]>
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. | |
4a0ee12a | 26 | * Copyright (c) 2012, 2016 by Delphix. All rights reserved. |
60101509 BB |
27 | */ |
28 | ||
29 | #include <sys/zfs_context.h> | |
e771de53 | 30 | #include <sys/spa_impl.h> |
60101509 BB |
31 | #include <sys/vdev_disk.h> |
32 | #include <sys/vdev_impl.h> | |
a6255b7f | 33 | #include <sys/abd.h> |
60101509 BB |
34 | #include <sys/fs/zfs.h> |
35 | #include <sys/zio.h> | |
e771de53 | 36 | #include <linux/mod_compat.h> |
60101509 | 37 | |
6839eed2 | 38 | char *zfs_vdev_scheduler = VDEV_SCHEDULER; |
8128bd89 | 39 | static void *zfs_vdev_holder = VDEV_HOLDER; |
6839eed2 | 40 | |
60101509 BB |
41 | /* |
42 | * Virtual device vector for disks. | |
43 | */ | |
44 | typedef struct dio_request { | |
60101509 | 45 | zio_t *dr_zio; /* Parent ZIO */ |
aa159afb | 46 | atomic_t dr_ref; /* References */ |
60101509 BB |
47 | int dr_error; /* Bio error */ |
48 | int dr_bio_count; /* Count of bio's */ | |
d1d7e268 | 49 | struct bio *dr_bio[0]; /* Attached bio's */ |
60101509 BB |
50 | } dio_request_t; |
51 | ||
52 | ||
53 | #ifdef HAVE_OPEN_BDEV_EXCLUSIVE | |
54 | static fmode_t | |
55 | vdev_bdev_mode(int smode) | |
56 | { | |
57 | fmode_t mode = 0; | |
58 | ||
59 | ASSERT3S(smode & (FREAD | FWRITE), !=, 0); | |
60 | ||
61 | if (smode & FREAD) | |
62 | mode |= FMODE_READ; | |
63 | ||
64 | if (smode & FWRITE) | |
65 | mode |= FMODE_WRITE; | |
66 | ||
d1d7e268 | 67 | return (mode); |
60101509 BB |
68 | } |
69 | #else | |
70 | static int | |
71 | vdev_bdev_mode(int smode) | |
72 | { | |
73 | int mode = 0; | |
74 | ||
75 | ASSERT3S(smode & (FREAD | FWRITE), !=, 0); | |
76 | ||
77 | if ((smode & FREAD) && !(smode & FWRITE)) | |
78 | mode = MS_RDONLY; | |
79 | ||
d1d7e268 | 80 | return (mode); |
60101509 BB |
81 | } |
82 | #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */ | |
83 | ||
84 | static uint64_t | |
85 | bdev_capacity(struct block_device *bdev) | |
86 | { | |
87 | struct hd_struct *part = bdev->bd_part; | |
88 | ||
89 | /* The partition capacity referenced by the block device */ | |
90 | if (part) | |
f74fae8b | 91 | return (part->nr_sects << 9); |
60101509 BB |
92 | |
93 | /* Otherwise assume the full device capacity */ | |
f74fae8b | 94 | return (get_capacity(bdev->bd_disk) << 9); |
60101509 BB |
95 | } |
96 | ||
d148e951 BB |
97 | static void |
98 | vdev_disk_error(zio_t *zio) | |
99 | { | |
100 | #ifdef ZFS_DEBUG | |
7e980733 | 101 | printk(KERN_WARNING "ZFS: zio error=%d type=%d offset=%llu size=%llu " |
193a37cb | 102 | "flags=%x\n", zio->io_error, zio->io_type, |
d148e951 | 103 | (u_longlong_t)zio->io_offset, (u_longlong_t)zio->io_size, |
193a37cb | 104 | zio->io_flags); |
d148e951 BB |
105 | #endif |
106 | } | |
107 | ||
6839eed2 BB |
108 | /* |
109 | * Use the Linux 'noop' elevator for zfs managed block devices. This | |
110 | * strikes the ideal balance by allowing the zfs elevator to do all | |
111 | * request ordering and prioritization. While allowing the Linux | |
112 | * elevator to do the maximum front/back merging allowed by the | |
113 | * physical device. This yields the largest possible requests for | |
114 | * the device with the lowest total overhead. | |
6839eed2 | 115 | */ |
