1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (C) 2011 Lawrence Livermore National Security, LLC.
24  * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
25  * Written by Brian Behlendorf <behlendorf1@llnl.gov>.
26  * LLNL-CODE-403049.
27  */
28 
29 #ifndef _ZFS_BLKDEV_H
30 #define	_ZFS_BLKDEV_H
31 
32 #include <linux/blkdev.h>
33 #include <linux/backing-dev.h>
34 #include <linux/hdreg.h>
35 #include <linux/major.h>
36 #include <linux/msdos_fs.h>	/* for SECTOR_* */
37 #include <linux/bio.h>
38 
39 #ifdef HAVE_BLK_MQ
40 #include <linux/blk-mq.h>
41 #endif
42 
43 #ifndef HAVE_BLK_QUEUE_FLAG_SET
44 static inline void
45 blk_queue_flag_set(unsigned int flag, struct request_queue *q)
46 {
47 	queue_flag_set(flag, q);
48 }
49 #endif
50 
51 #ifndef HAVE_BLK_QUEUE_FLAG_CLEAR
52 static inline void
53 blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
54 {
55 	queue_flag_clear(flag, q);
56 }
57 #endif
58 
59 /*
60  * 4.7 API,
61  * The blk_queue_write_cache() interface has replaced blk_queue_flush()
62  * interface.  However, the new interface is GPL-only thus we implement
63  * our own trivial wrapper when the GPL-only version is detected.
64  *
65  * 2.6.36 - 4.6 API,
66  * The blk_queue_flush() interface has replaced blk_queue_ordered()
67  * interface.  However, while the old interface was available to all the
68  * new one is GPL-only.   Thus if the GPL-only version is detected we
69  * implement our own trivial helper.
70  */
71 static inline void
72 blk_queue_set_write_cache(struct request_queue *q, bool wc, bool fua)
73 {
74 #if defined(HAVE_BLK_QUEUE_WRITE_CACHE_GPL_ONLY)
75 	if (wc)
76 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
77 	else
78 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
79 	if (fua)
80 		blk_queue_flag_set(QUEUE_FLAG_FUA, q);
81 	else
82 		blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
83 #elif defined(HAVE_BLK_QUEUE_WRITE_CACHE)
84 	blk_queue_write_cache(q, wc, fua);
85 #elif defined(HAVE_BLK_QUEUE_FLUSH_GPL_ONLY)
86 	if (wc)
87 		q->flush_flags |= REQ_FLUSH;
88 	if (fua)
89 		q->flush_flags |= REQ_FUA;
90 #elif defined(HAVE_BLK_QUEUE_FLUSH)
91 	blk_queue_flush(q, (wc ? REQ_FLUSH : 0) | (fua ? REQ_FUA : 0));
92 #else
93 #error "Unsupported kernel"
94 #endif
95 }
96 
97 static inline void
98 blk_queue_set_read_ahead(struct request_queue *q, unsigned long ra_pages)
99 {
100 #if !defined(HAVE_BLK_QUEUE_UPDATE_READAHEAD) && \
101 	!defined(HAVE_DISK_UPDATE_READAHEAD)
102 #ifdef HAVE_BLK_QUEUE_BDI_DYNAMIC
103 	q->backing_dev_info->ra_pages = ra_pages;
104 #else
105 	q->backing_dev_info.ra_pages = ra_pages;
106 #endif
107 #endif
108 }
109 
110 #ifdef HAVE_BIO_BVEC_ITER
111 #define	BIO_BI_SECTOR(bio)	(bio)->bi_iter.bi_sector
112 #define	BIO_BI_SIZE(bio)	(bio)->bi_iter.bi_size
113 #define	BIO_BI_IDX(bio)		(bio)->bi_iter.bi_idx
114 #define	BIO_BI_SKIP(bio)	(bio)->bi_iter.bi_bvec_done
115 #define	bio_for_each_segment4(bv, bvp, b, i)	\
116 	bio_for_each_segment((bv), (b), (i))
117 typedef struct bvec_iter bvec_iterator_t;
118 #else
119 #define	BIO_BI_SECTOR(bio)	(bio)->bi_sector
120 #define	BIO_BI_SIZE(bio)	(bio)->bi_size
121 #define	BIO_BI_IDX(bio)		(bio)->bi_idx
122 #define	BIO_BI_SKIP(bio)	(0)
123 #define	bio_for_each_segment4(bv, bvp, b, i)	\
124 	bio_for_each_segment((bvp), (b), (i))
125 typedef int bvec_iterator_t;
126 #endif
127 
128 static inline void
129 bio_set_flags_failfast(struct block_device *bdev, int *flags, bool dev,
130     bool transport, bool driver)
131 {
132 #ifdef CONFIG_BUG
133 	/*
134 	 * Disable FAILFAST for loopback devices because of the
135 	 * following incorrect BUG_ON() in loop_make_request().
