]> git.proxmox.com Git - mirror_zfs.git/blame - include/linux/blkdev_compat.h
Improve ZVOL queue behavior.
[mirror_zfs.git] / include / linux / blkdev_compat.h
CommitLineData
60101509
BB
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 http://www.opensolaris.org/os/licensing.
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 */
61e90960 21
60101509 22/*
61e90960 23 * Copyright (C) 2011 Lawrence Livermore National Security, LLC.
60101509
BB
24 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
25 * Written by Brian Behlendorf <behlendorf1@llnl.gov>.
26 * LLNL-CODE-403049.
27 */
28
61e90960
BB
29#ifndef _ZFS_BLKDEV_H
30#define _ZFS_BLKDEV_H
60101509
BB
31
32#include <linux/blkdev.h>
33#include <linux/elevator.h>
34
35#ifndef HAVE_FMODE_T
36typedef unsigned __bitwise__ fmode_t;
37#endif /* HAVE_FMODE_T */
38
39#ifndef HAVE_BLK_FETCH_REQUEST
40static inline struct request *
41blk_fetch_request(struct request_queue *q)
42{
43 struct request *req;
44
45 req = elv_next_request(q);
46 if (req)
47 blkdev_dequeue_request(req);
48
49 return req;
50}
51#endif /* HAVE_BLK_FETCH_REQUEST */
52
53#ifndef HAVE_BLK_REQUEUE_REQUEST
54static inline void
55blk_requeue_request(request_queue_t *q, struct request *req)
56{
57 elv_requeue_request(q, req);
58}
59#endif /* HAVE_BLK_REQUEUE_REQUEST */
60
61#ifndef HAVE_BLK_END_REQUEST
62static inline bool
63__blk_end_request(struct request *req, int error, unsigned int nr_bytes)
64{
65 LIST_HEAD(list);
66
67 /*
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.
71 */
72 list_add(&req->queuelist, &list);
73
74 /*
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.
78 */
79 req->hard_cur_sectors = nr_bytes >> 9;
80 end_request(req, ((error == 0) ? 1 : error));
81
82 return 0;
83}
84
85static inline bool
86blk_end_request(struct request *req, int error, unsigned int nr_bytes)
87{
88 struct request_queue *q = req->q;
89 bool rc;
90
91 spin_lock_irq(q->queue_lock);
92 rc = __blk_end_request(req, error, nr_bytes);
93 spin_unlock_irq(q->queue_lock);
94
95 return rc;
96}
97#else
98# ifdef HAVE_BLK_END_REQUEST_GPL_ONLY
99/*
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.
103 */
104# define __blk_end_request __blk_end_request_x
105# define blk_end_request blk_end_request_x
106
107static inline bool
108__blk_end_request_x(struct request *req, int error, unsigned int nr_bytes)
109{
110 /*
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.
114 */
115 req->hard_cur_sectors = nr_bytes >> 9;
116 end_request(req, ((error == 0) ? 1 : error));
117
118 return 0;
119}
120static inline bool
121blk_end_request_x(struct request *req, int error, unsigned int nr_bytes)
122{
123 struct request_queue *q = req->q;
124 bool rc;
125
126 spin_lock_irq(q->queue_lock);
127 rc = __blk_end_request_x(req, error, nr_bytes);
128 spin_unlock_irq(q->queue_lock);
129
130 return rc;
131}
132# endif /* HAVE_BLK_END_REQUEST_GPL_ONLY */
133#endif /* HAVE_BLK_END_REQUEST */
134
b18019d2
ED
135/*
136 * 2.6.36 API change,
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.
