xref: /linux/drivers/target/target_core_iblock.c (revision 8e07e0e3964ca4e23ce7b68e2096fe660a888942)
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 bdev_handle *bdev_handle;
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_handle = bdev_open_by_path(ib_dev->ibd_udev_path, mode, ib_dev,
121 					NULL);
122 	if (IS_ERR(bdev_handle)) {
123 		ret = PTR_ERR(bdev_handle);
124 		goto out_free_bioset;
125 	}
126 	ib_dev->ibd_bdev_handle = bdev_handle;
127 	ib_dev->ibd_bd = bd = bdev_handle->bdev;
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 		struct bio_set *bs = &ib_dev->ibd_bio_set;
153 
154 		if (!strcmp(bi->profile->name, "T10-DIF-TYPE3-IP") ||
155 		    !strcmp(bi->profile->name, "T10-DIF-TYPE1-IP")) {
156 			pr_err("IBLOCK export of blk_integrity: %s not"
157 			       " supported\n", bi->profile->name);
158 			ret = -ENOSYS;
159 			goto out_blkdev_put;
160 		}
161 
162 		if (!strcmp(bi->profile->name, "T10-DIF-TYPE3-CRC")) {
163 			dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE3_PROT;
164 		} else if (!strcmp(bi->profile->name, "T10-DIF-TYPE1-CRC")) {
165 			dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE1_PROT;
166 		}
167 
168 		if (dev->dev_attrib.pi_prot_type) {
169 			if (bioset_integrity_create(bs, IBLOCK_BIO_POOL_SIZE) < 0) {
170 				pr_err("Unable to allocate bioset for PI\n");
171 				ret = -ENOMEM;
172 				goto out_blkdev_put;
173 			}
174 			pr_debug("IBLOCK setup BIP bs->bio_integrity_pool: %p\n",
175 				 &bs->bio_integrity_pool);
176 		}
177 		dev->dev_attrib.hw_pi_prot_type = dev->dev_attrib.pi_prot_type;
178 	}
179 
180 	return 0;
181 
182 out_blkdev_put:
183 	bdev_release(ib_dev->ibd_bdev_handle);
184 out_free_bioset:
185 	bioset_exit(&ib_dev->ibd_bio_set);
186 out:
187 	return ret;
188 }
189 
190 static void iblock_dev_call_rcu(struct rcu_head *p)
191 {
192 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
193 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
194 
195 	kfree(ib_dev->ibd_plug);
196 	kfree(ib_dev);
197 }
198 
199 static void iblock_free_device(struct se_device *dev)
200 {
201 	call_rcu(&dev->rcu_head, iblock_dev_call_rcu);
202 }
203 
204 static void iblock_destroy_device(struct se_device *dev)
205 {
206 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
207 
208 	if (ib_dev->ibd_bdev_handle)
209 		bdev_release(ib_dev->ibd_bdev_handle);
210 	bioset_exit(&ib_dev->ibd_bio_set);
211 }
212 
213 static struct se_dev_plug *iblock_plug_device(struct se_device *se_dev)
214 {
215 	struct iblock_dev *ib_dev = IBLOCK_DEV(se_dev);
216 	struct iblock_dev_plug *ib_dev_plug;
217 
218 	/*
219 	 * Each se_device has a per cpu work this can be run from. We
220 	 * shouldn't have multiple threads on the same cpu calling this
221 	 * at the same time.
