xref: /linux/drivers/target/target_core_iblock.c (revision 9b960d8cd6f712cb2c03e2bdd4d5ca058238037f)
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 
IBLOCK_DEV(struct se_device * dev)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 
iblock_attach_hba(struct se_hba * hba,u32 host_id)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 
iblock_detach_hba(struct se_hba * hba)54 static void iblock_detach_hba(struct se_hba *hba)
55 {
56 }
57 
iblock_alloc_device(struct se_hba * hba,const char * name)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 
iblock_configure_unmap(struct se_device * dev)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 
iblock_configure_device(struct se_device * dev)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 	dev->dev_attrib.hw_pi_prot_type = dev->dev_attrib.pi_prot_type;
171 	return 0;
172 
173 out_blkdev_put:
174 	fput(ib_dev->ibd_bdev_file);
175 out_free_bioset:
176 	bioset_exit(&ib_dev->ibd_bio_set);
177 out:
178 	return ret;
179 }
180 
iblock_dev_call_rcu(struct rcu_head * p)181 static void iblock_dev_call_rcu(struct rcu_head *p)
182 {
183 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
184 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
185 
186 	kfree(ib_dev->ibd_plug);
187 	kfree(ib_dev);
188 }
189 
iblock_free_device(struct se_device * dev)190 static void iblock_free_device(struct se_device *dev)
191 {
192 	call_rcu(&dev->rcu_head, iblock_dev_call_rcu);
193 }
194 
iblock_destroy_device(struct se_device * dev)195 static void iblock_destroy_device(struct se_device *dev)
196 {
197 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
198 
199 	if (ib_dev->ibd_bdev_file)
200 		fput(ib_dev->ibd_bdev_file);
201 	bioset_exit(&ib_dev->ibd_bio_set);
202 }
203 
iblock_plug_device(struct se_device * se_dev)204 static struct se_dev_plug *iblock_plug_device(struct se_device *se_dev)
205 {
206 	struct iblock_dev *ib_dev = IBLOCK_DEV(se_dev);
207 	struct iblock_dev_plug *ib_dev_plug;
208 
209 	/*
210 	 * Each se_device has a per cpu work this can be run from. We
211 	 * shouldn't have multiple threads on the same cpu calling this
212 	 * at the same time.
213 	 */
214 	ib_dev_plug = &ib_dev->ibd_plug[raw_smp_processor_id()];
215 	if (test_and_set_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags))
216 		return NULL;
217 
218 	blk_start_plug(&ib_dev_plug->blk_plug);
219 	return &ib_dev_plug->se_plug;
220 }
221 
iblock_unplug_device(struct se_dev_plug * se_plug)222 static void iblock_unplug_device(struct se_dev_plug *se_plug)
223 {
224 	struct iblock_dev_plug *ib_dev_plug = container_of(se_plug,
225 					struct iblock_dev_plug, se_plug);
226 
227 	blk_finish_plug(&ib_dev_plug->blk_plug);
228 	clear_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags);
229 }
230 
iblock_get_blocks(struct se_device * dev)231 static sector_t iblock_get_blocks(struct se_device *dev)
232 {
233 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
234 	u32 block_size = bdev_logical_block_size(ib_dev->ibd_bd);
235 	unsigned long long blocks_long =
236 		div_u64(bdev_nr_bytes(ib_dev->ibd_bd), block_size) - 1;
237 
238 	if (block_size == dev->dev_attrib.block_size)
239 		return blocks_long;
240 
241 	switch (block_size) {
242 	case 4096:
243 		switch (dev->dev_attrib.block_size) {
244 		case 2048:
245 			blocks_long <<= 1;
246 			break;
247 		case 1024:
248 			blocks_long <<= 2;
249 			break;
250 		case 512:
251 			blocks_long <<= 3;
252 			break;
253 		default:
254 			break;
255 		}
256 		break;
257 	case 2048:
258 		switch (dev->dev_attrib.block_size) {
259 		case 4096:
260 			blocks_long >>= 1;
261 			break;
262 		case 1024:
263 			blocks_long <<= 1;
264 			break;
265 		case 512:
266 			blocks_long <<= 2;
267 			break;
268 		default:
269 			break;
270 		}
271 		break;
272 	case 1024:
273 		switch (dev->dev_attrib.block_size) {
274 		case 4096:
275 			blocks_long >>= 2;
276 			break;
277 		case 2048:
278 			blocks_long >>= 1;
279 			break;
280 		case 512:
281 			blocks_long <<= 1;
282 			break;
283 		default:
284 			break;
285 		}
286 		break;
287 	case 512:
288 		switch (dev->dev_attrib.block_size) {
289 		case 4096:
290 			blocks_long >>= 3;
291 			break;
292 		case 2048:
293 			blocks_long >>= 2;
294 			break;
295 		case 1024:
296 			blocks_long >>= 1;
297 			break;
298 		default:
299 			break;
300 		}
301 		break;
302 	default:
303 		break;
304 	}
305 
306 	return blocks_long;
307 }
308 
iblock_complete_cmd(struct se_cmd * cmd,blk_status_t blk_status)309 static void iblock_complete_cmd(struct se_cmd *cmd, blk_status_t blk_status)
310 {
311 	struct iblock_req *ibr = cmd->priv;
312 	u8 status;
313 
314 	if (!refcount_dec_and_test(&ibr->pending))
315 		return;
316 
317 	if (blk_status == BLK_STS_RESV_CONFLICT)
318 		status = SAM_STAT_RESERVATION_CONFLICT;
319 	else if (atomic_read(&ibr->ib_bio_err_cnt))
320 		status = SAM_STAT_CHECK_CONDITION;
321 	else
322 		status = SAM_STAT_GOOD;
323 
324 	target_complete_cmd(cmd, status);
325 	kfree(ibr);
326 }
327 
iblock_bio_done(struct bio * bio)328 static void iblock_bio_done(struct bio *bio)
329 {
330 	struct se_cmd *cmd = bio->bi_private;
331 	struct iblock_req *ibr = cmd->priv;
332 	blk_status_t blk_status = bio->bi_status;
333 
334 	if (bio->bi_status) {
335 		pr_err("bio error: %p,  err: %d\n", bio, bio->bi_status);
336 		/*
337 		 * Bump the ib_bio_err_cnt and release bio.