142 */
143#if defined(HAVE_BLK_QUEUE_FLUSH) && defined(HAVE_BLK_QUEUE_FLUSH_GPL_ONLY)
144#define blk_queue_flush __blk_queue_flush
145static inline void
146__blk_queue_flush(struct request_queue *q, unsigned int flags)
147{
148 q->flush_flags = flags & (REQ_FLUSH | REQ_FUA);
149}
150#endif /* HAVE_BLK_QUEUE_FLUSH && HAVE_BLK_QUEUE_FLUSH_GPL_ONLY */
151
60101509
BB
152#ifndef HAVE_BLK_RQ_POS
153static inline sector_t
154blk_rq_pos(struct request *req)
155{
156 return req->sector;
157}
158#endif /* HAVE_BLK_RQ_POS */
159
160#ifndef HAVE_BLK_RQ_SECTORS
161static inline unsigned int
162blk_rq_sectors(struct request *req)
163{
164 return req->nr_sectors;
165}
166#endif /* HAVE_BLK_RQ_SECTORS */
167
168#if !defined(HAVE_BLK_RQ_BYTES) || defined(HAVE_BLK_RQ_BYTES_GPL_ONLY)
169/*
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.
173 */
174#define blk_rq_bytes __blk_rq_bytes
175static inline unsigned int
176__blk_rq_bytes(struct request *req)
177{
178 return blk_rq_sectors(req) << 9;
179}
180#endif /* !HAVE_BLK_RQ_BYTES || HAVE_BLK_RQ_BYTES_GPL_ONLY */
181
8326eb46
BB
182/*
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.
187 */
188#ifndef blk_fs_request
189#define blk_fs_request(rq) ((rq)->cmd_type == REQ_TYPE_FS)
190#endif
191
3517f0b7
BB
192/*
193 * 2.6.27 API change,
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.
198 */
199#ifndef blk_queue_stackable
200#define blk_queue_stackable(q) ((q)->request_fn == NULL)
201#endif
202
34037afe
ED
203/*
204 * 2.6.34 API change,
205 * The blk_queue_max_hw_sectors() function replaces blk_queue_max_sectors().
206 */
207#ifndef HAVE_BLK_QUEUE_MAX_HW_SECTORS
208#define blk_queue_max_hw_sectors __blk_queue_max_hw_sectors
209static inline void
210__blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
211{
212 blk_queue_max_sectors(q, max_hw_sectors);
213}
214#endif
215
216/*
217 * 2.6.34 API change,
218 * The blk_queue_max_segments() function consolidates
219 * blk_queue_max_hw_segments() and blk_queue_max_phys_segments().
220 */
221#ifndef HAVE_BLK_QUEUE_MAX_SEGMENTS
222#define blk_queue_max_segments __blk_queue_max_segments
223static inline void
224__blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
225{
226 blk_queue_max_phys_segments(q, max_segments);
227 blk_queue_max_hw_segments(q, max_segments);
228}
229#endif
230
231/*
232 * 2.6.30 API change,
233 * The blk_queue_physical_block_size() function was introduced to
234 * indicate the smallest I/O the device can write without incurring
235 * a read-modify-write penalty. For older kernels this is a no-op.
236 */
237#ifndef HAVE_BLK_QUEUE_PHYSICAL_BLOCK_SIZE
238#define blk_queue_physical_block_size(q, x) ((void)(0))
239#endif
240
241/*
242 * 2.6.30 API change,
243 * The blk_queue_io_opt() function was added to indicate the optimal
244 * I/O size for the device. For older kernels this is a no-op.
245 */
246#ifndef HAVE_BLK_QUEUE_IO_OPT
247#define blk_queue_io_opt(q, x) ((void)(0))
248#endif
249
60101509
BB
250#ifndef HAVE_GET_DISK_RO
251static inline int
252get_disk_ro(struct gendisk *disk)
253{
254 int policy = 0;
255
256 if (disk->part[0])
257 policy = disk->part[0]->policy;
258
259 return policy;
260}
261#endif /* HAVE_GET_DISK_RO */
262
263#ifndef HAVE_RQ_IS_SYNC
264static inline bool
265rq_is_sync(struct request *req)
266{
267 return (req->flags & REQ_RW_SYNC);
268}
269#endif /* HAVE_RQ_IS_SYNC */
270
271#ifndef HAVE_RQ_FOR_EACH_SEGMENT
272struct req_iterator {
273 int i;
274 struct bio *bio;
275};
276
277# define for_each_bio(_bio) \
278 for (; _bio; _bio = _bio->bi_next)
279
280# define __rq_for_each_bio(_bio, rq) \
281 if ((rq->bio)) \
282 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
283
284# define rq_for_each_segment(bvl, _rq, _iter) \
285 __rq_for_each_bio(_iter.bio, _rq) \
286 bio_for_each_segment(bvl, _iter.bio, _iter.i)
287#endif /* HAVE_RQ_FOR_EACH_SEGMENT */
288
61e90960
BB
289/*
290 * Portable helper for correctly setting the FAILFAST flags. The
291 * correct usage has changed 3 times from 2.6.12 to 2.6.38.