136 	 * This support is also disabled for md devices because the
137 	 * test suite layers md devices on top of loopback devices.
138 	 * This may be removed when the loopback driver is fixed.
139 	 *
140 	 *   BUG_ON(!lo || (rw != READ && rw != WRITE));
141 	 */
142 	if ((MAJOR(bdev->bd_dev) == LOOP_MAJOR) ||
143 	    (MAJOR(bdev->bd_dev) == MD_MAJOR))
144 		return;
145 
146 #ifdef BLOCK_EXT_MAJOR
147 	if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
148 		return;
149 #endif /* BLOCK_EXT_MAJOR */
150 #endif /* CONFIG_BUG */
151 
152 	if (dev)
153 		*flags |= REQ_FAILFAST_DEV;
154 	if (transport)
155 		*flags |= REQ_FAILFAST_TRANSPORT;
156 	if (driver)
157 		*flags |= REQ_FAILFAST_DRIVER;
158 }
159 
160 /*
161  * Maximum disk label length, it may be undefined for some kernels.
162  */
163 #if !defined(DISK_NAME_LEN)
164 #define	DISK_NAME_LEN	32
165 #endif /* DISK_NAME_LEN */
166 
167 #ifdef HAVE_BIO_BI_STATUS
168 static inline int
169 bi_status_to_errno(blk_status_t status)
170 {
171 	switch (status)	{
172 	case BLK_STS_OK:
173 		return (0);
174 	case BLK_STS_NOTSUPP:
175 		return (EOPNOTSUPP);
176 	case BLK_STS_TIMEOUT:
177 		return (ETIMEDOUT);
178 	case BLK_STS_NOSPC:
179 		return (ENOSPC);
180 	case BLK_STS_TRANSPORT:
181 		return (ENOLINK);
182 	case BLK_STS_TARGET:
183 		return (EREMOTEIO);
184 #ifdef HAVE_BLK_STS_RESV_CONFLICT
185 	case BLK_STS_RESV_CONFLICT:
186 #else
187 	case BLK_STS_NEXUS:
188 #endif
189 		return (EBADE);
190 	case BLK_STS_MEDIUM:
191 		return (ENODATA);
192 	case BLK_STS_PROTECTION:
193 		return (EILSEQ);
194 	case BLK_STS_RESOURCE:
195 		return (ENOMEM);
196 	case BLK_STS_AGAIN:
197 		return (EAGAIN);
198 	case BLK_STS_IOERR:
199 		return (EIO);
200 	default:
201 		return (EIO);
202 	}
203 }
204 
205 static inline blk_status_t
206 errno_to_bi_status(int error)
207 {
208 	switch (error) {
209 	case 0:
210 		return (BLK_STS_OK);
211 	case EOPNOTSUPP:
212 		return (BLK_STS_NOTSUPP);
213 	case ETIMEDOUT:
214 		return (BLK_STS_TIMEOUT);
215 	case ENOSPC:
216 		return (BLK_STS_NOSPC);
217 	case ENOLINK:
218 		return (BLK_STS_TRANSPORT);
219 	case EREMOTEIO:
220 		return (BLK_STS_TARGET);
221 	case EBADE:
222 #ifdef HAVE_BLK_STS_RESV_CONFLICT
223 		return (BLK_STS_RESV_CONFLICT);
224 #else
225 		return (BLK_STS_NEXUS);
226 #endif
227 	case ENODATA:
228 		return (BLK_STS_MEDIUM);
229 	case EILSEQ:
230 		return (BLK_STS_PROTECTION);
231 	case ENOMEM:
232 		return (BLK_STS_RESOURCE);
233 	case EAGAIN:
234 		return (BLK_STS_AGAIN);
235 	case EIO:
236 		return (BLK_STS_IOERR);
237 	default:
238 		return (BLK_STS_IOERR);
239 	}
240 }
241 #endif /* HAVE_BIO_BI_STATUS */
242 
243 /*
244  * 4.3 API change
245  * The bio_endio() prototype changed slightly.  These are helper
246  * macro's to ensure the prototype and invocation are handled.
