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_obj(struct iblock_dev);
63 if (!ib_dev) {
64 pr_err("Unable to allocate struct iblock_dev\n");
65 return NULL;
66 }
67 ib_dev->ibd_exclusive = true;
68
69 ib_dev->ibd_plug = kzalloc_objs(*ib_dev->ibd_plug, nr_cpu_ids);
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_bdev(&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 void *holder = ib_dev;
99 unsigned int max_write_zeroes_sectors;
100 int ret;
101
102 if (!(ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)) {
103 pr_err("Missing udev_path= parameters for IBLOCK\n");
104 return -EINVAL;
105 }
106
107 ret = bioset_init(&ib_dev->ibd_bio_set, IBLOCK_BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
108 if (ret) {
109 pr_err("IBLOCK: Unable to create bioset\n");
110 goto out;
111 }
112
113 pr_debug("IBLOCK: Claiming struct block_device: %s: %d\n",
114 ib_dev->ibd_udev_path, ib_dev->ibd_exclusive);
115
116 if (!ib_dev->ibd_readonly)
117 mode |= BLK_OPEN_WRITE;
118 else
119 dev->dev_flags |= DF_READ_ONLY;
120
121 if (!ib_dev->ibd_exclusive)
122 holder = NULL;
123
124 bdev_file = bdev_file_open_by_path(ib_dev->ibd_udev_path, mode, holder,
125 NULL);
126 if (IS_ERR(bdev_file)) {
127 ret = PTR_ERR(bdev_file);
128 goto out_free_bioset;
129 }
130 ib_dev->ibd_bdev_file = bdev_file;
131 ib_dev->ibd_bd = bd = file_bdev(bdev_file);
132
133 q = bdev_get_queue(bd);
134
135 dev->dev_attrib.hw_block_size = bdev_logical_block_size(bd);
136 dev->dev_attrib.hw_max_sectors = mult_frac(queue_max_hw_sectors(q),
137 SECTOR_SIZE,
138 dev->dev_attrib.hw_block_size);
139 dev->dev_attrib.hw_queue_depth = q->nr_requests;
140
141 /*
142 * Enable write same emulation for IBLOCK and use 0xFFFF as
143 * the smaller WRITE_SAME(10) only has a two-byte block count.
144 */
145 max_write_zeroes_sectors = bdev_write_zeroes_sectors(bd);
146 if (max_write_zeroes_sectors)
147 dev->dev_attrib.max_write_same_len = max_write_zeroes_sectors;
148 else
149 dev->dev_attrib.max_write_same_len = 0xFFFF;
150
151 if (bdev_nonrot(bd))
152 dev->dev_attrib.is_nonrot = 1;
153
154 target_configure_write_atomic_from_bdev(&dev->dev_attrib, bd);
155
156 bi = bdev_get_integrity(bd);
157 if (!bi)
158 return 0;
159
160 switch (bi->csum_type) {
161 case BLK_INTEGRITY_CSUM_IP:
162 pr_err("IBLOCK export of blk_integrity: %s not supported\n",
163 blk_integrity_profile_name(bi));
164 ret = -ENOSYS;
165 goto out_blkdev_put;
166 case BLK_INTEGRITY_CSUM_CRC:
167 if (bi->flags & BLK_INTEGRITY_REF_TAG)
168 dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE1_PROT;
169 else
170 dev->dev_attrib.pi_prot_type = TARGET_DIF_TYPE3_PROT;
171 break;
172 default:
173 break;
174 }
175
176 dev->dev_attrib.hw_pi_prot_type = dev->dev_attrib.pi_prot_type;
177 return 0;
178
179 out_blkdev_put:
180 fput(ib_dev->ibd_bdev_file);
181 out_free_bioset:
182 bioset_exit(&ib_dev->ibd_bio_set);
183 out:
184 return ret;
185 }
186
iblock_dev_call_rcu(struct rcu_head * p)187 static void iblock_dev_call_rcu(struct rcu_head *p)
188 {
189 struct se_device *dev = container_of(p, struct se_device, rcu_head);
190 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
191
192 kfree(ib_dev->ibd_plug);
193 kfree(ib_dev);
194 }
195
iblock_free_device(struct se_device * dev)196 static void iblock_free_device(struct se_device *dev)
197 {
198 call_rcu(&dev->rcu_head, iblock_dev_call_rcu);
199 }
200
iblock_destroy_device(struct se_device * dev)201 static void iblock_destroy_device(struct se_device *dev)
202 {
203 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
204
205 if (ib_dev->ibd_bdev_file)
206 fput(ib_dev->ibd_bdev_file);
207 bioset_exit(&ib_dev->ibd_bio_set);
208 }
209
iblock_plug_device(struct se_device * se_dev)210 static struct se_dev_plug *iblock_plug_device(struct se_device *se_dev)
211 {
212 struct iblock_dev *ib_dev = IBLOCK_DEV(se_dev);
213 struct iblock_dev_plug *ib_dev_plug;
214
215 /*
216 * Each se_device has a per cpu work this can be run from. We
217 * shouldn't have multiple threads on the same cpu calling this
218 * at the same time.