338 		 */
339 		atomic_inc(&ibr->ib_bio_err_cnt);
340 		smp_mb__after_atomic();
341 	}
342 
343 	bio_put(bio);
344 
345 	iblock_complete_cmd(cmd, blk_status);
346 }
347 
iblock_get_bio(struct se_cmd * cmd,sector_t lba,u32 sg_num,blk_opf_t opf)348 static struct bio *iblock_get_bio(struct se_cmd *cmd, sector_t lba, u32 sg_num,
349 				  blk_opf_t opf)
350 {
351 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
352 	struct bio *bio;
353 
354 	/*
355 	 * Only allocate as many vector entries as the bio code allows us to,
356 	 * we'll loop later on until we have handled the whole request.
357 	 */
358 	bio = bio_alloc_bioset(ib_dev->ibd_bd, bio_max_segs(sg_num), opf,
359 			       GFP_NOIO, &ib_dev->ibd_bio_set);
360 	if (!bio) {
361 		pr_err("Unable to allocate memory for bio\n");
362 		return NULL;
363 	}
364 
365 	bio->bi_private = cmd;
366 	bio->bi_end_io = &iblock_bio_done;
367 	bio->bi_iter.bi_sector = lba;
368 
369 	return bio;
370 }
371 
iblock_submit_bios(struct bio_list * list)372 static void iblock_submit_bios(struct bio_list *list)
373 {
374 	struct blk_plug plug;
375 	struct bio *bio;
376 	/*
377 	 * The block layer handles nested plugs, so just plug/unplug to handle
378 	 * fabric drivers that didn't support batching and multi bio cmds.
379 	 */
380 	blk_start_plug(&plug);
381 	while ((bio = bio_list_pop(list)))
382 		submit_bio(bio);
383 	blk_finish_plug(&plug);
384 }
385 
iblock_end_io_flush(struct bio * bio)386 static void iblock_end_io_flush(struct bio *bio)
387 {
388 	struct se_cmd *cmd = bio->bi_private;
389 
390 	if (bio->bi_status)
391 		pr_err("IBLOCK: cache flush failed: %d\n", bio->bi_status);
392 
393 	if (cmd) {
394 		if (bio->bi_status)
395 			target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
396 		else
397 			target_complete_cmd(cmd, SAM_STAT_GOOD);
398 	}
399 
400 	bio_put(bio);
401 }
402 
403 /*
404  * Implement SYCHRONIZE CACHE.  Note that we can't handle lba ranges and must
405  * always flush the whole cache.
406  */
407 static sense_reason_t
iblock_execute_sync_cache(struct se_cmd * cmd)408 iblock_execute_sync_cache(struct se_cmd *cmd)
409 {
410 	struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
411 	int immed = (cmd->t_task_cdb[1] & 0x2);
412 	struct bio *bio;
413 
414 	/*
415 	 * If the Immediate bit is set, queue up the GOOD response
416 	 * for this SYNCHRONIZE_CACHE op.