292 */
2959d94a
BB
293static inline void
294bio_set_flags_failfast(struct block_device *bdev, int *flags)
295{
f4af6bb7 296#ifdef CONFIG_BUG
2959d94a 297 /*
f4af6bb7
BB
298 * Disable FAILFAST for loopback devices because of the
299 * following incorrect BUG_ON() in loop_make_request().
2959d94a
BB
300 * This support is also disabled for md devices because the
301 * test suite layers md devices on top of loopback devices.
302 * This may be removed when the loopback driver is fixed.
303 *
304 * BUG_ON(!lo || (rw != READ && rw != WRITE));
305 */
2959d94a
BB
306 if ((MAJOR(bdev->bd_dev) == LOOP_MAJOR) ||
307 (MAJOR(bdev->bd_dev) == MD_MAJOR))
308 return;
309
310#ifdef BLOCK_EXT_MAJOR
311 if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
312 return;
313#endif /* BLOCK_EXT_MAJOR */
314#endif /* CONFIG_BUG */
f4af6bb7
BB
315
316#ifdef HAVE_BIO_RW_FAILFAST_DTD
317 /* BIO_RW_FAILFAST_* preferred interface from 2.6.28 - 2.6.35 */
2959d94a
BB
318 *flags |=
319 ((1 << BIO_RW_FAILFAST_DEV) |
320 (1 << BIO_RW_FAILFAST_TRANSPORT) |
321 (1 << BIO_RW_FAILFAST_DRIVER));
f4af6bb7
BB
322#else
323# ifdef HAVE_BIO_RW_FAILFAST
324 /* BIO_RW_FAILFAST preferred interface from 2.6.12 - 2.6.27 */
2959d94a 325 *flags |= (1 << BIO_RW_FAILFAST);
f4af6bb7
BB
326# else
327# ifdef HAVE_REQ_FAILFAST_MASK
328 /* REQ_FAILFAST_* preferred interface from 2.6.36 - 2.6.xx,
329 * the BIO_* and REQ_* flags were unified under REQ_* flags. */
330 *flags |= REQ_FAILFAST_MASK;
331# endif /* HAVE_REQ_FAILFAST_MASK */
332# endif /* HAVE_BIO_RW_FAILFAST */
333#endif /* HAVE_BIO_RW_FAILFAST_DTD */
2959d94a
BB
334}
335
61e90960
BB
336/*
337 * Maximum disk label length, it may be undefined for some kernels.
338 */
60101509
BB
339#ifndef DISK_NAME_LEN
340#define DISK_NAME_LEN 32
341#endif /* DISK_NAME_LEN */
342
61e90960
BB
343/*
344 * 2.6.24 API change,
345 * The bio_end_io() prototype changed slightly. These are helper
346 * macro's to ensure the prototype and return value are handled.
347 */
348#ifdef HAVE_2ARGS_BIO_END_IO_T
349# define BIO_END_IO_PROTO(fn, x, y, z) static void fn(struct bio *x, int z)
350# define BIO_END_IO_RETURN(rc) return
351#else
352# define BIO_END_IO_PROTO(fn, x, y, z) static int fn(struct bio *x, \
353 unsigned int y, int z)
354# define BIO_END_IO_RETURN(rc) return rc
355#endif /* HAVE_2ARGS_BIO_END_IO_T */
356
357/*
45066d1f
BB
358 * 2.6.38 - 2.6.x API,
359 * blkdev_get_by_path()
360 * blkdev_put()
361 *
362 * 2.6.28 - 2.6.37 API,
363 * open_bdev_exclusive()
364 * close_bdev_exclusive()
365 *
366 * 2.6.12 - 2.6.27 API,
367 * open_bdev_excl()
368 * close_bdev_excl()
369 *
61e90960
BB
370 * Used to exclusively open a block device from within the kernel.