247  */
248 #ifdef HAVE_1ARG_BIO_END_IO_T
249 #ifdef HAVE_BIO_BI_STATUS
250 #define	BIO_END_IO_ERROR(bio)		bi_status_to_errno(bio->bi_status)
251 #define	BIO_END_IO_PROTO(fn, x, z)	static void fn(struct bio *x)
252 #define	BIO_END_IO(bio, error)		bio_set_bi_status(bio, error)
253 static inline void
254 bio_set_bi_status(struct bio *bio, int error)
255 {
256 	ASSERT3S(error, <=, 0);
257 	bio->bi_status = errno_to_bi_status(-error);
258 	bio_endio(bio);
259 }
260 #else
261 #define	BIO_END_IO_ERROR(bio)		(-(bio->bi_error))
262 #define	BIO_END_IO_PROTO(fn, x, z)	static void fn(struct bio *x)
263 #define	BIO_END_IO(bio, error)		bio_set_bi_error(bio, error)
264 static inline void
265 bio_set_bi_error(struct bio *bio, int error)
266 {
267 	ASSERT3S(error, <=, 0);
268 	bio->bi_error = error;
269 	bio_endio(bio);
270 }
271 #endif /* HAVE_BIO_BI_STATUS */
272 
273 #else
274 #define	BIO_END_IO_PROTO(fn, x, z)	static void fn(struct bio *x, int z)
275 #define	BIO_END_IO(bio, error)		bio_endio(bio, error);
276 #endif /* HAVE_1ARG_BIO_END_IO_T */
277 
278 /*
279  * 5.15 MACRO,
280  *   GD_DEAD
281  *
282  * 2.6.36 - 5.14 MACRO,
283  *   GENHD_FL_UP
284  *
285  * Check the disk status and return B_TRUE if alive
286  * otherwise B_FALSE
287  */
288 static inline boolean_t
289 zfs_check_disk_status(struct block_device *bdev)
290 {
291 #if defined(GENHD_FL_UP)
292 	return (!!(bdev->bd_disk->flags & GENHD_FL_UP));
293 #elif defined(GD_DEAD)
294 	return (!test_bit(GD_DEAD, &bdev->bd_disk->state));
295 #else
296 /*
297  * This is encountered if neither GENHD_FL_UP nor GD_DEAD is available in
298  * the kernel - likely due to an MACRO change that needs to be chased down.
299  */
300 #error "Unsupported kernel: no usable disk status check"
301 #endif
302 }
303 
304 /*
305  * 4.1 API,
306  * 3.10.0 CentOS 7.x API,
307  *   blkdev_reread_part()
308  *
309  * For older kernels trigger a re-reading of the partition table by calling
310  * check_disk_change() which calls flush_disk() to invalidate the device.
311  *
312  * For newer kernels (as of 5.10), bdev_check_media_change is used, in favor of
313  * check_disk_change(), with the modification that invalidation is no longer
314  * forced.
315  */
316 #ifdef HAVE_CHECK_DISK_CHANGE
317 #define	zfs_check_media_change(bdev)	check_disk_change(bdev)
318 #ifdef HAVE_BLKDEV_REREAD_PART
319 #define	vdev_bdev_reread_part(bdev)	blkdev_reread_part(bdev)
320 #else
321 #define	vdev_bdev_reread_part(bdev)	check_disk_change(bdev)
322 #endif /* HAVE_BLKDEV_REREAD_PART */
323 #else
324 #ifdef HAVE_BDEV_CHECK_MEDIA_CHANGE
325 static inline int
326 zfs_check_media_change(struct block_device *bdev)
327 {
328 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_REVALIDATE_DISK
329 	struct gendisk *gd = bdev->bd_disk;
330 	const struct block_device_operations *bdo = gd->fops;
331 #endif
332 
333 	if (!bdev_check_media_change(bdev))
334 		return (0);
335 
336 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_REVALIDATE_DISK
337 	/*
338 	 * Force revalidation, to mimic the old behavior of
339 	 * check_disk_change()
340 	 */
341 	if (bdo->revalidate_disk)
342 		bdo->revalidate_disk(gd);
343 #endif
344 
345 	return (0);
346 }
347 #define	vdev_bdev_reread_part(bdev)	zfs_check_media_change(bdev)
348 #elif defined(HAVE_DISK_CHECK_MEDIA_CHANGE)
349 #define	vdev_bdev_reread_part(bdev)	disk_check_media_change(bdev->bd_disk)
350 #define	zfs_check_media_change(bdev)	disk_check_media_change(bdev->bd_disk)
351 #else
352 /*
353  * This is encountered if check_disk_change() and bdev_check_media_change()
354  * are not available in the kernel - likely due to an API change that needs
355  * to be chased down.
