]>
Commit | Line | Data |
---|---|---|
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 | */ | |
21 | /* | |
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>. | |
25 | * LLNL-CODE-403049. | |
9bd274dd | 26 | * Copyright (c) 2012, 2014 by Delphix. All rights reserved. |
60101509 BB |
27 | */ |
28 | ||
29 | #include <sys/zfs_context.h> | |
30 | #include <sys/spa.h> | |
31 | #include <sys/vdev_disk.h> | |
32 | #include <sys/vdev_impl.h> | |
33 | #include <sys/fs/zfs.h> | |
34 | #include <sys/zio.h> | |
35 | #include <sys/sunldi.h> | |
36 | ||
6839eed2 | 37 | char *zfs_vdev_scheduler = VDEV_SCHEDULER; |
8128bd89 | 38 | static void *zfs_vdev_holder = VDEV_HOLDER; |
6839eed2 | 39 | |
60101509 BB |
40 | /* |
41 | * Virtual device vector for disks. | |
42 | */ | |
43 | typedef struct dio_request { | |
44 | struct completion dr_comp; /* Completion for sync IO */ | |
45 | atomic_t dr_ref; /* References */ | |
46 | zio_t *dr_zio; /* Parent ZIO */ | |
47 | int dr_rw; /* Read/Write */ | |
48 | int dr_error; /* Bio error */ | |
49 | int dr_bio_count; /* Count of bio's */ | |
d1d7e268 | 50 | struct bio *dr_bio[0]; /* Attached bio's */ |
60101509 BB |
51 | } dio_request_t; |
52 | ||
53 | ||
54 | #ifdef HAVE_OPEN_BDEV_EXCLUSIVE | |
55 | static fmode_t | |
56 | vdev_bdev_mode(int smode) | |
57 | { | |
58 | fmode_t mode = 0; | |
59 | ||
60 | ASSERT3S(smode & (FREAD | FWRITE), !=, 0); | |
61 | ||
62 | if (smode & FREAD) | |
63 | mode |= FMODE_READ; | |
64 | ||
65 | if (smode & FWRITE) | |
66 | mode |= FMODE_WRITE; | |
67 | ||
d1d7e268 | 68 | return (mode); |
60101509 BB |
69 | } |
70 | #else | |
71 | static int | |
72 | vdev_bdev_mode(int smode) | |
73 | { | |
74 | int mode = 0; | |
75 | ||
76 | ASSERT3S(smode & (FREAD | FWRITE), !=, 0); | |
77 | ||
78 | if ((smode & FREAD) && !(smode & FWRITE)) | |
79 | mode = MS_RDONLY; | |
80 | ||
d1d7e268 | 81 | return (mode); |
60101509 BB |
82 | } |
83 | #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */ | |
84 | ||
85 | static uint64_t | |
86 | bdev_capacity(struct block_device *bdev) | |
87 | { | |
88 | struct hd_struct *part = bdev->bd_part; | |
89 | ||
90 | /* The partition capacity referenced by the block device */ | |
91 | if (part) | |
f74fae8b | 92 | return (part->nr_sects << 9); |
60101509 BB |
93 | |
94 | /* Otherwise assume the full device capacity */ | |
f74fae8b | 95 | return (get_capacity(bdev->bd_disk) << 9); |
60101509 BB |
96 | } |
97 | ||
d148e951 BB |
98 | static void |
99 | vdev_disk_error(zio_t *zio) | |
100 | { | |
101 | #ifdef ZFS_DEBUG | |
a69052be BB |
102 | printk("ZFS: zio error=%d type=%d offset=%llu size=%llu " |
103 | "flags=%x delay=%llu\n", zio->io_error, zio->io_type, | |
d148e951 | 104 | (u_longlong_t)zio->io_offset, (u_longlong_t)zio->io_size, |
a69052be | 105 | zio->io_flags, (u_longlong_t)zio->io_delay); |
d148e951 BB |
106 | #endif |
107 | } | |
108 | ||
6839eed2 BB |
109 | /* |
110 | * Use the Linux 'noop' elevator for zfs managed block devices. This | |
111 | * strikes the ideal balance by allowing the zfs elevator to do all | |
112 | * request ordering and prioritization. While allowing the Linux | |
113 | * elevator to do the maximum front/back merging allowed by the | |
114 | * physical device. This yields the largest possible requests for | |
115 | * the device with the lowest total overhead. | |
6839eed2 BB |
116 | */ |
117 | static int | |
fdcd952b | 118 | vdev_elevator_switch(vdev_t *v, char *elevator) |
6839eed2 | 119 | { |
fdcd952b BB |
120 | vdev_disk_t *vd = v->vdev_tsd; |
121 | struct block_device *bdev = vd->vd_bdev; | |
122 | struct request_queue *q = bdev_get_queue(bdev); | |
123 | char *device = bdev->bd_disk->disk_name; | |
e2448b0e | 124 | int error; |
fdcd952b | 125 | |
84daadde PS |
126 | /* |
127 | * Skip devices which are not whole disks (partitions). | |
