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]
22 * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC.
23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
28 #include <sys/zfs_context.h>
30 #include <sys/vdev_disk.h>
31 #include <sys/vdev_impl.h>
32 #include <sys/fs/zfs.h>
34 #include <sys/sunldi.h>
37 * Virtual device vector for disks.
39 typedef struct dio_request
{
40 struct completion dr_comp
; /* Completion for sync IO */
41 atomic_t dr_ref
; /* References */
42 zio_t
*dr_zio
; /* Parent ZIO */
43 int dr_rw
; /* Read/Write */
44 int dr_error
; /* Bio error */
45 int dr_bio_count
; /* Count of bio's */
46 struct bio
*dr_bio
[0]; /* Attached bio's */
50 #ifdef HAVE_OPEN_BDEV_EXCLUSIVE
52 vdev_bdev_mode(int smode
)
56 ASSERT3S(smode
& (FREAD
| FWRITE
), !=, 0);
68 vdev_bdev_mode(int smode
)
72 ASSERT3S(smode
& (FREAD
| FWRITE
), !=, 0);
74 if ((smode
& FREAD
) && !(smode
& FWRITE
))
79 #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */
82 bdev_capacity(struct block_device
*bdev
)
84 struct hd_struct
*part
= bdev
->bd_part
;
86 /* The partition capacity referenced by the block device */
88 return part
->nr_sects
;
90 /* Otherwise assume the full device capacity */
91 return get_capacity(bdev
->bd_disk
);
95 vdev_disk_open(vdev_t
*v
, uint64_t *psize
, uint64_t *ashift
)
97 struct block_device
*bdev
;
101 /* Must have a pathname and it must be absolute. */
102 if (v
->vdev_path
== NULL
|| v
->vdev_path
[0] != '/') {
103 v
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
107 vd
= kmem_zalloc(sizeof(vdev_disk_t
), KM_SLEEP
);
112 * Devices are always opened by the path provided at configuration
113 * time. This means that if the provided path is a udev by-id path
114 * then drives may be recabled without an issue. If the provided
115 * path is a udev by-path path then the physical location information
116 * will be preserved. This can be critical for more complicated
117 * configurations where drives are located in specific physical
118 * locations to maximize the systems tolerence to component failure.
119 * Alternately you can provide your own udev rule to flexibly map
120 * the drives as you see fit. It is not advised that you use the
121 * /dev/[hd]d devices which may be reorder due to probing order.
122 * Devices in the wrong locations will be detected by the higher
123 * level vdev validation.
125 mode
= spa_mode(v
->vdev_spa
);
126 bdev
= vdev_bdev_open(v
->vdev_path
, vdev_bdev_mode(mode
), vd
);
128 kmem_free(vd
, sizeof(vdev_disk_t
));
129 return -PTR_ERR(bdev
);
134 block_size
= vdev_bdev_block_size(bdev
);
136 /* We think the wholedisk property should always be set when this
137 * function is called. ASSERT here so if any legitimate cases exist
138 * where it's not set, we'll find them during debugging. If we never
139 * hit the ASSERT, this and the following conditional statement can be
141 ASSERT3S(v
->vdev_wholedisk
, !=, -1ULL);
143 /* The wholedisk property was initialized to -1 in vdev_alloc() if it
144 * was unspecified. In that case, check if this is a whole device.
