush the whole cache.
412  */
413 static sense_reason_t
414 iblock_execute_sync_cache(struct se_cmd *cmd)
415 {
416 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
417 	int immed = (cmd->t_task_cdb[1] & 0x2);
418 	struct bio *bio;
419 
420 	/*
421 	 * If the Immediate bit is set, queue up the GOOD response
422 	 * for this SYNCHRONIZE_CACHE op.
423 	 */
424 	if (immed)
425 		target_complete_cmd(cmd, SAM_STAT_GOOD);
426 
427 	bio = bio_alloc(ib_dev->ibd_bd, 0, REQ_OP_WRITE | REQ_PREFLUSH,
428 			GFP_KERNEL);
429 	bio->bi_end_io = iblock_end_io_flush;
430 	if (!immed)
431 		bio->bi_private = cmd;
432 	submit_bio(bio);
433 	return 0;
434 }
435 
436 static sense_reason_t
437 iblock_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
438 {
439 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
440 	struct se_device *dev = cmd->se_dev;
441 	int ret;
442 
443 	ret = blkdev_issue_discard(bdev,
444 				   target_to_linux_sector(dev, lba),
445 				   target_to_linux_sector(dev,  nolb),
446 				   GFP_KERNEL);
447 	if (ret < 0) {
448 		pr_err("blkdev_issue_discard() failed: %d\n", ret);
449 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
450 	}
451 
452 	return 0;
453 }
454 
455 static sense_reason_t
456 iblock_execute_zero_out(struct block_device *bdev, struct se_cmd *cmd)
457 {
458 	struct se_device *dev = cmd->se_dev;
459 	struct scatterlist *sg = &cmd->t_data_sg[0];
460 	unsigned char *buf, *not_zero;
461 	int ret;
462 
463 	buf = kmap(sg_page(sg)) + sg->offset;
464 	if (!buf)
465 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
466 	/*
467 	 * Fall back to block_execute_write_same() slow-path if
468 	 * incoming WRITE_SAME payload does not contain zeros.
469 	 */
470 	not_zero = memchr_inv(buf, 0x00, cmd->data_length);
471 	kunmap(sg_page(sg));
472 
473 	if (not_zero)
474 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
475 
476 	ret = blkdev_issue_zeroout(bdev,
477 				target_to_linux_sector(dev, cmd->t_task_lba),
478 				target_to_linux_sector(dev,
479 					sbc_get_write_same_sectors(cmd)),
480 				GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
481 	if (ret)
482 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
483 
484 	target_complete_cmd(cmd, SAM_STAT_GOOD);
485 	return 0;
486 }
487 
488 static sense_reason_t
489 iblock_execute_write_same(struct se_cmd *cmd)
490 {
491 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
492 	struct iblock_req *ibr;
493 	struct scatterlist *sg;
494 	struct bio *bio;
495 	struct bio_list list;
496 	struct se_device *dev = cmd->se_dev;
497 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
498 	sector_t sectors = target_to_linux_sector(dev,
499 					sbc_get_write_same_sectors(cmd));
500 
501 	if (cmd->prot_op) {
502 		pr_err("WRITE_SAME: Protection information with IBLOCK"
503 		       " backends not supported\n");
504 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
505 	}
506 
507 	if (!cmd->t_data_nents)
508 		return TCM_INVALID_CDB_FIELD;
509 
510 	sg = &cmd->t_data_sg[0];
511 
512 	if (cmd->t_data_nents > 1 ||
513 	    sg->length != cmd->se_dev->dev_attrib.block_size) {
514 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
515 			" block_size: %u\n", cmd->t_data_nents, sg->length,
516 			cmd->se_dev->dev_attrib.block_size);