128 | * Device-mapper devices are excepted since they may be whole | |
129 | * disks despite the vdev_wholedisk flag, in which case we can | |
130 | * and should switch the elevator. If the device-mapper device | |
131 | * does not have an elevator (i.e. dm-raid, dm-crypt, etc.) the | |
132 | * "Skip devices without schedulers" check below will fail. | |
133 | */ | |
134 | if (!v->vdev_wholedisk && strncmp(device, "dm-", 3) != 0) | |
04516a45 BB |
135 | return (0); |
136 | ||
fdcd952b BB |
137 | /* Skip devices without schedulers (loop, ram, dm, etc) */ |
138 | if (!q->elevator || !blk_queue_stackable(q)) | |
139 | return (0); | |
6839eed2 | 140 | |
fdcd952b | 141 | /* Leave existing scheduler when set to "none" */ |
d1d7e268 | 142 | if (strncmp(elevator, "none", 4) && (strlen(elevator) == 4) == 0) |
6839eed2 BB |
143 | return (0); |
144 | ||
6d1d976b BB |
145 | #ifdef HAVE_ELEVATOR_CHANGE |
146 | error = elevator_change(q, elevator); | |
147 | #else | |
d1d7e268 MK |
148 | /* |
149 | * For pre-2.6.36 kernels elevator_change() is not available. | |
6d1d976b BB |
150 | * Therefore we fall back to using a usermodehelper to echo the |
151 | * elevator into sysfs; This requires /bin/echo and sysfs to be | |
152 | * mounted which may not be true early in the boot process. | |
153 | */ | |
d1d7e268 | 154 | #define SET_SCHEDULER_CMD \ |
6d1d976b BB |
155 | "exec 0</dev/null " \ |
156 | " 1>/sys/block/%s/queue/scheduler " \ | |
157 | " 2>/dev/null; " \ | |
158 | "echo %s" | |
159 | ||
160 | { | |
161 | char *argv[] = { "/bin/sh", "-c", NULL, NULL }; | |
162 | char *envp[] = { NULL }; | |
163 | ||
164 | argv[2] = kmem_asprintf(SET_SCHEDULER_CMD, device, elevator); | |
761394b3 | 165 | error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); |
6d1d976b BB |
166 | strfree(argv[2]); |
167 | } | |
168 | #endif /* HAVE_ELEVATOR_CHANGE */ | |
6839eed2 BB |
169 | if (error) |
170 | printk("ZFS: Unable to set \"%s\" scheduler for %s (%s): %d\n", | |
d1d7e268 | 171 | elevator, v->vdev_path, device, error); |
6839eed2 BB |
172 | |
173 | return (error); | |
174 | } | |
175 | ||
b5a28807 ED |
176 | /* |
177 | * Expanding a whole disk vdev involves invoking BLKRRPART on the | |
178 | * whole disk device. This poses a problem, because BLKRRPART will | |
179 | * return EBUSY if one of the disk's partitions is open. That's why | |
180 | * we have to do it here, just before opening the data partition. | |
181 | * Unfortunately, BLKRRPART works by dropping all partitions and | |
182 | * recreating them, which means that for a short time window, all | |
183 | * /dev/sdxN device files disappear (until udev recreates them). | |
184 | * This means two things: | |
185 | * - When we open the data partition just after a BLKRRPART, we | |
186 | * can't do it using the normal device file path because of the | |
187 | * obvious race condition with udev. Instead, we use reliable | |
188 | * kernel APIs to get a handle to the new partition device from | |
189 | * the whole disk device. | |
190 | * - Because vdev_disk_open() initially needs to find the device | |
191 | * using its path, multiple vdev_disk_open() invocations in | |
192 | * short succession on the same disk with BLKRRPARTs in the | |
193 | * middle have a high probability of failure (because of the | |
194 | * race condition with udev). A typical situation where this | |
195 | * might happen is when the zpool userspace tool does a | |
196 | * TRYIMPORT immediately followed by an IMPORT. For this | |
197 | * reason, we only invoke BLKRRPART in the module when strictly | |
198 | * necessary (zpool online -e case), and rely on userspace to | |
199 | * do it when possible. | |
200 | */ | |
201 | static struct block_device * | |
202 | vdev_disk_rrpart(const char *path, int mode, vdev_disk_t *vd) | |
203 | { | |
204 | #if defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) | |
