xref: /linux/drivers/target/target_core_iblock.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 <linux/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 file *bdev_file;
95 	struct block_device *bd;
96 	struct blk_integrity *bi;
97 	blk_mode_t mode = BLK_OPEN_READ;
98 	unsigned int max_write_zeroes_sectors;
99 	int ret;
100 
101 	if (!(ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)) {
102 		pr_err("Missing udev_path= parameters for IBLOCK\n");
103 		return -EINVAL;
104 	}
105 
106 	ret = bioset_init(&ib_dev->ibd_bio_set, IBLOCK_BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
107 	if (ret) {
108 		pr_err("IBLOCK: Unable to create bioset\n");
109 		goto out;
110 	}
111 
112 	pr_debug( "IBLOCK: Claiming struct block_device: %s\n",
113 			ib_dev->ibd_udev_path);
114 
115 	if (!ib_dev->ibd_readonly)
116 		mode |= BLK_OPEN_WRITE;
117 	else
118 		dev->dev_flags |= DF_READ_ONLY;
119 
120 	bdev_file = bdev_file_open_by_path(ib_dev->ibd_udev_path, mode, ib_dev,
121 					NULL);
122 	if (IS_ERR(bdev_file)) {
123 		ret = PTR_ERR(bdev_file);
124 		goto out_free_bioset;
125 	}
126 	ib_dev->ibd_bdev_file = bdev_file;
127 	ib_dev->ibd_bd = bd = file_bdev(bdev_file);
128 
129 	q = bdev_get_queue(bd);
130 
131 	dev->dev_attrib.hw_block_size = bdev_logical_block_size(bd);
132 	dev->dev_attrib.hw_max_sectors = mult_frac(queue_max_hw_sectors(q),
133 			SECTOR_SIZE,
134 			dev->dev_attrib.hw_block_size);
135 	dev->dev_attrib.hw_queue_depth = q->nr_requests;
136 
137 	/*
138 	 * Enable write same emulation for IBLOCK and use 0xFFFF as
139 	 * the smaller WRITE_SAME(10) only has a two-byte block count.
140 	 */
141 	max_write_zeroes_sectors = bdev_write_zeroes_sectors(bd);
142 	if (max_write_zeroes_sectors)
143 		dev->dev_attrib.max_write_same_len = max_write_zeroes_sectors;
144 	else
145 		dev->dev_attrib.max_write_same_len = 0xFFFF;
146 
147 	if (bdev_nonrot(bd))
148 		dev->dev_attrib.is_nonrot = 1;
149 
150 	bi = bdev_get_integrity(bd);
151 	if (!bi)
152 		return 0;
153 
154 	switch (bi->csum_type) {
155 	case BLK_INTEGRITY_CSUM_IP:
156 		pr_err("IBLOCK export of blk_integrity: %s not supported\n",
157 			blk_integrity_profile_name(bi));
158 		ret = -ENOSYS;
159 		goto out_blkdev_put;
160 	case BLK_INTEGRITY_CSUM_CRC:
161 		if (bi->flags & BLK_INTEGRITY_REF_TAG)
162 			dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE1_PROT;
163 		else
164 			dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE3_PROT;
165 		break;
166 	default:
167 		break;
168 	}
169 
170 	if (dev->dev_attrib.pi_prot_type) {
171 		struct bio_set *bs = &ib_dev->ibd_bio_set;
172 
173 		if (bioset_integrity_create(bs, IBLOCK_BIO_POOL_SIZE) < 0) {
174 			pr_err("Unable to allocate bioset for PI\n");
175 			ret = -ENOMEM;
176 			goto out_blkdev_put;
177 		}
178 		pr_debug("IBLOCK setup BIP bs->bio_integrity_pool: %p\n",
179 			 &bs->bio_integrity_pool);
180 	}
181 
182 	dev->dev_attrib.hw_pi_prot_type = dev->dev_attrib.pi_prot_type;
183 	return 0;
184 
185 out_blkdev_put:
186 	fput(ib_dev->ibd_bdev_file);
187 out_free_bioset:
188 	bioset_exit(&ib_dev->ibd_bio_set);
189 out:
190 	return ret;
191 }
192 
193 static void iblock_dev_call_rcu(struct rcu_head *p)
194 {
195 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
196 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
197 
198 	kfree(ib_dev->ibd_plug);
199 	kfree(ib_dev);
200 }
201 
202 static void iblock_free_device(struct se_device *dev)
203 {
204 	call_rcu(&dev->rcu_head, iblock_dev_call_rcu);
205 }
206 
207 static void iblock_destroy_device(struct se_device *dev)
208 {
209 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
210 
211 	if (ib_dev->ibd_bdev_file)
212 		fput(ib_dev->ibd_bdev_file);
213 	bioset_exit(&ib_dev->ibd_bio_set);
214 }
215 
216 static struct se_dev_plug *iblock_plug_device(struct se_device *se_dev)
217 {
218 	struct iblock_dev *ib_dev = IBLOCK_DEV(se_dev);
219 	struct iblock_dev_plug *ib_dev_plug;
220 
221 	/*
222 	 * Each se_device has a per cpu work this can be run from. We
223 	 * shouldn't have multiple threads on the same cpu calling this
224 	 * at the same time.