517 		return TCM_INVALID_CDB_FIELD;
518 	}
519 
520 	if (bdev_write_zeroes_sectors(bdev)) {
521 		if (!iblock_execute_zero_out(bdev, cmd))
522 			return 0;
523 	}
524 
525 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
526 	if (!ibr)
527 		goto fail;
528 	cmd->priv = ibr;
529 
530 	bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
531 	if (!bio)
532 		goto fail_free_ibr;
533 
534 	bio_list_init(&list);
535 	bio_list_add(&list, bio);
536 
537 	refcount_set(&ibr->pending, 1);
538 
539 	while (sectors) {
540 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
541 				!= sg->length) {
542 
543 			bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
544 			if (!bio)
545 				goto fail_put_bios;
546 
547 			refcount_inc(&ibr->pending);
548 			bio_list_add(&list, bio);
549 		}
550 
551 		/* Always in 512 byte units for Linux/Block */
552 		block_lba += sg->length >> SECTOR_SHIFT;
553 		sectors -= sg->length >> SECTOR_SHIFT;
554 	}
555 
556 	iblock_submit_bios(&list);
557 	return 0;
558 
559 fail_put_bios:
560 	while ((bio = bio_list_pop(&list)))
561 		bio_put(bio);
562 fail_free_ibr:
563 	kfree(ibr);
564 fail:
565 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
566 }
567 
568 enum {
569 	Opt_udev_path, Opt_readonly, Opt_force, Opt_err
570 };
571 
572 static match_table_t tokens = {
573 	{Opt_udev_path, "udev_path=%s"},
574 	{Opt_readonly, "readonly=%d"},
575 	{Opt_force, "force=%d"},
576 	{Opt_err, NULL}
577 };
578 
579 static ssize_t iblock_set_configfs_dev_params(struct se_device *dev,
580 		const char *page, ssize_t count)
581 {
582 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
583 	char *orig, *ptr, *arg_p, *opts;
584 	substring_t args[MAX_OPT_ARGS];
585 	int ret = 0, token;
586 	unsigned long tmp_readonly;
587 
588 	opts = kstrdup(page, GFP_KERNEL);
589 	if (!opts)
590 		return -ENOMEM;
591 
592 	orig = opts;
593 
594 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
595 		if (!*ptr)
596 			continue;
597 
598 		token = match_token(ptr, tokens, args);
599 		switch (token) {
600 		case Opt_udev_path:
601 			if (ib_dev->ibd_bd) {
602 				pr_err("Unable to set udev_path= while"
603 					" ib_dev->ibd_bd exists\n");
604 				ret = -EEXIST;
605 				goto out;
606 			}
607 			if (match_strlcpy(ib_dev->ibd_udev_path, &args[0],
608 				SE_UDEV_PATH_LEN) == 0) {
609 				ret = -EINVAL;
610 				break;
611 			}
612 			pr_debug("IBLOCK: Referencing UDEV path: %s\n",
613 					ib_dev->ibd_udev_path);
614 			ib_dev->ibd_flags |= IBDF_HAS_UDEV_PATH;
615 			break;
616 		case Opt_readonly:
617 			arg_p = match_strdup(&args[0]);
618 			if (!arg_p) {
619 				ret = -ENOMEM;
620 				break;
621 			}
622 			ret = kstrtoul(arg_p, 0, &tmp_readonly);
623 			kfree(arg_p);
624 			if (ret < 0) {
625 				pr_err("kstrtoul() failed for"
626 						" readonly=\n");
627 				goto out;
628 			}
629 			ib_dev->ibd_readonly = tmp_readonly;
630 			pr_debug("IBLOCK: readonly: %d\n", ib_dev->ibd_readonly);
631 			break;
632 		case Opt_force:
633 			break;
634 		default:
635 			break;
636 		}
637 	}
638 
639 out:
640 	kfree(orig);
641 	return (!ret) ? count : ret;
642 }
643 
644 static ssize_t iblock_show_configfs_dev_params(struct se_device *dev, char *b)
645 {
646 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