205 | struct block_device *bdev, *result = ERR_PTR(-ENXIO); | |
206 | struct gendisk *disk; | |
207 | int error, partno; | |
208 | ||
8128bd89 | 209 | bdev = vdev_bdev_open(path, vdev_bdev_mode(mode), zfs_vdev_holder); |
b5a28807 | 210 | if (IS_ERR(bdev)) |
d1d7e268 | 211 | return (bdev); |
b5a28807 ED |
212 | |
213 | disk = get_gendisk(bdev->bd_dev, &partno); | |
214 | vdev_bdev_close(bdev, vdev_bdev_mode(mode)); | |
215 | ||
216 | if (disk) { | |
217 | bdev = bdget(disk_devt(disk)); | |
218 | if (bdev) { | |
219 | error = blkdev_get(bdev, vdev_bdev_mode(mode), vd); | |
220 | if (error == 0) | |
221 | error = ioctl_by_bdev(bdev, BLKRRPART, 0); | |
222 | vdev_bdev_close(bdev, vdev_bdev_mode(mode)); | |
223 | } | |
224 | ||
225 | bdev = bdget_disk(disk, partno); | |
226 | if (bdev) { | |
227 | error = blkdev_get(bdev, | |
228 | vdev_bdev_mode(mode) | FMODE_EXCL, vd); | |
229 | if (error == 0) | |
230 | result = bdev; | |
231 | } | |
232 | put_disk(disk); | |
233 | } | |
234 | ||
d1d7e268 | 235 | return (result); |
b5a28807 | 236 | #else |
d1d7e268 | 237 | return (ERR_PTR(-EOPNOTSUPP)); |
b5a28807 ED |
238 | #endif /* defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) */ |
239 | } | |
240 | ||
60101509 | 241 | static int |
1bd201e7 CS |
242 | vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize, |
243 | uint64_t *ashift) | |
60101509 | 244 | { |
b5a28807 | 245 | struct block_device *bdev = ERR_PTR(-ENXIO); |
60101509 BB |
246 | vdev_disk_t *vd; |
247 | int mode, block_size; | |
248 | ||
249 | /* Must have a pathname and it must be absolute. */ | |
250 | if (v->vdev_path == NULL || v->vdev_path[0] != '/') { | |
251 | v->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; | |
d1d7e268 | 252 | return (EINVAL); |
60101509 BB |
253 | } |
254 | ||
0d8103d9 BB |
255 | /* |
256 | * Reopen the device if it's not currently open. Otherwise, | |
257 | * just update the physical size of the device. | |
258 | */ | |
259 | if (v->vdev_tsd != NULL) { | |
260 | ASSERT(v->vdev_reopening); | |
261 | vd = v->vdev_tsd; | |
262 | goto skip_open; | |
263 | } | |
264 | ||
d1d7e268 | 265 | vd = kmem_zalloc(sizeof (vdev_disk_t), KM_PUSHPAGE); |
60101509 | 266 | if (vd == NULL) |
d1d7e268 | 267 | return (ENOMEM); |
60101509 BB |
268 | |
269 | /* | |
270 | * Devices are always opened by the path provided at configuration | |
271 | * time. This means that if the provided path is a udev by-id path | |
272 | * then drives may be recabled without an issue. If the provided | |
4e95cc99 | 273 | * path is a udev by-path path, then the physical location information |
60101509 BB |
274 | * will be preserved. This can be critical for more complicated |
275 | * configurations where drives are located in specific physical | |
276 | * locations to maximize the systems tolerence to component failure. | |
4e95cc99 | 277 | * Alternatively, you can provide your own udev rule to flexibly map |
60101509 | 278 | * the drives as you see fit. It is not advised that you use the |
4e95cc99 | 279 | * /dev/[hd]d devices which may be reordered due to probing order. |
60101509 BB |
280 | * Devices in the wrong locations will be detected by the higher |
281 | * level vdev validation. | |
282 | */ | |
283 | mode = spa_mode(v->vdev_spa); | |
b5a28807 ED |
284 | if (v->vdev_wholedisk && v->vdev_expanding) |
285 | bdev = vdev_disk_rrpart(v->vdev_path, mode, vd); | |
286 | if (IS_ERR(bdev)) | |
8128bd89 BB |
287 | bdev = vdev_bdev_open(v->vdev_path, |
288 | vdev_bdev_mode(mode), zfs_vdev_holder); | |
60101509 | 289 | if (IS_ERR(bdev)) { |
d1d7e268 MK |
290 | kmem_free(vd, sizeof (vdev_disk_t)); |
291 | return (-PTR_ERR(bdev)); | |
60101509 BB |
292 | } |
293 | ||
294 | v->vdev_tsd = vd; | |
295 | vd->vd_bdev = bdev; | |
0d8103d9 BB |
296 | |
297 | skip_open: | |
298 | /* Determine the physical block size */ | |