222 	 */
223 	ib_dev_plug = &ib_dev->ibd_plug[raw_smp_processor_id()];
224 	if (test_and_set_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags))
225 		return NULL;
226 
227 	blk_start_plug(&ib_dev_plug->blk_plug);
228 	return &ib_dev_plug->se_plug;
229 }
230 
231 static void iblock_unplug_device(struct se_dev_plug *se_plug)
232 {
233 	struct iblock_dev_plug *ib_dev_plug = container_of(se_plug,
234 					struct iblock_dev_plug, se_plug);
235 
236 	blk_finish_plug(&ib_dev_plug->blk_plug);
237 	clear_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags);
238 }
239 
240 static sector_t iblock_get_blocks(struct se_device *dev)
241 {
242 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
243 	u32 block_size = bdev_logical_block_size(ib_dev->ibd_bd);
244 	unsigned long long blocks_long =
245 		div_u64(bdev_nr_bytes(ib_dev->ibd_bd), block_size) - 1;
246 
247 	if (block_size == dev->dev_attrib.block_size)
248 		return blocks_long;
249 
250 	switch (block_size) {
251 	case 4096:
252 		switch (dev->dev_attrib.block_size) {
253 		case 2048:
254 			blocks_long <<= 1;
255 			break;
256 		case 1024:
257 			blocks_long <<= 2;
258 			break;
259 		case 512:
260 			blocks_long <<= 3;
261 			break;
262 		default:
263 			break;
264 		}
265 		break;
266 	case 2048:
267 		switch (dev->dev_attrib.block_size) {
268 		case 4096:
269 			blocks_long >>= 1;
270 			break;
271 		case 1024:
272 			blocks_long <<= 1;
273 			break;
274 		case 512:
275 			blocks_long <<= 2;
276 			break;
277 		default:
278 			break;
279 		}
280 		break;
281 	case 1024:
282 		switch (dev->dev_attrib.block_size) {
283 		case 4096:
284 			blocks_long >>= 2;
285 			break;
286 		case 2048:
287 			blocks_long >>= 1;
288 			break;
289 		case 512:
290 			blocks_long <<= 1;
291 			break;
292 		default:
293 			break;
294 		}
295 		break;
296 	case 512:
297 		switch (dev->dev_attrib.block_size) {
298 		case 4096:
299 			blocks_long >>= 3;
300 			break;
301 		case 2048:
302 			blocks_long >>= 2;
303 			break;
304 		case 1024:
305 			blocks_long >>= 1;
306 			break;
307 		default:
308 			break;
309 		}
310 		break;
311 	default:
312 		break;
313 	}
314 
315 	return blocks_long;
316 }
317 
318 static void iblock_complete_cmd(struct se_cmd *cmd, blk_status_t blk_status)
319 {
320 	struct iblock_req *ibr = cmd->priv;
321 	u8 status;
322 
323 	if (!refcount_dec_and_test(&ibr->pending))
324 		return;
325 
326 	if (blk_status == BLK_STS_RESV_CONFLICT)
327 		status = SAM_STAT_RESERVATION_CONFLICT;
328 	else if (atomic_read(&ibr->ib_bio_err_cnt))
329 		status = SAM_STAT_CHECK_CONDITION;
330 	else
331 		status = SAM_STAT_GOOD;
332 
333 	target_complete_cmd(cmd, status);
334 	kfree(ibr);
335 }
336 
337 static void iblock_bio_done(struct bio *bio)
338 {
339 	struct se_cmd *cmd = bio->bi_private;
340 	struct iblock_req *ibr = cmd->priv;
341 	blk_status_t blk_status = bio->bi_status;
342 
343 	if (bio->bi_status) {
344 		pr_err("bio error: %p,  err: %d\n", bio, bio->bi_status);
345 		/*
346 		 * Bump the ib_bio_err_cnt and release bio.
347 		 */
348 		atomic_inc(&ibr->ib_bio_err_cnt);
349 		smp_mb__after_atomic();
350 	}
351 
352 	bio_put(bio);
353 
354 	iblock_complete_cmd(cmd, blk_status);
355 }
356 
357 static struct bio *iblock_get_bio(struct se_cmd *cmd, sector_t lba, u32 sg_num,
358 				  blk_opf_t opf)
359 {
360 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
361 	struct bio *bio;
362 
363 	/*
364 	 * Only allocate as many vector entries as the bio code allows us to,
365 	 * we'll loop later on until we have handled the whole request.
366 	 */
367 	bio = bio_alloc_bioset(ib_dev->ibd_bd, bio_max_segs(sg_num), opf,
368 			       GFP_NOIO, &ib_dev->ibd_bio_set);
369 	if (!bio) {
370 		pr_err("Unable to allocate memory for bio\n");
371 		return NULL;
372 	}
373 
374 	bio->bi_private = cmd;
375 	bio->bi_end_io = &iblock_bio_done;
376 	bio->bi_iter.bi_sector = lba;
377 
378 	return bio;
379 }
380 
381 static void iblock_submit_bios(struct bio_list *list)
382 {
383 	struct blk_plug plug;
384 	struct bio *bio;
385 	/*
386 	 * The block layer handles nested plugs, so just plug/unplug to handle
387 	 * fabric drivers that didn't support batching and multi bio cmds.
388 	 */
389 	blk_start_plug(&plug);
390 	while ((bio = bio_list_pop(list)))
391 		submit_bio(bio);
392 	blk_finish_plug(&plug);
393 }
394 
395 static void iblock_end_io_flush(struct bio *bio)
396 {
397 	struct se_cmd *cmd = bio->bi_private;
398 
399 	if (bio->bi_status)
400 		pr_err("IBLOCK: cache flush failed: %d\n", bio->bi_status);
401 
402 	if (cmd) {
403 		if (bio->bi_status)
404 			target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
405 		else
406 			target_complete_cmd(cmd, SAM_STAT_GOOD);
407 	}
408 
409 	bio_put(bio);
410 }
411 
412 /*
413  * Implement SYCHRONIZE CACHE.  Note that we can't handle lba ranges and must
414  * always flush the whole cache.