371 */
45066d1f
BB
372#if defined(HAVE_BLKDEV_GET_BY_PATH)
373# define vdev_bdev_open(path, md, hld) blkdev_get_by_path(path, \
374 (md) | FMODE_EXCL, hld)
375# define vdev_bdev_close(bdev, md) blkdev_put(bdev, (md) | FMODE_EXCL)
376#elif defined(HAVE_OPEN_BDEV_EXCLUSIVE)
61e90960
BB
377# define vdev_bdev_open(path, md, hld) open_bdev_exclusive(path, md, hld)
378# define vdev_bdev_close(bdev, md) close_bdev_exclusive(bdev, md)
379#else
380# define vdev_bdev_open(path, md, hld) open_bdev_excl(path, md, hld)
381# define vdev_bdev_close(bdev, md) close_bdev_excl(bdev)
45066d1f 382#endif /* HAVE_BLKDEV_GET_BY_PATH | HAVE_OPEN_BDEV_EXCLUSIVE */
61e90960
BB
383
384/*
385 * 2.6.22 API change
386 * The function invalidate_bdev() lost it's second argument because
387 * it was unused.
388 */
389#ifdef HAVE_1ARG_INVALIDATE_BDEV
390# define vdev_bdev_invalidate(bdev) invalidate_bdev(bdev)
391#else
392# define vdev_bdev_invalidate(bdev) invalidate_bdev(bdev, 1)
393#endif /* HAVE_1ARG_INVALIDATE_BDEV */
394
395/*
396 * 2.6.30 API change
397 * Change to make it explicit there this is the logical block size.
398 */
399#ifdef HAVE_BDEV_LOGICAL_BLOCK_SIZE
400# define vdev_bdev_block_size(bdev) bdev_logical_block_size(bdev)
401#else
402# define vdev_bdev_block_size(bdev) bdev_hardsect_size(bdev)
403#endif
404
96801d29
BB
405/*
406 * 2.6.37 API change
407 * The WRITE_FLUSH, WRITE_FUA, and WRITE_FLUSH_FUA flags have been
408 * introduced as a replacement for WRITE_BARRIER. This was done to
409 * allow richer semantics to be expressed to the block layer. It is
410 * the block layers responsibility to choose the correct way to
411 * implement these semantics.
b18019d2
ED
412 *
413 * The existence of these flags implies that REQ_FLUSH an REQ_FUA are
414 * defined. Thus we can safely define VDEV_REQ_FLUSH and VDEV_REQ_FUA
415 * compatibility macros.
96801d29
BB
416 */
417#ifdef WRITE_FLUSH_FUA
418# define VDEV_WRITE_FLUSH_FUA WRITE_FLUSH_FUA
b18019d2
ED
419# define VDEV_REQ_FLUSH REQ_FLUSH
420# define VDEV_REQ_FUA REQ_FUA
96801d29
BB
421#else
422# define VDEV_WRITE_FLUSH_FUA WRITE_BARRIER
b18019d2
ED
423# define VDEV_REQ_FLUSH REQ_HARDBARRIER
424# define VDEV_REQ_FUA REQ_HARDBARRIER
96801d29
BB
425#endif
426
61e90960
BB
427/*
428 * Default Linux IO Scheduler,
429 * Setting the scheduler to noop will allow the Linux IO scheduler to
430 * still perform front and back merging, while leaving the request
431 * ordering and prioritization to the ZFS IO scheduler.
432 */
433#define VDEV_SCHEDULER "noop"
60101509 434
61e90960 435#endif /* _ZFS_BLKDEV_H */