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.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
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]
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>.
32 #include <linux/blkdev.h>
33 #include <linux/elevator.h>
36 typedef unsigned __bitwise__ fmode_t;
37 #endif /* HAVE_FMODE_T */
39 #ifndef HAVE_BLK_FETCH_REQUEST
40 static inline struct request *
41 blk_fetch_request(struct request_queue *q)
45 req = elv_next_request(q);
47 blkdev_dequeue_request(req);
51 #endif /* HAVE_BLK_FETCH_REQUEST */
53 #ifndef HAVE_BLK_REQUEUE_REQUEST
55 blk_requeue_request(request_queue_t *q, struct request *req)
57 elv_requeue_request(q, req);
59 #endif /* HAVE_BLK_REQUEUE_REQUEST */
61 #ifndef HAVE_BLK_END_REQUEST
63 __blk_end_request(struct request *req, int error, unsigned int nr_bytes)
68 * Request has already been dequeued but 2.6.18 version of
69 * end_request() unconditionally dequeues the request so we
70 * add it to a local list to prevent hitting the BUG_ON.
72 list_add(&req->queuelist, &list);
75 * The old API required the driver to end each segment and not
76 * the entire request. In our case we always need to end the
77 * entire request partial requests are not supported.
79 req->hard_cur_sectors = nr_bytes >> 9;
80 end_request(req, ((error == 0) ? 1 : error));
86 blk_end_request(struct request *req, int error, unsigned int nr_bytes)
88 struct request_queue *q = req->q;
91 spin_lock_irq(q->queue_lock);
92 rc = __blk_end_request(req, error, nr_bytes);
93 spin_unlock_irq(q->queue_lock);
98 #ifdef HAVE_BLK_END_REQUEST_GPL_ONLY
100 * Define required to avoid conflicting 2.6.29 non-static prototype for a
101 * GPL-only version of the helper. As of 2.6.31 the helper is available
102 * to non-GPL modules and is not explicitly exported GPL-only.
104 #define __blk_end_request __blk_end_request_x
105 #define blk_end_request blk_end_request_x
108 __blk_end_request_x(struct request *req, int error, unsigned int nr_bytes)
111 * The old API required the driver to end each segment and not
112 * the entire request. In our case we always need to end the
113 * entire request partial requests are not supported.
115 req->hard_cur_sectors = nr_bytes >> 9;
116 end_request(req, ((error == 0) ? 1 : error));
121 blk_end_request_x(struct request *req, int error, unsigned int nr_bytes)
123 struct request_queue *q = req->q;
126 spin_lock_irq(q->queue_lock);
127 rc = __blk_end_request_x(req, error, nr_bytes);
128 spin_unlock_irq(q->queue_lock);
132 #endif /* HAVE_BLK_END_REQUEST_GPL_ONLY */
133 #endif /* HAVE_BLK_END_REQUEST */
137 * The blk_queue_flush() interface has replaced blk_queue_ordered()
138 * interface. However, while the old interface was available to all the
139 * new one is GPL-only. Thus if the GPL-only version is detected we
140 * implement our own trivial helper compatibility funcion. The hope is
141 * that long term this function will be opened up.
143 #if defined(HAVE_BLK_QUEUE_FLUSH) && defined(HAVE_BLK_QUEUE_FLUSH_GPL_ONLY)
144 #define blk_queue_flush __blk_queue_flush
146 __blk_queue_flush(struct request_queue *q, unsigned int flags)
148 q->flush_flags = flags & (REQ_FLUSH | REQ_FUA);
150 #endif /* HAVE_BLK_QUEUE_FLUSH && HAVE_BLK_QUEUE_FLUSH_GPL_ONLY */
152 #ifndef HAVE_BLK_RQ_POS
153 static inline sector_t
154 blk_rq_pos(struct request *req)
156 return (req->sector);
158 #endif /* HAVE_BLK_RQ_POS */
160 #ifndef HAVE_BLK_RQ_SECTORS
161 static inline unsigned int
162 blk_rq_sectors(struct request *req)
164 return (req->nr_sectors);
166 #endif /* HAVE_BLK_RQ_SECTORS */
168 #if !defined(HAVE_BLK_RQ_BYTES) || defined(HAVE_BLK_RQ_BYTES_GPL_ONLY)
170 * Define required to avoid conflicting 2.6.29 non-static prototype for a
171 * GPL-only version of the helper. As of 2.6.31 the helper is available
172 * to non-GPL modules in the form of a static inline in the header.