647 	struct block_device *bd = ib_dev->ibd_bd;
648 	ssize_t bl = 0;
649 
650 	if (bd)
651 		bl += sprintf(b + bl, "iBlock device: %pg", bd);
652 	if (ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)
653 		bl += sprintf(b + bl, "  UDEV PATH: %s",
654 				ib_dev->ibd_udev_path);
655 	bl += sprintf(b + bl, "  readonly: %d\n", ib_dev->ibd_readonly);
656 
657 	bl += sprintf(b + bl, "        ");
658 	if (bd) {
659 		bl += sprintf(b + bl, "Major: %d Minor: %d  %s\n",
660 			MAJOR(bd->bd_dev), MINOR(bd->bd_dev),
661 			"CLAIMED: IBLOCK");
662 	} else {
663 		bl += sprintf(b + bl, "Major: 0 Minor: 0\n");
664 	}
665 
666 	return bl;
667 }
668 
669 static int
670 iblock_alloc_bip(struct se_cmd *cmd, struct bio *bio,
671 		 struct sg_mapping_iter *miter)
672 {
673 	struct se_device *dev = cmd->se_dev;
674 	struct blk_integrity *bi;
675 	struct bio_integrity_payload *bip;
676 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
677 	int rc;
678 	size_t resid, len;
679 
680 	bi = bdev_get_integrity(ib_dev->ibd_bd);
681 	if (!bi) {
682 		pr_err("Unable to locate bio_integrity\n");
683 		return -ENODEV;
684 	}
685 
686 	bip = bio_integrity_alloc(bio, GFP_NOIO, bio_max_segs(cmd->t_prot_nents));
687 	if (IS_ERR(bip)) {
688 		pr_err("Unable to allocate bio_integrity_payload\n");
689 		return PTR_ERR(bip);
690 	}
691 
692 	/* virtual start sector must be in integrity interval units */
693 	bip_set_seed(bip, bio->bi_iter.bi_sector >>
694 				  (bi->interval_exp - SECTOR_SHIFT));
695 
696 	pr_debug("IBLOCK BIP Size: %u Sector: %llu\n", bip->bip_iter.bi_size,
697 		 (unsigned long long)bip->bip_iter.bi_sector);
698 
699 	resid = bio_integrity_bytes(bi, bio_sectors(bio));
700 	while (resid > 0 && sg_miter_next(miter)) {
701 
702 		len = min_t(size_t, miter->length, resid);
703 		rc = bio_integrity_add_page(bio, miter->page, len,
704 					    offset_in_page(miter->addr));
705 		if (rc != len) {
706 			pr_err("bio_integrity_add_page() failed; %d\n", rc);
707 			sg_miter_stop(miter);
708 			return -ENOMEM;
709 		}
710 
711 		pr_debug("Added bio integrity page: %p length: %zu offset: %lu\n",
712 			  miter->page, len, offset_in_page(miter->addr));
713 
714 		resid -= len;
715 		if (len < miter->length)
716 			miter->consumed -= miter->length - len;
717 	}
718 	sg_miter_stop(miter);
719 
720 	return 0;
721 }
722 
723 static sense_reason_t
724 iblock_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
725 		  enum dma_data_direction data_direction)
726 {
727 	struct se_device *dev = cmd->se_dev;
728 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
729 	struct iblock_req *ibr;
730 	struct bio *bio;
731 	struct bio_list list;
732 	struct scatterlist *sg;
733 	u32 sg_num = sgl_nents;
734 	blk_opf_t opf;
735 	unsigned bio_cnt;
736 	int i, rc;
737 	struct sg_mapping_iter prot_miter;
738 	unsigned int miter_dir;
739 
740 	if (data_direction == DMA_TO_DEVICE) {
741 		struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
742 
743 		/*
744 		 * Set bits to indicate WRITE_ODIRECT so we are not throttled
745 		 * by WBT.
746 		 */
747 		opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
748 		/*
749 		 * Force writethrough using REQ_FUA if a volatile write cache
750 		 * is not enabled, or if initiator set the Force Unit Access bit.