415  */
416 static sense_reason_t
417 iblock_execute_sync_cache(struct se_cmd *cmd)
418 {
419 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
420 	int immed = (cmd->t_task_cdb[1] & 0x2);
421 	struct bio *bio;
422 
423 	/*
424 	 * If the Immediate bit is set, queue up the GOOD response
425 	 * for this SYNCHRONIZE_CACHE op.
426 	 */
427 	if (immed)
428 		target_complete_cmd(cmd, SAM_STAT_GOOD);
429 
430 	bio = bio_alloc(ib_dev->ibd_bd, 0, REQ_OP_WRITE | REQ_PREFLUSH,
431 			GFP_KERNEL);
432 	bio->bi_end_io = iblock_end_io_flush;
433 	if (!immed)
434 		bio->bi_private = cmd;
435 	submit_bio(bio);
436 	return 0;
437 }
438 
439 static sense_reason_t
440 iblock_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
441 {
442 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
443 	struct se_device *dev = cmd->se_dev;
444 	int ret;
445 
446 	ret = blkdev_issue_discard(bdev,
447 				   target_to_linux_sector(dev, lba),
448 				   target_to_linux_sector(dev,  nolb),
449 				   GFP_KERNEL);
450 	if (ret < 0) {
451 		pr_err("blkdev_issue_discard() failed: %d\n", ret);
452 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
453 	}
454 
455 	return 0;
456 }
457 
458 static sense_reason_t
459 iblock_execute_zero_out(struct block_device *bdev, struct se_cmd *cmd)
460 {
461 	struct se_device *dev = cmd->se_dev;
462 	struct scatterlist *sg = &cmd->t_data_sg[0];
463 	unsigned char *buf, *not_zero;
464 	int ret;
465 
466 	buf = kmap(sg_page(sg)) + sg->offset;
467 	if (!buf)
468 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
469 	/*
470 	 * Fall back to block_execute_write_same() slow-path if
471 	 * incoming WRITE_SAME payload does not contain zeros.
472 	 */
473 	not_zero = memchr_inv(buf, 0x00, cmd->data_length);
474 	kunmap(sg_page(sg));
475 
476 	if (not_zero)
477 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
478 
479 	ret = blkdev_issue_zeroout(bdev,
480 				target_to_linux_sector(dev, cmd->t_task_lba),
481 				target_to_linux_sector(dev,
482 					sbc_get_write_same_sectors(cmd)),
483 				GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
484 	if (ret)
485 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
486 
487 	target_complete_cmd(cmd, SAM_STAT_GOOD);
488 	return 0;
489 }
490 
491 static sense_reason_t
492 iblock_execute_write_same(struct se_cmd *cmd)
493 {
494 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
495 	struct iblock_req *ibr;
496 	struct scatterlist *sg;
497 	struct bio *bio;
498 	struct bio_list list;
499 	struct se_device *dev = cmd->se_dev;
500 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
501 	sector_t sectors = target_to_linux_sector(dev,
502 					sbc_get_write_same_sectors(cmd));
503 
504 	if (cmd->prot_op) {
505 		pr_err("WRITE_SAME: Protection information with IBLOCK"
506 		       " backends not supported\n");
507 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
508 	}
509 
510 	if (!cmd->t_data_nents)
511 		return TCM_INVALID_CDB_FIELD;
512 
513 	sg = &cmd->t_data_sg[0];
514 
515 	if (cmd->t_data_nents > 1 ||
516 	    sg->length != cmd->se_dev->dev_attrib.block_size) {
517 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
518 			" block_size: %u\n", cmd->t_data_nents, sg->length,
519 			cmd->se_dev->dev_attrib.block_size);
520 		return TCM_INVALID_CDB_FIELD;
521 	}
522 
523 	if (bdev_write_zeroes_sectors(bdev)) {
524 		if (!iblock_execute_zero_out(bdev, cmd))
525 			return 0;
526 	}
527 
528 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
529 	if (!ibr)
530 		goto fail;
531 	cmd->priv = ibr;
532 
533 	bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
534 	if (!bio)
535 		goto fail_free_ibr;
536 
537 	bio_list_init(&list);