174 #define blk_rq_bytes __blk_rq_bytes
175 static inline unsigned int
176 __blk_rq_bytes(struct request *req)
178 return (blk_rq_sectors(req) << 9);
180 #endif /* !HAVE_BLK_RQ_BYTES || HAVE_BLK_RQ_BYTES_GPL_ONLY */
183 * Most of the blk_* macros were removed in 2.6.36. Ostensibly this was
184 * done to improve readability and allow easier grepping. However, from
185 * a portability stand point the macros are helpful. Therefore the needed
186 * macros are redefined here if they are missing from the kernel.
188 #ifndef blk_fs_request
189 #define blk_fs_request(rq) ((rq)->cmd_type == REQ_TYPE_FS)
194 * The blk_queue_stackable() queue flag was added in 2.6.27 to handle dm
195 * stacking drivers. Prior to this request stacking drivers were detected
196 * by checking (q->request_fn == NULL), for earlier kernels we revert to
197 * this legacy behavior.
199 #ifndef blk_queue_stackable
200 #define blk_queue_stackable(q) ((q)->request_fn == NULL)
205 * The blk_queue_max_hw_sectors() function replaces blk_queue_max_sectors().
207 #ifndef HAVE_BLK_QUEUE_MAX_HW_SECTORS
208 #define blk_queue_max_hw_sectors __blk_queue_max_hw_sectors
210 __blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
212 blk_queue_max_sectors(q, max_hw_sectors);
218 * The blk_queue_max_segments() function consolidates
219 * blk_queue_max_hw_segments() and blk_queue_max_phys_segments().
221 #ifndef HAVE_BLK_QUEUE_MAX_SEGMENTS
222 #define blk_queue_max_segments __blk_queue_max_segments
224 __blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
226 blk_queue_max_phys_segments(q, max_segments);
227 blk_queue_max_hw_segments(q, max_segments);
231 #ifndef HAVE_GET_DISK_RO
233 get_disk_ro(struct gendisk *disk)
238 policy = disk->part[0]->policy;
242 #endif /* HAVE_GET_DISK_RO */
244 #ifndef HAVE_RQ_IS_SYNC
246 rq_is_sync(struct request *req)
248 return (req->flags & REQ_RW_SYNC);
250 #endif /* HAVE_RQ_IS_SYNC */
252 #ifndef HAVE_RQ_FOR_EACH_SEGMENT
253 struct req_iterator {
258 #define for_each_bio(_bio) \
259 for (; _bio; _bio = _bio->bi_next)
261 #define __rq_for_each_bio(_bio, rq) \
263 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
265 #define rq_for_each_segment(bvl, _rq, _iter) \
266 __rq_for_each_bio(_iter.bio, _rq) \
267 bio_for_each_segment(bvl, _iter.bio, _iter.i)
269 #define HAVE_RQ_FOR_EACH_SEGMENT_BVP 1
270 #endif /* HAVE_RQ_FOR_EACH_SEGMENT */
274 * rq_for_each_segment changed from taking bio_vec * to taking bio_vec.
275 * We provide rq_for_each_segment4 which takes both.
276 * You should not modify the fields in @bv and @bvp.
278 * Note: the if-else is just to inject the assignment before the loop body.