219 */
220 ib_dev_plug = &ib_dev->ibd_plug[raw_smp_processor_id()];
221 if (test_and_set_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags))
222 return NULL;
223
224 blk_start_plug(&ib_dev_plug->blk_plug);
225 return &ib_dev_plug->se_plug;
226 }
227
iblock_unplug_device(struct se_dev_plug * se_plug)228 static void iblock_unplug_device(struct se_dev_plug *se_plug)
229 {
230 struct iblock_dev_plug *ib_dev_plug = container_of(se_plug,
231 struct iblock_dev_plug, se_plug);
232
233 blk_finish_plug(&ib_dev_plug->blk_plug);
234 clear_bit(IBD_PLUGF_PLUGGED, &ib_dev_plug->flags);
235 }
236
iblock_get_blocks(struct se_device * dev)237 static sector_t iblock_get_blocks(struct se_device *dev)
238 {
239 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
240 u32 block_size = bdev_logical_block_size(ib_dev->ibd_bd);
241 unsigned long long blocks_long =
242 div_u64(bdev_nr_bytes(ib_dev->ibd_bd), block_size) - 1;
243
244 if (block_size == dev->dev_attrib.block_size)
245 return blocks_long;
246
247 switch (block_size) {
248 case 4096:
249 switch (dev->dev_attrib.block_size) {
250 case 2048:
251 blocks_long <<= 1;
252 break;
253 case 1024:
254 blocks_long <<= 2;
255 break;
256 case 512:
257 blocks_long <<= 3;
258 break;
259 default:
260 break;
261 }
262 break;
263 case 2048:
264 switch (dev->dev_attrib.block_size) {
265 case 4096:
266 blocks_long >>= 1;
267 break;
268 case 1024:
269 blocks_long <<= 1;
270 break;
271 case 512:
272 blocks_long <<= 2;
273 break;
274 default:
275 break;
276 }
277 break;
278 case 1024:
279 switch (dev->dev_attrib.block_size) {
280 case 4096:
281 blocks_long >>= 2;
282 break;
283 case 2048:
284 blocks_long >>= 1;
285 break;
286 case 512:
287 blocks_long <<= 1;
288 break;
289 default:
290 break;
291 }
292 break;
293 case 512:
294 switch (dev->dev_attrib.block_size) {
295 case 4096:
296 blocks_long >>= 3;
297 break;
298 case 2048:
299 blocks_long >>= 2;
300 break;
301 case 1024:
302 blocks_long >>= 1;
303 break;
304 default:
305 break;
306 }
307 break;
308 default:
309 break;
310 }
311
312 return blocks_long;
313 }
314
iblock_complete_cmd(struct se_cmd * cmd,blk_status_t blk_status)315 static void iblock_complete_cmd(struct se_cmd *cmd, blk_status_t blk_status)
316 {
317 struct iblock_req *ibr = cmd->priv;
318 u8 status;
319
320 if (!refcount_dec_and_test(&ibr->pending))
321 return;
322
323 if (blk_status == BLK_STS_RESV_CONFLICT)
324 status = SAM_STAT_RESERVATION_CONFLICT;
325 else if (atomic_read(&ibr->ib_bio_err_cnt))
326 status = SAM_STAT_CHECK_CONDITION;
327 else
328 status = SAM_STAT_GOOD;
329
330 target_complete_cmd(cmd, status);
331 kfree(ibr);
332 }
333
iblock_bio_done(struct bio * bio)334 static void iblock_bio_done(struct bio *bio)
335 {
336 struct se_cmd *cmd = bio->bi_private;
337 struct iblock_req *ibr = cmd->priv;
338 blk_status_t blk_status = bio->bi_status;
339
340 if (bio->bi_status) {
341 pr_err("bio error: %p, err: %d\n", bio, bio->bi_status);
342 /*
343 * Bump the ib_bio_err_cnt and release bio.
344 */
345 atomic_inc(&ibr->ib_bio_err_cnt);
346 smp_mb__after_atomic();
347 }
348
349 bio_put(bio);
350
351 iblock_complete_cmd(cmd, blk_status);
352 }
353
iblock_get_bio(struct se_cmd * cmd,sector_t lba,u32 sg_num,blk_opf_t opf)354 static struct bio *iblock_get_bio(struct se_cmd *cmd, sector_t lba, u32 sg_num,
355 blk_opf_t opf)
356 {
357 struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
358 struct bio *bio;
359
360 /*
361 * Only allocate as many vector entries as the bio code allows us to,
362 * we'll loop later on until we have handled the whole request.
363 */
364 bio = bio_alloc_bioset(ib_dev->ibd_bd, bio_max_segs(sg_num), opf,
365 GFP_NOIO, &ib_dev->ibd_bio_set);
366 if (!bio) {
367 pr_err("Unable to allocate memory for bio\n");
368 return NULL;
369 }
370
371 bio->bi_private = cmd;
372 bio->bi_end_io = &iblock_bio_done;
373 bio->bi_iter.bi_sector = lba;
374
375 return bio;
376 }
377
iblock_submit_bios(struct bio_list * list)378 static void iblock_submit_bios(struct bio_list *list)
379 {
380 struct blk_plug plug;
381 struct bio *bio;
382 /*
383 * The block layer handles nested plugs, so just plug/unplug to handle
384 * fabric drivers that didn't support batching and multi bio cmds.