299 | block_size = vdev_bdev_block_size(vd->vd_bdev); | |
60101509 | 300 | |
60101509 BB |
301 | /* Clear the nowritecache bit, causes vdev_reopen() to try again. */ |
302 | v->vdev_nowritecache = B_FALSE; | |
303 | ||
304 | /* Physical volume size in bytes */ | |
0d8103d9 | 305 | *psize = bdev_capacity(vd->vd_bdev); |
60101509 | 306 | |
1bd201e7 CS |
307 | /* TODO: report possible expansion size */ |
308 | *max_psize = *psize; | |
309 | ||
60101509 | 310 | /* Based on the minimum sector size set the block size */ |
9bd274dd | 311 | *ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1; |
60101509 | 312 | |
6839eed2 | 313 | /* Try to set the io scheduler elevator algorithm */ |
fdcd952b | 314 | (void) vdev_elevator_switch(v, zfs_vdev_scheduler); |
6839eed2 | 315 | |
d1d7e268 | 316 | return (0); |
60101509 BB |
317 | } |
318 | ||
319 | static void | |
320 | vdev_disk_close(vdev_t *v) | |
321 | { | |
322 | vdev_disk_t *vd = v->vdev_tsd; | |
323 | ||
0d8103d9 | 324 | if (v->vdev_reopening || vd == NULL) |
60101509 BB |
325 | return; |
326 | ||
327 | if (vd->vd_bdev != NULL) | |
328 | vdev_bdev_close(vd->vd_bdev, | |
d1d7e268 | 329 | vdev_bdev_mode(spa_mode(v->vdev_spa))); |
60101509 | 330 | |
d1d7e268 | 331 | kmem_free(vd, sizeof (vdev_disk_t)); |
60101509 BB |
332 | v->vdev_tsd = NULL; |
333 | } | |
334 | ||
335 | static dio_request_t * | |
336 | vdev_disk_dio_alloc(int bio_count) | |
337 | { | |
338 | dio_request_t *dr; | |
339 | int i; | |
340 | ||
d1d7e268 MK |
341 | dr = kmem_zalloc(sizeof (dio_request_t) + |
342 | sizeof (struct bio *) * bio_count, KM_PUSHPAGE); | |
60101509 BB |
343 | if (dr) { |
344 | init_completion(&dr->dr_comp); | |
345 | atomic_set(&dr->dr_ref, 0); | |
346 | dr->dr_bio_count = bio_count; | |
347 | dr->dr_error = 0; | |
348 | ||
349 | for (i = 0; i < dr->dr_bio_count; i++) | |
350 | dr->dr_bio[i] = NULL; | |
351 | } | |
352 | ||
d1d7e268 | 353 | return (dr); |
60101509 BB |
354 | } |
355 | ||
356 | static void | |
357 | vdev_disk_dio_free(dio_request_t *dr) | |
358 | { | |
359 | int i; | |
360 | ||
361 | for (i = 0; i < dr->dr_bio_count; i++) | |
362 | if (dr->dr_bio[i]) | |
363 | bio_put(dr->dr_bio[i]); | |
364 | ||
d1d7e268 MK |
365 | kmem_free(dr, sizeof (dio_request_t) + |
366 | sizeof (struct bio *) * dr->dr_bio_count); | |
60101509 BB |
367 | } |
368 | ||
675de5aa BB |
369 | static int |
370 | vdev_disk_dio_is_sync(dio_request_t *dr) | |
371 | { | |
372 | #ifdef HAVE_BIO_RW_SYNC | |
373 | /* BIO_RW_SYNC preferred interface from 2.6.12-2.6.29 */ | |
d1d7e268 | 374 | return (dr->dr_rw & (1 << BIO_RW_SYNC)); |
675de5aa | 375 | #else |
d1d7e268 | 376 | #ifdef HAVE_BIO_RW_SYNCIO |
675de5aa | 377 | /* BIO_RW_SYNCIO preferred interface from 2.6.30-2.6.35 */ |
d1d7e268 MK |
378 | return (dr->dr_rw & (1 << BIO_RW_SYNCIO)); |
379 | #else | |
380 | #ifdef HAVE_REQ_SYNC | |
675de5aa | 381 | /* REQ_SYNC preferred interface from 2.6.36-2.6.xx */ |
d1d7e268 MK |
382 | return (dr->dr_rw & REQ_SYNC); |
383 | #else | |
384 | #error "Unable to determine bio sync flag" | |
385 | #endif /* HAVE_REQ_SYNC */ | |
386 | #endif /* HAVE_BIO_RW_SYNC */ | |
675de5aa BB |
387 | #endif /* HAVE_BIO_RW_SYNCIO */ |
388 | } | |
389 | ||
60101509 BB |
390 | static void |
391 | vdev_disk_dio_get(dio_request_t *dr) | |
392 | { | |
393 | atomic_inc(&dr->dr_ref); | |
394 | } | |
395 | ||
396 | static int | |
397 | vdev_disk_dio_put(dio_request_t *dr) | |
398 | { | |
399 | int rc = atomic_dec_return(&dr->dr_ref); | |
400 | ||
401 | /* | |
402 | * Free the dio_request when the last reference is dropped and | |
403 | * ensure zio_interpret is called only once with the correct zio | |
404 | */ | |
405 | if (rc == 0) { | |
406 | zio_t *zio = dr->dr_zio; | |
407 | int error = dr->dr_error; | |
408 | ||
409 | vdev_disk_dio_free(dr); | |
410 | ||
411 | if (zio) { | |
cc92e9d0 | 412 | zio->io_delay = jiffies_64 - zio->io_delay; |