e771de53 | 116 | static void |
fdcd952b | 117 | vdev_elevator_switch(vdev_t *v, char *elevator) |
6839eed2 | 118 | { |
fdcd952b | 119 | vdev_disk_t *vd = v->vdev_tsd; |
e771de53 BB |
120 | struct request_queue *q; |
121 | char *device; | |
e2448b0e | 122 | int error; |
fdcd952b | 123 | |
e771de53 BB |
124 | for (int c = 0; c < v->vdev_children; c++) |
125 | vdev_elevator_switch(v->vdev_child[c], elevator); | |
126 | ||
127 | if (!v->vdev_ops->vdev_op_leaf || vd->vd_bdev == NULL) | |
128 | return; | |
129 | ||
130 | q = bdev_get_queue(vd->vd_bdev); | |
131 | device = vd->vd_bdev->bd_disk->disk_name; | |
132 | ||
84daadde PS |
133 | /* |
134 | * Skip devices which are not whole disks (partitions). | |
135 | * Device-mapper devices are excepted since they may be whole | |
136 | * disks despite the vdev_wholedisk flag, in which case we can | |
137 | * and should switch the elevator. If the device-mapper device | |
138 | * does not have an elevator (i.e. dm-raid, dm-crypt, etc.) the | |
139 | * "Skip devices without schedulers" check below will fail. | |
140 | */ | |
141 | if (!v->vdev_wholedisk && strncmp(device, "dm-", 3) != 0) | |
e771de53 | 142 | return; |
04516a45 | 143 | |
fdcd952b BB |
144 | /* Skip devices without schedulers (loop, ram, dm, etc) */ |
145 | if (!q->elevator || !blk_queue_stackable(q)) | |
e771de53 | 146 | return; |
6839eed2 | 147 | |
fdcd952b | 148 | /* Leave existing scheduler when set to "none" */ |
4903926f | 149 | if ((strncmp(elevator, "none", 4) == 0) && (strlen(elevator) == 4)) |
e771de53 | 150 | return; |
6839eed2 | 151 | |
6d1d976b BB |
152 | #ifdef HAVE_ELEVATOR_CHANGE |
153 | error = elevator_change(q, elevator); | |
154 | #else | |
d1d7e268 MK |
155 | /* |
156 | * For pre-2.6.36 kernels elevator_change() is not available. | |
6d1d976b BB |
157 | * Therefore we fall back to using a usermodehelper to echo the |
158 | * elevator into sysfs; This requires /bin/echo and sysfs to be | |
159 | * mounted which may not be true early in the boot process. | |
160 | */ | |
d1d7e268 | 161 | #define SET_SCHEDULER_CMD \ |
6d1d976b BB |
162 | "exec 0</dev/null " \ |
163 | " 1>/sys/block/%s/queue/scheduler " \ | |
164 | " 2>/dev/null; " \ | |
165 | "echo %s" | |
166 | ||
e771de53 BB |
167 | char *argv[] = { "/bin/sh", "-c", NULL, NULL }; |
168 | char *envp[] = { NULL }; | |
6d1d976b | 169 | |
e771de53 BB |
170 | argv[2] = kmem_asprintf(SET_SCHEDULER_CMD, device, elevator); |
171 | error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); | |
172 | strfree(argv[2]); | |
6d1d976b | 173 | #endif /* HAVE_ELEVATOR_CHANGE */ |
6839eed2 | 174 | if (error) |
7e980733 D |
175 | printk(KERN_NOTICE "ZFS: Unable to set \"%s\" scheduler" |
176 | " for %s (%s): %d\n", elevator, v->vdev_path, device, | |
177 | error); | |
6839eed2 BB |
178 | } |
179 | ||
b5a28807 ED |
180 | /* |
181 | * Expanding a whole disk vdev involves invoking BLKRRPART on the | |
182 | * whole disk device. This poses a problem, because BLKRRPART will | |
183 | * return EBUSY if one of the disk's partitions is open. That's why | |
184 | * we have to do it here, just before opening the data partition. | |
185 | * Unfortunately, BLKRRPART works by dropping all partitions and | |
186 | * recreating them, which means that for a short time window, all | |