385 */
386 blk_start_plug(&plug);
387 while ((bio = bio_list_pop(list)))
388 submit_bio(bio);
389 blk_finish_plug(&plug);
390 }
391
iblock_end_io_flush(struct bio * bio)392 static void iblock_end_io_flush(struct bio *bio)
393 {
394 struct se_cmd *cmd = bio->bi_private;
395
396 if (bio->bi_status)
397 pr_err("IBLOCK: cache flush failed: %d\n", bio->bi_status);
398
399 if (cmd) {
400 if (bio->bi_status)
401 target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
402 else
403 target_complete_cmd(cmd, SAM_STAT_GOOD);
404 }
405
406 bio_put(bio);
407 }
408
409 /*
410 * Implement SYCHRONIZE CACHE. Note that we can't handle lba ranges and must
411 * always flush the whole cache.
412 */
413 static sense_reason_t
iblock_execute_sync_cache(struct se_cmd * cmd)414 iblock_execute_sync_cache(struct se_cmd *cmd)
415 {
416 struct iblock_dev *ib_dev = IBLOCK_DEV(cmd->se_dev);
417 int immed = (cmd->t_task_cdb[1] & 0x2);
418 struct bio *bio;
419
420 /*
421 * If the Immediate bit is set, queue up the GOOD response
422 * for this SYNCHRONIZE_CACHE op.
423 */
424 if (immed)
425 target_complete_cmd(cmd, SAM_STAT_GOOD);
426
427 bio = bio_alloc(ib_dev->ibd_bd, 0, REQ_OP_WRITE | REQ_PREFLUSH,
428 GFP_KERNEL);
429 bio->bi_end_io = iblock_end_io_flush;
430 if (!immed)
431 bio->bi_private = cmd;
432 submit_bio(bio);
433 return 0;
434 }
435
436 static sense_reason_t
iblock_execute_unmap(struct se_cmd * cmd,sector_t lba,sector_t nolb)437 iblock_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
438 {
439 struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
440 struct se_device *dev = cmd->se_dev;
441 int ret;
442
443 ret = blkdev_issue_discard(bdev,
444 target_to_linux_sector(dev, lba),
445 target_to_linux_sector(dev, nolb),
446 GFP_KERNEL);
447 if (ret < 0) {
448 pr_err("blkdev_issue_discard() failed: %d\n", ret);
449 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
450 }
451
452 return 0;
453 }
454
455 static sense_reason_t
iblock_execute_zero_out(struct block_device * bdev,struct se_cmd * cmd)456 iblock_execute_zero_out(struct block_device *bdev, struct se_cmd *cmd)
457 {
458 struct se_device *dev = cmd->se_dev;
459 struct scatterlist *sg = &cmd->t_data_sg[0];
460 unsigned char *buf, *not_zero;
461 int ret;
462
463 buf = kmap(sg_page(sg)) + sg->offset;
464 if (!buf)
465 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
466 /*
467 * Fall back to block_execute_write_same() slow-path if
468 * incoming WRITE_SAME payload does not contain zeros.
469 */
470 not_zero = memchr_inv(buf, 0x00, cmd->data_length);
471 kunmap(sg_page(sg));
472
473 if (not_zero)
474 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
475
476 ret = blkdev_issue_zeroout(bdev,
477 target_to_linux_sector(dev, cmd->t_task_lba),
478 target_to_linux_sector(dev,
479 sbc_get_write_same_sectors(cmd)),
480 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
481 if (ret)
482 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
483
484 target_complete_cmd(cmd, SAM_STAT_GOOD);
485 return 0;
486 }
487
488 static sense_reason_t
iblock_execute_write_same(struct se_cmd * cmd)489 iblock_execute_write_same(struct se_cmd *cmd)
490 {
491 struct block_device *bdev = IBLOCK_DEV(cmd->se_dev)->ibd_bd;
492 struct iblock_req *ibr;
493 struct scatterlist *sg;
494 struct bio *bio;
495 struct bio_list list;
496 struct se_device *dev = cmd->se_dev;
497 sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
498 sector_t sectors = target_to_linux_sector(dev,
499 sbc_get_write_same_sectors(cmd));
500
501 if (cmd->prot_op) {
502 pr_err("WRITE_SAME: Protection information with IBLOCK"
503 " backends not supported\n");
504 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
505 }
506
507 if (!cmd->t_data_nents)
508 return TCM_INVALID_CDB_FIELD;
509
510 sg = &cmd->t_data_sg[0];
511
512 if (cmd->t_data_nents > 1 ||
513 sg->length != cmd->se_dev->dev_attrib.block_size) {
514 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
515 " block_size: %u\n", cmd->t_data_nents, sg->length,
516 cmd->se_dev->dev_attrib.block_size);
517 return TCM_INVALID_CDB_FIELD;