225 	 */
226 	ib_dev_plug = &ib_dev->ibd_plug[raw_smp_processor_id()];
227 	if (test_and_set_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags))
228 		return NULL;
229 
230 	blk_start_plug(&ib_dev_plug->blk_plug);
231 	return &ib_dev_plug->se_plug;
232 }
233 
234 static void iblock_unplug_device(struct se_dev_plug *se_plug)
235 {
236 	struct iblock_dev_plug *ib_dev_plug = container_of(se_plug,
237 					struct iblock_dev_plug, se_plug);
238 
239 	blk_finish_plug(&ib_dev_plug->blk_plug);
240 	clear_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags);
241 }
242 
243 static sector_t iblock_get_blocks(struct se_device *dev)
244 {
245 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
246 	u32 block_size = bdev_logical_block_size(ib_dev->ibd_bd);
247 	unsigned long long blocks_long =
248 		div_u64(bdev_nr_bytes(ib_dev->ibd_bd), block_size) - 1;
249 
250 	if (block_size == dev->dev_attrib.block_size)
251 		return blocks_long;
252 
253 	switch (block_size) {
254 	case 4096:
255 		switch (dev->dev_attrib.block_size) {
256 		case 2048:
257 			blocks_long <<= 1;
258 			break;
259 		case 1024:
260 			blocks_long <<= 2;
261 			break;
262 		case 512:
263 			blocks_long <<= 3;
264 			break;
265 		default:
266 			break;
267 		}
268 		break;
269 	case 2048:
270 		switch (dev->dev_attrib.block_size) {
271 		case 4096:
272 			blocks_long >>= 1;
273 			break;
274 		case 1024:
275 			blocks_long <<= 1;
276 			break;
277 		case 512:
278 			blocks_long <<= 2;
279 			break;
280 		default:
281 			break;
282 		}
283 		break;
284 	case 1024:
285 		switch (dev->dev_attrib.block_size) {
286 		case 4096:
287 			blocks_long >>= 2;
288 			break;
289 		case 2048:
290 			blocks_long >>= 1;
291 			break;
292 		case 512:
293 			blocks_long <<= 1;
294 			break;
295 		default:
296 			break;
297 		}
298 		break;
299 	case 512:
300 		switch (dev->dev_attrib.block_size) {
301 		case 4096:
302 			blocks_long >>= 3;
303 			break;
304 		case 2048:
305 			blocks_long >>= 2;
306 			break;
307 		case 1024:
308 			blocks_long >>= 1;
309 			break;
310 		default:
311 			break;
312 		}
313 		break;
314 	default:
315 		break;
316 	}
317 
318 	return blocks_long;
319 }
320 
321 static void iblock_complete_cmd(struct se_cmd *cmd, blk_status_t blk_status)
322 {
323 	struct iblock_req *ibr = cmd->priv;
324 	u8 status;
325 
326 	if (!refcount_dec_and_test(&ibr->pending))
327 		return;
328 
329 	if (blk_status == BLK_STS_RESV_CONFLICT)
330 		status = SAM_STAT_RESERVATION_CONFLICT;
331 	else if (atomic_read(&ibr->ib_bio_err_cnt))
332 		status = SAM_STAT_CHECK_CONDITION;
333 	else
334 		status = SAM_STAT_GOOD;
335 
336 	target_complete_cmd(cmd, status);
337 	kfree(ibr);
338 }
339 
340 static void iblock_bio_done(struct bio *bio)
341 {
342 	struct se_cmd *cmd = bio->bi_private;
343 	struct iblock_req *ibr = cmd->priv;
344 	blk_status_t blk_status = bio->bi_status;
345 
346 	if (bio->bi_status) {
347 		pr_err("bio error: %p,  err: %d\n", bio, bio->bi_status);
348 		/*
349 		 * Bump the ib_bio_err_cnt and release bio.
350 		 */
351 		atomic_inc(&ibr->ib_bio_err_cnt);
352 		smp_mb__after_atomic();
353 	}
354 
355 	bio_put(bio);
356 
357 	iblock_complete_cmd(cmd, blk_status);
358 }
359 
360 static struct bio *iblock_get_bio(struct se_cmd *cmd, sector_t lba, u32 sg_num,
361 				  blk_opf_t opf)
362 {
363 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
364 	struct bio *bio;
365 
366 	/*
367 	 * Only allocate as many vector entries as the bio code allows us to,
368 	 * we'll loop later on until we have handled the whole request.