145 * When bdev->bd_contains == bdev we have a whole device and not simply
147 if (v
->vdev_wholedisk
== -1ULL)
148 v
->vdev_wholedisk
= (bdev
->bd_contains
== bdev
);
150 /* Clear the nowritecache bit, causes vdev_reopen() to try again. */
151 v
->vdev_nowritecache
= B_FALSE
;
153 /* Physical volume size in bytes */
154 *psize
= bdev_capacity(bdev
) * block_size
;
156 /* Based on the minimum sector size set the block size */
157 *ashift
= highbit(MAX(block_size
, SPA_MINBLOCKSIZE
)) - 1;
163 vdev_disk_close(vdev_t
*v
)
165 vdev_disk_t
*vd
= v
->vdev_tsd
;
170 if (vd
->vd_bdev
!= NULL
)
171 vdev_bdev_close(vd
->vd_bdev
,
172 vdev_bdev_mode(spa_mode(v
->vdev_spa
)));
174 kmem_free(vd
, sizeof(vdev_disk_t
));
178 static dio_request_t
*
179 vdev_disk_dio_alloc(int bio_count
)
184 dr
= kmem_zalloc(sizeof(dio_request_t
) +
185 sizeof(struct bio
*) * bio_count
, KM_SLEEP
);
187 init_completion(&dr
->dr_comp
);
188 atomic_set(&dr
->dr_ref
, 0);
189 dr
->dr_bio_count
= bio_count
;
192 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
193 dr
->dr_bio
[i
] = NULL
;
200 vdev_disk_dio_free(dio_request_t
*dr
)
204 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
206 bio_put(dr
->dr_bio
[i
]);
208 kmem_free(dr
, sizeof(dio_request_t
) +
209 sizeof(struct bio
*) * dr
->dr_bio_count
);
213 vdev_disk_dio_get(dio_request_t
*dr
)
215 atomic_inc(&dr
->dr_ref
);
219 vdev_disk_dio_put(dio_request_t
*dr
)
221 int rc
= atomic_dec_return(&dr
->dr_ref
);
224 * Free the dio_request when the last reference is dropped and
225 * ensure zio_interpret is called only once with the correct zio
228 zio_t
*zio
= dr
->dr_zio
;
229 int error
= dr
->dr_error
;
231 vdev_disk_dio_free(dr
);
234 zio
->io_error
= error
;
242 BIO_END_IO_PROTO(vdev_disk_physio_completion
, bio
, size
, error
)
244 dio_request_t
*dr
= bio
->bi_private
;
247 /* Fatal error but print some useful debugging before asserting */
249 PANIC("dr == NULL, bio->bi_private == NULL\n"
250 "bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d\n"
251 "bi_idx: %d, bi_size: %d, bi_end_io: %p, bi_cnt: %d\n",
252 bio
->bi_next
, bio
->bi_flags
, bio
->bi_rw
, bio
->bi_vcnt
,
253 bio
->bi_idx
, bio
->bi_size
, bio
->bi_end_io
,
254 atomic_read(&bio
->bi_cnt
));
256 #ifndef HAVE_2ARGS_BIO_END_IO_T
259 #endif /* HAVE_2ARGS_BIO_END_IO_T */
261 if (error
== 0 && !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
264 if (dr
->dr_error
== 0)
265 dr
->dr_error
= error
;
267 /* Drop reference aquired by __vdev_disk_physio */
268 rc
= vdev_disk_dio_put(dr
);
270 /* Wake up synchronous waiter this is the last outstanding bio */
271 if ((rc
== 1) && (dr
->dr_rw
& (1 << DIO_RW_SYNCIO
)))
272 complete(&dr
->dr_comp
);
274 BIO_END_IO_RETURN(0);
277 static inline unsigned long
278 bio_nr_pages(void *bio_ptr
, unsigned int bio_size
)
280 return ((((unsigned long)bio_ptr
+ bio_size
+ PAGE_SIZE
- 1) >>
281 PAGE_SHIFT
) - ((unsigned long)bio_ptr
>> PAGE_SHIFT
));
285 bio_map(struct bio
*bio
, void *bio_ptr
, unsigned int bio_size
)
287 unsigned int offset
, size
, i
;
290 offset
= offset_in_page(bio_ptr
);
291 for (i
= 0; i
< bio
->bi_max_vecs
; i
++) {
292 size
= PAGE_SIZE
- offset
;
300 if (kmem_virt(bio_ptr
))
301 page
= vmalloc_to_page(bio_ptr
);
303 page
= virt_to_page(bio_ptr
);
305 if (bio_add_page(bio
, page
, size
, offset
) != size
)
317 __vdev_disk_physio(struct block_device
*bdev
, zio_t
*zio
, caddr_t kbuf_ptr
,
318 size_t kbuf_size
, uint64_t kbuf_offset
, int flags
)
323 int bio_size
, bio_count
= 16;
324 int i
= 0, error
= 0, block_size
;
327 dr
= vdev_disk_dio_alloc(bio_count
);
333 block_size
= vdev_bdev_block_size(bdev
);
335 #ifdef BIO_RW_FAILFAST
336 if (flags
& (1 << BIO_RW_FAILFAST
))
337 dr
->dr_rw
|= 1 << BIO_RW_FAILFAST
;
338 #endif /* BIO_RW_FAILFAST */
341 * When the IO size exceeds the maximum bio size for the request
342 * queue we are forced to break the IO in multiple bio's and wait
343 * for them all to complete. Ideally, all pool users will set
344 * their volume block size to match the maximum request size and
345 * the common case will be one bio per vdev IO request.