417 	 */
418 	if (immed)
419 		target_complete_cmd(cmd, SAM_STAT_GOOD);
420 
421 	bio = bio_alloc(ib_dev->ibd_bd, 0, REQ_OP_WRITE | REQ_PREFLUSH,
422 			GFP_KERNEL);
423 	bio->bi_end_io = iblock_end_io_flush;
424 	if (!immed)
425 		bio->bi_private = cmd;
426 	submit_bio(bio);
427 	return 0;
428 }
429 
430 static sense_reason_t
iblock_execute_unmap(struct se_cmd * cmd,sector_t lba,sector_t nolb)431 iblock_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
432 {
433 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
434 	struct se_device *dev = cmd->se_dev;
435 	int ret;
436 
437 	ret = blkdev_issue_discard(bdev,
438 				   target_to_linux_sector(dev, lba),
439 				   target_to_linux_sector(dev,  nolb),
440 				   GFP_KERNEL);
441 	if (ret < 0) {
442 		pr_err("blkdev_issue_discard() failed: %d\n", ret);
443 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
444 	}
445 
446 	return 0;
447 }
448 
449 static sense_reason_t
iblock_execute_zero_out(struct block_device * bdev,struct se_cmd * cmd)450 iblock_execute_zero_out(struct block_device *bdev, struct se_cmd *cmd)
451 {
452 	struct se_device *dev = cmd->se_dev;
453 	struct scatterlist *sg = &cmd->t_data_sg[0];
454 	unsigned char *buf, *not_zero;
455 	int ret;
456 
457 	buf = kmap(sg_page(sg)) + sg->offset;
458 	if (!buf)
459 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
460 	/*
461 	 * Fall back to block_execute_write_same() slow-path if
462 	 * incoming WRITE_SAME payload does not contain zeros.
463 	 */
464 	not_zero = memchr_inv(buf, 0x00, cmd->data_length);
465 	kunmap(sg_page(sg));
466 
467 	if (not_zero)
468 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
469 
470 	ret = blkdev_issue_zeroout(bdev,
471 				target_to_linux_sector(dev, cmd->t_task_lba),
472 				target_to_linux_sector(dev,
473 					sbc_get_write_same_sectors(cmd)),
474 				GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
475 	if (ret)
476 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
477 
478 	target_complete_cmd(cmd, SAM_STAT_GOOD);
479 	return 0;
480 }
481 
482 static sense_reason_t
iblock_execute_write_same(struct se_cmd * cmd)483 iblock_execute_write_same(struct se_cmd *cmd)
484 {
485 	struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
486 	struct iblock_req *ibr;
487 	struct scatterlist *sg;
488 	struct bio *bio;
489 	struct bio_list list;
490 	struct se_device *dev = cmd->se_dev;
491 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
492 	sector_t sectors = target_to_linux_sector(dev,
493 					sbc_get_write_same_sectors(cmd));
494 
495 	if (cmd->prot_op) {
496 		pr_err("WRITE_SAME: Protection information with IBLOCK"
497 		       " backends not supported\n");
498 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
499 	}
500 
501 	if (!cmd->t_data_nents)
502 		return TCM_INVALID_CDB_FIELD;
503 
504 	sg = &cmd->t_data_sg[0];
505 
506 	if (cmd->t_data_nents > 1 ||
507 	    sg->length != cmd->se_dev->dev_attrib.block_size) {
508 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
509 			" block_size: %u\n", cmd->t_data_nents, sg->length,
510 			cmd->se_dev->dev_attrib.block_size);
511 		return TCM_INVALID_CDB_FIELD;
512 	}
513 
514 	if (bdev_write_zeroes_sectors(bdev)) {
515 		if (!iblock_execute_zero_out(bdev, cmd))
516 			return 0;
517 	}
518 
519 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
520 	if (!ibr)
521 		goto fail;
522 	cmd->priv = ibr;
523 
524 	bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
525 	if (!bio)
526 		goto fail_free_ibr;
527 
528 	bio_list_init(&list);
529 	bio_list_add(&list, bio);
530 
531 	refcount_set(&ibr->pending, 1);
532 
533 	while (sectors) {
534 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
535 				!= sg->length) {
536 
537 			bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
538 			if (!bio)
539 				goto fail_put_bios;
540 
541 			refcount_inc(&ibr->pending);
542 			bio_list_add(&list, bio);
543 		}
544 
545 		/* Always in 512 byte units for Linux/Block */
546 		block_lba += sg->length >> SECTOR_SHIFT;
547 		sectors -= sg->length >> SECTOR_SHIFT;
548 	}
549 
550 	iblock_submit_bios(&list);
551 	return 0;
552 
553 fail_put_bios:
554 	while ((bio = bio_list_pop(&list)))
555 		bio_put(bio);
556 fail_free_ibr:
557 	kfree(ibr);
558 fail:
559 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
560 }
561 
562 enum {
563 	Opt_udev_path, Opt_readonly, Opt_force, Opt_err
564 };
565 
566 static match_table_t tokens = {
567 	{Opt_udev_path, "udev_path=%s"},
568 	{Opt_readonly, "readonly=%d"},
569 	{Opt_force, "force=%d"},
570 	{Opt_err, NULL}
571 };
572 
iblock_set_configfs_dev_params(struct se_device * dev,const char * page,ssize_t count)573 static ssize_t iblock_set_configfs_dev_params(struct se_device *dev,
574 		const char *page, ssize_t count)
575 {
576 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
577 	char *orig, *ptr, *arg_p, *opts;
578 	substring_t args[MAX_OPT_ARGS];