187 | * /dev/sdxN device files disappear (until udev recreates them). | |
188 | * This means two things: | |
189 | * - When we open the data partition just after a BLKRRPART, we | |
190 | * can't do it using the normal device file path because of the | |
191 | * obvious race condition with udev. Instead, we use reliable | |
192 | * kernel APIs to get a handle to the new partition device from | |
193 | * the whole disk device. | |
194 | * - Because vdev_disk_open() initially needs to find the device | |
195 | * using its path, multiple vdev_disk_open() invocations in | |
196 | * short succession on the same disk with BLKRRPARTs in the | |
197 | * middle have a high probability of failure (because of the | |
198 | * race condition with udev). A typical situation where this | |
199 | * might happen is when the zpool userspace tool does a | |
200 | * TRYIMPORT immediately followed by an IMPORT. For this | |
201 | * reason, we only invoke BLKRRPART in the module when strictly | |
202 | * necessary (zpool online -e case), and rely on userspace to | |
203 | * do it when possible. | |
204 | */ | |
205 | static struct block_device * | |
206 | vdev_disk_rrpart(const char *path, int mode, vdev_disk_t *vd) | |
207 | { | |
208 | #if defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) | |
209 | struct block_device *bdev, *result = ERR_PTR(-ENXIO); | |
210 | struct gendisk *disk; | |
211 | int error, partno; | |
212 | ||
8128bd89 | 213 | bdev = vdev_bdev_open(path, vdev_bdev_mode(mode), zfs_vdev_holder); |
b5a28807 | 214 | if (IS_ERR(bdev)) |
d1d7e268 | 215 | return (bdev); |
b5a28807 ED |
216 | |
217 | disk = get_gendisk(bdev->bd_dev, &partno); | |
218 | vdev_bdev_close(bdev, vdev_bdev_mode(mode)); | |
219 | ||
220 | if (disk) { | |
221 | bdev = bdget(disk_devt(disk)); | |
222 | if (bdev) { | |
223 | error = blkdev_get(bdev, vdev_bdev_mode(mode), vd); | |
224 | if (error == 0) | |
225 | error = ioctl_by_bdev(bdev, BLKRRPART, 0); | |
226 | vdev_bdev_close(bdev, vdev_bdev_mode(mode)); | |
227 | } | |
228 | ||
229 | bdev = bdget_disk(disk, partno); | |
230 | if (bdev) { | |
231 | error = blkdev_get(bdev, | |
232 | vdev_bdev_mode(mode) | FMODE_EXCL, vd); | |
233 | if (error == 0) | |
234 | result = bdev; | |
235 | } | |
236 | put_disk(disk); | |
237 | } | |
238 | ||
d1d7e268 | 239 | return (result); |
b5a28807 | 240 | #else |
d1d7e268 | 241 | return (ERR_PTR(-EOPNOTSUPP)); |
b5a28807 ED |
242 | #endif /* defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) */ |
243 | } | |
244 | ||
60101509 | 245 | static int |
1bd201e7 CS |
246 | vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize, |
247 | uint64_t *ashift) | |
60101509 | 248 | { |
b5a28807 | 249 | struct block_device *bdev = ERR_PTR(-ENXIO); |
60101509 | 250 | vdev_disk_t *vd; |
2d82ea8b | 251 | int count = 0, mode, block_size; |
60101509 BB |
252 | |
253 | /* Must have a pathname and it must be absolute. */ | |
254 | if (v->vdev_path == NULL || v->vdev_path[0] != '/') { | |
255 | v->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; | |
4a0ee12a PZ |
256 | vdev_dbgmsg(v, "vdev_disk_open: invalid " |
257 | "vdev_path '%s'", v->vdev_path); | |
2d82ea8b | 258 | return (SET_ERROR(EINVAL)); |
