1 /*******************************************************************************
2 * Filename: target_core_iblock.c
4 * This file contains the Storage Engine <-> Linux BlockIO transport
7 * (c) Copyright 2003-2013 Datera, Inc.
9 * Nicholas A. Bellinger <nab@kernel.org>
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
25 ******************************************************************************/
27 #include <linux/string.h>
28 #include <linux/parser.h>
29 #include <linux/timer.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/spinlock.h>
34 #include <linux/bio.h>
35 #include <linux/genhd.h>
36 #include <linux/file.h>
37 #include <linux/module.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <asm/unaligned.h>
42 #include <target/target_core_base.h>
43 #include <target/target_core_backend.h>
45 #include "target_core_iblock.h"
47 #define IBLOCK_MAX_BIO_PER_TASK 32 /* max # of bios to submit at a time */
48 #define IBLOCK_BIO_POOL_SIZE 128
50 static inline struct iblock_dev
*IBLOCK_DEV(struct se_device
*dev
)
52 return container_of(dev
, struct iblock_dev
, dev
);
56 static struct se_subsystem_api iblock_template
;
58 /* iblock_attach_hba(): (Part of se_subsystem_api_t template)
62 static int iblock_attach_hba(struct se_hba
*hba
, u32 host_id
)
64 pr_debug("CORE_HBA[%d] - TCM iBlock HBA Driver %s on"
65 " Generic Target Core Stack %s\n", hba
->hba_id
,
66 IBLOCK_VERSION
, TARGET_CORE_MOD_VERSION
);
70 static void iblock_detach_hba(struct se_hba
*hba
)
74 static struct se_device
*iblock_alloc_device(struct se_hba
*hba
, const char *name
)
76 struct iblock_dev
*ib_dev
= NULL
;
78 ib_dev
= kzalloc(sizeof(struct iblock_dev
), GFP_KERNEL
);
80 pr_err("Unable to allocate struct iblock_dev\n");
84 pr_debug( "IBLOCK: Allocated ib_dev for %s\n", name
);
89 static int iblock_configure_device(struct se_device
*dev
)
91 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
92 struct request_queue
*q
;
93 struct block_device
*bd
= NULL
;
97 if (!(ib_dev
->ibd_flags
& IBDF_HAS_UDEV_PATH
)) {
98 pr_err("Missing udev_path= parameters for IBLOCK\n");
102 ib_dev
->ibd_bio_set
= bioset_create(IBLOCK_BIO_POOL_SIZE
, 0);
103 if (!ib_dev
->ibd_bio_set
) {
104 pr_err("IBLOCK: Unable to create bioset\n");
108 pr_debug( "IBLOCK: Claiming struct block_device: %s\n",
109 ib_dev
->ibd_udev_path
);
111 mode
= FMODE_READ
|FMODE_EXCL
;
112 if (!ib_dev
->ibd_readonly
)
115 bd
= blkdev_get_by_path(ib_dev
->ibd_udev_path
, mode
, ib_dev
);
118 goto out_free_bioset
;
122 q
= bdev_get_queue(bd
);
124 dev
->dev_attrib
.hw_block_size
= bdev_logical_block_size(bd
);
125 dev
->dev_attrib
.hw_max_sectors
= UINT_MAX
;
126 dev
->dev_attrib
.hw_queue_depth
= q
->nr_requests
;
129 * Check if the underlying struct block_device request_queue supports
130 * the QUEUE_FLAG_DISCARD bit for UNMAP/WRITE_SAME in SCSI + TRIM
131 * in ATA and we need to set TPE=1
133 if (blk_queue_discard(q
)) {
134 dev
->dev_attrib
.max_unmap_lba_count
=
135 q
->limits
.max_discard_sectors
;
138 * Currently hardcoded to 1 in Linux/SCSI code..
140 dev
->dev_attrib
.max_unmap_block_desc_count
= 1;
141 dev
->dev_attrib
.unmap_granularity
=
142 q
->limits
.discard_granularity
>> 9;
143 dev
->dev_attrib
.unmap_granularity_alignment
=
144 q
->limits
.discard_alignment
;
146 pr_debug("IBLOCK: BLOCK Discard support available,"
147 " disabled by default\n");
150 * Enable write same emulation for IBLOCK and use 0xFFFF as
151 * the smaller WRITE_SAME(10) only has a two-byte block count.