579 	int ret = 0, token;
580 	unsigned long tmp_readonly;
581 
582 	opts = kstrdup(page, GFP_KERNEL);
583 	if (!opts)
584 		return -ENOMEM;
585 
586 	orig = opts;
587 
588 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
589 		if (!*ptr)
590 			continue;
591 
592 		token = match_token(ptr, tokens, args);
593 		switch (token) {
594 		case Opt_udev_path:
595 			if (ib_dev->ibd_bd) {
596 				pr_err("Unable to set udev_path= while"
597 					" ib_dev->ibd_bd exists\n");
598 				ret = -EEXIST;
599 				goto out;
600 			}
601 			if (match_strlcpy(ib_dev->ibd_udev_path, &args[0],
602 				SE_UDEV_PATH_LEN) == 0) {
603 				ret = -EINVAL;
604 				break;
605 			}
606 			pr_debug("IBLOCK: Referencing UDEV path: %s\n",
607 					ib_dev->ibd_udev_path);
608 			ib_dev->ibd_flags |= IBDF_HAS_UDEV_PATH;
609 			break;
610 		case Opt_readonly:
611 			arg_p = match_strdup(&args[0]);
612 			if (!arg_p) {
613 				ret = -ENOMEM;
614 				break;
615 			}
616 			ret = kstrtoul(arg_p, 0, &tmp_readonly);
617 			kfree(arg_p);
618 			if (ret < 0) {
619 				pr_err("kstrtoul() failed for"
620 						" readonly=\n");
621 				goto out;
622 			}
623 			ib_dev->ibd_readonly = tmp_readonly;
624 			pr_debug("IBLOCK: readonly: %d\n", ib_dev->ibd_readonly);
625 			break;
626 		case Opt_force:
627 			break;
628 		default:
629 			break;
630 		}
631 	}
632 
633 out:
634 	kfree(orig);
635 	return (!ret) ? count : ret;
636 }
637 
iblock_show_configfs_dev_params(struct se_device * dev,char * b)638 static ssize_t iblock_show_configfs_dev_params(struct se_device *dev, char *b)
639 {
640 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
641 	struct block_device *bd = ib_dev->ibd_bd;
642 	ssize_t bl = 0;
643 
644 	if (bd)
645 		bl += sprintf(b + bl, "iBlock device: %pg", bd);
646 	if (ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)
647 		bl += sprintf(b + bl, "  UDEV PATH: %s",
648 				ib_dev->ibd_udev_path);
649 	bl += sprintf(b + bl, "  readonly: %d\n", ib_dev->ibd_readonly);
650 
651 	bl += sprintf(b + bl, "        ");
652 	if (bd) {
653 		bl += sprintf(b + bl, "Major: %d Minor: %d  %s\n",
654 			MAJOR(bd->bd_dev), MINOR(bd->bd_dev),
655 			"CLAIMED: IBLOCK");
656 	} else {
657 		bl += sprintf(b + bl, "Major: 0 Minor: 0\n");
658 	}
659 
660 	return bl;
661 }
662 
663 static int
iblock_alloc_bip(struct se_cmd * cmd,struct bio * bio,struct sg_mapping_iter * miter)664 iblock_alloc_bip(struct se_cmd *cmd, struct bio *bio,
665 		 struct sg_mapping_iter *miter)
666 {
667 	struct se_device *dev = cmd->se_dev;
668 	struct blk_integrity *bi;
669 	struct bio_integrity_payload *bip;
670 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
671 	int rc;
672 	size_t resid, len;
673 
674 	bi = bdev_get_integrity(ib_dev->ibd_bd);
675 	if (!bi) {
676 		pr_err("Unable to locate bio_integrity\n");
677 		return -ENODEV;
678 	}
679 
680 	bip = bio_integrity_alloc(bio, GFP_NOIO, bio_max_segs(cmd->t_prot_nents));
681 	if (IS_ERR(bip)) {
682 		pr_err("Unable to allocate bio_integrity_payload\n");
683 		return PTR_ERR(bip);
684 	}
685 
686 	/* virtual start sector must be in integrity interval units */
687 	bip_set_seed(bip, bio->bi_iter.bi_sector >>
688 				  (bi->interval_exp - SECTOR_SHIFT));
689 
690 	pr_debug("IBLOCK BIP Size: %u Sector: %llu\n", bip->bip_iter.bi_size,
691 		 (unsigned long long)bip->bip_iter.bi_sector);
692 
693 	resid = bio_integrity_bytes(bi, bio_sectors(bio));
694 	while (resid > 0 && sg_miter_next(miter)) {
695 
696 		len = min_t(size_t, miter->length, resid);
697 		rc = bio_integrity_add_page(bio, miter->page, len,
698 					    offset_in_page(miter->addr));
699 		if (rc != len) {
700 			pr_err("bio_integrity_add_page() failed; %d\n", rc);
701 			sg_miter_stop(miter);
702 			return -ENOMEM;
703 		}
704 
705 		pr_debug("Added bio integrity page: %p length: %zu offset: %lu\n",
706 			  miter->page, len, offset_in_page(miter->addr));
707 
708 		resid -= len;
709 		if (len < miter->length)
710 			miter->consumed -= miter->length - len;
711 	}
712 	sg_miter_stop(miter);
713 
714 	return 0;
715 }
716 
717 static sense_reason_t
iblock_execute_rw(struct se_cmd * cmd,struct scatterlist * sgl,u32 sgl_nents,enum dma_data_direction data_direction)718 iblock_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
719 		  enum dma_data_direction data_direction)
720 {
721 	struct se_device *dev = cmd->se_dev;
722 	sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
723 	struct iblock_req *ibr;
724 	struct bio *bio;
725 	struct bio_list list;
726 	struct scatterlist *sg;
727 	u32 sg_num = sgl_nents;
728 	blk_opf_t opf;
729 	unsigned bio_cnt;
730 	int i, rc;
731 	struct sg_mapping_iter prot_miter;
732 	unsigned int miter_dir;
733 
734 	if (data_direction == DMA_TO_DEVICE) {
735 		struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
736 
737 		/*
738 		 * Set bits to indicate WRITE_ODIRECT so we are not throttled
739 		 * by WBT.