751 		 */
752 		miter_dir = SG_MITER_TO_SG;
753 		if (bdev_fua(ib_dev->ibd_bd)) {
754 			if (cmd->se_cmd_flags & SCF_FUA)
755 				opf |= REQ_FUA;
756 			else if (!bdev_write_cache(ib_dev->ibd_bd))
757 				opf |= REQ_FUA;
758 		}
759 	} else {
760 		opf = REQ_OP_READ;
761 		miter_dir = SG_MITER_FROM_SG;
762 	}
763 
764 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
765 	if (!ibr)
766 		goto fail;
767 	cmd->priv = ibr;
768 
769 	if (!sgl_nents) {
770 		refcount_set(&ibr->pending, 1);
771 		iblock_complete_cmd(cmd, BLK_STS_OK);
772 		return 0;
773 	}
774 
775 	bio = iblock_get_bio(cmd, block_lba, sgl_nents, opf);
776 	if (!bio)
777 		goto fail_free_ibr;
778 
779 	bio_list_init(&list);
780 	bio_list_add(&list, bio);
781 
782 	refcount_set(&ibr->pending, 2);
783 	bio_cnt = 1;
784 
785 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type)
786 		sg_miter_start(&prot_miter, cmd->t_prot_sg, cmd->t_prot_nents,
787 			       miter_dir);
788 
789 	for_each_sg(sgl, sg, sgl_nents, i) {
790 		/*
791 		 * XXX: if the length the device accepts is shorter than the
792 		 *	length of the S/G list entry this will cause and
793 		 *	endless loop.  Better hope no driver uses huge pages.
794 		 */
795 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
796 				!= sg->length) {
797 			if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
798 				rc = iblock_alloc_bip(cmd, bio, &prot_miter);
799 				if (rc)
800 					goto fail_put_bios;
801 			}
802 
803 			if (bio_cnt >= IBLOCK_MAX_BIO_PER_TASK) {
804 				iblock_submit_bios(&list);
805 				bio_cnt = 0;
806 			}
807 
808 			bio = iblock_get_bio(cmd, block_lba, sg_num, opf);
809 			if (!bio)
810 				goto fail_put_bios;
811 
812 			refcount_inc(&ibr->pending);
813 			bio_list_add(&list, bio);
814 			bio_cnt++;
815 		}
816 
817 		/* Always in 512 byte units for Linux/Block */
818 		block_lba += sg->length >> SECTOR_SHIFT;
819 		sg_num--;
820 	}
821 
822 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
823 		rc = iblock_alloc_bip(cmd, bio, &prot_miter);
824 		if (rc)
825 			goto fail_put_bios;
826 	}
827 
828 	iblock_submit_bios(&list);
829 	iblock_complete_cmd(cmd, BLK_STS_OK);
830 	return 0;
831 
832 fail_put_bios:
833 	while ((bio = bio_list_pop(&list)))
834 		bio_put(bio);
835 fail_free_ibr:
836 	kfree(ibr);
837 fail:
838 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
839 }
840 
841 static sense_reason_t iblock_execute_pr_out(struct se_cmd *cmd, u8 sa, u64 key,
842 					    u64 sa_key, u8 type, bool aptpl)
843 {
844 	struct se_device *dev = cmd->se_dev;
845 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
846 	struct block_device *bdev = ib_dev->ibd_bd;
847 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
848 	int ret;
849 
850 	if (!ops) {
851 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
852 		return TCM_UNSUPPORTED_SCSI_OPCODE;
853 	}
854 
855 	switch (sa) {
856 	case PRO_REGISTER:
857 	case PRO_REGISTER_AND_IGNORE_EXISTING_KEY:
858 		if (!ops->pr_register) {
859 			pr_err("block device does not support pr_register.\n");
860 			return TCM_UNSUPPORTED_SCSI_OPCODE;
861 		}
862 
863 		/* The block layer pr ops always enables aptpl */
864 		if (!aptpl)
865 			pr_info("APTPL not set by initiator, but will be used.\n");
866 
867 		ret = ops->pr_register(bdev, key, sa_key,
868 				sa == PRO_REGISTER ? 0 : PR_FL_IGNORE_KEY);
869 		break;
870 	case PRO_RESERVE:
871 		if (!ops->pr_reserve) {
872 			pr_err("block_device does not support pr_reserve.\n");
873 			return TCM_UNSUPPORTED_SCSI_OPCODE;
874 		}
875 
876 		ret = ops->pr_reserve(bdev, key, scsi_pr_type_to_block(type), 0);
877 		break;
878 	case PRO_CLEAR:
879 		if (!ops->pr_clear) {