369 	 */
370 	bio = bio_alloc_bioset(ib_dev->ibd_bd, bio_max_segs(sg_num), opf,
371 			       GFP_NOIO, &ib_dev->ibd_bio_set);
372 	if (!bio) {
373 		pr_err("Unable to allocate memory for bio\n");
374 		return NULL;
375 	}
376 
377 	bio->bi_private = cmd;
378 	bio->bi_end_io = &iblock_bio_done;
379 	bio->bi_iter.bi_sector = lba;
380 
381 	return bio;
382 }
383 
384 static void iblock_submit_bios(struct bio_list *list)
385 {
386 	struct blk_plug plug;
387 	struct bio *bio;
388 	/*
389 	 * The block layer handles nested plugs, so just plug/unplug to handle
390 	 * fabric drivers that didn't support batching and multi bio cmds.
391 	 */
392 	blk_start_plug(&plug);
393 	while ((bio = bio_list_pop(list)))
394 		submit_bio(bio);
395 	blk_finish_plug(&plug);
396 }
397 
398 static void iblock_end_io_flush(struct bio *bio)
399 {
400 	struct se_cmd *cmd = bio->bi_private;
401 
402 	if (bio->bi_status)
403 		pr_err("IBLOCK: cache flush failed: %d\n", bio->bi_status);
404 
405 	if (cmd) {
406 		if (bio->bi_status)
407 			target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
408 		else
409 			target_complete_cmd(cmd, SAM_STAT_GOOD);
410 	}
411 
412 	bio_put(bio);
413 }
414 
415 /*
416  * Implement SYCHRONIZE CACHE.  Note that we can't handle lba ranges and must
417  * always flush the whole cache.
418  */
419 static sense_reason_t
420 iblock_execute_sync_cache(struct se_cmd *cmd)
421 {
422 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
423 	int immed = (cmd->t_task_cdb[1] & 0x2);
424 	struct bio *bio;
425 
426 	/*
427 	 * If the Immediate bit is set, queue up the GOOD response
428 	 * for this SYNCHRONIZE_CACHE op.
429 	 */
430 	if (immed)
431 		target_complete_cmd(cmd, SAM_STAT_GOOD);
432 
433 	bio = bio_alloc(ib_dev->ibd_bd, 0, REQ_OP_WRITE | REQ_PREFLUSH,
434 			GFP_KERNEL);
435 	bio->bi_end_io = iblock_end_io_flush;
436 	if (!immed)
437 		bio->bi_private = cmd;
438 	submit_bio(bio);
439 	return 0;
440 }
441 
442 static sense_reason_t
443 iblock_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
444 {
445 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
446 	struct se_device *dev = cmd->se_dev;
447 	int ret;
448 
449 	ret = blkdev_issue_discard(bdev,
450 				   target_to_linux_sector(dev, lba),
451 				   target_to_linux_sector(dev,  nolb),
452 				   GFP_KERNEL);
453 	if (ret < 0) {
454 		pr_err("blkdev_issue_discard() failed: %d\n", ret);
455 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
456 	}
457 
458 	return 0;
459 }
460 
461 static sense_reason_t
462 iblock_execute_zero_out(struct block_device *bdev, struct se_cmd *cmd)
463 {
464 	struct se_device *dev = cmd->se_dev;
465 	struct scatterlist *sg = &cmd->t_data_sg[0];
466 	unsigned char *buf, *not_zero;
467 	int ret;
468 
469 	buf = kmap(sg_page(sg)) + sg->offset;
470 	if (!buf)
471 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
472 	/*
473 	 * Fall back to block_execute_write_same() slow-path if
474 	 * incoming WRITE_SAME payload does not contain zeros.
475 	 */
476 	not_zero = memchr_inv(buf, 0x00, cmd->data_length);
477 	kunmap(sg_page(sg));
478 
479 	if (not_zero)
480 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
481 
482 	ret = blkdev_issue_zeroout(bdev,
483 				target_to_linux_sector(dev, cmd->t_task_lba),
484 				target_to_linux_sector(dev,
485 					sbc_get_write_same_sectors(cmd)),
486 				GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
487 	if (ret)
488 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
489 
490 	target_complete_cmd(cmd, SAM_STAT_GOOD);
491 	return 0;
492 }
493 
494 static sense_reason_t
495 iblock_execute_write_same(struct se_cmd *cmd)
496 {
497 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
498 	struct iblock_req *ibr;
499 	struct scatterlist *sg;
500 	struct bio *bio;
501 	struct bio_list list;
502 	struct se_device *dev = cmd->se_dev;
503 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
504 	sector_t sectors = target_to_linux_sector(dev,
505 					sbc_get_write_same_sectors(cmd));
506 
507 	if (cmd->prot_op) {
508 		pr_err("WRITE_SAME: Protection information with IBLOCK"