60101509 | 413 | zio->io_error = error; |
d148e951 BB |
414 | ASSERT3S(zio->io_error, >=, 0); |
415 | if (zio->io_error) | |
416 | vdev_disk_error(zio); | |
60101509 BB |
417 | zio_interrupt(zio); |
418 | } | |
419 | } | |
420 | ||
d1d7e268 | 421 | return (rc); |
60101509 BB |
422 | } |
423 | ||
424 | BIO_END_IO_PROTO(vdev_disk_physio_completion, bio, size, error) | |
425 | { | |
426 | dio_request_t *dr = bio->bi_private; | |
427 | int rc; | |
428 | ||
429 | /* Fatal error but print some useful debugging before asserting */ | |
430 | if (dr == NULL) | |
431 | PANIC("dr == NULL, bio->bi_private == NULL\n" | |
432 | "bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d\n" | |
433 | "bi_idx: %d, bi_size: %d, bi_end_io: %p, bi_cnt: %d\n", | |
434 | bio->bi_next, bio->bi_flags, bio->bi_rw, bio->bi_vcnt, | |
d4541210 | 435 | BIO_BI_IDX(bio), BIO_BI_SIZE(bio), bio->bi_end_io, |
60101509 BB |
436 | atomic_read(&bio->bi_cnt)); |
437 | ||
438 | #ifndef HAVE_2ARGS_BIO_END_IO_T | |
d4541210 | 439 | if (BIO_BI_SIZE(bio)) |
d1d7e268 | 440 | return (1); |
60101509 BB |
441 | #endif /* HAVE_2ARGS_BIO_END_IO_T */ |
442 | ||
443 | if (error == 0 && !test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
d1d7e268 | 444 | error = (-EIO); |
60101509 BB |
445 | |
446 | if (dr->dr_error == 0) | |
d148e951 | 447 | dr->dr_error = -error; |
60101509 BB |
448 | |
449 | /* Drop reference aquired by __vdev_disk_physio */ | |
450 | rc = vdev_disk_dio_put(dr); | |
451 | ||
452 | /* Wake up synchronous waiter this is the last outstanding bio */ | |
675de5aa | 453 | if ((rc == 1) && vdev_disk_dio_is_sync(dr)) |
60101509 BB |
454 | complete(&dr->dr_comp); |
455 | ||
456 | BIO_END_IO_RETURN(0); | |
457 | } | |
458 | ||
459 | static inline unsigned long | |
460 | bio_nr_pages(void *bio_ptr, unsigned int bio_size) | |
461 | { | |
462 | return ((((unsigned long)bio_ptr + bio_size + PAGE_SIZE - 1) >> | |
d1d7e268 | 463 | PAGE_SHIFT) - ((unsigned long)bio_ptr >> PAGE_SHIFT)); |
60101509 BB |
464 | } |
465 | ||
466 | static unsigned int | |
467 | bio_map(struct bio *bio, void *bio_ptr, unsigned int bio_size) | |
468 | { | |
469 | unsigned int offset, size, i; | |
470 | struct page *page; | |
471 | ||
472 | offset = offset_in_page(bio_ptr); | |
473 | for (i = 0; i < bio->bi_max_vecs; i++) { | |
474 | size = PAGE_SIZE - offset; | |
475 | ||
476 | if (bio_size <= 0) | |
477 | break; | |
478 | ||
479 | if (size > bio_size) | |
480 | size = bio_size; | |
481 | ||
71f8548e | 482 | if (is_vmalloc_addr(bio_ptr)) |
60101509 BB |
483 | page = vmalloc_to_page(bio_ptr); |
484 | else | |
485 | page = virt_to_page(bio_ptr); | |
486 | ||
17584980 CC |
487 | /* |
488 | * Some network related block device uses tcp_sendpage, which | |
489 | * doesn't behave well when using 0-count page, this is a | |
490 | * safety net to catch them. | |
491 | */ | |
492 | ASSERT3S(page_count(page), >, 0); | |
493 | ||
60101509 BB |
494 | if (bio_add_page(bio, page, size, offset) != size) |
495 | break; | |
496 | ||
497 | bio_ptr += size; | |
498 | bio_size -= size; | |
499 | offset = 0; | |
500 | } | |
501 | ||
d1d7e268 | 502 | return (bio_size); |
60101509 BB |
503 | } |
504 | ||
505 | static int | |
506 | __vdev_disk_physio(struct block_device *bdev, zio_t *zio, caddr_t kbuf_ptr, | |
d1d7e268 | 507 | size_t kbuf_size, uint64_t kbuf_offset, int flags) |
60101509 | 508 | { |
d1d7e268 | 509 | dio_request_t *dr; |
60101509 BB |
510 | caddr_t bio_ptr; |
511 | uint64_t bio_offset; | |
512 | int bio_size, bio_count = 16; | |
f74fae8b | 513 | int i = 0, error = 0; |
60101509 | 514 | |
e06be586 NB |
515 | ASSERT3U(kbuf_offset + kbuf_size, <=, bdev->bd_inode->i_size); |
516 | ||
60101509 BB |
517 | retry: |
518 | dr = vdev_disk_dio_alloc(bio_count); | |
519 | if (dr == NULL) | |
d1d7e268 | 520 | return (ENOMEM); |
60101509 | 521 | |