518 }
519
520 if (bdev_write_zeroes_sectors(bdev)) {
521 if (!iblock_execute_zero_out(bdev, cmd))
522 return 0;
523 }
524
525 ibr = kzalloc_obj(struct iblock_req);
526 if (!ibr)
527 goto fail;
528 cmd->priv = ibr;
529
530 bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
531 if (!bio)
532 goto fail_free_ibr;
533
534 bio_list_init(&list);
535 bio_list_add(&list, bio);
536
537 refcount_set(&ibr->pending, 1);
538
539 while (sectors) {
540 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
541 != sg->length) {
542
543 bio = iblock_get_bio(cmd, block_lba, 1, REQ_OP_WRITE);
544 if (!bio)
545 goto fail_put_bios;
546
547 refcount_inc(&ibr->pending);
548 bio_list_add(&list, bio);
549 }
550
551 /* Always in 512 byte units for Linux/Block */
552 block_lba += sg->length >> SECTOR_SHIFT;
553 sectors -= sg->length >> SECTOR_SHIFT;
554 }
555
556 iblock_submit_bios(&list);
557 return 0;
558
559 fail_put_bios:
560 while ((bio = bio_list_pop(&list)))
561 bio_put(bio);
562 fail_free_ibr:
563 kfree(ibr);
564 fail:
565 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
566 }
567
568 enum {
569 Opt_udev_path, Opt_readonly, Opt_force, Opt_exclusive, Opt_err,
570 };
571
572 static match_table_t tokens = {
573 {Opt_udev_path, "udev_path=%s"},
574 {Opt_readonly, "readonly=%d"},
575 {Opt_force, "force=%d"},
576 {Opt_exclusive, "exclusive=%d"},
577 {Opt_err, NULL}
578 };
579
iblock_set_configfs_dev_params(struct se_device * dev,const char * page,ssize_t count)580 static ssize_t iblock_set_configfs_dev_params(struct se_device *dev,
581 const char *page, ssize_t count)
582 {
583 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
584 char *orig, *ptr, *arg_p, *opts;
585 substring_t args[MAX_OPT_ARGS];
586 int ret = 0, token, tmp_exclusive;
587 unsigned long tmp_readonly;
588
589 opts = kstrdup(page, GFP_KERNEL);
590 if (!opts)
591 return -ENOMEM;
592
593 orig = opts;
594
595 while ((ptr = strsep(&opts, ",\n")) != NULL) {
596 if (!*ptr)
597 continue;
598
599 token = match_token(ptr, tokens, args);
600 switch (token) {
601 case Opt_udev_path:
602 if (ib_dev->ibd_bd) {
603 pr_err("Unable to set udev_path= while"
604 " ib_dev->ibd_bd exists\n");
605 ret = -EEXIST;
606 goto out;
607 }
608 if (match_strlcpy(ib_dev->ibd_udev_path, &args[0],
609 SE_UDEV_PATH_LEN) == 0) {
610 ret = -EINVAL;
611 break;
612 }
613 pr_debug("IBLOCK: Referencing UDEV path: %s\n",
614 ib_dev->ibd_udev_path);
615 ib_dev->ibd_flags |= IBDF_HAS_UDEV_PATH;
616 break;
617 case Opt_readonly:
618 arg_p = match_strdup(&args[0]);
619 if (!arg_p) {
620 ret = -ENOMEM;
621 break;
622 }
623 ret = kstrtoul(arg_p, 0, &tmp_readonly);
624 kfree(arg_p);
625 if (ret < 0) {
626 pr_err("kstrtoul() failed for"
627 " readonly=\n");
628 goto out;
629 }
630 ib_dev->ibd_readonly = tmp_readonly;
631 pr_debug("IBLOCK: readonly: %d\n", ib_dev->ibd_readonly);
632 break;
633 case Opt_exclusive:
634 arg_p = match_strdup(&args[0]);
635 if (!arg_p) {
636 ret = -ENOMEM;
637 break;
638 }
639 ret = kstrtoint(arg_p, 0, &tmp_exclusive);
640 kfree(arg_p);
641 if (ret < 0) {
642 pr_err("kstrtoul() failed for exclusive=\n");
643 goto out;
644 }
645 ib_dev->ibd_exclusive = tmp_exclusive;
646 pr_debug("IBLOCK: exclusive: %d\n",
647 ib_dev->ibd_exclusive);
648 break;
649 case Opt_force:
650 break;
651 default:
652 break;
653 }
654 }
655
656 out:
657 kfree(orig);
658 return (!ret) ? count : ret;
659 }
660
iblock_show_configfs_dev_params(struct se_device * dev,char * b)661 static ssize_t iblock_show_configfs_dev_params(struct se_device *dev, char *b)
662 {
663 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
664 struct block_device *bd = ib_dev->ibd_bd;
665 ssize_t bl = 0;
666
667 if (bd)
668 bl += sprintf(b + bl, "iBlock device: %pg", bd);
669 if (ib_dev->ibd_flags & IBDF_HAS_UDEV_PATH)
670 bl += sprintf(b + bl, " UDEV PATH: %s",
671 ib_dev->ibd_udev_path);
672 bl += sprintf(b + bl, " readonly: %d\n", ib_dev->ibd_readonly);
673 bl += sprintf(b + bl, " exclusive: %d\n", ib_dev->ibd_exclusive);
674
675 bl += sprintf(b + bl, " ");
676 if (bd) {