60101509 BB |
259 | } |
260 | ||
0d8103d9 BB |
261 | /* |
262 | * Reopen the device if it's not currently open. Otherwise, | |
263 | * just update the physical size of the device. | |
264 | */ | |
265 | if (v->vdev_tsd != NULL) { | |
266 | ASSERT(v->vdev_reopening); | |
267 | vd = v->vdev_tsd; | |
268 | goto skip_open; | |
269 | } | |
270 | ||
79c76d5b | 271 | vd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP); |
60101509 | 272 | if (vd == NULL) |
2d82ea8b | 273 | return (SET_ERROR(ENOMEM)); |
60101509 BB |
274 | |
275 | /* | |
276 | * Devices are always opened by the path provided at configuration | |
277 | * time. This means that if the provided path is a udev by-id path | |
278 | * then drives may be recabled without an issue. If the provided | |
4e95cc99 | 279 | * path is a udev by-path path, then the physical location information |
60101509 BB |
280 | * will be preserved. This can be critical for more complicated |
281 | * configurations where drives are located in specific physical | |
282 | * locations to maximize the systems tolerence to component failure. | |
4e95cc99 | 283 | * Alternatively, you can provide your own udev rule to flexibly map |
60101509 | 284 | * the drives as you see fit. It is not advised that you use the |
4e95cc99 | 285 | * /dev/[hd]d devices which may be reordered due to probing order. |
60101509 BB |
286 | * Devices in the wrong locations will be detected by the higher |
287 | * level vdev validation. | |
2d82ea8b BB |
288 | * |
289 | * The specified paths may be briefly removed and recreated in | |
290 | * response to udev events. This should be exceptionally unlikely | |
291 | * because the zpool command makes every effort to verify these paths | |
292 | * have already settled prior to reaching this point. Therefore, | |
293 | * a ENOENT failure at this point is highly likely to be transient | |
294 | * and it is reasonable to sleep and retry before giving up. In | |
295 | * practice delays have been observed to be on the order of 100ms. | |
60101509 BB |
296 | */ |
297 | mode = spa_mode(v->vdev_spa); | |
b5a28807 ED |
298 | if (v->vdev_wholedisk && v->vdev_expanding) |
299 | bdev = vdev_disk_rrpart(v->vdev_path, mode, vd); | |
2d82ea8b BB |
300 | |
301 | while (IS_ERR(bdev) && count < 50) { | |
8128bd89 BB |
302 | bdev = vdev_bdev_open(v->vdev_path, |
303 | vdev_bdev_mode(mode), zfs_vdev_holder); | |
2d82ea8b BB |
304 | if (unlikely(PTR_ERR(bdev) == -ENOENT)) { |
305 | msleep(10); | |
306 | count++; | |
307 | } else if (IS_ERR(bdev)) { | |
308 | break; | |
309 | } | |
310 | } | |
311 | ||
60101509 | 312 | if (IS_ERR(bdev)) { |
2d82ea8b BB |
313 | dprintf("failed open v->vdev_path=%s, error=%d count=%d\n", |
314 | v->vdev_path, -PTR_ERR(bdev), count); | |
d1d7e268 | 315 | kmem_free(vd, sizeof (vdev_disk_t)); |
2d82ea8b | 316 | return (SET_ERROR(-PTR_ERR(bdev))); |
60101509 BB |
317 | } |
318 | ||
319 | v->vdev_tsd = vd; | |
320 | vd->vd_bdev = bdev; | |
0d8103d9 BB |
321 | |
322 | skip_open: | |
323 | /* Determine the physical block size */ | |
324 | block_size = vdev_bdev_block_size(vd->vd_bdev); | |
60101509 | 325 | |
60101509 BB |
326 | /* Clear the nowritecache bit, causes vdev_reopen() to try again. */ |
327 | v->vdev_nowritecache = B_FALSE; | |