2959d94a | 522 | if (zio && !(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD))) |
29b763cd | 523 | bio_set_flags_failfast(bdev, &flags); |
2959d94a | 524 | |
60101509 BB |
525 | dr->dr_zio = zio; |
526 | dr->dr_rw = flags; | |
60101509 | 527 | |
60101509 BB |
528 | /* |
529 | * When the IO size exceeds the maximum bio size for the request | |
530 | * queue we are forced to break the IO in multiple bio's and wait | |
531 | * for them all to complete. Ideally, all pool users will set | |
532 | * their volume block size to match the maximum request size and | |
533 | * the common case will be one bio per vdev IO request. | |
534 | */ | |
535 | bio_ptr = kbuf_ptr; | |
536 | bio_offset = kbuf_offset; | |
537 | bio_size = kbuf_size; | |
538 | for (i = 0; i <= dr->dr_bio_count; i++) { | |
539 | ||
540 | /* Finished constructing bio's for given buffer */ | |
541 | if (bio_size <= 0) | |
542 | break; | |
543 | ||
544 | /* | |
545 | * By default only 'bio_count' bio's per dio are allowed. | |
546 | * However, if we find ourselves in a situation where more | |
547 | * are needed we allocate a larger dio and warn the user. | |
548 | */ | |
549 | if (dr->dr_bio_count == i) { | |
550 | vdev_disk_dio_free(dr); | |
551 | bio_count *= 2; | |
60101509 BB |
552 | goto retry; |
553 | } | |
554 | ||
555 | dr->dr_bio[i] = bio_alloc(GFP_NOIO, | |
d1d7e268 | 556 | bio_nr_pages(bio_ptr, bio_size)); |
29b763cd IH |
557 | /* bio_alloc() with __GFP_WAIT never returns NULL */ |
558 | if (unlikely(dr->dr_bio[i] == NULL)) { | |
60101509 | 559 | vdev_disk_dio_free(dr); |
d1d7e268 | 560 | return (ENOMEM); |
60101509 BB |
561 | } |
562 | ||
563 | /* Matching put called by vdev_disk_physio_completion */ | |
564 | vdev_disk_dio_get(dr); | |
565 | ||
566 | dr->dr_bio[i]->bi_bdev = bdev; | |
d4541210 | 567 | BIO_BI_SECTOR(dr->dr_bio[i]) = bio_offset >> 9; |
60101509 BB |
568 | dr->dr_bio[i]->bi_rw = dr->dr_rw; |
569 | dr->dr_bio[i]->bi_end_io = vdev_disk_physio_completion; | |
570 | dr->dr_bio[i]->bi_private = dr; | |
571 | ||
572 | /* Remaining size is returned to become the new size */ | |
573 | bio_size = bio_map(dr->dr_bio[i], bio_ptr, bio_size); | |
574 | ||
575 | /* Advance in buffer and construct another bio if needed */ | |
d4541210 CC |
576 | bio_ptr += BIO_BI_SIZE(dr->dr_bio[i]); |
577 | bio_offset += BIO_BI_SIZE(dr->dr_bio[i]); | |
60101509 BB |
578 | } |
579 | ||
580 | /* Extra reference to protect dio_request during submit_bio */ | |
581 | vdev_disk_dio_get(dr); | |
a69052be BB |
582 | if (zio) |
583 | zio->io_delay = jiffies_64; | |
60101509 BB |
584 | |
585 | /* Submit all bio's associated with this dio */ | |
586 | for (i = 0; i < dr->dr_bio_count; i++) | |
587 | if (dr->dr_bio[i]) | |
588 | submit_bio(dr->dr_rw, dr->dr_bio[i]); | |
589 | ||
590 | /* | |
591 | * On synchronous blocking requests we wait for all bio the completion | |
592 | * callbacks to run. We will be woken when the last callback runs | |
593 | * for this dio. We are responsible for putting the last dio_request | |
594 | * reference will in turn put back the last bio references. The | |
595 | * only synchronous consumer is vdev_disk_read_rootlabel() all other | |
596 | * IO originating from vdev_disk_io_start() is asynchronous. | |
597 | */ | |
675de5aa | 598 | if (vdev_disk_dio_is_sync(dr)) { |
60101509 BB |
599 | wait_for_completion(&dr->dr_comp); |
600 | error = dr->dr_error; | |
601 | ASSERT3S(atomic_read(&dr->dr_ref), ==, 1); | |
602 | } | |
603 | ||
d1d7e268 | 604 | (void) vdev_disk_dio_put(dr); |
60101509 | 605 | |
d1d7e268 | 606 | return (error); |
60101509 BB |
607 | } |
608 | ||
609 | int | |
610 | vdev_disk_physio(struct block_device *bdev, caddr_t kbuf, | |
d1d7e268 | 611 | size_t size, uint64_t offset, int flags) |
60101509 | 612 | { |
2959d94a | 613 | bio_set_flags_failfast(bdev, &flags); |