348 bio_offset
= kbuf_offset
;
349 bio_size
= kbuf_size
;
350 for (i
= 0; i
<= dr
->dr_bio_count
; i
++) {
352 /* Finished constructing bio's for given buffer */
357 * By default only 'bio_count' bio's per dio are allowed.
358 * However, if we find ourselves in a situation where more
359 * are needed we allocate a larger dio and warn the user.
361 if (dr
->dr_bio_count
== i
) {
362 vdev_disk_dio_free(dr
);
364 printk("WARNING: Resized bio's/dio to %d\n",bio_count
);
368 dr
->dr_bio
[i
] = bio_alloc(GFP_NOIO
,
369 bio_nr_pages(bio_ptr
, bio_size
));
370 if (dr
->dr_bio
[i
] == NULL
) {
371 vdev_disk_dio_free(dr
);
375 /* Matching put called by vdev_disk_physio_completion */
376 vdev_disk_dio_get(dr
);
378 dr
->dr_bio
[i
]->bi_bdev
= bdev
;
379 dr
->dr_bio
[i
]->bi_sector
= bio_offset
/ block_size
;
380 dr
->dr_bio
[i
]->bi_rw
= dr
->dr_rw
;
381 dr
->dr_bio
[i
]->bi_end_io
= vdev_disk_physio_completion
;
382 dr
->dr_bio
[i
]->bi_private
= dr
;
384 /* Remaining size is returned to become the new size */
385 bio_size
= bio_map(dr
->dr_bio
[i
], bio_ptr
, bio_size
);
387 /* Advance in buffer and construct another bio if needed */
388 bio_ptr
+= dr
->dr_bio
[i
]->bi_size
;
389 bio_offset
+= dr
->dr_bio
[i
]->bi_size
;
392 /* Extra reference to protect dio_request during submit_bio */
393 vdev_disk_dio_get(dr
);
395 /* Submit all bio's associated with this dio */
396 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
398 submit_bio(dr
->dr_rw
, dr
->dr_bio
[i
]);
401 * On synchronous blocking requests we wait for all bio the completion
402 * callbacks to run. We will be woken when the last callback runs
403 * for this dio. We are responsible for putting the last dio_request
404 * reference will in turn put back the last bio references. The
405 * only synchronous consumer is vdev_disk_read_rootlabel() all other
406 * IO originating from vdev_disk_io_start() is asynchronous.
408 if (dr
->dr_rw
& (1 << DIO_RW_SYNCIO
)) {
409 wait_for_completion(&dr
->dr_comp
);
410 error
= dr
->dr_error
;
411 ASSERT3S(atomic_read(&dr
->dr_ref
), ==, 1);
414 (void)vdev_disk_dio_put(dr
);
420 vdev_disk_physio(struct block_device
*bdev
, caddr_t kbuf
,
421 size_t size
, uint64_t offset
, int flags
)
423 return __vdev_disk_physio(bdev
, NULL
, kbuf
, size
, offset
, flags
);
426 /* 2.6.24 API change */
427 #ifdef HAVE_BIO_EMPTY_BARRIER
428 BIO_END_IO_PROTO(vdev_disk_io_flush_completion
, bio
, size
, rc
)
430 zio_t
*zio
= bio
->bi_private
;
433 if (rc
&& (rc
== -EOPNOTSUPP
))
434 zio
->io_vd
->vdev_nowritecache
= B_TRUE
;
439 BIO_END_IO_RETURN(0);
443 vdev_disk_io_flush(struct block_device
*bdev
, zio_t
*zio
)
445 struct request_queue
*q
;
448 q
= bdev_get_queue(bdev
);
452 bio
= bio_alloc(GFP_KERNEL
, 0);
456 bio
->bi_end_io
= vdev_disk_io_flush_completion
;
457 bio
->bi_private
= zio
;
459 submit_bio(WRITE_BARRIER
, bio
);
465 vdev_disk_io_flush(struct block_device
*bdev
, zio_t
*zio
)
469 #endif /* HAVE_BIO_EMPTY_BARRIER */
472 vdev_disk_io_start(zio_t
*zio
)
474 vdev_t
*v
= zio
->io_vd
;
475 vdev_disk_t
*vd
= v
->vdev_tsd
;
478 switch (zio
->io_type
) {
481 if (!vdev_readable(v
)) {
482 zio
->io_error
= ENXIO
;
483 return ZIO_PIPELINE_CONTINUE
;
486 switch (zio
->io_cmd
) {
487 case DKIOCFLUSHWRITECACHE
:
489 if (zfs_nocacheflush