153 dev
->dev_attrib
.max_write_same_len
= 0xFFFF;
155 if (blk_queue_nonrot(q
))
156 dev
->dev_attrib
.is_nonrot
= 1;
161 bioset_free(ib_dev
->ibd_bio_set
);
162 ib_dev
->ibd_bio_set
= NULL
;
167 static void iblock_free_device(struct se_device
*dev
)
169 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
171 if (ib_dev
->ibd_bd
!= NULL
)
172 blkdev_put(ib_dev
->ibd_bd
, FMODE_WRITE
|FMODE_READ
|FMODE_EXCL
);
173 if (ib_dev
->ibd_bio_set
!= NULL
)
174 bioset_free(ib_dev
->ibd_bio_set
);
178 static unsigned long long iblock_emulate_read_cap_with_block_size(
179 struct se_device
*dev
,
180 struct block_device
*bd
,
181 struct request_queue
*q
)
183 unsigned long long blocks_long
= (div_u64(i_size_read(bd
->bd_inode
),
184 bdev_logical_block_size(bd
)) - 1);
185 u32 block_size
= bdev_logical_block_size(bd
);
187 if (block_size
== dev
->dev_attrib
.block_size
)
190 switch (block_size
) {
192 switch (dev
->dev_attrib
.block_size
) {
206 switch (dev
->dev_attrib
.block_size
) {
221 switch (dev
->dev_attrib
.block_size
) {
236 switch (dev
->dev_attrib
.block_size
) {
257 static void iblock_complete_cmd(struct se_cmd
*cmd
)
259 struct iblock_req
*ibr
= cmd
->priv
;
262 if (!atomic_dec_and_test(&ibr
->pending
))
265 if (atomic_read(&ibr
->ib_bio_err_cnt
))
266 status
= SAM_STAT_CHECK_CONDITION
;
268 status
= SAM_STAT_GOOD
;
270 target_complete_cmd(cmd
, status
);
274 static void iblock_bio_done(struct bio
*bio
, int err
)
276 struct se_cmd
*cmd
= bio
->bi_private
;
277 struct iblock_req
*ibr
= cmd
->priv
;
280 * Set -EIO if !BIO_UPTODATE and the passed is still err=0
282 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) && !err
)
286 pr_err("test_bit(BIO_UPTODATE) failed for bio: %p,"
287 " err: %d\n", bio
, err
);
289 * Bump the ib_bio_err_cnt and release bio.
291 atomic_inc(&ibr
->ib_bio_err_cnt
);
292 smp_mb__after_atomic_inc();
297 iblock_complete_cmd(cmd
);
301 iblock_get_bio(struct se_cmd
*cmd
, sector_t lba
, u32 sg_num
)
303 struct iblock_dev
*ib_dev
= IBLOCK_DEV(cmd
->se_dev
);
307 * Only allocate as many vector entries as the bio code allows us to,
308 * we'll loop later on until we have handled the whole request.
310 if (sg_num
> BIO_MAX_PAGES
)
311 sg_num
= BIO_MAX_PAGES
;
313 bio
= bio_alloc_bioset(GFP_NOIO
, sg_num
, ib_dev
->ibd_bio_set
);
315 pr_err("Unable to allocate memory for bio\n");
319 bio
->bi_bdev
= ib_dev
->ibd_bd
;
320 bio
->bi_private
= cmd
;
321 bio
->bi_end_io
= &iblock_bio_done
;
322 bio
->bi_sector
= lba
;
327 static void iblock_submit_bios(struct bio_list
*list
, int rw
)
329 struct blk_plug plug
;
332 blk_start_plug(&plug
);
333 while ((bio
= bio_list_pop(list
)))
335 blk_finish_plug(&plug
);
338 static void iblock_end_io_flush(struct bio
*bio
, int err
)
340 struct se_cmd
*cmd
= bio
->bi_private
;
343 pr_err("IBLOCK: cache flush failed: %d\n", err
);
347 target_complete_cmd(cmd
, SAM_STAT_CHECK_CONDITION
);
349 target_complete_cmd(cmd
, SAM_STAT_GOOD
);
356 * Implement SYCHRONIZE CACHE. Note that we can't handle lba ranges and must
357 * always flush the whole cache.