880 			pr_err("block_device does not support pr_clear.\n");
881 			return TCM_UNSUPPORTED_SCSI_OPCODE;
882 		}
883 
884 		ret = ops->pr_clear(bdev, key);
885 		break;
886 	case PRO_PREEMPT:
887 	case PRO_PREEMPT_AND_ABORT:
888 		if (!ops->pr_clear) {
889 			pr_err("block_device does not support pr_preempt.\n");
890 			return TCM_UNSUPPORTED_SCSI_OPCODE;
891 		}
892 
893 		ret = ops->pr_preempt(bdev, key, sa_key,
894 				      scsi_pr_type_to_block(type),
895 				      sa == PRO_PREEMPT_AND_ABORT);
896 		break;
897 	case PRO_RELEASE:
898 		if (!ops->pr_clear) {
899 			pr_err("block_device does not support pr_pclear.\n");
900 			return TCM_UNSUPPORTED_SCSI_OPCODE;
901 		}
902 
903 		ret = ops->pr_release(bdev, key, scsi_pr_type_to_block(type));
904 		break;
905 	default:
906 		pr_err("Unknown PERSISTENT_RESERVE_OUT SA: 0x%02x\n", sa);
907 		return TCM_UNSUPPORTED_SCSI_OPCODE;
908 	}
909 
910 	if (!ret)
911 		return TCM_NO_SENSE;
912 	else if (ret == PR_STS_RESERVATION_CONFLICT)
913 		return TCM_RESERVATION_CONFLICT;
914 	else
915 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
916 }
917 
918 static void iblock_pr_report_caps(unsigned char *param_data)
919 {
920 	u16 len = 8;
921 
922 	put_unaligned_be16(len, &param_data[0]);
923 	/*
924 	 * When using the pr_ops passthrough method we only support exporting
925 	 * the device through one target port because from the backend module
926 	 * level we can't see the target port config. As a result we only
927 	 * support registration directly from the I_T nexus the cmd is sent
928 	 * through and do not set ATP_C here.
929 	 *
930 	 * The block layer pr_ops do not support passing in initiators so
931 	 * we don't set SIP_C here.
932 	 */
933 	/* PTPL_C: Persistence across Target Power Loss bit */
934 	param_data[2] |= 0x01;
935 	/*
936 	 * We are filling in the PERSISTENT RESERVATION TYPE MASK below, so
937 	 * set the TMV: Task Mask Valid bit.
938 	 */
939 	param_data[3] |= 0x80;
940 	/*
941 	 * Change ALLOW COMMANDs to 0x20 or 0x40 later from Table 166
942 	 */
943 	param_data[3] |= 0x10; /* ALLOW COMMANDs field 001b */
944 	/*
945 	 * PTPL_A: Persistence across Target Power Loss Active bit. The block
946 	 * layer pr ops always enables this so report it active.
947 	 */
948 	param_data[3] |= 0x01;
949 	/*
950 	 * Setup the PERSISTENT RESERVATION TYPE MASK from Table 212 spc4r37.
951 	 */
952 	param_data[4] |= 0x80; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
953 	param_data[4] |= 0x40; /* PR_TYPE_EXCLUSIVE_ACCESS_REGONLY */
954 	param_data[4] |= 0x20; /* PR_TYPE_WRITE_EXCLUSIVE_REGONLY */
955 	param_data[4] |= 0x08; /* PR_TYPE_EXCLUSIVE_ACCESS */
956 	param_data[4] |= 0x02; /* PR_TYPE_WRITE_EXCLUSIVE */
957 	param_data[5] |= 0x01; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
958 }
959 
960 static sense_reason_t iblock_pr_read_keys(struct se_cmd *cmd,
961 					  unsigned char *param_data)
962 {
963 	struct se_device *dev = cmd->se_dev;
964 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
965 	struct block_device *bdev = ib_dev->ibd_bd;
966 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
967 	int i, len, paths, data_offset;
968 	struct pr_keys *keys;
969 	sense_reason_t ret;
970 
971 	if (!ops) {
972 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
973 		return TCM_UNSUPPORTED_SCSI_OPCODE;
974 	}
975 
976 	if (!ops->pr_read_keys) {
977 		pr_err("Block device does not support read_keys.\n");
978 		return TCM_UNSUPPORTED_SCSI_OPCODE;
979 	}
980 
981 	/*
982 	 * We don't know what's under us, but dm-multipath will register every
983 	 * path with the same key, so start off with enough space for 16 paths.