280 #ifdef HAVE_RQ_FOR_EACH_SEGMENT_BVP
281 #define rq_for_each_segment4(bv, bvp, rq, iter) \
282 rq_for_each_segment(bvp, rq, iter) \
283 if ((bv = *bvp), 0) \
287 #define rq_for_each_segment4(bv, bvp, rq, iter) \
288 rq_for_each_segment(bv, rq, iter) \
289 if ((bvp = &bv), 0) \
294 #ifdef HAVE_BIO_BVEC_ITER
295 #define BIO_BI_SECTOR(bio) (bio)->bi_iter.bi_sector
296 #define BIO_BI_SIZE(bio) (bio)->bi_iter.bi_size
297 #define BIO_BI_IDX(bio) (bio)->bi_iter.bi_idx
299 #define BIO_BI_SECTOR(bio) (bio)->bi_sector
300 #define BIO_BI_SIZE(bio) (bio)->bi_size
301 #define BIO_BI_IDX(bio) (bio)->bi_idx
305 * Portable helper for correctly setting the FAILFAST flags. The
306 * correct usage has changed 3 times from 2.6.12 to 2.6.38.
309 bio_set_flags_failfast(struct block_device *bdev, int *flags)
313 * Disable FAILFAST for loopback devices because of the
314 * following incorrect BUG_ON() in loop_make_request().
315 * This support is also disabled for md devices because the
316 * test suite layers md devices on top of loopback devices.
317 * This may be removed when the loopback driver is fixed.
319 * BUG_ON(!lo || (rw != READ && rw != WRITE));
321 if ((MAJOR(bdev->bd_dev) == LOOP_MAJOR) ||
322 (MAJOR(bdev->bd_dev) == MD_MAJOR))
325 #ifdef BLOCK_EXT_MAJOR
326 if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
328 #endif /* BLOCK_EXT_MAJOR */
329 #endif /* CONFIG_BUG */
331 #if defined(HAVE_BIO_RW_FAILFAST_DTD)
332 /* BIO_RW_FAILFAST_* preferred interface from 2.6.28 - 2.6.35 */
334 (1 << BIO_RW_FAILFAST_DEV) |
335 (1 << BIO_RW_FAILFAST_TRANSPORT) |
336 (1 << BIO_RW_FAILFAST_DRIVER));
337 #elif defined(HAVE_REQ_FAILFAST_MASK)
339 * REQ_FAILFAST_* preferred interface from 2.6.36 - 2.6.xx,
340 * the BIO_* and REQ_* flags were unified under REQ_* flags.
342 *flags |= REQ_FAILFAST_MASK;
344 #error "Undefined block IO FAILFAST interface."
349 * Maximum disk label length, it may be undefined for some kernels.
351 #ifndef DISK_NAME_LEN
352 #define DISK_NAME_LEN 32
353 #endif /* DISK_NAME_LEN */
357 * The bio_end_io() prototype changed slightly. These are helper
358 * macro's to ensure the prototype and return value are handled.
360 #ifdef HAVE_2ARGS_BIO_END_IO_T
361 #define BIO_END_IO_PROTO(fn, x, y, z) static void fn(struct bio *x, int z)
362 #define BIO_END_IO_RETURN(rc) return
364 #define BIO_END_IO_PROTO(fn, x, y, z) static int fn( \
368 #define BIO_END_IO_RETURN(rc) return rc
369 #endif /* HAVE_2ARGS_BIO_END_IO_T */
372 * 2.6.38 - 2.6.x API,
373 * blkdev_get_by_path()
376 * 2.6.28 - 2.6.37 API,
377 * open_bdev_exclusive()
378 * close_bdev_exclusive()
380 * 2.6.12 - 2.6.27 API,
384 * Used to exclusively open a block device from within the kernel.
386 #if defined(HAVE_BLKDEV_GET_BY_PATH)
387 #define vdev_bdev_open(path, md, hld) blkdev_get_by_path(path, \
388 (md) | FMODE_EXCL, hld)
389 #define vdev_bdev_close(bdev, md) blkdev_put(bdev, (md) | FMODE_EXCL)
390 #elif defined(HAVE_OPEN_BDEV_EXCLUSIVE)
391 #define vdev_bdev_open(path, md, hld) open_bdev_exclusive(path, md, hld)
392 #define vdev_bdev_close(bdev, md) close_bdev_exclusive(bdev, md)
394 #define vdev_bdev_open(path, md, hld) open_bdev_excl(path, md, hld)
395 #define vdev_bdev_close(bdev, md) close_bdev_excl(bdev)
396 #endif /* HAVE_BLKDEV_GET_BY_PATH | HAVE_OPEN_BDEV_EXCLUSIVE */
400 * The function invalidate_bdev() lost it's second argument because
403 #ifdef HAVE_1ARG_INVALIDATE_BDEV
404 #define vdev_bdev_invalidate(bdev) invalidate_bdev(bdev)
406 #define vdev_bdev_invalidate(bdev) invalidate_bdev(bdev, 1)
407 #endif /* HAVE_1ARG_INVALIDATE_BDEV */
411 * The function was exported for use, prior to this it existed by the
412 * symbol was not exported.