740 		 */
741 		opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
742 		/*
743 		 * Force writethrough using REQ_FUA if a volatile write cache
744 		 * is not enabled, or if initiator set the Force Unit Access bit.
745 		 */
746 		miter_dir = SG_MITER_TO_SG;
747 		if (bdev_fua(ib_dev->ibd_bd)) {
748 			if (cmd->se_cmd_flags & SCF_FUA)
749 				opf |= REQ_FUA;
750 			else if (!bdev_write_cache(ib_dev->ibd_bd))
751 				opf |= REQ_FUA;
752 		}
753 	} else {
754 		opf = REQ_OP_READ;
755 		miter_dir = SG_MITER_FROM_SG;
756 	}
757 
758 	ibr = kzalloc(sizeof(struct iblock_req), GFP_KERNEL);
759 	if (!ibr)
760 		goto fail;
761 	cmd->priv = ibr;
762 
763 	if (!sgl_nents) {
764 		refcount_set(&ibr->pending, 1);
765 		iblock_complete_cmd(cmd, BLK_STS_OK);
766 		return 0;
767 	}
768 
769 	bio = iblock_get_bio(cmd, block_lba, sgl_nents, opf);
770 	if (!bio)
771 		goto fail_free_ibr;
772 
773 	bio_list_init(&list);
774 	bio_list_add(&list, bio);
775 
776 	refcount_set(&ibr->pending, 2);
777 	bio_cnt = 1;
778 
779 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type)
780 		sg_miter_start(&prot_miter, cmd->t_prot_sg, cmd->t_prot_nents,
781 			       miter_dir);
782 
783 	for_each_sg(sgl, sg, sgl_nents, i) {
784 		/*
785 		 * XXX: if the length the device accepts is shorter than the
786 		 *	length of the S/G list entry this will cause and
787 		 *	endless loop.  Better hope no driver uses huge pages.
788 		 */
789 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
790 				!= sg->length) {
791 			if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
792 				rc = iblock_alloc_bip(cmd, bio, &prot_miter);
793 				if (rc)
794 					goto fail_put_bios;
795 			}
796 
797 			if (bio_cnt >= IBLOCK_MAX_BIO_PER_TASK) {
798 				iblock_submit_bios(&list);
799 				bio_cnt = 0;
800 			}
801 
802 			bio = iblock_get_bio(cmd, block_lba, sg_num, opf);
803 			if (!bio)
804 				goto fail_put_bios;
805 
806 			refcount_inc(&ibr->pending);
807 			bio_list_add(&list, bio);
808 			bio_cnt++;
809 		}
810 
811 		/* Always in 512 byte units for Linux/Block */
812 		block_lba += sg->length >> SECTOR_SHIFT;
813 		sg_num--;
814 	}
815 
816 	if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
817 		rc = iblock_alloc_bip(cmd, bio, &prot_miter);
818 		if (rc)
819 			goto fail_put_bios;
820 	}
821 
822 	iblock_submit_bios(&list);
823 	iblock_complete_cmd(cmd, BLK_STS_OK);
824 	return 0;
825 
826 fail_put_bios:
827 	while ((bio = bio_list_pop(&list)))
828 		bio_put(bio);
829 fail_free_ibr:
830 	kfree(ibr);
831 fail:
832 	return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
833 }
834 
iblock_execute_pr_out(struct se_cmd * cmd,u8 sa,u64 key,u64 sa_key,u8 type,bool aptpl)835 static sense_reason_t iblock_execute_pr_out(struct se_cmd *cmd, u8 sa, u64 key,
836 					    u64 sa_key, u8 type, bool aptpl)
837 {
838 	struct se_device *dev = cmd->se_dev;
839 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
840 	struct block_device *bdev = ib_dev->ibd_bd;
841 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
842 	int ret;
843 
844 	if (!ops) {
845 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
846 		return TCM_UNSUPPORTED_SCSI_OPCODE;
847 	}
848 
849 	switch (sa) {