538 	bio_list_add(&list, bio);
539 
540 	refcount_set(&ibr->pending, 1);
541 
542 	while (sectors) {
543 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
544 				!= sg->length) {
545 
546 			bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
547 			if (!bio)
548 				goto fail_put_bios;
549 
550 			refcount_inc(&ibr->pending);
551 			bio_list_add(&list, bio);
552 		}
553 
554 		/* Always in 512 byte units for Linux/Block */
555 		block_lba += sg->length >> SECTOR_SHIFT;
556 		sectors -= sg->length >> SECTOR_SHIFT;
557 	}
558 
559 	iblock_submit_bios(&list);
560 	return 0;
561 
562 fail_put_bios:
563 	while ((bio = bio_list_pop(&list)))
564 		bio_put(bio);
565 fail_free_ibr:
566 	kfree(ibr);
567 fail:
568 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
569 }
570 
571 enum {
572 	Opt_udev_path, Opt_readonly, Opt_force, Opt_err
573 };
574 
575 static match_table_t tokens = {
576 	{Opt_udev_path, "udev_path=%s"},
577 	{Opt_readonly, "readonly=%d"},
578 	{Opt_force, "force=%d"},
579 	{Opt_err, NULL}
580 };
581 
582 static ssize_t iblock_set_configfs_dev_params(struct se_device *dev,
583 		const char *page, ssize_t count)
584 {
585 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
586 	char *orig, *ptr, *arg_p, *opts;
587 	substring_t args[MAX_OPT_ARGS];
588 	int ret = 0, token;
589 	unsigned long tmp_readonly;
590 
591 	opts = kstrdup(page, GFP_KERNEL);
592 	if (!opts)
593 		return -ENOMEM;
594 
595 	orig = opts;
596 
597 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
598 		if (!*ptr)
599 			continue;
600 
601 		token = match_token(ptr, tokens, args);
602 		switch (token) {
603 		case Opt_udev_path:
604 			if (ib_dev->ibd_bd) {
605 				pr_err("Unable to set udev_path= while"
606 					" ib_dev->ibd_bd exists\n");
607 				ret = -EEXIST;
608 				goto out;
609 			}
610 			if (match_strlcpy(ib_dev->ibd_udev_path, &args[0],
611 				SE_UDEV_PATH_LEN) == 0) {
612 				ret = -EINVAL;
613 				break;
614 			}
615 			pr_debug("IBLOCK: Referencing UDEV path: %s\n",
616 					ib_dev->ibd_udev_path);
617 			ib_dev->ibd_flags |= IBDF_HAS_UDEV_PATH;
618 			break;
619 		case Opt_readonly:
620 			arg_p = match_strdup(&args[0]);
621 			if (!arg_p) {
622 				ret = -ENOMEM;
623 				break;
624 			}
625 			ret = kstrtoul(arg_p, 0, &tmp_readonly);
626 			kfree(arg_p);
627 			if (ret < 0) {
628 				pr_err("kstrtoul() failed for"
629 						" readonly=\n");
630 				goto out;
631 			}
632 			ib_dev->ibd_readonly = tmp_readonly;
633 			pr_debug("IBLOCK: readonly: %d\n", ib_dev->ibd_readonly);
634 			break;
635 		case Opt_force:
636 			break;
637 		default:
638 			break;
639 		}
640 	}
641 
642 out:
643 	kfree(orig);
644 	return (!ret) ? count : ret;
645 }
646 
647 static ssize_t iblock_show_configfs_dev_params(struct se_device *dev, char *b)
648 {
649 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
650 	struct block_device *bd = ib_dev->ibd_bd;
651 	ssize_t bl = 0;
652 
653 	if (bd)
654 		bl += sprintf(b + bl, "iBlock device: %pg", bd);
655 	if (ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)
656 		bl += sprintf(b + bl, "  UDEV PATH: %s",
657 				ib_dev->ibd_udev_path);
658 	bl += sprintf(b + bl, "  readonly: %d\n", ib_dev->ibd_readonly);
659 
660 	bl += sprintf(b + bl, "        ");
661 	if (bd) {
662 		bl += sprintf(b + bl, "Major: %d Minor: %d  %s\n",
663 			MAJOR(bd->bd_dev), MINOR(bd->bd_dev),
664 			"CLAIMED: IBLOCK");
665 	} else {
666 		bl += sprintf(b + bl, "Major: 0 Minor: 0\n");
667 	}
668 
669 	return bl;
670 }
671 
672 static int
673 iblock_alloc_bip(struct se_cmd *cmd, struct bio *bio,
674 		 struct sg_mapping_iter *miter)
675 {
676 	struct se_device *dev = cmd->se_dev;