d1d7e268 | 614 | return (__vdev_disk_physio(bdev, NULL, kbuf, size, offset, flags)); |
60101509 BB |
615 | } |
616 | ||
60101509 BB |
617 | BIO_END_IO_PROTO(vdev_disk_io_flush_completion, bio, size, rc) |
618 | { | |
619 | zio_t *zio = bio->bi_private; | |
620 | ||
cc92e9d0 | 621 | zio->io_delay = jiffies_64 - zio->io_delay; |
60101509 BB |
622 | zio->io_error = -rc; |
623 | if (rc && (rc == -EOPNOTSUPP)) | |
624 | zio->io_vd->vdev_nowritecache = B_TRUE; | |
625 | ||
626 | bio_put(bio); | |
d148e951 BB |
627 | ASSERT3S(zio->io_error, >=, 0); |
628 | if (zio->io_error) | |
629 | vdev_disk_error(zio); | |
60101509 BB |
630 | zio_interrupt(zio); |
631 | ||
632 | BIO_END_IO_RETURN(0); | |
633 | } | |
634 | ||
635 | static int | |
636 | vdev_disk_io_flush(struct block_device *bdev, zio_t *zio) | |
637 | { | |
638 | struct request_queue *q; | |
639 | struct bio *bio; | |
640 | ||
641 | q = bdev_get_queue(bdev); | |
642 | if (!q) | |
d1d7e268 | 643 | return (ENXIO); |
60101509 | 644 | |
abc41ac7 | 645 | bio = bio_alloc(GFP_NOIO, 0); |
29b763cd IH |
646 | /* bio_alloc() with __GFP_WAIT never returns NULL */ |
647 | if (unlikely(bio == NULL)) | |
d1d7e268 | 648 | return (ENOMEM); |
60101509 BB |
649 | |
650 | bio->bi_end_io = vdev_disk_io_flush_completion; | |
651 | bio->bi_private = zio; | |
652 | bio->bi_bdev = bdev; | |
a69052be | 653 | zio->io_delay = jiffies_64; |
96801d29 | 654 | submit_bio(VDEV_WRITE_FLUSH_FUA, bio); |
cecb7487 | 655 | invalidate_bdev(bdev); |
60101509 | 656 | |
d1d7e268 | 657 | return (0); |
60101509 | 658 | } |
60101509 BB |
659 | |
660 | static int | |
661 | vdev_disk_io_start(zio_t *zio) | |
662 | { | |
663 | vdev_t *v = zio->io_vd; | |
664 | vdev_disk_t *vd = v->vdev_tsd; | |
665 | int flags, error; | |
666 | ||
667 | switch (zio->io_type) { | |
668 | case ZIO_TYPE_IOCTL: | |
669 | ||
670 | if (!vdev_readable(v)) { | |
2e528b49 | 671 | zio->io_error = SET_ERROR(ENXIO); |
d1d7e268 | 672 | return (ZIO_PIPELINE_CONTINUE); |
60101509 BB |
673 | } |
674 | ||
675 | switch (zio->io_cmd) { | |
676 | case DKIOCFLUSHWRITECACHE: | |
677 | ||
678 | if (zfs_nocacheflush) | |
679 | break; | |
680 | ||
681 | if (v->vdev_nowritecache) { | |
2e528b49 | 682 | zio->io_error = SET_ERROR(ENOTSUP); |
60101509 BB |
683 | break; |
684 | } | |
685 | ||
686 | error = vdev_disk_io_flush(vd->vd_bdev, zio); | |
687 | if (error == 0) | |
d1d7e268 | 688 | return (ZIO_PIPELINE_STOP); |
60101509 BB |
689 | |
690 | zio->io_error = error; | |
691 | if (error == ENOTSUP) | |
692 | v->vdev_nowritecache = B_TRUE; | |
693 | ||
694 | break; | |
695 | ||
696 | default: | |
2e528b49 | 697 | zio->io_error = SET_ERROR(ENOTSUP); |
60101509 BB |
698 | } |
699 | ||
d1d7e268 | 700 | return (ZIO_PIPELINE_CONTINUE); |
60101509 BB |
701 | |
702 | case ZIO_TYPE_WRITE: | |
703 | flags = WRITE; | |
704 | break; | |
705 | ||
706 | case ZIO_TYPE_READ: | |
707 | flags = READ; | |
708 | break; | |
709 | ||
710 | default: | |
2e528b49 | 711 | zio->io_error = SET_ERROR(ENOTSUP); |
d1d7e268 | 712 | return (ZIO_PIPELINE_CONTINUE); |
60101509 BB |
713 | } |
714 | ||
60101509 | 715 | error = __vdev_disk_physio(vd->vd_bdev, zio, zio->io_data, |
d1d7e268 | 716 | zio->io_size, zio->io_offset, flags); |
60101509 BB |
717 | if (error) { |
718 | zio->io_error = error; | |
d1d7e268 | 719 | return (ZIO_PIPELINE_CONTINUE); |
60101509 BB |
720 | } |
721 | ||
d1d7e268 | 722 | return (ZIO_PIPELINE_STOP); |
60101509 BB |
723 | } |
724 | ||
725 | static void | |
726 | vdev_disk_io_done(zio_t *zio) | |
727 | { | |
728 | /* | |
729 | * If the device returned EIO, we revalidate the media. If it is | |
730 | * determined the media has changed this triggers the asynchronous | |
731 | * removal of the device from the configuration. | |
732 | */ | |
733 | if (zio->io_error == EIO) { | |
d1d7e268 | 734 | vdev_t *v = zio->io_vd; |
60101509 BB |