509 		       " backends not supported\n");
510 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
511 	}
512 
513 	if (!cmd->t_data_nents)
514 		return TCM_INVALID_CDB_FIELD;
515 
516 	sg = &cmd->t_data_sg[0];
517 
518 	if (cmd->t_data_nents > 1 ||
519 	    sg->length != cmd->se_dev->dev_attrib.block_size) {
520 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
521 			" block_size: %u\n", cmd->t_data_nents, sg->length,
522 			cmd->se_dev->dev_attrib.block_size);
523 		return TCM_INVALID_CDB_FIELD;
524 	}
525 
526 	if (bdev_write_zeroes_sectors(bdev)) {
527 		if (!iblock_execute_zero_out(bdev, cmd))
528 			return 0;
529 	}
530 
531 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
532 	if (!ibr)
533 		goto fail;
534 	cmd->priv = ibr;
535 
536 	bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
537 	if (!bio)
538 		goto fail_free_ibr;
539 
540 	bio_list_init(&list);
541 	bio_list_add(&list, bio);
542 
543 	refcount_set(&ibr->pending, 1);
544 
545 	while (sectors) {
546 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
547 				!= sg->length) {
548 
549 			bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
550 			if (!bio)
551 				goto fail_put_bios;
552 
553 			refcount_inc(&ibr->pending);
554 			bio_list_add(&list, bio);
555 		}
556 
557 		/* Always in 512 byte units for Linux/Block */
558 		block_lba += sg->length >> SECTOR_SHIFT;
559 		sectors -= sg->length >> SECTOR_SHIFT;
560 	}
561 
562 	iblock_submit_bios(&list);
563 	return 0;
564 
565 fail_put_bios:
566 	while ((bio = bio_list_pop(&list)))
567 		bio_put(bio);
568 fail_free_ibr:
569 	kfree(ibr);
570 fail:
571 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
572 }
573 
574 enum {
575 	Opt_udev_path, Opt_readonly, Opt_force, Opt_err
576 };
577 
578 static match_table_t tokens = {
579 	{Opt_udev_path, "udev_path=%s"},
580 	{Opt_readonly, "readonly=%d"},
581 	{Opt_force, "force=%d"},
582 	{Opt_err, NULL}
583 };
584 
585 static ssize_t iblock_set_configfs_dev_params(struct se_device *dev,
586 		const char *page, ssize_t count)
587 {
588 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
589 	char *orig, *ptr, *arg_p, *opts;
590 	substring_t args[MAX_OPT_ARGS];
591 	int ret = 0, token;
592 	unsigned long tmp_readonly;
593 
594 	opts = kstrdup(page, GFP_KERNEL);
595 	if (!opts)
596 		return -ENOMEM;
597 
598 	orig = opts;
599 
600 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
601 		if (!*ptr)
602 			continue;
603 
604 		token = match_token(ptr, tokens, args);
605 		switch (token) {
606 		case Opt_udev_path:
607 			if (ib_dev->ibd_bd) {
608 				pr_err("Unable to set udev_path= while"
609 					" ib_dev->ibd_bd exists\n");
610 				ret = -EEXIST;
611 				goto out;
612 			}
613 			if (match_strlcpy(ib_dev->ibd_udev_path, &args[0],
614 				SE_UDEV_PATH_LEN) == 0) {
615 				ret = -EINVAL;
616 				break;
617 			}
618 			pr_debug("IBLOCK: Referencing UDEV path: %s\n",
619 					ib_dev->ibd_udev_path);
620 			ib_dev->ibd_flags |= IBDF_HAS_UDEV_PATH;
621 			break;
622 		case Opt_readonly:
623 			arg_p = match_strdup(&args[0]);
624 			if (!arg_p) {
625 				ret = -ENOMEM;
626 				break;
627 			}
628 			ret = kstrtoul(arg_p, 0, &tmp_readonly);
629 			kfree(arg_p);
630 			if (ret < 0) {
631 				pr_err("kstrtoul() failed for"
632 						" readonly=\n");
633 				goto out;
634 			}
635 			ib_dev->ibd_readonly = tmp_readonly;
636 			pr_debug("IBLOCK: readonly: %d\n", ib_dev->ibd_readonly);
637 			break;
638 		case Opt_force:
639 			break;
640 		default:
641 			break;
642 		}
643 	}
644 
645 out:
646 	kfree(orig);
647 	return (!ret) ? count : ret;
648 }
649 
650 static ssize_t iblock_show_configfs_dev_params(struct se_device *dev, char *b)
651 {
652 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
653 	struct block_device *bd = ib_dev->ibd_bd;
654 	ssize_t bl = 0;
655 
656 	if (bd)
657 		bl += sprintf(b + bl, "iBlock device: %pg", bd);
658 	if (ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)
659 		bl += sprintf(b + bl, "  UDEV PATH: %s",
660 				ib_dev->ibd_udev_path);
661 	bl += sprintf(b + bl, "  readonly: %d\n", ib_dev->ibd_readonly);
662 