677 	struct blk_integrity *bi;
678 	struct bio_integrity_payload *bip;
679 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
680 	int rc;
681 	size_t resid, len;
682 
683 	bi = bdev_get_integrity(ib_dev->ibd_bd);
684 	if (!bi) {
685 		pr_err("Unable to locate bio_integrity\n");
686 		return -ENODEV;
687 	}
688 
689 	bip = bio_integrity_alloc(bio, GFP_NOIO, bio_max_segs(cmd->t_prot_nents));
690 	if (IS_ERR(bip)) {
691 		pr_err("Unable to allocate bio_integrity_payload\n");
692 		return PTR_ERR(bip);
693 	}
694 
695 	/* virtual start sector must be in integrity interval units */
696 	bip_set_seed(bip, bio->bi_iter.bi_sector >>
697 				  (bi->interval_exp - SECTOR_SHIFT));
698 
699 	pr_debug("IBLOCK BIP Size: %u Sector: %llu\n", bip->bip_iter.bi_size,
700 		 (unsigned long long)bip->bip_iter.bi_sector);
701 
702 	resid = bio_integrity_bytes(bi, bio_sectors(bio));
703 	while (resid > 0 && sg_miter_next(miter)) {
704 
705 		len = min_t(size_t, miter->length, resid);
706 		rc = bio_integrity_add_page(bio, miter->page, len,
707 					    offset_in_page(miter->addr));
708 		if (rc != len) {
709 			pr_err("bio_integrity_add_page() failed; %d\n", rc);
710 			sg_miter_stop(miter);
711 			return -ENOMEM;
712 		}
713 
714 		pr_debug("Added bio integrity page: %p length: %zu offset: %lu\n",
715 			  miter->page, len, offset_in_page(miter->addr));
716 
717 		resid -= len;
718 		if (len < miter->length)
719 			miter->consumed -= miter->length - len;
720 	}
721 	sg_miter_stop(miter);
722 
723 	return 0;
724 }
725 
726 static sense_reason_t
727 iblock_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
728 		  enum dma_data_direction data_direction)
729 {
730 	struct se_device *dev = cmd->se_dev;
731 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
732 	struct iblock_req *ibr;
733 	struct bio *bio;
734 	struct bio_list list;
735 	struct scatterlist *sg;
736 	u32 sg_num = sgl_nents;
737 	blk_opf_t opf;
738 	unsigned bio_cnt;
739 	int i, rc;
740 	struct sg_mapping_iter prot_miter;
741 	unsigned int miter_dir;
742 
743 	if (data_direction == DMA_TO_DEVICE) {
744 		struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
745 
746 		/*
747 		 * Set bits to indicate WRITE_ODIRECT so we are not throttled
748 		 * by WBT.
749 		 */
750 		opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
751 		/*
752 		 * Force writethrough using REQ_FUA if a volatile write cache
753 		 * is not enabled, or if initiator set the Force Unit Access bit.
754 		 */
755 		miter_dir = SG_MITER_TO_SG;
756 		if (bdev_fua(ib_dev->ibd_bd)) {
757 			if (cmd->se_cmd_flags & SCF_FUA)
758 				opf |= REQ_FUA;
759 			else if (!bdev_write_cache(ib_dev->ibd_bd))
760 				opf |= REQ_FUA;
761 		}
762 	} else {
763 		opf = REQ_OP_READ;
764 		miter_dir = SG_MITER_FROM_SG;
765 	}
766 
767 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
768 	if (!ibr)
769 		goto fail;
770 	cmd->priv = ibr;
771 
772 	if (!sgl_nents) {
773 		refcount_set(&ibr->pending, 1);
774 		iblock_complete_cmd(cmd, BLK_STS_OK);
775 		return 0;
776 	}
777 
778 	bio = iblock_get_bio(cmd, block_lba, sgl_nents, opf);
779 	if (!bio)
780 		goto fail_free_ibr;
781 
782 	bio_list_init(&list);
783 	bio_list_add(&list, bio);
784 
785 	refcount_set(&ibr->pending, 2);
786 	bio_cnt = 1;
787 
788 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type)
789 		sg_miter_start(&prot_miter, cmd->t_prot_sg, cmd->t_prot_nents,
790 			       miter_dir);
791 
792 	for_each_sg(sgl, sg, sgl_nents, i) {
793 		/*
794 		 * XXX: if the length the device accepts is shorter than the
795 		 *	length of the S/G list entry this will cause and
796 		 *	endless loop.  Better hope no driver uses huge pages.