356  */
357 #error "Unsupported kernel: no usable disk change check"
358 #endif /* HAVE_BDEV_CHECK_MEDIA_CHANGE */
359 #endif /* HAVE_CHECK_DISK_CHANGE */
360 
361 /*
362  * 2.6.27 API change
363  * The function was exported for use, prior to this it existed but the
364  * symbol was not exported.
365  *
366  * 4.4.0-6.21 API change for Ubuntu
367  * lookup_bdev() gained a second argument, FMODE_*, to check inode permissions.
368  *
369  * 5.11 API change
370  * Changed to take a dev_t argument which is set on success and return a
371  * non-zero error code on failure.
372  */
373 static inline int
374 vdev_lookup_bdev(const char *path, dev_t *dev)
375 {
376 #if defined(HAVE_DEVT_LOOKUP_BDEV)
377 	return (lookup_bdev(path, dev));
378 #elif defined(HAVE_1ARG_LOOKUP_BDEV)
379 	struct block_device *bdev = lookup_bdev(path);
380 	if (IS_ERR(bdev))
381 		return (PTR_ERR(bdev));
382 
383 	*dev = bdev->bd_dev;
384 	bdput(bdev);
385 
386 	return (0);
387 #elif defined(HAVE_MODE_LOOKUP_BDEV)
388 	struct block_device *bdev = lookup_bdev(path, FMODE_READ);
389 	if (IS_ERR(bdev))
390 		return (PTR_ERR(bdev));
391 
392 	*dev = bdev->bd_dev;
393 	bdput(bdev);
394 
395 	return (0);
396 #else
397 #error "Unsupported kernel"
398 #endif
399 }
400 
401 #if defined(HAVE_BLK_MODE_T)
402 #define	blk_mode_is_open_write(flag)	((flag) & BLK_OPEN_WRITE)
403 #else
404 #define	blk_mode_is_open_write(flag)	((flag) & FMODE_WRITE)
405 #endif
406 
407 /*
408  * Kernels without bio_set_op_attrs use bi_rw for the bio flags.
409  */
410 #if !defined(HAVE_BIO_SET_OP_ATTRS)
411 static inline void
412 bio_set_op_attrs(struct bio *bio, unsigned rw, unsigned flags)
413 {
414 #if defined(HAVE_BIO_BI_OPF)
415 	bio->bi_opf = rw | flags;
416 #else
417 	bio->bi_rw |= rw | flags;
418 #endif /* HAVE_BIO_BI_OPF */
419 }
420 #endif
421 
422 /*
423  * bio_set_flush - Set the appropriate flags in a bio to guarantee
424  * data are on non-volatile media on completion.
425  *
426  * 2.6.37 - 4.8 API,
427  *   Introduce WRITE_FLUSH, WRITE_FUA, and WRITE_FLUSH_FUA flags as a
428  *   replacement for WRITE_BARRIER to allow expressing richer semantics
429  *   to the block layer.  It's up to the block layer to implement the
430  *   semantics correctly. Use the WRITE_FLUSH_FUA flag combination.
431  *
432  * 4.8 - 4.9 API,
433  *   REQ_FLUSH was renamed to REQ_PREFLUSH.  For consistency with previous
434  *   OpenZFS releases, prefer the WRITE_FLUSH_FUA flag set if it's available.
435  *
436  * 4.10 API,
437  *   The read/write flags and their modifiers, including WRITE_FLUSH,
438  *   WRITE_FUA and WRITE_FLUSH_FUA were removed from fs.h in
439  *   torvalds/linux@70fd7614 and replaced by direct flag modification
440  *   of the REQ_ flags in bio->bi_opf.  Use REQ_PREFLUSH.