735 | vdev_disk_t *vd = v->vdev_tsd; |
736 | ||
737 | if (check_disk_change(vd->vd_bdev)) { | |
738 | vdev_bdev_invalidate(vd->vd_bdev); | |
739 | v->vdev_remove_wanted = B_TRUE; | |
740 | spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE); | |
741 | } | |
742 | } | |
743 | } | |
744 | ||
745 | static void | |
746 | vdev_disk_hold(vdev_t *vd) | |
747 | { | |
748 | ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); | |
749 | ||
750 | /* We must have a pathname, and it must be absolute. */ | |
751 | if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') | |
752 | return; | |
753 | ||
754 | /* | |
755 | * Only prefetch path and devid info if the device has | |
756 | * never been opened. | |
757 | */ | |
758 | if (vd->vdev_tsd != NULL) | |
759 | return; | |
760 | ||
761 | /* XXX: Implement me as a vnode lookup for the device */ | |
762 | vd->vdev_name_vp = NULL; | |
763 | vd->vdev_devid_vp = NULL; | |
764 | } | |
765 | ||
766 | static void | |
767 | vdev_disk_rele(vdev_t *vd) | |
768 | { | |
769 | ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); | |
770 | ||
771 | /* XXX: Implement me as a vnode rele for the device */ | |
772 | } | |
773 | ||
774 | vdev_ops_t vdev_disk_ops = { | |
775 | vdev_disk_open, | |
776 | vdev_disk_close, | |
777 | vdev_default_asize, | |
778 | vdev_disk_io_start, | |
779 | vdev_disk_io_done, | |
780 | NULL, | |
781 | vdev_disk_hold, | |
782 | vdev_disk_rele, | |
783 | VDEV_TYPE_DISK, /* name of this vdev type */ | |
784 | B_TRUE /* leaf vdev */ | |
785 | }; | |
786 | ||
787 | /* | |
788 | * Given the root disk device devid or pathname, read the label from | |
789 | * the device, and construct a configuration nvlist. | |
790 | */ | |
791 | int | |
792 | vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config) | |
793 | { | |
794 | struct block_device *bdev; | |
795 | vdev_label_t *label; | |
796 | uint64_t s, size; | |
797 | int i; | |
798 | ||
8128bd89 | 799 | bdev = vdev_bdev_open(devpath, vdev_bdev_mode(FREAD), zfs_vdev_holder); |
60101509 | 800 | if (IS_ERR(bdev)) |
d1d7e268 | 801 | return (-PTR_ERR(bdev)); |
60101509 | 802 | |
f74fae8b | 803 | s = bdev_capacity(bdev); |
60101509 BB |
804 | if (s == 0) { |
805 | vdev_bdev_close(bdev, vdev_bdev_mode(FREAD)); | |
d1d7e268 | 806 | return (EIO); |
60101509 BB |
807 | } |
808 | ||
d1d7e268 MK |
809 | size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t); |
810 | label = vmem_alloc(sizeof (vdev_label_t), KM_PUSHPAGE); | |
60101509 BB |
811 | |
812 | for (i = 0; i < VDEV_LABELS; i++) { | |
d1d7e268 | 813 | uint64_t offset, state, txg = 0; |
60101509 BB |
814 | |
815 | /* read vdev label */ | |
816 | offset = vdev_label_offset(size, i, 0); | |
817 | if (vdev_disk_physio(bdev, (caddr_t)label, | |
818 | VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, READ_SYNC) != 0) | |
819 | continue; | |
820 | ||
821 | if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist, | |
822 | sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) { | |
823 | *config = NULL; | |
824 | continue; | |
825 | } | |
826 | ||
827 | if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE, | |
828 | &state) != 0 || state >= POOL_STATE_DESTROYED) { | |
829 | nvlist_free(*config); | |
830 | *config = NULL; | |
831 | continue; | |
832 | } | |
833 | ||
834 | if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG, | |
835 | &txg) != 0 || txg == 0) { | |
836 | nvlist_free(*config); | |
837 | *config = NULL; | |
838 | continue; | |
839 | } | |
840 | ||
841 | break; | |
842 | } | |
843 | ||
d1d7e268 | 844 | vmem_free(label, sizeof (vdev_label_t)); |
60101509 BB |
845 | vdev_bdev_close(bdev, vdev_bdev_mode(FREAD)); |
846 | ||
d1d7e268 | 847 | return (0); |
60101509 | 848 | } |
6839eed2 BB |
849 | |
850 | module_param(zfs_vdev_scheduler, charp, 0644); | |
c409e464 | 851 | MODULE_PARM_DESC(zfs_vdev_scheduler, "I/O scheduler"); |