328 | ||
fb40095f RY |
329 | /* Inform the ZIO pipeline that we are non-rotational */ |
330 | v->vdev_nonrot = blk_queue_nonrot(bdev_get_queue(vd->vd_bdev)); | |
331 | ||
60101509 | 332 | /* Physical volume size in bytes */ |
0d8103d9 | 333 | *psize = bdev_capacity(vd->vd_bdev); |
60101509 | 334 | |
1bd201e7 CS |
335 | /* TODO: report possible expansion size */ |
336 | *max_psize = *psize; | |
337 | ||
60101509 | 338 | /* Based on the minimum sector size set the block size */ |
9bd274dd | 339 | *ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1; |
60101509 | 340 | |
6839eed2 | 341 | /* Try to set the io scheduler elevator algorithm */ |
fdcd952b | 342 | (void) vdev_elevator_switch(v, zfs_vdev_scheduler); |
6839eed2 | 343 | |
d1d7e268 | 344 | return (0); |
60101509 BB |
345 | } |
346 | ||
347 | static void | |
348 | vdev_disk_close(vdev_t *v) | |
349 | { | |
350 | vdev_disk_t *vd = v->vdev_tsd; | |
351 | ||
0d8103d9 | 352 | if (v->vdev_reopening || vd == NULL) |
60101509 BB |
353 | return; |
354 | ||
355 | if (vd->vd_bdev != NULL) | |
356 | vdev_bdev_close(vd->vd_bdev, | |
d1d7e268 | 357 | vdev_bdev_mode(spa_mode(v->vdev_spa))); |
60101509 | 358 | |
d1d7e268 | 359 | kmem_free(vd, sizeof (vdev_disk_t)); |
60101509 BB |
360 | v->vdev_tsd = NULL; |
361 | } | |
362 | ||
363 | static dio_request_t * | |
364 | vdev_disk_dio_alloc(int bio_count) | |
365 | { | |
366 | dio_request_t *dr; | |
367 | int i; | |
368 | ||
d1d7e268 | 369 | dr = kmem_zalloc(sizeof (dio_request_t) + |
79c76d5b | 370 | sizeof (struct bio *) * bio_count, KM_SLEEP); |
60101509 | 371 | if (dr) { |
60101509 BB |
372 | atomic_set(&dr->dr_ref, 0); |
373 | dr->dr_bio_count = bio_count; | |
374 | dr->dr_error = 0; | |
375 | ||
376 | for (i = 0; i < dr->dr_bio_count; i++) | |
377 | dr->dr_bio[i] = NULL; | |
378 | } | |
379 | ||
d1d7e268 | 380 | return (dr); |
60101509 BB |
381 | } |
382 | ||
383 | static void | |
384 | vdev_disk_dio_free(dio_request_t *dr) | |
385 | { | |
386 | int i; | |
387 | ||
388 | for (i = 0; i < dr->dr_bio_count; i++) | |
389 | if (dr->dr_bio[i]) | |
390 | bio_put(dr->dr_bio[i]); | |
391 | ||
d1d7e268 MK |
392 | kmem_free(dr, sizeof (dio_request_t) + |
393 | sizeof (struct bio *) * dr->dr_bio_count); | |
60101509 BB |
394 | } |
395 | ||
396 | static void | |
397 | vdev_disk_dio_get(dio_request_t *dr) | |
398 | { | |
399 | atomic_inc(&dr->dr_ref); | |
400 | } | |
401 | ||
402 | static int | |
403 | vdev_disk_dio_put(dio_request_t *dr) | |
404 | { | |
405 | int rc = atomic_dec_return(&dr->dr_ref); | |
406 | ||
407 | /* | |
408 | * Free the dio_request when the last reference is dropped and | |
409 | * ensure zio_interpret is called only once with the correct zio | |
410 | */ | |
411 | if (rc == 0) { | |
412 | zio_t *zio = dr->dr_zio; | |
413 | int error = dr->dr_error; | |
414 | ||
415 | vdev_disk_dio_free(dr); | |
416 | ||
417 | if (zio) { | |
418 | zio->io_error = error; | |
d148e951 BB |
419 | ASSERT3S(zio->io_error, >=, 0); |
420 | if (zio->io_error) | |
421 | vdev_disk_error(zio); | |
a6255b7f | 422 | |
26ef0cc7 | 423 | zio_delay_interrupt(zio); |
60101509 BB |
424 | } |
425 | } | |
426 | ||
d1d7e268 | 427 | return (rc); |
60101509 BB |
428 | } |
429 | ||