441  */
442 static inline void
443 bio_set_flush(struct bio *bio)
444 {
445 #if defined(HAVE_REQ_PREFLUSH)	/* >= 4.10 */
446 	bio_set_op_attrs(bio, 0, REQ_PREFLUSH | REQ_OP_WRITE);
447 #elif defined(WRITE_FLUSH_FUA)	/* >= 2.6.37 and <= 4.9 */
448 	bio_set_op_attrs(bio, 0, WRITE_FLUSH_FUA);
449 #else
450 #error	"Allowing the build will cause bio_set_flush requests to be ignored."
451 #endif
452 }
453 
454 /*
455  * 4.8 API,
456  *   REQ_OP_FLUSH
457  *
458  * 4.8-rc0 - 4.8-rc1,
459  *   REQ_PREFLUSH
460  *
461  * 2.6.36 - 4.7 API,
462  *   REQ_FLUSH
463  *
464  * in all cases but may have a performance impact for some kernels.  It
465  * has the advantage of minimizing kernel specific changes in the zvol code.
466  *
467  */
468 static inline boolean_t
469 bio_is_flush(struct bio *bio)
470 {
471 #if defined(HAVE_REQ_OP_FLUSH) && defined(HAVE_BIO_BI_OPF)
472 	return ((bio_op(bio) == REQ_OP_FLUSH) || (bio->bi_opf & REQ_PREFLUSH));
473 #elif defined(HAVE_REQ_PREFLUSH) && defined(HAVE_BIO_BI_OPF)
474 	return (bio->bi_opf & REQ_PREFLUSH);
475 #elif defined(HAVE_REQ_PREFLUSH) && !defined(HAVE_BIO_BI_OPF)
476 	return (bio->bi_rw & REQ_PREFLUSH);
477 #elif defined(HAVE_REQ_FLUSH)
478 	return (bio->bi_rw & REQ_FLUSH);
479 #else
480 #error	"Unsupported kernel"
481 #endif
482 }
483 
484 /*
485  * 4.8 API,
486  *   REQ_FUA flag moved to bio->bi_opf
487  *
488  * 2.6.x - 4.7 API,
489  *   REQ_FUA
490  */
491 static inline boolean_t
492 bio_is_fua(struct bio *bio)
493 {
494 #if defined(HAVE_BIO_BI_OPF)
495 	return (bio->bi_opf & REQ_FUA);
496 #elif defined(REQ_FUA)
497 	return (bio->bi_rw & REQ_FUA);
498 #else
499 #error	"Allowing the build will cause fua requests to be ignored."
500 #endif
501 }
502 
503 /*
504  * 4.8 API,
505  *   REQ_OP_DISCARD
506  *
507  * 2.6.36 - 4.7 API,
508  *   REQ_DISCARD
509  *
510  * In all cases the normal I/O path is used for discards.  The only
511  * difference is how the kernel tags individual I/Os as discards.
512  */
513 static inline boolean_t
514 bio_is_discard(struct bio *bio)
515 {
516 #if defined(HAVE_REQ_OP_DISCARD)
517 	return (bio_op(bio) == REQ_OP_DISCARD);
518 #elif defined(HAVE_REQ_DISCARD)
519 	return (bio->bi_rw & REQ_DISCARD);
520 #else
521 #error "Unsupported kernel"
522 #endif
523 }
524 
525 /*
526  * 4.8 API,
527  *   REQ_OP_SECURE_ERASE
528  *
529  * 2.6.36 - 4.7 API,
530  *   REQ_SECURE
531  */
532 static inline boolean_t
533 bio_is_secure_erase(struct bio *bio)
534 {
535 #if defined(HAVE_REQ_OP_SECURE_ERASE)
536 	return (bio_op(bio) == REQ_OP_SECURE_ERASE);
537 #elif defined(REQ_SECURE)
538 	return (bio->bi_rw & REQ_SECURE);
539 #else
540 	return (0);
541 #endif
542 }
543 
544 /*
545  * 2.6.33 API change
546  * Discard granularity and alignment restrictions may now be set.  For
547  * older kernels which do not support this it is safe to skip it.