663 	bl += sprintf(b + bl, "        ");
664 	if (bd) {
665 		bl += sprintf(b + bl, "Major: %d Minor: %d  %s\n",
666 			MAJOR(bd->bd_dev), MINOR(bd->bd_dev),
667 			"CLAIMED: IBLOCK");
668 	} else {
669 		bl += sprintf(b + bl, "Major: 0 Minor: 0\n");
670 	}
671 
672 	return bl;
673 }
674 
675 static int
676 iblock_alloc_bip(struct se_cmd *cmd, struct bio *bio,
677 		 struct sg_mapping_iter *miter)
678 {
679 	struct se_device *dev = cmd->se_dev;
680 	struct blk_integrity *bi;
681 	struct bio_integrity_payload *bip;
682 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
683 	int rc;
684 	size_t resid, len;
685 
686 	bi = bdev_get_integrity(ib_dev->ibd_bd);
687 	if (!bi) {
688 		pr_err("Unable to locate bio_integrity\n");
689 		return -ENODEV;
690 	}
691 
692 	bip = bio_integrity_alloc(bio, GFP_NOIO, bio_max_segs(cmd->t_prot_nents));
693 	if (IS_ERR(bip)) {
694 		pr_err("Unable to allocate bio_integrity_payload\n");
695 		return PTR_ERR(bip);
696 	}
697 
698 	/* virtual start sector must be in integrity interval units */
699 	bip_set_seed(bip, bio->bi_iter.bi_sector >>
700 				  (bi->interval_exp - SECTOR_SHIFT));
701 
702 	pr_debug("IBLOCK BIP Size: %u Sector: %llu\n", bip->bip_iter.bi_size,
703 		 (unsigned long long)bip->bip_iter.bi_sector);
704 
705 	resid = bio_integrity_bytes(bi, bio_sectors(bio));
706 	while (resid > 0 && sg_miter_next(miter)) {
707 
708 		len = min_t(size_t, miter->length, resid);
709 		rc = bio_integrity_add_page(bio, miter->page, len,
710 					    offset_in_page(miter->addr));
711 		if (rc != len) {
712 			pr_err("bio_integrity_add_page() failed; %d\n", rc);
713 			sg_miter_stop(miter);
714 			return -ENOMEM;
715 		}
716 
717 		pr_debug("Added bio integrity page: %p length: %zu offset: %lu\n",
718 			  miter->page, len, offset_in_page(miter->addr));
719 
720 		resid -= len;
721 		if (len < miter->length)
722 			miter->consumed -= miter->length - len;
723 	}
724 	sg_miter_stop(miter);
725 
726 	return 0;
727 }
728 
729 static sense_reason_t
730 iblock_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
731 		  enum dma_data_direction data_direction)
732 {
733 	struct se_device *dev = cmd->se_dev;
734 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
735 	struct iblock_req *ibr;
736 	struct bio *bio;
737 	struct bio_list list;
738 	struct scatterlist *sg;
739 	u32 sg_num = sgl_nents;
740 	blk_opf_t opf;
741 	unsigned bio_cnt;
742 	int i, rc;
743 	struct sg_mapping_iter prot_miter;
744 	unsigned int miter_dir;
745 
746 	if (data_direction == DMA_TO_DEVICE) {
747 		struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
748 
749 		/*
750 		 * Set bits to indicate WRITE_ODIRECT so we are not throttled
751 		 * by WBT.
752 		 */
753 		opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
754 		/*
755 		 * Force writethrough using REQ_FUA if a volatile write cache
756 		 * is not enabled, or if initiator set the Force Unit Access bit.
757 		 */
758 		miter_dir = SG_MITER_TO_SG;
759 		if (bdev_fua(ib_dev->ibd_bd)) {
760 			if (cmd->se_cmd_flags & SCF_FUA)
761 				opf |= REQ_FUA;
762 			else if (!bdev_write_cache(ib_dev->ibd_bd))
763 				opf |= REQ_FUA;
764 		}
765 	} else {
766 		opf = REQ_OP_READ;
767 		miter_dir = SG_MITER_FROM_SG;
768 	}
769 
770 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
771 	if (!ibr)
772 		goto fail;
773 	cmd->priv = ibr;
774 
775 	if (!sgl_nents) {
776 		refcount_set(&ibr->pending, 1);
777 		iblock_complete_cmd(cmd, BLK_STS_OK);
778 		return 0;
779 	}
780 
781 	bio = iblock_get_bio(cmd, block_lba, sgl_nents, opf);
782 	if (!bio)
783 		goto fail_free_ibr;
784 
785 	bio_list_init(&list);
786 	bio_list_add(&list, bio);
787 
788 	refcount_set(&ibr->pending, 2);
789 	bio_cnt = 1;
790 
791 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type)
792 		sg_miter_start(&prot_miter, cmd->t_prot_sg, cmd->t_prot_nents,
793 			       miter_dir);
794 
795 	for_each_sg(sgl, sg, sgl_nents, i) {
796 		/*
797 		 * XXX: if the length the device accepts is shorter than the
798 		 *	length of the S/G list entry this will cause and
799 		 *	endless loop.  Better hope no driver uses huge pages.