677 bl += sprintf(b + bl, "Major: %d Minor: %d %s\n",
678 MAJOR(bd->bd_dev), MINOR(bd->bd_dev),
679 "CLAIMED: IBLOCK");
680 } else {
681 bl += sprintf(b + bl, "Major: 0 Minor: 0\n");
682 }
683
684 return bl;
685 }
686
687 static int
iblock_alloc_bip(struct se_cmd * cmd,struct bio * bio,struct sg_mapping_iter * miter)688 iblock_alloc_bip(struct se_cmd *cmd, struct bio *bio,
689 struct sg_mapping_iter *miter)
690 {
691 struct se_device *dev = cmd->se_dev;
692 struct blk_integrity *bi;
693 struct bio_integrity_payload *bip;
694 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
695 int rc;
696 size_t resid, len;
697
698 bi = bdev_get_integrity(ib_dev->ibd_bd);
699 if (!bi) {
700 pr_err("Unable to locate bio_integrity\n");
701 return -ENODEV;
702 }
703
704 bip = bio_integrity_alloc(bio, GFP_NOIO, bio_max_segs(cmd->t_prot_nents));
705 if (IS_ERR(bip)) {
706 pr_err("Unable to allocate bio_integrity_payload\n");
707 return PTR_ERR(bip);
708 }
709
710 /* virtual start sector must be in integrity interval units */
711 bip_set_seed(bip, bio->bi_iter.bi_sector >>
712 (bi->interval_exp - SECTOR_SHIFT));
713
714 pr_debug("IBLOCK BIP Size: %u Sector: %llu\n", bip->bip_iter.bi_size,
715 (unsigned long long)bip->bip_iter.bi_sector);
716
717 resid = bio_integrity_bytes(bi, bio_sectors(bio));
718 while (resid > 0 && sg_miter_next(miter)) {
719
720 len = min_t(size_t, miter->length, resid);
721 rc = bio_integrity_add_page(bio, miter->page, len,
722 offset_in_page(miter->addr));
723 if (rc != len) {
724 pr_err("bio_integrity_add_page() failed; %d\n", rc);
725 sg_miter_stop(miter);
726 return -ENOMEM;
727 }
728
729 pr_debug("Added bio integrity page: %p length: %zu offset: %lu\n",
730 miter->page, len, offset_in_page(miter->addr));
731
732 resid -= len;
733 if (len < miter->length)
734 miter->consumed -= miter->length - len;
735 }
736 sg_miter_stop(miter);
737
738 return 0;
739 }
740
741 static sense_reason_t
iblock_execute_rw(struct se_cmd * cmd,struct scatterlist * sgl,u32 sgl_nents,enum dma_data_direction data_direction)742 iblock_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
743 enum dma_data_direction data_direction)
744 {
745 struct se_device *dev = cmd->se_dev;
746 sector_t block_lba = target_to_linux_sector(dev, cmd->t_task_lba);
747 struct iblock_req *ibr;
748 struct bio *bio;
749 struct bio_list list;
750 struct scatterlist *sg;
751 u32 sg_num = sgl_nents;
752 blk_opf_t opf;
753 unsigned bio_cnt;
754 int i, rc;
755 struct sg_mapping_iter prot_miter;
756 unsigned int miter_dir;
757
758 if (data_direction == DMA_TO_DEVICE) {
759 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
760
761 /*
762 * Set bits to indicate WRITE_ODIRECT so we are not throttled
763 * by WBT.
764 */
765 opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
766 /*
767 * Force writethrough using REQ_FUA if a volatile write cache
768 * is not enabled, or if initiator set the Force Unit Access bit.
769 */
770 miter_dir = SG_MITER_TO_SG;
771 if (bdev_fua(ib_dev->ibd_bd)) {
772 if (cmd->se_cmd_flags & SCF_FUA)
773 opf |= REQ_FUA;
774 else if (!bdev_write_cache(ib_dev->ibd_bd))
775 opf |= REQ_FUA;
776 }
777
778 if (cmd->se_cmd_flags & SCF_ATOMIC)
779 opf |= REQ_ATOMIC;
780 } else {
781 opf = REQ_OP_READ;
782 miter_dir = SG_MITER_FROM_SG;
783 }
784
785 ibr = kzalloc_obj(struct iblock_req);
786 if (!ibr)
787 goto fail;
788 cmd->priv = ibr;
789
790 if (!sgl_nents) {
791 refcount_set(&ibr->pending, 1);
792 iblock_complete_cmd(cmd, BLK_STS_OK);
793 return 0;
794 }
795
796 bio = iblock_get_bio(cmd, block_lba, sgl_nents, opf);
797 if (!bio)
798 goto fail_free_ibr;
799
800 bio_list_init(&list);
801 bio_list_add(&list, bio);
802
803 refcount_set(&ibr->pending, 2);
804 bio_cnt = 1;
805
806 if (cmd->prot_type && dev->dev_attrib.pi_prot_type)
807 sg_miter_start(&prot_miter, cmd->t_prot_sg, cmd->t_prot_nents,
808 miter_dir);
809
810 for_each_sg(sgl, sg, sgl_nents, i) {
811 /*
812 * XXX: if the length the device accepts is shorter than the
813 * length of the S/G list entry this will cause and
814 * endless loop. Better hope no driver uses huge pages.