359 static sense_reason_t
360 iblock_execute_sync_cache(struct se_cmd
*cmd
)
362 struct iblock_dev
*ib_dev
= IBLOCK_DEV(cmd
->se_dev
);
363 int immed
= (cmd
->t_task_cdb
[1] & 0x2);
367 * If the Immediate bit is set, queue up the GOOD response
368 * for this SYNCHRONIZE_CACHE op.
371 target_complete_cmd(cmd
, SAM_STAT_GOOD
);
373 bio
= bio_alloc(GFP_KERNEL
, 0);
374 bio
->bi_end_io
= iblock_end_io_flush
;
375 bio
->bi_bdev
= ib_dev
->ibd_bd
;
377 bio
->bi_private
= cmd
;
378 submit_bio(WRITE_FLUSH
, bio
);
382 static sense_reason_t
383 iblock_do_unmap(struct se_cmd
*cmd
, void *priv
,
384 sector_t lba
, sector_t nolb
)
386 struct block_device
*bdev
= priv
;
389 ret
= blkdev_issue_discard(bdev
, lba
, nolb
, GFP_KERNEL
, 0);
391 pr_err("blkdev_issue_discard() failed: %d\n", ret
);
392 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
;
398 static sense_reason_t
399 iblock_execute_unmap(struct se_cmd
*cmd
)
401 struct block_device
*bdev
= IBLOCK_DEV(cmd
->se_dev
)->ibd_bd
;
403 return sbc_execute_unmap(cmd
, iblock_do_unmap
, bdev
);
406 static sense_reason_t
407 iblock_execute_write_same_unmap(struct se_cmd
*cmd
)
409 struct block_device
*bdev
= IBLOCK_DEV(cmd
->se_dev
)->ibd_bd
;
410 sector_t lba
= cmd
->t_task_lba
;
411 sector_t nolb
= sbc_get_write_same_sectors(cmd
);
414 ret
= iblock_do_unmap(cmd
, bdev
, lba
, nolb
);
418 target_complete_cmd(cmd
, GOOD
);
422 static sense_reason_t
423 iblock_execute_write_same(struct se_cmd
*cmd
)
425 struct iblock_req
*ibr
;
426 struct scatterlist
*sg
;
428 struct bio_list list
;
429 sector_t block_lba
= cmd
->t_task_lba
;
430 sector_t sectors
= sbc_get_write_same_sectors(cmd
);
432 sg
= &cmd
->t_data_sg
[0];
434 if (cmd
->t_data_nents
> 1 ||
435 sg
->length
!= cmd
->se_dev
->dev_attrib
.block_size
) {
436 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
437 " block_size: %u\n", cmd
->t_data_nents
, sg
->length
,
438 cmd
->se_dev
->dev_attrib
.block_size
);
439 return TCM_INVALID_CDB_FIELD
;
442 ibr
= kzalloc(sizeof(struct iblock_req
), GFP_KERNEL
);
447 bio
= iblock_get_bio(cmd
, block_lba
, 1);
451 bio_list_init(&list
);
452 bio_list_add(&list
, bio
);
454 atomic_set(&ibr
->pending
, 1);
457 while (bio_add_page(bio
, sg_page(sg
), sg
->length
, sg
->offset
)
460 bio
= iblock_get_bio(cmd
, block_lba
, 1);
464 atomic_inc(&ibr
->pending
);
465 bio_list_add(&list
, bio
);
468 /* Always in 512 byte units for Linux/Block */
469 block_lba
+= sg
->length
>> IBLOCK_LBA_SHIFT
;
473 iblock_submit_bios(&list
, WRITE
);
477 while ((bio
= bio_list_pop(&list
)))
482 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
;
486 Opt_udev_path
, Opt_readonly
, Opt_force
, Opt_err
489 static match_table_t tokens
= {
490 {Opt_udev_path