)
492 if (v
->vdev_nowritecache
) {
493 zio
->io_error
= ENOTSUP
;
497 error
= vdev_disk_io_flush(vd
->vd_bdev
, zio
);
499 return ZIO_PIPELINE_STOP
;
501 zio
->io_error
= error
;
502 if (error
== ENOTSUP
)
503 v
->vdev_nowritecache
= B_TRUE
;
508 zio
->io_error
= ENOTSUP
;
511 return ZIO_PIPELINE_CONTINUE
;
522 zio
->io_error
= ENOTSUP
;
523 return ZIO_PIPELINE_CONTINUE
;
526 #ifdef BIO_RW_FAILFAST
527 if (zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
))
528 flags
|= (1 << BIO_RW_FAILFAST
);
529 #endif /* BIO_RW_FAILFAST */
531 error
= __vdev_disk_physio(vd
->vd_bdev
, zio
, zio
->io_data
,
532 zio
->io_size
, zio
->io_offset
, flags
);
534 zio
->io_error
= error
;
535 return ZIO_PIPELINE_CONTINUE
;
538 return ZIO_PIPELINE_STOP
;
542 vdev_disk_io_done(zio_t
*zio
)
545 * If the device returned EIO, we revalidate the media. If it is
546 * determined the media has changed this triggers the asynchronous
547 * removal of the device from the configuration.
549 if (zio
->io_error
== EIO
) {
550 vdev_t
*v
= zio
->io_vd
;
551 vdev_disk_t
*vd
= v
->vdev_tsd
;
553 if (check_disk_change(vd
->vd_bdev
)) {
554 vdev_bdev_invalidate(vd
->vd_bdev
);
555 v
->vdev_remove_wanted
= B_TRUE
;
556 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
562 vdev_disk_hold(vdev_t
*vd
)
564 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
566 /* We must have a pathname, and it must be absolute. */
567 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
571 * Only prefetch path and devid info if the device has
574 if (vd
->vdev_tsd
!= NULL
)
577 /* XXX: Implement me as a vnode lookup for the device */
578 vd
->vdev_name_vp
= NULL
;
579 vd
->vdev_devid_vp
= NULL
;
583 vdev_disk_rele(vdev_t
*vd
)
585 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
587 /* XXX: Implement me as a vnode rele for the device */
590 vdev_ops_t vdev_disk_ops
= {
599 VDEV_TYPE_DISK
, /* name of this vdev type */
600 B_TRUE
/* leaf vdev */
604 * Given the root disk device devid or pathname, read the label from
605 * the device, and construct a configuration nvlist.
608 vdev_disk_read_rootlabel(char *devpath
, char *devid
, nvlist_t
**config
)
610 struct block_device
*bdev
;
615 bdev
= vdev_bdev_open(devpath
, vdev_bdev_mode(FREAD
), NULL
);
617 return -PTR_ERR(bdev
);
619 s
= bdev_capacity(bdev
) * vdev_bdev_block_size(bdev
);
621 vdev_bdev_close(bdev
, vdev_bdev_mode(FREAD
));
625 size
= P2ALIGN_TYPED(s
, sizeof(vdev_label_t
), uint64_t);
626 label
= vmem_alloc(sizeof(vdev_label_t
), KM_SLEEP
);
628 for (i
= 0; i
< VDEV_LABELS
; i
++) {
629 uint64_t offset
, state
, txg
= 0;
631 /* read vdev label */
632 offset
= vdev_label_offset(size
, i
, 0);
633 if (vdev_disk_physio(bdev
, (caddr_t
)label
,
634 VDEV_SKIP_SIZE
+ VDEV_PHYS_SIZE
, offset
, READ_SYNC
) != 0)
637 if (nvlist_unpack(label
->vl_vdev_phys
.vp_nvlist
,
638 sizeof (label
->vl_vdev_phys
.vp_nvlist
), config
, 0) != 0) {
643 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_STATE
,
644 &state
) != 0 || state
>= POOL_STATE_DESTROYED
) {
645 nvlist_free(*config
);
650 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_TXG
,
651 &txg
) != 0 || txg
== 0) {
652 nvlist_free(*config
);
660 vmem_free(label
, sizeof(vdev_label_t
));
661 vdev_bdev_close(bdev
, vdev_bdev_mode(FREAD
));