800 		 */
801 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
802 				!= sg->length) {
803 			if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
804 				rc = iblock_alloc_bip(cmd, bio, &prot_miter);
805 				if (rc)
806 					goto fail_put_bios;
807 			}
808 
809 			if (bio_cnt >= IBLOCK_MAX_BIO_PER_TASK) {
810 				iblock_submit_bios(&list);
811 				bio_cnt = 0;
812 			}
813 
814 			bio = iblock_get_bio(cmd, block_lba, sg_num, opf);
815 			if (!bio)
816 				goto fail_put_bios;
817 
818 			refcount_inc(&ibr->pending);
819 			bio_list_add(&list, bio);
820 			bio_cnt++;
821 		}
822 
823 		/* Always in 512 byte units for Linux/Block */
824 		block_lba += sg->length >> SECTOR_SHIFT;
825 		sg_num--;
826 	}
827 
828 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
829 		rc = iblock_alloc_bip(cmd, bio, &prot_miter);
830 		if (rc)
831 			goto fail_put_bios;
832 	}
833 
834 	iblock_submit_bios(&list);
835 	iblock_complete_cmd(cmd, BLK_STS_OK);
836 	return 0;
837 
838 fail_put_bios:
839 	while ((bio = bio_list_pop(&list)))
840 		bio_put(bio);
841 fail_free_ibr:
842 	kfree(ibr);
843 fail:
844 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
845 }
846 
847 static sense_reason_t iblock_execute_pr_out(struct se_cmd *cmd, u8 sa, u64 key,
848 					    u64 sa_key, u8 type, bool aptpl)
849 {
850 	struct se_device *dev = cmd->se_dev;
851 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
852 	struct block_device *bdev = ib_dev->ibd_bd;
853 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
854 	int ret;
855 
856 	if (!ops) {
857 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
858 		return TCM_UNSUPPORTED_SCSI_OPCODE;
859 	}
860 
861 	switch (sa) {
862 	case PRO_REGISTER:
863 	case PRO_REGISTER_AND_IGNORE_EXISTING_KEY:
864 		if (!ops->pr_register) {
865 			pr_err("block device does not support pr_register.\n");
866 			return TCM_UNSUPPORTED_SCSI_OPCODE;
867 		}
868 
869 		/* The block layer pr ops always enables aptpl */
870 		if (!aptpl)
871 			pr_info("APTPL not set by initiator, but will be used.\n");
872 
873 		ret = ops->pr_register(bdev, key, sa_key,
874 				sa == PRO_REGISTER ? 0 : PR_FL_IGNORE_KEY);
875 		break;
876 	case PRO_RESERVE:
877 		if (!ops->pr_reserve) {
878 			pr_err("block_device does not support pr_reserve.\n");
879 			return TCM_UNSUPPORTED_SCSI_OPCODE;
880 		}
881 
882 		ret = ops->pr_reserve(bdev, key, scsi_pr_type_to_block(type), 0);
883 		break;
884 	case PRO_CLEAR:
885 		if (!ops->pr_clear) {
886 			pr_err("block_device does not support pr_clear.\n");
887 			return TCM_UNSUPPORTED_SCSI_OPCODE;
888 		}
889 
890 		ret = ops->pr_clear(bdev, key);
891 		break;
892 	case PRO_PREEMPT:
893 	case PRO_PREEMPT_AND_ABORT:
894 		if (!ops->pr_clear) {
895 			pr_err("block_device does not support pr_preempt.\n");
896 			return TCM_UNSUPPORTED_SCSI_OPCODE;
897 		}
898 
899 		ret = ops->pr_preempt(bdev, key, sa_key,
900 				      scsi_pr_type_to_block(type),
901 				      sa == PRO_PREEMPT_AND_ABORT);
902 		break;
903 	case PRO_RELEASE:
904 		if (!ops->pr_clear) {
905 			pr_err("block_device does not support pr_pclear.\n");
906 			return TCM_UNSUPPORTED_SCSI_OPCODE;
907 		}
908 
909 		ret = ops->pr_release(bdev, key, scsi_pr_type_to_block(type));
910 		break;
911 	default:
912 		pr_err("Unknown PERSISTENT_RESERVE_OUT SA: 0x%02x\n", sa);
913 		return TCM_UNSUPPORTED_SCSI_OPCODE;
914 	}
915 
916 	if (!ret)
917 		return TCM_NO_SENSE;
918 	else if (ret == PR_STS_RESERVATION_CONFLICT)
919 		return TCM_RESERVATION_CONFLICT;
920 	else
921 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
922 }
923 
924 static void iblock_pr_report_caps(unsigned char *param_data)
925 {
926 	u16 len = 8;
927 
928 	put_unaligned_be16(len, &param_data[0]);
929 	/*
930 	 * When using the pr_ops passthrough method we only support exporting
931 	 * the device through one target port because from the backend module
932 	 * level we can't see the target port config. As a result we only
933 	 * support registration directly from the I_T nexus the cmd is sent
934 	 * through and do not set ATP_C here.
935 	 *
936 	 * The block layer pr_ops do not support passing in initiators so
937 	 * we don't set SIP_C here.
938 	 */
939 	/* PTPL_C: Persistence across Target Power Loss bit */
940 	param_data[2] |= 0x01;
941 	/*
942 	 * We are filling in the PERSISTENT RESERVATION TYPE MASK below, so
943 	 * set the TMV: Task Mask Valid bit.