850 	case PRO_REGISTER:
851 	case PRO_REGISTER_AND_IGNORE_EXISTING_KEY:
852 		if (!ops->pr_register) {
853 			pr_err("block device does not support pr_register.\n");
854 			return TCM_UNSUPPORTED_SCSI_OPCODE;
855 		}
856 
857 		/* The block layer pr ops always enables aptpl */
858 		if (!aptpl)
859 			pr_info("APTPL not set by initiator, but will be used.\n");
860 
861 		ret = ops->pr_register(bdev, key, sa_key,
862 				sa == PRO_REGISTER ? 0 : PR_FL_IGNORE_KEY);
863 		break;
864 	case PRO_RESERVE:
865 		if (!ops->pr_reserve) {
866 			pr_err("block_device does not support pr_reserve.\n");
867 			return TCM_UNSUPPORTED_SCSI_OPCODE;
868 		}
869 
870 		ret = ops->pr_reserve(bdev, key, scsi_pr_type_to_block(type), 0);
871 		break;
872 	case PRO_CLEAR:
873 		if (!ops->pr_clear) {
874 			pr_err("block_device does not support pr_clear.\n");
875 			return TCM_UNSUPPORTED_SCSI_OPCODE;
876 		}
877 
878 		ret = ops->pr_clear(bdev, key);
879 		break;
880 	case PRO_PREEMPT:
881 	case PRO_PREEMPT_AND_ABORT:
882 		if (!ops->pr_clear) {
883 			pr_err("block_device does not support pr_preempt.\n");
884 			return TCM_UNSUPPORTED_SCSI_OPCODE;
885 		}
886 
887 		ret = ops->pr_preempt(bdev, key, sa_key,
888 				      scsi_pr_type_to_block(type),
889 				      sa == PRO_PREEMPT_AND_ABORT);
890 		break;
891 	case PRO_RELEASE:
892 		if (!ops->pr_clear) {
893 			pr_err("block_device does not support pr_pclear.\n");
894 			return TCM_UNSUPPORTED_SCSI_OPCODE;
895 		}
896 
897 		ret = ops->pr_release(bdev, key, scsi_pr_type_to_block(type));
898 		break;
899 	default:
900 		pr_err("Unknown PERSISTENT_RESERVE_OUT SA: 0x%02x\n", sa);
901 		return TCM_UNSUPPORTED_SCSI_OPCODE;
902 	}
903 
904 	if (!ret)
905 		return TCM_NO_SENSE;
906 	else if (ret == PR_STS_RESERVATION_CONFLICT)
907 		return TCM_RESERVATION_CONFLICT;
908 	else
909 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
910 }
911 
iblock_pr_report_caps(unsigned char * param_data)912 static void iblock_pr_report_caps(unsigned char *param_data)
913 {
914 	u16 len = 8;
915 
916 	put_unaligned_be16(len, &param_data[0]);
917 	/*
918 	 * When using the pr_ops passthrough method we only support exporting
919 	 * the device through one target port because from the backend module
920 	 * level we can't see the target port config. As a result we only
921 	 * support registration directly from the I_T nexus the cmd is sent
922 	 * through and do not set ATP_C here.
923 	 *
924 	 * The block layer pr_ops do not support passing in initiators so
925 	 * we don't set SIP_C here.
926 	 */
927 	/* PTPL_C: Persistence across Target Power Loss bit */
928 	param_data[2] |= 0x01;
929 	/*
930 	 * We are filling in the PERSISTENT RESERVATION TYPE MASK below, so
931 	 * set the TMV: Task Mask Valid bit.
932 	 */
933 	param_data[3] |= 0x80;
934 	/*
935 	 * Change ALLOW COMMANDs to 0x20 or 0x40 later from Table 166
936 	 */
937 	param_data[3] |= 0x10; /* ALLOW COMMANDs field 001b */
938 	/*
939 	 * PTPL_A: Persistence across Target Power Loss Active bit. The block
940 	 * layer pr ops always enables this so report it active.
941 	 */
942 	param_data[3] |= 0x01;
943 	/*
944 	 * Setup the PERSISTENT RESERVATION TYPE MASK from Table 212 spc4r37.