, "udev_path=%s"},
491 {Opt_readonly
, "readonly=%d"},
492 {Opt_force
, "force=%d"},
496 static ssize_t
iblock_set_configfs_dev_params(struct se_device
*dev
,
497 const char *page
, ssize_t count
)
499 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
500 char *orig
, *ptr
, *arg_p
, *opts
;
501 substring_t args
[MAX_OPT_ARGS
];
503 unsigned long tmp_readonly
;
505 opts
= kstrdup(page
, GFP_KERNEL
);
511 while ((ptr
= strsep(&opts
, ",\n")) != NULL
) {
515 token
= match_token(ptr
, tokens
, args
);
518 if (ib_dev
->ibd_bd
) {
519 pr_err("Unable to set udev_path= while"
520 " ib_dev->ibd_bd exists\n");
524 if (match_strlcpy(ib_dev
->ibd_udev_path
, &args
[0],
525 SE_UDEV_PATH_LEN
) == 0) {
529 pr_debug("IBLOCK: Referencing UDEV path: %s\n",
530 ib_dev
->ibd_udev_path
);
531 ib_dev
->ibd_flags
|= IBDF_HAS_UDEV_PATH
;
534 arg_p
= match_strdup(&args
[0]);
539 ret
= kstrtoul(arg_p
, 0, &tmp_readonly
);
542 pr_err("kstrtoul() failed for"
546 ib_dev
->ibd_readonly
= tmp_readonly
;
547 pr_debug("IBLOCK: readonly: %d\n", ib_dev
->ibd_readonly
);
558 return (!ret
) ? count
: ret
;
561 static ssize_t
iblock_show_configfs_dev_params(struct se_device
*dev
, char *b
)
563 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
564 struct block_device
*bd
= ib_dev
->ibd_bd
;
565 char buf
[BDEVNAME_SIZE
];
569 bl
+= sprintf(b
+ bl
, "iBlock device: %s",
571 if (ib_dev
->ibd_flags
& IBDF_HAS_UDEV_PATH
)
572 bl
+= sprintf(b
+ bl
, " UDEV PATH: %s",
573 ib_dev
->ibd_udev_path
);
574 bl
+= sprintf(b
+ bl
, " readonly: %d\n", ib_dev
->ibd_readonly
);
576 bl
+= sprintf(b
+ bl
, " ");
578 bl
+= sprintf(b
+ bl
, "Major: %d Minor: %d %s\n",
579 MAJOR(bd
->bd_dev
), MINOR(bd
->bd_dev
), (!bd
->bd_contains
) ?
580 "" : (bd
->bd_holder
== ib_dev
) ?
581 "CLAIMED: IBLOCK" : "CLAIMED: OS");
583 bl
+= sprintf(b
+ bl
, "Major: 0 Minor: 0\n");
589 static sense_reason_t
590 iblock_execute_rw(struct se_cmd
*cmd
, struct scatterlist
*sgl
, u32 sgl_nents
,
591 enum dma_data_direction data_direction
)
593 struct se_device
*dev
= cmd
->se_dev
;
594 struct iblock_req
*ibr
;
596 struct bio_list list
;
597 struct scatterlist
*sg
;
598 u32 sg_num
= sgl_nents
;
604 if (data_direction
== DMA_TO_DEVICE
) {
605 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
606 struct request_queue
*q
= bdev_get_queue(ib_dev
->ibd_bd
);
608 * Force writethrough using WRITE_FUA if a volatile write cache
609 * is not enabled, or if initiator set the Force Unit Access bit.
611 if (q
->flush_flags
& REQ_FUA
) {
612 if (cmd
->se_cmd_flags
& SCF_FUA
)
614 else if (!(q
->flush_flags
& REQ_FLUSH
))
626 * Convert the blocksize advertised to the initiator to the 512 byte
627 * units unconditionally used by the Linux block layer.