548  */
549 static inline void
550 blk_queue_discard_granularity(struct request_queue *q, unsigned int dg)
551 {
552 	q->limits.discard_granularity = dg;
553 }
554 
555 /*
556  * 5.19 API,
557  *   bdev_max_discard_sectors()
558  *
559  * 2.6.32 API,
560  *   blk_queue_discard()
561  */
562 static inline boolean_t
563 bdev_discard_supported(struct block_device *bdev)
564 {
565 #if defined(HAVE_BDEV_MAX_DISCARD_SECTORS)
566 	return (!!bdev_max_discard_sectors(bdev));
567 #elif defined(HAVE_BLK_QUEUE_DISCARD)
568 	return (!!blk_queue_discard(bdev_get_queue(bdev)));
569 #else
570 #error "Unsupported kernel"
571 #endif
572 }
573 
574 /*
575  * 5.19 API,
576  *   bdev_max_secure_erase_sectors()
577  *
578  * 4.8 API,
579  *   blk_queue_secure_erase()
580  *
581  * 2.6.36 - 4.7 API,
582  *   blk_queue_secdiscard()
583  */
584 static inline boolean_t
585 bdev_secure_discard_supported(struct block_device *bdev)
586 {
587 #if defined(HAVE_BDEV_MAX_SECURE_ERASE_SECTORS)
588 	return (!!bdev_max_secure_erase_sectors(bdev));
589 #elif defined(HAVE_BLK_QUEUE_SECURE_ERASE)
590 	return (!!blk_queue_secure_erase(bdev_get_queue(bdev)));
591 #elif defined(HAVE_BLK_QUEUE_SECDISCARD)
592 	return (!!blk_queue_secdiscard(bdev_get_queue(bdev)));
593 #else
594 #error "Unsupported kernel"
595 #endif
596 }
597 
598 /*
599  * A common holder for vdev_bdev_open() is used to relax the exclusive open
600  * semantics slightly.  Internal vdev disk callers may pass VDEV_HOLDER to
601  * allow them to open the device multiple times.  Other kernel callers and
602  * user space processes which don't pass this value will get EBUSY.  This is
603  * currently required for the correct operation of hot spares.
604  */
605 #define	VDEV_HOLDER			((void *)0x2401de7)
606 
607 static inline unsigned long
608 blk_generic_start_io_acct(struct request_queue *q __attribute__((unused)),
609     struct gendisk *disk __attribute__((unused)),
610     int rw __attribute__((unused)), struct bio *bio)
611 {
612 #if defined(HAVE_BDEV_IO_ACCT_63)
613 	return (bdev_start_io_acct(bio->bi_bdev, bio_op(bio),
614 	    jiffies));
615 #elif defined(HAVE_BDEV_IO_ACCT_OLD)
616 	return (bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio),
617 	    bio_op(bio), jiffies));
618 #elif defined(HAVE_DISK_IO_ACCT)
619 	return (disk_start_io_acct(disk, bio_sectors(bio), bio_op(bio)));
620 #elif defined(HAVE_BIO_IO_ACCT)
621 	return (bio_start_io_acct(bio));
622 #elif defined(HAVE_GENERIC_IO_ACCT_3ARG)
623 	unsigned long start_time = jiffies;
624 	generic_start_io_acct(rw, bio_sectors(bio), &disk->part0);
625 	return (start_time);
626 #elif defined(HAVE_GENERIC_IO_ACCT_4ARG)
627 	unsigned long start_time = jiffies;
628 	generic_start_io_acct(q, rw, bio_sectors(bio), &disk->part0);
629 	return (start_time);
630 #else
631 	/* Unsupported */
632 	return (0);
633 #endif
634 }
635 
636 static inline void
637 blk_generic_end_io_acct(struct request_queue *q __attribute__((unused)),
638     struct gendisk *disk __attribute__((unused)),
639     int rw __attribute__((unused)), struct bio *bio, unsigned long start_time)
640 {
641 #if defined(HAVE_BDEV_IO_ACCT_63)
642 	bdev_end_io_acct(bio->bi_bdev, bio_op(bio), bio_sectors(bio),
643 	    start_time);
644 #elif defined(HAVE_BDEV_IO_ACCT_OLD)
645 	bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
646 #elif defined(HAVE_DISK_IO_ACCT)
647 	disk_end_io_acct(disk, bio_op(bio), start_time);
648 #elif defined(HAVE_BIO_IO_ACCT)
649 	bio_end_io_acct(bio, start_time);
650 #elif defined(HAVE_GENERIC_IO_ACCT_3ARG)
651 	generic_end_io_acct(rw, &disk->part0, start_time);
652 #elif defined(HAVE_GENERIC_IO_ACCT_4ARG)
653 	generic_end_io_acct(q, rw, &disk->part0, start_time);
654 #endif
655 }
656 
657 #ifndef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS
658 static inline struct request_queue *
659 blk_generic_alloc_queue(make_request_fn make_request, int node_id)
660 {
661 #if defined(HAVE_BLK_ALLOC_QUEUE_REQUEST_FN)
662 	return (blk_alloc_queue(make_request, node_id));
663 #elif defined(HAVE_BLK_ALLOC_QUEUE_REQUEST_FN_RH)
664 	return (blk_alloc_queue_rh(make_request, node_id));
665 #else
666 	struct request_queue *q = blk_alloc_queue(GFP_KERNEL);
667 	if (q != NULL)
668 		blk_queue_make_request(q, make_request);
669 
670 	return (q);
671 #endif
672 }
673 #endif /* !HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS */
674 
675 /*
676  * All the io_*() helper functions below can operate on a bio, or a rq, but
677  * not both.  The older submit_bio() codepath will pass a bio, and the
678  * newer blk-mq codepath will pass a rq.