944 	 */
945 	param_data[3] |= 0x80;
946 	/*
947 	 * Change ALLOW COMMANDs to 0x20 or 0x40 later from Table 166
948 	 */
949 	param_data[3] |= 0x10; /* ALLOW COMMANDs field 001b */
950 	/*
951 	 * PTPL_A: Persistence across Target Power Loss Active bit. The block
952 	 * layer pr ops always enables this so report it active.
953 	 */
954 	param_data[3] |= 0x01;
955 	/*
956 	 * Setup the PERSISTENT RESERVATION TYPE MASK from Table 212 spc4r37.
957 	 */
958 	param_data[4] |= 0x80; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
959 	param_data[4] |= 0x40; /* PR_TYPE_EXCLUSIVE_ACCESS_REGONLY */
960 	param_data[4] |= 0x20; /* PR_TYPE_WRITE_EXCLUSIVE_REGONLY */
961 	param_data[4] |= 0x08; /* PR_TYPE_EXCLUSIVE_ACCESS */
962 	param_data[4] |= 0x02; /* PR_TYPE_WRITE_EXCLUSIVE */
963 	param_data[5] |= 0x01; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
964 }
965 
966 static sense_reason_t iblock_pr_read_keys(struct se_cmd *cmd,
967 					  unsigned char *param_data)
968 {
969 	struct se_device *dev = cmd->se_dev;
970 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
971 	struct block_device *bdev = ib_dev->ibd_bd;
972 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
973 	int i, len, paths, data_offset;
974 	struct pr_keys *keys;
975 	sense_reason_t ret;
976 
977 	if (!ops) {
978 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
979 		return TCM_UNSUPPORTED_SCSI_OPCODE;
980 	}
981 
982 	if (!ops->pr_read_keys) {
983 		pr_err("Block device does not support read_keys.\n");
984 		return TCM_UNSUPPORTED_SCSI_OPCODE;
985 	}
986 
987 	/*
988 	 * We don't know what's under us, but dm-multipath will register every
989 	 * path with the same key, so start off with enough space for 16 paths.
990 	 * which is not a lot of memory and should normally be enough.
991 	 */
992 	paths = 16;
993 retry:
994 	len = 8 * paths;
995 	keys = kzalloc(sizeof(*keys) + len, GFP_KERNEL);
996 	if (!keys)
997 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
998 
999 	keys->num_keys = paths;
1000 	if (!ops->pr_read_keys(bdev, keys)) {
1001 		if (keys->num_keys > paths) {
1002 			kfree(keys);
1003 			paths *= 2;
1004 			goto retry;
1005 		}
1006 	} else {
1007 		ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1008 		goto free_keys;
1009 	}
1010 
1011 	ret = TCM_NO_SENSE;
1012 
1013 	put_unaligned_be32(keys->generation, &param_data[0]);
1014 	if (!keys->num_keys) {
1015 		put_unaligned_be32(0, &param_data[4]);
1016 		goto free_keys;
1017 	}
1018 
1019 	put_unaligned_be32(8 * keys->num_keys, &param_data[4]);
1020 
1021 	data_offset = 8;
1022 	for (i = 0; i < keys->num_keys; i++) {
1023 		if (data_offset + 8 > cmd->data_length)
1024 			break;
1025 
1026 		put_unaligned_be64(keys->keys[i], &param_data[data_offset]);
1027 		data_offset += 8;
1028 	}
1029 
1030 free_keys:
1031 	kfree(keys);
1032 	return ret;
1033 }
1034 
1035 static sense_reason_t iblock_pr_read_reservation(struct se_cmd *cmd,
1036 						 unsigned char *param_data)
1037 {
1038 	struct se_device *dev = cmd->se_dev;
1039 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1040 	struct block_device *bdev = ib_dev->ibd_bd;
1041 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
1042 	struct pr_held_reservation rsv = { };
1043 
1044 	if (!ops) {
1045 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
1046 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1047 	}
1048 
1049 	if (!ops->pr_read_reservation) {
1050 		pr_err("Block device does not support read_keys.\n");
1051 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1052 	}
1053 
1054 	if (ops->pr_read_reservation(bdev, &rsv))
1055 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1056 
1057 	put_unaligned_be32(rsv.generation, &param_data[0]);
1058 	if (!block_pr_type_to_scsi(rsv.type)) {
1059 		put_unaligned_be32(0, &param_data[4]);
1060 		return TCM_NO_SENSE;
1061 	}
1062 
1063 	put_unaligned_be32(16, &param_data[4]);
1064 
1065 	if (cmd->data_length < 16)
1066 		return TCM_NO_SENSE;
1067 	put_unaligned_be64(rsv.key, &param_data[8]);
1068 
1069 	if (cmd->data_length < 22)
1070 		return TCM_NO_SENSE;
1071 	param_data[21] = block_pr_type_to_scsi(rsv.type);
1072 
1073 	return TCM_NO_SENSE;