797 		 */
798 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
799 				!= sg->length) {
800 			if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
801 				rc = iblock_alloc_bip(cmd, bio, &prot_miter);
802 				if (rc)
803 					goto fail_put_bios;
804 			}
805 
806 			if (bio_cnt >= IBLOCK_MAX_BIO_PER_TASK) {
807 				iblock_submit_bios(&list);
808 				bio_cnt = 0;
809 			}
810 
811 			bio = iblock_get_bio(cmd, block_lba, sg_num, opf);
812 			if (!bio)
813 				goto fail_put_bios;
814 
815 			refcount_inc(&ibr->pending);
816 			bio_list_add(&list, bio);
817 			bio_cnt++;
818 		}
819 
820 		/* Always in 512 byte units for Linux/Block */
821 		block_lba += sg->length >> SECTOR_SHIFT;
822 		sg_num--;
823 	}
824 
825 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
826 		rc = iblock_alloc_bip(cmd, bio, &prot_miter);
827 		if (rc)
828 			goto fail_put_bios;
829 	}
830 
831 	iblock_submit_bios(&list);
832 	iblock_complete_cmd(cmd, BLK_STS_OK);
833 	return 0;
834 
835 fail_put_bios:
836 	while ((bio = bio_list_pop(&list)))
837 		bio_put(bio);
838 fail_free_ibr:
839 	kfree(ibr);
840 fail:
841 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
842 }
843 
844 static sense_reason_t iblock_execute_pr_out(struct se_cmd *cmd, u8 sa, u64 key,
845 					    u64 sa_key, u8 type, bool aptpl)
846 {
847 	struct se_device *dev = cmd->se_dev;
848 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
849 	struct block_device *bdev = ib_dev->ibd_bd;
850 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
851 	int ret;
852 
853 	if (!ops) {
854 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
855 		return TCM_UNSUPPORTED_SCSI_OPCODE;
856 	}
857 
858 	switch (sa) {
859 	case PRO_REGISTER:
860 	case PRO_REGISTER_AND_IGNORE_EXISTING_KEY:
861 		if (!ops->pr_register) {
862 			pr_err("block device does not support pr_register.\n");
863 			return TCM_UNSUPPORTED_SCSI_OPCODE;
864 		}
865 
866 		/* The block layer pr ops always enables aptpl */
867 		if (!aptpl)
868 			pr_info("APTPL not set by initiator, but will be used.\n");
869 
870 		ret = ops->pr_register(bdev, key, sa_key,
871 				sa == PRO_REGISTER ? 0 : PR_FL_IGNORE_KEY);
872 		break;
873 	case PRO_RESERVE:
874 		if (!ops->pr_reserve) {
875 			pr_err("block_device does not support pr_reserve.\n");
876 			return TCM_UNSUPPORTED_SCSI_OPCODE;
877 		}
878 
879 		ret = ops->pr_reserve(bdev, key, scsi_pr_type_to_block(type), 0);
880 		break;
881 	case PRO_CLEAR:
882 		if (!ops->pr_clear) {
883 			pr_err("block_device does not support pr_clear.\n");
884 			return TCM_UNSUPPORTED_SCSI_OPCODE;
885 		}
886 
887 		ret = ops->pr_clear(bdev, key);
888 		break;
889 	case PRO_PREEMPT:
890 	case PRO_PREEMPT_AND_ABORT:
891 		if (!ops->pr_clear) {
892 			pr_err("block_device does not support pr_preempt.\n");
893 			return TCM_UNSUPPORTED_SCSI_OPCODE;
894 		}
895 
896 		ret = ops->pr_preempt(bdev, key, sa_key,
897 				      scsi_pr_type_to_block(type),
898 				      sa == PRO_PREEMPT_AND_ABORT);
899 		break;
900 	case PRO_RELEASE:
901 		if (!ops->pr_clear) {
902 			pr_err("block_device does not support pr_pclear.\n");
903 			return TCM_UNSUPPORTED_SCSI_OPCODE;
904 		}
905 
906 		ret = ops->pr_release(bdev, key, scsi_pr_type_to_block(type));
907 		break;
908 	default:
909 		pr_err("Unknown PERSISTENT_RESERVE_OUT SA: 0x%02x\n", sa);
910 		return TCM_UNSUPPORTED_SCSI_OPCODE;
911 	}
912 
913 	if (!ret)
914 		return TCM_NO_SENSE;
915 	else if (ret == PR_STS_RESERVATION_CONFLICT)
916 		return TCM_RESERVATION_CONFLICT;
917 	else
918 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
919 }
920 
921 static void iblock_pr_report_caps(unsigned char *param_data)
922 {
923 	u16 len = 8;
924 
925 	put_unaligned_be16(len, &param_data[0]);
926 	/*
927 	 * When using the pr_ops passthrough method we only support exporting
928 	 * the device through one target port because from the backend module
929 	 * level we can't see the target port config. As a result we only
930 	 * support registration directly from the I_T nexus the cmd is sent
931 	 * through and do not set ATP_C here.
932 	 *
933 	 * The block layer pr_ops do not support passing in initiators so
934 	 * we don't set SIP_C here.
935 	 */
936 	/* PTPL_C: Persistence across Target Power Loss bit */
937 	param_data[2] |= 0x01;
938 	/*
939 	 * We are filling in the PERSISTENT RESERVATION TYPE MASK below, so
940 	 * set the TMV: Task Mask Valid bit.
941 	 */
942 	param_data[3] |= 0x80;
943 	/*
944 	 * Change ALLOW COMMANDs to 0x20 or 0x40 later from Table 166
945 	 */
946 	param_data[3] |= 0x10; /* ALLOW COMMANDs field 001b */
947 	/*
948 	 * PTPL_A: Persistence across Target Power Loss Active bit. The block
949 	 * layer pr ops always enables this so report it active.