679  */
680 static inline int
681 io_data_dir(struct bio *bio, struct request *rq)
682 {
683 #ifdef HAVE_BLK_MQ
684 	if (rq != NULL) {
685 		if (op_is_write(req_op(rq))) {
686 			return (WRITE);
687 		} else {
688 			return (READ);
689 		}
690 	}
691 #else
692 	ASSERT3P(rq, ==, NULL);
693 #endif
694 	return (bio_data_dir(bio));
695 }
696 
697 static inline int
698 io_is_flush(struct bio *bio, struct request *rq)
699 {
700 #ifdef HAVE_BLK_MQ
701 	if (rq != NULL)
702 		return (req_op(rq) == REQ_OP_FLUSH);
703 #else
704 	ASSERT3P(rq, ==, NULL);
705 #endif
706 	return (bio_is_flush(bio));
707 }
708 
709 static inline int
710 io_is_discard(struct bio *bio, struct request *rq)
711 {
712 #ifdef HAVE_BLK_MQ
713 	if (rq != NULL)
714 		return (req_op(rq) == REQ_OP_DISCARD);
715 #else
716 	ASSERT3P(rq, ==, NULL);
717 #endif
718 	return (bio_is_discard(bio));
719 }
720 
721 static inline int
722 io_is_secure_erase(struct bio *bio, struct request *rq)
723 {
724 #ifdef HAVE_BLK_MQ
725 	if (rq != NULL)
726 		return (req_op(rq) == REQ_OP_SECURE_ERASE);
727 #else
728 	ASSERT3P(rq, ==, NULL);
729 #endif
730 	return (bio_is_secure_erase(bio));
731 }
732 
733 static inline int
734 io_is_fua(struct bio *bio, struct request *rq)
735 {
736 #ifdef HAVE_BLK_MQ
737 	if (rq != NULL)
738 		return (rq->cmd_flags & REQ_FUA);
739 #else
740 	ASSERT3P(rq, ==, NULL);
741 #endif
742 	return (bio_is_fua(bio));
743 }
744 
745 
746 static inline uint64_t
747 io_offset(struct bio *bio, struct request *rq)
748 {
749 #ifdef HAVE_BLK_MQ
750 	if (rq != NULL)
751 		return (blk_rq_pos(rq) << 9);
752 #else
753 	ASSERT3P(rq, ==, NULL);
754 #endif
755 	return (BIO_BI_SECTOR(bio) << 9);
756 }
757 
758 static inline uint64_t
759 io_size(struct bio *bio, struct request *rq)
760 {
761 #ifdef HAVE_BLK_MQ
762 	if (rq != NULL)
763 		return (blk_rq_bytes(rq));
764 #else
765 	ASSERT3P(rq, ==, NULL);
766 #endif
767 	return (BIO_BI_SIZE(bio));
768 }
769 
770 static inline int
771 io_has_data(struct bio *bio, struct request *rq)
772 {
773 #ifdef HAVE_BLK_MQ
774 	if (rq != NULL)
775 		return (bio_has_data(rq->bio));
776 #else
777 	ASSERT3P(rq, ==, NULL);
778 #endif
779 	return (bio_has_data(bio));
780 }
781 #endif /* _ZFS_BLKDEV_H */
782