815 */
816 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
817 != sg->length) {
818 if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
819 rc = iblock_alloc_bip(cmd, bio, &prot_miter);
820 if (rc)
821 goto fail_put_bios;
822 }
823
824 if (bio_cnt >= IBLOCK_MAX_BIO_PER_TASK) {
825 iblock_submit_bios(&list);
826 bio_cnt = 0;
827 }
828
829 bio = iblock_get_bio(cmd, block_lba, sg_num, opf);
830 if (!bio)
831 goto fail_put_bios;
832
833 refcount_inc(&ibr->pending);
834 bio_list_add(&list, bio);
835 bio_cnt++;
836 }
837
838 /* Always in 512 byte units for Linux/Block */
839 block_lba += sg->length >> SECTOR_SHIFT;
840 sg_num--;
841 }
842
843 if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
844 rc = iblock_alloc_bip(cmd, bio, &prot_miter);
845 if (rc)
846 goto fail_put_bios;
847 }
848
849 iblock_submit_bios(&list);
850 iblock_complete_cmd(cmd, BLK_STS_OK);
851 return 0;
852
853 fail_put_bios:
854 while ((bio = bio_list_pop(&list)))
855 bio_put(bio);
856 fail_free_ibr:
857 kfree(ibr);
858 fail:
859 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
860 }
861
iblock_execute_pr_out(struct se_cmd * cmd,u8 sa,u64 key,u64 sa_key,u8 type,bool aptpl)862 static sense_reason_t iblock_execute_pr_out(struct se_cmd *cmd, u8 sa, u64 key,
863 u64 sa_key, u8 type, bool aptpl)
864 {
865 struct se_device *dev = cmd->se_dev;
866 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
867 struct block_device *bdev = ib_dev->ibd_bd;
868 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
869 int ret;
870
871 if (!ops) {
872 pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
873 return TCM_UNSUPPORTED_SCSI_OPCODE;
874 }
875
876 switch (sa) {
877 case PRO_REGISTER:
878 case PRO_REGISTER_AND_IGNORE_EXISTING_KEY:
879 if (!ops->pr_register) {
880 pr_err("block device does not support pr_register.\n");
881 return TCM_UNSUPPORTED_SCSI_OPCODE;
882 }
883
884 /* The block layer pr ops always enables aptpl */
885 if (!aptpl)
886 pr_info("APTPL not set by initiator, but will be used.\n");
887
888 ret = ops->pr_register(bdev, key, sa_key,
889 sa == PRO_REGISTER ? 0 : PR_FL_IGNORE_KEY);
890 break;
891 case PRO_RESERVE:
892 if (!ops->pr_reserve) {
893 pr_err("block_device does not support pr_reserve.\n");
894 return TCM_UNSUPPORTED_SCSI_OPCODE;
895 }
896
897 ret = ops->pr_reserve(bdev, key, scsi_pr_type_to_block(type), 0);
898 break;
899 case PRO_CLEAR:
900 if (!ops->pr_clear) {
901 pr_err("block_device does not support pr_clear.\n");
902 return TCM_UNSUPPORTED_SCSI_OPCODE;
903 }
904
905 ret = ops->pr_clear(bdev, key);
906 break;
907 case PRO_PREEMPT:
908 case PRO_PREEMPT_AND_ABORT:
909 if (!ops->pr_clear) {
910 pr_err("block_device does not support pr_preempt.\n");
911 return TCM_UNSUPPORTED_SCSI_OPCODE;
912 }
913
914 ret = ops->pr_preempt(bdev, key, sa_key,
915 scsi_pr_type_to_block(type),
916 sa == PRO_PREEMPT_AND_ABORT);
917 break;
918 case PRO_RELEASE:
919 if (!ops->pr_clear) {
920 pr_err("block_device does not support pr_pclear.\n");
921 return TCM_UNSUPPORTED_SCSI_OPCODE;
922 }
923
924 ret = ops->pr_release(bdev, key, scsi_pr_type_to_block(type));
925 break;
926 default:
927 pr_err("Unknown PERSISTENT_RESERVE_OUT SA: 0x%02x\n", sa);
928 return TCM_UNSUPPORTED_SCSI_OPCODE;
929 }
930
931 if (!ret)
932 return TCM_NO_SENSE;
933 else if (ret == PR_STS_RESERVATION_CONFLICT)
934 return TCM_RESERVATION_CONFLICT;
935 else
936 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
937 }
938
iblock_pr_report_caps(unsigned char * param_data)939 static void iblock_pr_report_caps(unsigned char *param_data)
940 {
941 u16 len = 8;
942
943 put_unaligned_be16(len, ¶m_data[0]);
944 /*
945 * When using the pr_ops passthrough method we only support exporting
946 * the device through one target port because from the backend module
947 * level we can't see the target port config. As a result we only
948 * support registration directly from the I_T nexus the cmd is sent
949 * through and do not set ATP_C here.
950 *
951 * The block layer pr_ops do not support passing in initiators so
952 * we don't set SIP_C here.
953 */
954 /* PTPL_C: Persistence across Target Power Loss bit */
955 param_data[2] |= 0x01;
956 /*
957 * We are filling in the PERSISTENT RESERVATION TYPE MASK below, so
958 * set the TMV: Task Mask Valid bit.
959 */
960 param_data[3] |= 0x80;
961 /*
962 * Change ALLOW COMMANDs to 0x20 or 0x40 later from Table 166
963 */
964 param_data[3] |= 0x10; /* ALLOW COMMANDs field 001b */
965 /*
966 * PTPL_A: Persistence across Target Power Loss Active bit. The block
967 * layer pr ops always enables this so report it active.