414 #ifndef HAVE_LOOKUP_BDEV
415 #define lookup_bdev(path) ERR_PTR(-ENOTSUP)
420 * To ensure good performance preferentially use the physical block size
421 * for proper alignment. The physical size is supposed to be the internal
422 * sector size used by the device. This is often 4096 byte for AF devices,
423 * while a smaller 512 byte logical size is supported for compatibility.
425 * Unfortunately, many drives still misreport their physical sector size.
426 * For devices which are known to lie you may need to manually set this
427 * at pool creation time with 'zpool create -o ashift=12 ...'.
429 * When the physical block size interface isn't available, we fall back to
430 * the logical block size interface and then the older hard sector size.
432 #ifdef HAVE_BDEV_PHYSICAL_BLOCK_SIZE
433 #define vdev_bdev_block_size(bdev) bdev_physical_block_size(bdev)
435 #ifdef HAVE_BDEV_LOGICAL_BLOCK_SIZE
436 #define vdev_bdev_block_size(bdev) bdev_logical_block_size(bdev)
438 #define vdev_bdev_block_size(bdev) bdev_hardsect_size(bdev)
439 #endif /* HAVE_BDEV_LOGICAL_BLOCK_SIZE */
440 #endif /* HAVE_BDEV_PHYSICAL_BLOCK_SIZE */
444 * The WRITE_FLUSH, WRITE_FUA, and WRITE_FLUSH_FUA flags have been
445 * introduced as a replacement for WRITE_BARRIER. This was done to
446 * allow richer semantics to be expressed to the block layer. It is
447 * the block layers responsibility to choose the correct way to
448 * implement these semantics.
450 * The existence of these flags implies that REQ_FLUSH an REQ_FUA are
451 * defined. Thus we can safely define VDEV_REQ_FLUSH and VDEV_REQ_FUA
452 * compatibility macros.
454 #ifdef WRITE_FLUSH_FUA
455 #define VDEV_WRITE_FLUSH_FUA WRITE_FLUSH_FUA
456 #define VDEV_REQ_FLUSH REQ_FLUSH
457 #define VDEV_REQ_FUA REQ_FUA
459 #define VDEV_WRITE_FLUSH_FUA WRITE_BARRIER
460 #define VDEV_REQ_FLUSH REQ_HARDBARRIER
461 #define VDEV_REQ_FUA REQ_FUA
466 * Use the normal I/O patch for discards.
469 #define VDEV_REQ_DISCARD REQ_DISCARD
474 * Discard granularity and alignment restrictions may now be set. For
475 * older kernels which do not support this it is safe to skip it.
477 #ifdef HAVE_DISCARD_GRANULARITY
479 blk_queue_discard_granularity(struct request_queue *q, unsigned int dg)
481 q->limits.discard_granularity = dg;
484 #define blk_queue_discard_granularity(x, dg) ((void)0)
485 #endif /* HAVE_DISCARD_GRANULARITY */
488 * Default Linux IO Scheduler,
489 * Setting the scheduler to noop will allow the Linux IO scheduler to
490 * still perform front and back merging, while leaving the request
491 * ordering and prioritization to the ZFS IO scheduler.
493 #define VDEV_SCHEDULER "noop"
496 * A common holder for vdev_bdev_open() is used to relax the exclusive open
497 * semantics slightly. Internal vdev disk callers may pass VDEV_HOLDER to
498 * allow them to open the device multiple times. Other kernel callers and
499 * user space processes which don't pass this value will get EBUSY. This is
500 * currently required for the correct operation of hot spares.
502 #define VDEV_HOLDER ((void *)0x2401de7)
504 #endif /* _ZFS_BLKDEV_H */