950 	 */
951 	param_data[3] |= 0x01;
952 	/*
953 	 * Setup the PERSISTENT RESERVATION TYPE MASK from Table 212 spc4r37.
954 	 */
955 	param_data[4] |= 0x80; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
956 	param_data[4] |= 0x40; /* PR_TYPE_EXCLUSIVE_ACCESS_REGONLY */
957 	param_data[4] |= 0x20; /* PR_TYPE_WRITE_EXCLUSIVE_REGONLY */
958 	param_data[4] |= 0x08; /* PR_TYPE_EXCLUSIVE_ACCESS */
959 	param_data[4] |= 0x02; /* PR_TYPE_WRITE_EXCLUSIVE */
960 	param_data[5] |= 0x01; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
961 }
962 
963 static sense_reason_t iblock_pr_read_keys(struct se_cmd *cmd,
964 					  unsigned char *param_data)
965 {
966 	struct se_device *dev = cmd->se_dev;
967 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
968 	struct block_device *bdev = ib_dev->ibd_bd;
969 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
970 	int i, len, paths, data_offset;
971 	struct pr_keys *keys;
972 	sense_reason_t ret;
973 
974 	if (!ops) {
975 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
976 		return TCM_UNSUPPORTED_SCSI_OPCODE;
977 	}
978 
979 	if (!ops->pr_read_keys) {
980 		pr_err("Block device does not support read_keys.\n");
981 		return TCM_UNSUPPORTED_SCSI_OPCODE;
982 	}
983 
984 	/*
985 	 * We don't know what's under us, but dm-multipath will register every
986 	 * path with the same key, so start off with enough space for 16 paths.
987 	 * which is not a lot of memory and should normally be enough.
988 	 */
989 	paths = 16;
990 retry:
991 	len = 8 * paths;
992 	keys = kzalloc(sizeof(*keys) + len, GFP_KERNEL);
993 	if (!keys)
994 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
995 
996 	keys->num_keys = paths;
997 	if (!ops->pr_read_keys(bdev, keys)) {
998 		if (keys->num_keys > paths) {
999 			kfree(keys);
1000 			paths *= 2;
1001 			goto retry;
1002 		}
1003 	} else {
1004 		ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1005 		goto free_keys;
1006 	}
1007 
1008 	ret = TCM_NO_SENSE;
1009 
1010 	put_unaligned_be32(keys->generation, &param_data[0]);
1011 	if (!keys->num_keys) {
1012 		put_unaligned_be32(0, &param_data[4]);
1013 		goto free_keys;
1014 	}
1015 
1016 	put_unaligned_be32(8 * keys->num_keys, &param_data[4]);
1017 
1018 	data_offset = 8;
1019 	for (i = 0; i < keys->num_keys; i++) {
1020 		if (data_offset + 8 > cmd->data_length)
1021 			break;
1022 
1023 		put_unaligned_be64(keys->keys[i], &param_data[data_offset]);
1024 		data_offset += 8;
1025 	}
1026 
1027 free_keys:
1028 	kfree(keys);
1029 	return ret;
1030 }
1031 
1032 static sense_reason_t iblock_pr_read_reservation(struct se_cmd *cmd,
1033 						 unsigned char *param_data)
1034 {
1035 	struct se_device *dev = cmd->se_dev;
1036 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1037 	struct block_device *bdev = ib_dev->ibd_bd;
1038 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
1039 	struct pr_held_reservation rsv = { };
1040 
1041 	if (!ops) {
1042 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
1043 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1044 	}
1045 
1046 	if (!ops->pr_read_reservation) {
1047 		pr_err("Block device does not support read_keys.\n");
1048 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1049 	}
1050 
1051 	if (ops->pr_read_reservation(bdev, &rsv))
1052 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1053 
1054 	put_unaligned_be32(rsv.generation, &param_data[0]);
1055 	if (!block_pr_type_to_scsi(rsv.type)) {
1056 		put_unaligned_be32(0, &param_data[4]);
1057 		return TCM_NO_SENSE;
1058 	}
1059 
1060 	put_unaligned_be32(16, &param_data[4]);
1061 
1062 	if (cmd->data_length < 16)
1063 		return TCM_NO_SENSE;
1064 	put_unaligned_be64(rsv.key, &param_data[8]);
1065 
1066 	if (cmd->data_length < 22)
1067 		return TCM_NO_SENSE;
1068 	param_data[21] = block_pr_type_to_scsi(rsv.type);
1069 
1070 	return TCM_NO_SENSE;
1071 }
1072 
1073 static sense_reason_t iblock_execute_pr_in(struct se_cmd *cmd, u8 sa,
1074 					   unsigned char *param_data)
1075 {