629 if (dev
->dev_attrib
.block_size
== 4096)
630 block_lba
= (cmd
->t_task_lba
<< 3);
631 else if (dev
->dev_attrib
.block_size
== 2048)
632 block_lba
= (cmd
->t_task_lba
<< 2);
633 else if (dev
->dev_attrib
.block_size
== 1024)
634 block_lba
= (cmd
->t_task_lba
<< 1);
635 else if (dev
->dev_attrib
.block_size
== 512)
636 block_lba
= cmd
->t_task_lba
;
638 pr_err("Unsupported SCSI -> BLOCK LBA conversion:"
639 " %u\n", dev
->dev_attrib
.block_size
);
640 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
;
643 ibr
= kzalloc(sizeof(struct iblock_req
), GFP_KERNEL
);
649 atomic_set(&ibr
->pending
, 1);
650 iblock_complete_cmd(cmd
);
654 bio
= iblock_get_bio(cmd
, block_lba
, sgl_nents
);
658 bio_list_init(&list
);
659 bio_list_add(&list
, bio
);
661 atomic_set(&ibr
->pending
, 2);
664 for_each_sg(sgl
, sg
, sgl_nents
, i
) {
666 * XXX: if the length the device accepts is shorter than the
667 * length of the S/G list entry this will cause and
668 * endless loop. Better hope no driver uses huge pages.
670 while (bio_add_page(bio
, sg_page(sg
), sg
->length
, sg
->offset
)
672 if (bio_cnt
>= IBLOCK_MAX_BIO_PER_TASK
) {
673 iblock_submit_bios(&list
, rw
);
677 bio
= iblock_get_bio(cmd
, block_lba
, sg_num
);
681 atomic_inc(&ibr
->pending
);
682 bio_list_add(&list
, bio
);
686 /* Always in 512 byte units for Linux/Block */
687 block_lba
+= sg
->length
>> IBLOCK_LBA_SHIFT
;
691 iblock_submit_bios(&list
, rw
);
692 iblock_complete_cmd(cmd
);
696 while ((bio
= bio_list_pop(&list
)))
701 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE
;
704 static sector_t
iblock_get_blocks(struct se_device
*dev
)
706 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
707 struct block_device
*bd
= ib_dev
->ibd_bd
;
708 struct request_queue
*q
= bdev_get_queue(bd
);
710 return iblock_emulate_read_cap_with_block_size(dev
, bd
, q
);
713 static struct sbc_ops iblock_sbc_ops
= {
714 .execute_rw
= iblock_execute_rw
,
715 .execute_sync_cache
= iblock_execute_sync_cache
,
716 .execute_write_same
= iblock_execute_write_same
,
717 .execute_write_same_unmap
= iblock_execute_write_same_unmap
,
718 .execute_unmap
= iblock_execute_unmap
,
721 static sense_reason_t
722 iblock_parse_cdb(struct se_cmd
*cmd
)
724 return sbc_parse_cdb(cmd
, &iblock_sbc_ops
);
727 bool iblock_get_write_cache(struct se_device
*dev
)
729 struct iblock_dev
*ib_dev
= IBLOCK_DEV(dev
);
730 struct block_device
*bd
= ib_dev
->ibd_bd
;
731 struct request_queue
*q
= bdev_get_queue(bd
);
733 return q
->flush_flags
& REQ_FLUSH
;
736 static struct se_subsystem_api iblock_template
= {
738 .inquiry_prod
= "IBLOCK",
739 .inquiry_rev
= IBLOCK_VERSION
,
740 .owner
= THIS_MODULE
,
741 .transport_type
= TRANSPORT_PLUGIN_VHBA_PDEV
,
742 .attach_hba
= iblock_attach_hba
,
743 .detach_hba
= iblock_detach_hba
,
744 .alloc_device
= iblock_alloc_device
,
745 .configure_device
= iblock_configure_device
,
746 .free_device
= iblock_free_device
,
747 .parse_cdb
= iblock_parse_cdb
,
748 .set_configfs_dev_params
= iblock_set_configfs_dev_params
,
749 .show_configfs_dev_params
= iblock_show_configfs_dev_params
,
750 .get_device_type
= sbc_get_device_type
,
751 .get_blocks
= iblock_get_blocks
,
752 .get_write_cache
= iblock_get_write_cache
,
755 static int __init
iblock_module_init(void)
757 return transport_subsystem_register(&iblock_template
);
760 static void __exit
iblock_module_exit(void)
762 transport_subsystem_release(&iblock_template
);
765 MODULE_DESCRIPTION("TCM IBLOCK subsystem plugin");
766 MODULE_AUTHOR("nab@Linux-iSCSI.org");
767 MODULE_LICENSE("GPL");
769 module_init(iblock_module_init
);
770 module_exit(iblock_module_exit
);