968 */
969 param_data[3] |= 0x01;
970 /*
971 * Setup the PERSISTENT RESERVATION TYPE MASK from Table 212 spc4r37.
972 */
973 param_data[4] |= 0x80; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
974 param_data[4] |= 0x40; /* PR_TYPE_EXCLUSIVE_ACCESS_REGONLY */
975 param_data[4] |= 0x20; /* PR_TYPE_WRITE_EXCLUSIVE_REGONLY */
976 param_data[4] |= 0x08; /* PR_TYPE_EXCLUSIVE_ACCESS */
977 param_data[4] |= 0x02; /* PR_TYPE_WRITE_EXCLUSIVE */
978 param_data[5] |= 0x01; /* PR_TYPE_EXCLUSIVE_ACCESS_ALLREG */
979 }
980
iblock_pr_read_keys(struct se_cmd * cmd,unsigned char * param_data)981 static sense_reason_t iblock_pr_read_keys(struct se_cmd *cmd,
982 unsigned char *param_data)
983 {
984 struct se_device *dev = cmd->se_dev;
985 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
986 struct block_device *bdev = ib_dev->ibd_bd;
987 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
988 int i, len, paths, data_offset;
989 struct pr_keys *keys;
990 sense_reason_t ret;
991
992 if (!ops) {
993 pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
994 return TCM_UNSUPPORTED_SCSI_OPCODE;
995 }
996
997 if (!ops->pr_read_keys) {
998 pr_err("Block device does not support read_keys.\n");
999 return TCM_UNSUPPORTED_SCSI_OPCODE;
1000 }
1001
1002 /*
1003 * We don't know what's under us, but dm-multipath will register every
1004 * path with the same key, so start off with enough space for 16 paths.
1005 * which is not a lot of memory and should normally be enough.
1006 */
1007 paths = 16;
1008 retry:
1009 len = 8 * paths;
1010 keys = kzalloc(sizeof(*keys) + len, GFP_KERNEL);
1011 if (!keys)
1012 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1013
1014 keys->num_keys = paths;
1015 if (!ops->pr_read_keys(bdev, keys)) {
1016 if (keys->num_keys > paths) {
1017 kfree(keys);
1018 paths *= 2;
1019 goto retry;
1020 }
1021 } else {
1022 ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1023 goto free_keys;
1024 }
1025
1026 ret = TCM_NO_SENSE;
1027
1028 put_unaligned_be32(keys->generation, ¶m_data[0]);
1029 if (!keys->num_keys) {
1030 put_unaligned_be32(0, ¶m_data[4]);
1031 goto free_keys;
1032 }
1033
1034 put_unaligned_be32(8 * keys->num_keys, ¶m_data[4]);
1035
1036 data_offset = 8;
1037 for (i = 0; i < keys->num_keys; i++) {
1038 if (data_offset + 8 > cmd->data_length)
1039 break;
1040
1041 put_unaligned_be64(keys->keys[i], ¶m_data[data_offset]);
1042 data_offset += 8;
1043 }
1044
1045 free_keys:
1046 kfree(keys);
1047 return ret;
1048 }
1049
iblock_pr_read_reservation(struct se_cmd * cmd,unsigned char * param_data)1050 static sense_reason_t iblock_pr_read_reservation(struct se_cmd *cmd,
1051 unsigned char *param_data)
1052 {
1053 struct se_device *dev = cmd->se_dev;
1054 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1055 struct block_device *bdev = ib_dev->ibd_bd;
1056 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
1057 struct pr_held_reservation rsv = { };
1058
1059 if (!ops) {
1060 pr_err("Block device does not support pr_ops but iblock device has been configured for PR passthrough.\n");
1061 return TCM_UNSUPPORTED_SCSI_OPCODE;
1062 }
1063
1064 if (!ops->pr_read_reservation) {
1065 pr_err("Block device does not support read_keys.\n");
1066 return TCM_UNSUPPORTED_SCSI_OPCODE;
1067 }
1068
1069 if (ops->pr_read_reservation(bdev, &rsv))
1070 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1071
1072 put_unaligned_be32(rsv.generation, ¶m_data[0]);
1073 if (!block_pr_type_to_scsi(rsv.type)) {
1074 put_unaligned_be32(0, ¶m_data[4]);
1075 return TCM_NO_SENSE;
1076 }
1077
1078 put_unaligned_be32(16, ¶m_data[4]);
1079
1080 if (cmd->data_length < 16)
1081 return TCM_NO_SENSE;
1082 put_unaligned_be64(rsv.key, ¶m_data[8]);
1083
1084 if (cmd->data_length < 22)
1085 return TCM_NO_SENSE;
1086 param_data[21] = block_pr_type_to_scsi(rsv.type);
1087
1088 return TCM_NO_SENSE;
1089 }
1090
iblock_execute_pr_in(struct se_cmd * cmd,u8 sa,unsigned char * param_data)1091 static sense_reason_t iblock_execute_pr_in(struct se_cmd *cmd, u8 sa,
1092 unsigned char *param_data)
1093 {
1094 sense_reason_t ret = TCM_NO_SENSE;
1095
1096 switch (sa) {
1097 case PRI_REPORT_CAPABILITIES:
1098 iblock_pr_report_caps(param_data);
1099 break;
1100 case PRI_READ_KEYS:
1101 ret = iblock_pr_read_keys(cmd, param_data);
1102 break;
1103 case PRI_READ_RESERVATION:
1104 ret = iblock_pr_read_reservation(cmd, param_data);
1105 break;
1106 default:
1107 pr_err("Unknown PERSISTENT_RESERVE_IN SA: 0x%02x\n", sa);
1108 return TCM_UNSUPPORTED_SCSI_OPCODE;
1109 }
1110
1111 return ret;
1112 }
1113
iblock_get_alignment_offset_lbas(struct se_device * dev)1114 static sector_t iblock_get_alignment_offset_lbas(struct se_device *dev)
1115 {
1116 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1117 struct block_device *bd = ib_dev->ibd_bd;
1118 int ret;
1119
1120 ret = bdev_alignment_offset(bd);
1121 if (ret == -1)
1122 return 0;
1123
1124 /* convert offset-bytes to offset-lbas */
1125 return ret / bdev_logical_block_size(bd);
1126 }
1127
iblock_get_lbppbe(struct se_device * dev)1128 static unsigned int iblock_get_lbppbe(struct se_device *dev)
1129 {
1130 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1131 struct block_device *bd = ib_dev->ibd_bd;
1132 unsigned int logs_per_phys =
1133 bdev_physical_block_size(bd) / bdev_logical_block_size(bd);
1134
1135 return ilog2(logs_per_phys);
1136 }
1137
iblock_get_io_min(struct se_device * dev)1138 static unsigned int iblock_get_io_min(struct se_device *dev)
1139 {
1140 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1141 struct block_device *bd = ib_dev->ibd_bd;
1142
1143 return bdev_io_min(bd);
1144 }
1145
iblock_get_io_opt(struct se_device * dev)1146 static unsigned int iblock_get_io_opt(struct se_device *dev)
1147 {
1148 struct iblock_dev *ib_dev = IBLOCK_DEV(dev);
1149 struct block_device *bd = ib_dev->ibd_bd;
1150
1151 return bdev_io_opt(bd);
1152 }
1153
1154 static struct exec_cmd_ops iblock_exec_cmd_ops = {
1155 .execute_rw = iblock_execute_rw,
1156 .execute_sync_cache = iblock_execute_sync_cache,
1157 .execute_write_same = iblock_execute_write_same,
1158 .execute_unmap = iblock_execute_unmap,
1159 .execute_pr_out = iblock_execute_pr_out,
1160 .execute_pr_in = iblock_execute_pr_in,
1161 };
1162
1163 static sense_reason_t
iblock_parse_cdb(struct se_cmd * cmd)1164 iblock_parse_cdb(struct se_cmd *cmd)
1165 {
1166 return sbc_parse_cdb(cmd, &iblock_exec_cmd_ops);
1167 }
1168
iblock_get_write_cache(struct se_device * dev)1169 static bool iblock_get_write_cache(struct se_device *dev)
1170 {
1171 return bdev_write_cache(IBLOCK_DEV(dev)->ibd_bd);
1172 }
1173
1174 static const struct target_backend_ops iblock_ops = {
1175 .name = "iblock",
1176 .inquiry_prod = "IBLOCK",
1177 .transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR,
1178 .inquiry_rev = IBLOCK_VERSION,
1179 .owner = THIS_MODULE,
1180 .attach_hba = iblock_attach_hba,
1181 .detach_hba = iblock_detach_hba,
1182 .alloc_device = iblock_alloc_device,
1183 .configure_device = iblock_configure_device,
1184 .destroy_device = iblock_destroy_device,
1185 .free_device = iblock_free_device,
1186 .configure_unmap = iblock_configure_unmap,
1187 .plug_device = iblock_plug_device,
1188 .unplug_device = iblock_unplug_device,
1189 .parse_cdb = iblock_parse_cdb,
1190 .set_configfs_dev_params = iblock_set_configfs_dev_params,
1191 .show_configfs_dev_params = iblock_show_configfs_dev_params,
1192 .get_device_type = sbc_get_device_type,
1193 .get_blocks = iblock_get_blocks,
1194 .get_alignment_offset_lbas = iblock_get_alignment_offset_lbas,
1195 .get_lbppbe = iblock_get_lbppbe,
1196 .get_io_min = iblock_get_io_min,
1197 .get_io_opt = iblock_get_io_opt,
1198 .get_write_cache = iblock_get_write_cache,
1199 .tb_dev_attrib_attrs = sbc_attrib_attrs,
1200 };
1201
iblock_module_init(void)1202 static int __init iblock_module_init(void)
1203 {
1204 return transport_backend_register(&iblock_ops);
1205 }
1206
iblock_module_exit(void)1207 static void __exit iblock_module_exit(void)
1208 {
1209 target_backend_unregister(&iblock_ops);
1210 }
1211
1212 MODULE_DESCRIPTION("TCM IBLOCK subsystem plugin");
1213 MODULE_AUTHOR("nab@Linux-iSCSI.org");
1214 MODULE_LICENSE("GPL");
1215
1216 module_init(iblock_module_init);
1217 module_exit(iblock_module_exit);
1218