1076 	sense_reason_t ret = TCM_NO_SENSE;
1077 
1078 	switch (sa) {
1079 	case PRI_REPORT_CAPABILITIES:
1080 		iblock_pr_report_caps(param_data);
1081 		break;
1082 	case PRI_READ_KEYS:
1083 		ret = iblock_pr_read_keys(cmd, param_data);
1084 		break;
1085 	case PRI_READ_RESERVATION:
1086 		ret = iblock_pr_read_reservation(cmd, param_data);
1087 		break;
1088 	default:
1089 		pr_err("Unknown PERSISTENT_RESERVE_IN SA: 0x%02x\n", sa);
1090 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1091 	}
1092 
1093 	return ret;
1094 }
1095 
1096 static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
1097 {
1098 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1099 	struct block_device *bd = ib_dev->ibd_bd;
1100 	int ret;
1101 
1102 	ret = bdev_alignment_offset(bd);
1103 	if (ret == -1)
1104 		return 0;
1105 
1106 	/* convert offset-bytes to offset-lbas */
1107 	return ret / bdev_logical_block_size(bd);
1108 }
1109 
1110 static unsigned int iblock_get_lbppbe(struct se_device *dev)
1111 {
1112 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1113 	struct block_device *bd = ib_dev->ibd_bd;
1114 	unsigned int logs_per_phys =
1115 		bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
1116 
1117 	return ilog2(logs_per_phys);
1118 }
1119 
1120 static unsigned int iblock_get_io_min(struct se_device *dev)
1121 {
1122 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1123 	struct block_device *bd = ib_dev->ibd_bd;
1124 
1125 	return bdev_io_min(bd);
1126 }
1127 
1128 static unsigned int iblock_get_io_opt(struct se_device *dev)
1129 {
1130 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1131 	struct block_device *bd = ib_dev->ibd_bd;
1132 
1133 	return bdev_io_opt(bd);
1134 }
1135 
1136 static struct exec_cmd_ops iblock_exec_cmd_ops = {
1137 	.execute_rw		= iblock_execute_rw,
1138 	.execute_sync_cache	= iblock_execute_sync_cache,
1139 	.execute_write_same	= iblock_execute_write_same,
1140 	.execute_unmap		= iblock_execute_unmap,
1141 	.execute_pr_out		= iblock_execute_pr_out,
1142 	.execute_pr_in		= iblock_execute_pr_in,
1143 };
1144 
1145 static sense_reason_t
1146 iblock_parse_cdb(struct se_cmd *cmd)
1147 {
1148 	return sbc_parse_cdb(cmd, &iblock_exec_cmd_ops);
1149 }
1150 
1151 static bool iblock_get_write_cache(struct se_device *dev)
1152 {
1153 	return bdev_write_cache(IBLOCK_DEV(dev)->ibd_bd);
1154 }
1155 
1156 static const struct target_backend_ops iblock_ops = {
1157 	.name			= "iblock",
1158 	.inquiry_prod		= "IBLOCK",
1159 	.transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR,
1160 	.inquiry_rev		= IBLOCK_VERSION,
1161 	.owner			= THIS_MODULE,
1162 	.attach_hba		= iblock_attach_hba,
1163 	.detach_hba		= iblock_detach_hba,
1164 	.alloc_device		= iblock_alloc_device,
1165 	.configure_device	= iblock_configure_device,
1166 	.destroy_device		= iblock_destroy_device,
1167 	.free_device		= iblock_free_device,
1168 	.configure_unmap	= iblock_configure_unmap,
1169 	.plug_device		= iblock_plug_device,
1170 	.unplug_device		= iblock_unplug_device,
1171 	.parse_cdb		= iblock_parse_cdb,
1172 	.set_configfs_dev_params = iblock_set_configfs_dev_params,
1173 	.show_configfs_dev_params = iblock_show_configfs_dev_params,
1174 	.get_device_type	= sbc_get_device_type,
1175 	.get_blocks		= iblock_get_blocks,
1176 	.get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
1177 	.get_lbppbe		= iblock_get_lbppbe,
1178 	.get_io_min		= iblock_get_io_min,
1179 	.get_io_opt		= iblock_get_io_opt,
1180 	.get_write_cache	= iblock_get_write_cache,
1181 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
1182 };
1183 
1184 static int __init iblock_module_init(void)
1185 {
1186 	return transport_backend_register(&iblock_ops);
1187 }
1188 
1189 static void __exit iblock_module_exit(void)
1190 {
1191 	target_backend_unregister(&iblock_ops);
1192 }
1193 
1194 MODULE_DESCRIPTION("TCM IBLOCK subsystem plugin");
1195 MODULE_AUTHOR("nab@Linux-iSCSI.org");
1196 MODULE_LICENSE("GPL");
1197 
1198 module_init(iblock_module_init);
1199 module_exit(iblock_module_exit);
1200