945 	 */
946 	param_data[4] |= 0x80; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
947 	param_data[4] |= 0x40; /* PR_TYPE_EXCLUSIVE_ACCESS_REGONLY */
948 	param_data[4] |= 0x20; /* PR_TYPE_WRITE_EXCLUSIVE_REGONLY */
949 	param_data[4] |= 0x08; /* PR_TYPE_EXCLUSIVE_ACCESS */
950 	param_data[4] |= 0x02; /* PR_TYPE_WRITE_EXCLUSIVE */
951 	param_data[5] |= 0x01; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
952 }
953 
iblock_pr_read_keys(struct se_cmd * cmd,unsigned char * param_data)954 static sense_reason_t iblock_pr_read_keys(struct se_cmd *cmd,
955 					  unsigned char *param_data)
956 {
957 	struct se_device *dev = cmd->se_dev;
958 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
959 	struct block_device *bdev = ib_dev->ibd_bd;
960 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
961 	int i, len, paths, data_offset;
962 	struct pr_keys *keys;
963 	sense_reason_t ret;
964 
965 	if (!ops) {
966 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
967 		return TCM_UNSUPPORTED_SCSI_OPCODE;
968 	}
969 
970 	if (!ops->pr_read_keys) {
971 		pr_err("Block device does not support read_keys.\n");
972 		return TCM_UNSUPPORTED_SCSI_OPCODE;
973 	}
974 
975 	/*
976 	 * We don't know what's under us, but dm-multipath will register every
977 	 * path with the same key, so start off with enough space for 16 paths.
978 	 * which is not a lot of memory and should normally be enough.
979 	 */
980 	paths = 16;
981 retry:
982 	len = 8 * paths;
983 	keys = kzalloc(sizeof(*keys) + len, GFP_KERNEL);
984 	if (!keys)
985 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
986 
987 	keys->num_keys = paths;
988 	if (!ops->pr_read_keys(bdev, keys)) {
989 		if (keys->num_keys > paths) {
990 			kfree(keys);
991 			paths *= 2;
992 			goto retry;
993 		}
994 	} else {
995 		ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
996 		goto free_keys;
997 	}
998 
999 	ret = TCM_NO_SENSE;
1000 
1001 	put_unaligned_be32(keys->generation, &param_data[0]);
1002 	if (!keys->num_keys) {
1003 		put_unaligned_be32(0, &param_data[4]);
1004 		goto free_keys;
1005 	}
1006 
1007 	put_unaligned_be32(8 * keys->num_keys, &param_data[4]);
1008 
1009 	data_offset = 8;
1010 	for (i = 0; i < keys->num_keys; i++) {
1011 		if (data_offset + 8 > cmd->data_length)
1012 			break;
1013 
1014 		put_unaligned_be64(keys->keys[i], &param_data[data_offset]);
1015 		data_offset += 8;
1016 	}
1017 
1018 free_keys:
1019 	kfree(keys);
1020 	return ret;
1021 }
1022 
iblock_pr_read_reservation(struct se_cmd * cmd,unsigned char * param_data)1023 static sense_reason_t iblock_pr_read_reservation(struct se_cmd *cmd,
1024 						 unsigned char *param_data)
1025 {
1026 	struct se_device *dev = cmd->se_dev;
1027 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1028 	struct block_device *bdev = ib_dev->ibd_bd;
1029 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
1030 	struct pr_held_reservation rsv = { };
1031 
1032 	if (!ops) {
1033 		pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
1034 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1035 	}
1036 
1037 	if (!ops->pr_read_reservation) {
1038 		pr_err("Block device does not support read_keys.\n");
1039 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1040 	}
1041 
1042 	if (ops->pr_read_reservation(bdev, &rsv))
1043 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1044 
1045 	put_unaligned_be32(rsv.generation, &param_data[0]);
1046 	if (!block_pr_type_to_scsi(rsv.type)) {
1047 		put_unaligned_be32(0, &param_data[4]);
1048 		return TCM_NO_SENSE;
1049 	}
1050 
1051 	put_unaligned_be32(16, &param_data[4]);
1052 
1053 	if (cmd->data_length < 16)
1054 		return TCM_NO_SENSE;
1055 	put_unaligned_be64(rsv.key, &param_data[8]);
1056 
1057 	if (cmd->data_length < 22)
1058 		return TCM_NO_SENSE;
1059 	param_data[21] = block_pr_type_to_scsi(rsv.type);
1060 
1061 	return TCM_NO_SENSE;
1062 }
1063 
iblock_execute_pr_in(struct se_cmd * cmd,u8 sa,unsigned char * param_data)1064 static sense_reason_t iblock_execute_pr_in(struct se_cmd *cmd, u8 sa,
1065 					   unsigned char *param_data)
1066 {
1067 	sense_reason_t ret = TCM_NO_SENSE;
1068 
1069 	switch (sa) {
1070 	case PRI_REPORT_CAPABILITIES:
1071 		iblock_pr_report_caps(param_data);
1072 		break;
1073 	case PRI_READ_KEYS:
1074 		ret = iblock_pr_read_keys(cmd, param_data);
1075 		break;
1076 	case PRI_READ_RESERVATION:
1077 		ret = iblock_pr_read_reservation(cmd, param_data);
1078 		break;
1079 	default:
1080 		pr_err("Unknown PERSISTENT_RESERVE_IN SA: 0x%02x\n", sa);
1081 		return TCM_UNSUPPORTED_SCSI_OPCODE;
1082 	}
1083 
1084 	return ret;
1085 }
1086 
iblock_get_alignment_offset_lbas(struct se_device * dev)1087 static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
1088 {
1089 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1090 	struct block_device *bd = ib_dev->ibd_bd;
1091 	int ret;
1092 
1093 	ret = bdev_alignment_offset(bd);
1094 	if (ret == -1)
1095 		return 0;
1096 
1097 	/* convert offset-bytes to offset-lbas */
1098 	return ret / bdev_logical_block_size(bd);
1099 }
1100 
iblock_get_lbppbe(struct se_device * dev)1101 static unsigned int iblock_get_lbppbe(struct se_device *dev)
1102 {
1103 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1104 	struct block_device *bd = ib_dev->ibd_bd;
1105 	unsigned int logs_per_phys =
1106 		bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
1107 
1108 	return ilog2(logs_per_phys);
1109 }
1110 
iblock_get_io_min(struct se_device * dev)1111 static unsigned int iblock_get_io_min(struct se_device *dev)
1112 {
1113 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1114 	struct block_device *bd = ib_dev->ibd_bd;
1115 
1116 	return bdev_io_min(bd);
1117 }
1118 
iblock_get_io_opt(struct se_device * dev)1119 static unsigned int iblock_get_io_opt(struct se_device *dev)
1120 {
1121 	struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1122 	struct block_device *bd = ib_dev->ibd_bd;
1123 
1124 	return bdev_io_opt(bd);
1125 }
1126 
1127 static struct exec_cmd_ops iblock_exec_cmd_ops = {
1128 	.execute_rw		= iblock_execute_rw,
1129 	.execute_sync_cache	= iblock_execute_sync_cache,
1130 	.execute_write_same	= iblock_execute_write_same,
1131 	.execute_unmap		= iblock_execute_unmap,
1132 	.execute_pr_out		= iblock_execute_pr_out,
1133 	.execute_pr_in		= iblock_execute_pr_in,
1134 };
1135 
1136 static sense_reason_t
iblock_parse_cdb(struct se_cmd * cmd)1137 iblock_parse_cdb(struct se_cmd *cmd)
1138 {
1139 	return sbc_parse_cdb(cmd, &iblock_exec_cmd_ops);
1140 }
1141 
iblock_get_write_cache(struct se_device * dev)1142 static bool iblock_get_write_cache(struct se_device *dev)
1143 {
1144 	return bdev_write_cache(IBLOCK_DEV(dev)->ibd_bd);
1145 }
1146 
1147 static const struct target_backend_ops iblock_ops = {
1148 	.name			= "iblock",
1149 	.inquiry_prod		= "IBLOCK",
1150 	.transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR,
1151 	.inquiry_rev		= IBLOCK_VERSION,
1152 	.owner			= THIS_MODULE,
1153 	.attach_hba		= iblock_attach_hba,
1154 	.detach_hba		= iblock_detach_hba,
1155 	.alloc_device		= iblock_alloc_device,
1156 	.configure_device	= iblock_configure_device,
1157 	.destroy_device		= iblock_destroy_device,
1158 	.free_device		= iblock_free_device,
1159 	.configure_unmap	= iblock_configure_unmap,
1160 	.plug_device		= iblock_plug_device,
1161 	.unplug_device		= iblock_unplug_device,
1162 	.parse_cdb		= iblock_parse_cdb,
1163 	.set_configfs_dev_params = iblock_set_configfs_dev_params,
1164 	.show_configfs_dev_params = iblock_show_configfs_dev_params,
1165 	.get_device_type	= sbc_get_device_type,
1166 	.get_blocks		= iblock_get_blocks,
1167 	.get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
1168 	.get_lbppbe		= iblock_get_lbppbe,
1169 	.get_io_min		= iblock_get_io_min,
1170 	.get_io_opt		= iblock_get_io_opt,
1171 	.get_write_cache	= iblock_get_write_cache,
1172 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
1173 };
1174 
iblock_module_init(void)1175 static int __init iblock_module_init(void)
1176 {
1177 	return transport_backend_register(&iblock_ops);
1178 }
1179 
iblock_module_exit(void)1180 static void __exit iblock_module_exit(void)
1181 {
1182 	target_backend_unregister(&iblock_ops);
1183 }
1184 
1185 MODULE_DESCRIPTION("TCM IBLOCK subsystem plugin");
1186 MODULE_AUTHOR("nab@Linux-iSCSI.org");
1187 MODULE_LICENSE("GPL");
1188 
1189 module_init(iblock_module_init);
1190 module_exit(iblock_module_exit);
1191