984 	 * which is not a lot of memory and should normally be enough.
985 	 */
986 	paths = 16;
987 retry:
988 	len = 8 * paths;
989 	keys = kzalloc(sizeof(*keys) + len, GFP_KERNEL);
990 	if (!keys)
991 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
992 
993 	keys->num_keys = paths;
994 	if (!ops->pr_read_keys(bdev, keys)) {
995 		if (keys->num_keys > paths) {
996 			kfree(keys);
997 			paths *= 2;
998 			goto retry;
999 		}
1000 	} else {
1001 		ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1002 		goto free_keys;
1003 	}
1004 
1005 	ret = TCM_NO_SENSE;
1006 
1007 	put_unaligned_be32(keys->generation, &param_data[0]);
1008 	if (!keys->num_keys) {
1009 		put_unaligned_be32(0, &param_data[4]);
1010 		goto free_keys;
1011 	}
1012 
1013 	put_unaligned_be32(8 * keys->num_keys, &param_data[4]);
1014 
1015 	data_offset = 8;
1016 	for (i = 0; i < keys->num_keys; i++) {
1017 		if (data_offset + 8 > cmd->data_length)
1018 			break;
1019 
1020 		put_unaligned_be64(keys->keys[i], &param_data[data_offset]);
1021 		data_offset += 8;
1022 	}
1023 
1024 free_keys:
1025 	kfree(keys);
1026 	return ret;
1027 }
1028 
1029 static sense_reason_t iblock_pr_read_reservation(struct se_cmd *cmd,
1030 						 unsigned char *param_data)
1031 {
1032 	struct se_device *dev = cmd->se_dev;
1033 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1034 	struct block_device *bdev = ib_dev->ibd_bd;
1035 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
1036 	struct pr_held_reservation rsv = { };
1037 
1038 	if (!ops) {
1039 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
1040 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1041 	}
1042 
1043 	if (!ops->pr_read_reservation) {
1044 		pr_err("Block device does not support read_keys.\n");
1045 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1046 	}
1047 
1048 	if (ops->pr_read_reservation(bdev, &rsv))
1049 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1050 
1051 	put_unaligned_be32(rsv.generation, &param_data[0]);
1052 	if (!block_pr_type_to_scsi(rsv.type)) {
1053 		put_unaligned_be32(0, &param_data[4]);
1054 		return TCM_NO_SENSE;
1055 	}
1056 
1057 	put_unaligned_be32(16, &param_data[4]);
1058 
1059 	if (cmd->data_length < 16)
1060 		return TCM_NO_SENSE;
1061 	put_unaligned_be64(rsv.key, &param_data[8]);
1062 
1063 	if (cmd->data_length < 22)
1064 		return TCM_NO_SENSE;
1065 	param_data[21] = block_pr_type_to_scsi(rsv.type);
1066 
1067 	return TCM_NO_SENSE;
1068 }
1069 
1070 static sense_reason_t iblock_execute_pr_in(struct se_cmd *cmd, u8 sa,
1071 					   unsigned char *param_data)
1072 {
1073 	sense_reason_t ret = TCM_NO_SENSE;
1074 
1075 	switch (sa) {
1076 	case PRI_REPORT_CAPABILITIES:
1077 		iblock_pr_report_caps(param_data);
1078 		break;
1079 	case PRI_READ_KEYS:
1080 		ret = iblock_pr_read_keys(cmd, param_data);
1081 		break;
1082 	case PRI_READ_RESERVATION:
1083 		ret = iblock_pr_read_reservation(cmd, param_data);
1084 		break;
1085 	default:
1086 		pr_err("Unknown PERSISTENT_RESERVE_IN SA: 0x%02x\n", sa);
1087 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1088 	}
1089 
1090 	return ret;
1091 }
1092 
1093 static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
1094 {
1095 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1096 	struct block_device *bd = ib_dev->ibd_bd;