784a7fe5 | 430 | BIO_END_IO_PROTO(vdev_disk_physio_completion, bio, error) |
60101509 BB |
431 | { |
432 | dio_request_t *dr = bio->bi_private; | |
433 | int rc; | |
434 | ||
784a7fe5 LW |
435 | if (dr->dr_error == 0) { |
436 | #ifdef HAVE_1ARG_BIO_END_IO_T | |
36ba27e9 | 437 | dr->dr_error = BIO_END_IO_ERROR(bio); |
784a7fe5 LW |
438 | #else |
439 | if (error) | |
440 | dr->dr_error = -(error); | |
441 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
442 | dr->dr_error = EIO; | |
443 | #endif | |
444 | } | |
60101509 | 445 | |
b0be93e8 | 446 | /* Drop reference acquired by __vdev_disk_physio */ |
60101509 | 447 | rc = vdev_disk_dio_put(dr); |
60101509 BB |
448 | } |
449 | ||
60101509 BB |
450 | static unsigned int |
451 | bio_map(struct bio *bio, void *bio_ptr, unsigned int bio_size) | |
452 | { | |
453 | unsigned int offset, size, i; | |
454 | struct page *page; | |
455 | ||
456 | offset = offset_in_page(bio_ptr); | |
457 | for (i = 0; i < bio->bi_max_vecs; i++) { | |
458 | size = PAGE_SIZE - offset; | |
459 | ||
460 | if (bio_size <= 0) | |
461 | break; | |
462 | ||
463 | if (size > bio_size) | |
464 | size = bio_size; | |
465 | ||
71f8548e | 466 | if (is_vmalloc_addr(bio_ptr)) |
60101509 BB |
467 | page = vmalloc_to_page(bio_ptr); |
468 | else | |
469 | page = virt_to_page(bio_ptr); | |
470 | ||
17584980 CC |
471 | /* |
472 | * Some network related block device uses tcp_sendpage, which | |
473 | * doesn't behave well when using 0-count page, this is a | |
474 | * safety net to catch them. | |
475 | */ | |
476 | ASSERT3S(page_count(page), >, 0); | |
477 | ||
60101509 BB |
478 | if (bio_add_page(bio, page, size, offset) != size) |
479 | break; | |
480 | ||
481 | bio_ptr += size; | |
482 | bio_size -= size; | |
483 | offset = 0; | |
484 | } | |
485 | ||
d1d7e268 | 486 | return (bio_size); |
60101509 BB |
487 | } |
488 | ||
b0be93e8 IH |
489 | static unsigned int |
490 | bio_map_abd_off(struct bio *bio, abd_t *abd, unsigned int size, size_t off) | |
491 | { | |
492 | if (abd_is_linear(abd)) | |
493 | return (bio_map(bio, ((char *)abd_to_buf(abd)) + off, size)); | |
494 | ||
495 | return (abd_scatter_bio_map_off(bio, abd, size, off)); | |
496 | } | |
497 | ||
bbb1b6ce | 498 | static inline void |
3b86aeb2 | 499 | vdev_submit_bio_impl(struct bio *bio) |
bbb1b6ce BB |
500 | { |
501 | #ifdef HAVE_1ARG_SUBMIT_BIO | |
bbb1b6ce BB |
502 | submit_bio(bio); |
503 | #else | |
3b86aeb2 | 504 | submit_bio(0, bio); |
bbb1b6ce BB |
505 | #endif |
506 | } | |
507 | ||
787acae0 GDN |
508 | #ifndef HAVE_BIO_SET_DEV |
509 | static inline void | |
510 | bio_set_dev(struct bio *bio, struct block_device *bdev) | |
511 | { | |
512 | bio->bi_bdev = bdev; | |
513 | } | |
514 | #endif /* !HAVE_BIO_SET_DEV */ | |
515 | ||
37f9dac5 | 516 | static inline void |
3b86aeb2 | 517 | vdev_submit_bio(struct bio *bio) |
37f9dac5 RY |
518 | { |
519 | #ifdef HAVE_CURRENT_BIO_TAIL | |
520 | struct bio **bio_tail = current->bio_tail; | |
521 | current->bio_tail = NULL; | |
3b86aeb2 | 522 | vdev_submit_bio_impl(bio); |
37f9dac5 RY |
523 | current->bio_tail = bio_tail; |
524 | #else | |
525 | struct bio_list *bio_list = current->bio_list; | |
526 | current->bio_list = NULL; | |
3b86aeb2 | 527 | vdev_submit_bio_impl(bio); |
37f9dac5 RY |