1074 }
1075 
1076 static sense_reason_t iblock_execute_pr_in(struct se_cmd *cmd, u8 sa,
1077 					   unsigned char *param_data)
1078 {
1079 	sense_reason_t ret = TCM_NO_SENSE;
1080 
1081 	switch (sa) {
1082 	case PRI_REPORT_CAPABILITIES:
1083 		iblock_pr_report_caps(param_data);
1084 		break;
1085 	case PRI_READ_KEYS:
1086 		ret = iblock_pr_read_keys(cmd, param_data);
1087 		break;
1088 	case PRI_READ_RESERVATION:
1089 		ret = iblock_pr_read_reservation(cmd, param_data);
1090 		break;
1091 	default:
1092 		pr_err("Unknown PERSISTENT_RESERVE_IN SA: 0x%02x\n", sa);
1093 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1094 	}
1095 
1096 	return ret;
1097 }
1098 
1099 static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
1100 {
1101 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1102 	struct block_device *bd = ib_dev->ibd_bd;
1103 	int ret;
1104 
1105 	ret = bdev_alignment_offset(bd);
1106 	if (ret == -1)
1107 		return 0;
1108 
1109 	/* convert offset-bytes to offset-lbas */
1110 	return ret / bdev_logical_block_size(bd);
1111 }
1112 
1113 static unsigned int iblock_get_lbppbe(struct se_device *dev)
1114 {
1115 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1116 	struct block_device *bd = ib_dev->ibd_bd;
1117 	unsigned int logs_per_phys =
1118 		bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
1119 
1120 	return ilog2(logs_per_phys);
1121 }
1122 
1123 static unsigned int iblock_get_io_min(struct se_device *dev)
1124 {
1125 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1126 	struct block_device *bd = ib_dev->ibd_bd;
1127 
1128 	return bdev_io_min(bd);
1129 }
1130 
1131 static unsigned int iblock_get_io_opt(struct se_device *dev)
1132 {
1133 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1134 	struct block_device *bd = ib_dev->ibd_bd;
1135 
1136 	return bdev_io_opt(bd);
1137 }
1138 
1139 static struct exec_cmd_ops iblock_exec_cmd_ops = {
1140 	.execute_rw		= iblock_execute_rw,
1141 	.execute_sync_cache	= iblock_execute_sync_cache,
1142 	.execute_write_same	= iblock_execute_write_same,
1143 	.execute_unmap		= iblock_execute_unmap,
1144 	.execute_pr_out		= iblock_execute_pr_out,
1145 	.execute_pr_in		= iblock_execute_pr_in,
1146 };
1147 
1148 static sense_reason_t
1149 iblock_parse_cdb(struct se_cmd *cmd)
1150 {
1151 	return sbc_parse_cdb(cmd, &iblock_exec_cmd_ops);
1152 }
1153 
1154 static bool iblock_get_write_cache(struct se_device *dev)
1155 {
1156 	return bdev_write_cache(IBLOCK_DEV(dev)->ibd_bd);
1157 }
1158 
1159 static const struct target_backend_ops iblock_ops = {
1160 	.name			= "iblock",
1161 	.inquiry_prod		= "IBLOCK",
1162 	.transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR,
1163 	.inquiry_rev		= IBLOCK_VERSION,
1164 	.owner			= THIS_MODULE,
1165 	.attach_hba		= iblock_attach_hba,
1166 	.detach_hba		= iblock_detach_hba,
1167 	.alloc_device		= iblock_alloc_device,
1168 	.configure_device	= iblock_configure_device,
1169 	.destroy_device		= iblock_destroy_device,
1170 	.free_device		= iblock_free_device,
1171 	.configure_unmap	= iblock_configure_unmap,
1172 	.plug_device		= iblock_plug_device,
1173 	.unplug_device		= iblock_unplug_device,
1174 	.parse_cdb		= iblock_parse_cdb,
1175 	.set_configfs_dev_params = iblock_set_configfs_dev_params,
1176 	.show_configfs_dev_params = iblock_show_configfs_dev_params,
1177 	.get_device_type	= sbc_get_device_type,
1178 	.get_blocks		= iblock_get_blocks,
1179 	.get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
1180 	.get_lbppbe		= iblock_get_lbppbe,
1181 	.get_io_min		= iblock_get_io_min,
1182 	.get_io_opt		= iblock_get_io_opt,
1183 	.get_write_cache	= iblock_get_write_cache,
1184 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
1185 };
1186 
1187 static int __init iblock_module_init(void)
1188 {
1189 	return transport_backend_register(&iblock_ops);
1190 }
1191 
1192 static void __exit iblock_module_exit(void)
1193 {
1194 	target_backend_unregister(&iblock_ops);
1195 }
1196 
1197 MODULE_DESCRIPTION("TCM IBLOCK subsystem plugin");
1198 MODULE_AUTHOR("nab@Linux-iSCSI.org");
1199 MODULE_LICENSE("GPL");
1200 
1201 module_init(iblock_module_init);
1202 module_exit(iblock_module_exit);
1203