1097 	int ret;
1098 
1099 	ret = bdev_alignment_offset(bd);
1100 	if (ret == -1)
1101 		return 0;
1102 
1103 	/* convert offset-bytes to offset-lbas */
1104 	return ret / bdev_logical_block_size(bd);
1105 }
1106 
1107 static unsigned int iblock_get_lbppbe(struct se_device *dev)
1108 {
1109 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1110 	struct block_device *bd = ib_dev->ibd_bd;
1111 	unsigned int logs_per_phys =
1112 		bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
1113 
1114 	return ilog2(logs_per_phys);
1115 }
1116 
1117 static unsigned int iblock_get_io_min(struct se_device *dev)
1118 {
1119 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1120 	struct block_device *bd = ib_dev->ibd_bd;
1121 
1122 	return bdev_io_min(bd);
1123 }
1124 
1125 static unsigned int iblock_get_io_opt(struct se_device *dev)
1126 {
1127 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1128 	struct block_device *bd = ib_dev->ibd_bd;
1129 
1130 	return bdev_io_opt(bd);
1131 }
1132 
1133 static struct exec_cmd_ops iblock_exec_cmd_ops = {
1134 	.execute_rw		= iblock_execute_rw,
1135 	.execute_sync_cache	= iblock_execute_sync_cache,
1136 	.execute_write_same	= iblock_execute_write_same,
1137 	.execute_unmap		= iblock_execute_unmap,
1138 	.execute_pr_out		= iblock_execute_pr_out,
1139 	.execute_pr_in		= iblock_execute_pr_in,
1140 };
1141 
1142 static sense_reason_t
1143 iblock_parse_cdb(struct se_cmd *cmd)
1144 {
1145 	return sbc_parse_cdb(cmd, &iblock_exec_cmd_ops);
1146 }
1147 
1148 static bool iblock_get_write_cache(struct se_device *dev)
1149 {
1150 	return bdev_write_cache(IBLOCK_DEV(dev)->ibd_bd);
1151 }
1152 
1153 static const struct target_backend_ops iblock_ops = {
1154 	.name			= "iblock",
1155 	.inquiry_prod		= "IBLOCK",
1156 	.transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR,
1157 	.inquiry_rev		= IBLOCK_VERSION,
1158 	.owner			= THIS_MODULE,
1159 	.attach_hba		= iblock_attach_hba,
1160 	.detach_hba		= iblock_detach_hba,
1161 	.alloc_device		= iblock_alloc_device,
1162 	.configure_device	= iblock_configure_device,
1163 	.destroy_device		= iblock_destroy_device,
1164 	.free_device		= iblock_free_device,
1165 	.configure_unmap	= iblock_configure_unmap,
1166 	.plug_device		= iblock_plug_device,
1167 	.unplug_device		= iblock_unplug_device,
1168 	.parse_cdb		= iblock_parse_cdb,
1169 	.set_configfs_dev_params = iblock_set_configfs_dev_params,
1170 	.show_configfs_dev_params = iblock_show_configfs_dev_params,
1171 	.get_device_type	= sbc_get_device_type,
1172 	.get_blocks		= iblock_get_blocks,
1173 	.get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
1174 	.get_lbppbe		= iblock_get_lbppbe,
1175 	.get_io_min		= iblock_get_io_min,
1176 	.get_io_opt		= iblock_get_io_opt,
1177 	.get_write_cache	= iblock_get_write_cache,
1178 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
1179 };
1180 
1181 static int __init iblock_module_init(void)
1182 {
1183 	return transport_backend_register(&iblock_ops);
1184 }
1185 
1186 static void __exit iblock_module_exit(void)
1187 {
1188 	target_backend_unregister(&iblock_ops);
1189 }
1190 
1191 MODULE_DESCRIPTION("TCM IBLOCK subsystem plugin");
1192 MODULE_AUTHOR("nab@Linux-iSCSI.org");
1193 MODULE_LICENSE("GPL");
1194 
1195 module_init(iblock_module_init);
1196 module_exit(iblock_module_exit);
1197