528 | current->bio_list = bio_list; |
529 | #endif | |
530 | } | |
531 | ||
60101509 | 532 | static int |
b0be93e8 IH |
533 | __vdev_disk_physio(struct block_device *bdev, zio_t *zio, |
534 | size_t io_size, uint64_t io_offset, int rw, int flags) | |
60101509 | 535 | { |
d1d7e268 | 536 | dio_request_t *dr; |
b0be93e8 | 537 | uint64_t abd_offset; |
60101509 | 538 | uint64_t bio_offset; |
3b86aeb2 | 539 | int bio_size, bio_count = 16; |
f74fae8b | 540 | int i = 0, error = 0; |
e8ac4557 IH |
541 | #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG) |
542 | struct blk_plug plug; | |
543 | #endif | |
60101509 | 544 | |
b0be93e8 IH |
545 | ASSERT(zio != NULL); |
546 | ASSERT3U(io_offset + io_size, <=, bdev->bd_inode->i_size); | |
e06be586 | 547 | |
60101509 BB |
548 | retry: |
549 | dr = vdev_disk_dio_alloc(bio_count); | |
550 | if (dr == NULL) | |
ecb2b7dc | 551 | return (SET_ERROR(ENOMEM)); |
60101509 | 552 | |
2959d94a | 553 | if (zio && !(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD))) |
29b763cd | 554 | bio_set_flags_failfast(bdev, &flags); |
2959d94a | 555 | |
60101509 | 556 | dr->dr_zio = zio; |
60101509 | 557 | |
60101509 BB |
558 | /* |
559 | * When the IO size exceeds the maximum bio size for the request | |
560 | * queue we are forced to break the IO in multiple bio's and wait | |
561 | * for them all to complete. Ideally, all pool users will set | |
562 | * their volume block size to match the maximum request size and | |
563 | * the common case will be one bio per vdev IO request. | |
564 | */ | |
a6255b7f | 565 | |
b0be93e8 IH |
566 | abd_offset = 0; |
567 | bio_offset = io_offset; | |
568 | bio_size = io_size; | |
60101509 BB |
569 | for (i = 0; i <= dr->dr_bio_count; i++) { |
570 | ||
571 | /* Finished constructing bio's for given buffer */ | |
572 | if (bio_size <= 0) | |
573 | break; | |
574 | ||
575 | /* | |
576 | * By default only 'bio_count' bio's per dio are allowed. | |
577 | * However, if we find ourselves in a situation where more | |
578 | * are needed we allocate a larger dio and warn the user. | |
579 | */ | |
580 | if (dr->dr_bio_count == i) { | |
581 | vdev_disk_dio_free(dr); | |
582 | bio_count *= 2; | |
60101509 BB |
583 | goto retry; |
584 | } | |
585 | ||
29b763cd | 586 | /* bio_alloc() with __GFP_WAIT never returns NULL */ |
f1512ee6 | 587 | dr->dr_bio[i] = bio_alloc(GFP_NOIO, |
b0be93e8 | 588 | MIN(abd_nr_pages_off(zio->io_abd, bio_size, abd_offset), |
02730c33 | 589 | BIO_MAX_PAGES)); |
29b763cd | 590 | if (unlikely(dr->dr_bio[i] == NULL)) { |
60101509 | 591 | vdev_disk_dio_free(dr); |
ecb2b7dc | 592 | return (SET_ERROR(ENOMEM)); |
60101509 BB |
593 | } |
594 | ||
595 | /* Matching put called by vdev_disk_physio_completion */ | |
596 | vdev_disk_dio_get(dr); | |
597 | ||
787acae0 | 598 | bio_set_dev(dr->dr_bio[i], bdev); |
d4541210 | 599 | BIO_BI_SECTOR(dr->dr_bio[i]) = bio_offset >> 9; |
60101509 BB |
600 | dr->dr_bio[i]->bi_end_io = vdev_disk_physio_completion; |
601 | dr->dr_bio[i]->bi_private = dr; | |
3b86aeb2 | 602 | bio_set_op_attrs(dr->dr_bio[i], rw, flags); |
60101509 BB |
603 | |
604 | /* Remaining size is returned to become the new size */ | |
b0be93e8 | 605 | bio_size = bio_map_abd_off(dr->dr_bio[i], zio->io_abd, |
02730c33 | 606 | bio_size, abd_offset); |
60101509 BB |
607 | |
608 | /* Advance in buffer and construct another bio if needed */ | |
b0be93e8 | 609 | abd_offset += BIO_BI_SIZE(dr->dr_bio[i]); |
d4541210 | 610 | bio_offset += BIO_BI_SIZE(dr->dr_bio[i]); |
60101509 BB |
611 | } |
612 | ||
37f9dac5 | 613 | /* Extra reference to protect dio_request during vdev_submit_bio */ |
60101509 BB |
614 | vdev_disk_dio_get(dr); |
615 | ||
e8ac4557 IH |
616 | #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG) |
617 | if (dr->dr_bio_count > 1) | |
618 | blk_start_plug(&plug); | |
619 | #endif | |
620 | ||
60101509 BB |
621 | /* Submit all bio's associated with this dio */ |
622 | for (i = 0; i < dr->dr_bio_count; i++) | |
623 | if (dr->dr_bio[i]) | |
3b86aeb2 | 624 | vdev_submit_bio(dr->dr_bio[i]); |
60101509 | 625 | |
e8ac4557 IH |
626 | #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG) |
627 | if (dr->dr_bio_count > 1) | |
628 | blk_finish_plug(&plug); | |
629 | #endif | |
630 | ||
d1d7e268 | 631 | (void) vdev_disk_dio_put(dr); |
60101509 | 632 | |
d1d7e268 | 633 | return (error); |
60101509 BB |
634 | } |
635 | ||
36ba27e9 | 636 | BIO_END_IO_PROTO(vdev_disk_io_flush_completion, bio, error) |
60101509 BB |
637 | { |
638 | zio_t *zio = bio->bi_private; | |
784a7fe5 | 639 | #ifdef HAVE_1ARG_BIO_END_IO_T |
36ba27e9 BB |
640 | zio->io_error = BIO_END_IO_ERROR(bio); |
641 | #else | |
642 | zio->io_error = -error; | |
784a7fe5 | 643 | #endif |
60101509 | 644 | |
36ba27e9 | 645 | if (zio->io_error && (zio->io_error == EOPNOTSUPP)) |
60101509 BB |
646 | zio->io_vd->vdev_nowritecache = B_TRUE; |
647 | ||
648 | bio_put(bio); | |
d148e951 BB |
649 | ASSERT3S(zio->io_error, >=, 0); |
650 | if (zio->io_error) | |
651 | vdev_disk_error(zio); | |
60101509 | 652 | zio_interrupt(zio); |
60101509 BB |
653 | } |
654 | ||
655 | static int | |
656 | vdev_disk_io_flush(struct block_device *bdev, zio_t *zio) | |
657 | { | |
658 | struct request_queue *q; | |
659 | struct bio *bio; | |
660 | ||
661 | q = bdev_get_queue(bdev); | |
662 | if (!q) | |
ecb2b7dc | 663 | return (SET_ERROR(ENXIO)); |
60101509 | 664 | |
abc41ac7 | 665 | bio = bio_alloc(GFP_NOIO, 0); |
29b763cd IH |
666 | /* bio_alloc() with __GFP_WAIT never returns NULL */ |
667 | if (unlikely(bio == NULL)) | |
ecb2b7dc | 668 | return (SET_ERROR(ENOMEM)); |
60101509 BB |
669 | |
670 | bio->bi_end_io = vdev_disk_io_flush_completion; | |
671 | bio->bi_private = zio; | |
787acae0 | 672 | bio_set_dev(bio, bdev); |
a5e046ea | 673 | bio_set_flush(bio); |
3b86aeb2 | 674 | vdev_submit_bio(bio); |
cecb7487 | 675 | invalidate_bdev(bdev); |
60101509 | 676 | |
d1d7e268 | 677 | return (0); |
60101509 | 678 | } |
60101509 | 679 | |
98b25418 | 680 | static void |
60101509 BB |
681 | vdev_disk_io_start(zio_t *zio) |
682 | { | |
683 | vdev_t *v = zio->io_vd; | |
684 | vdev_disk_t *vd = v->vdev_tsd; | |
3b86aeb2 | 685 | int rw, flags, error; |
60101509 BB |
686 | |
687 | switch (zio->io_type) { | |
688 | case ZIO_TYPE_IOCTL: | |
689 | ||
690 | if (!vdev_readable(v)) { | |
2e528b49 | 691 | zio->io_error = SET_ERROR(ENXIO); |
98b25418 GW |
692 | zio_interrupt(zio); |
693 | return; | |
60101509 BB |
694 | } |
695 | ||
696 | switch (zio->io_cmd) { | |
697 | case DKIOCFLUSHWRITECACHE: | |
698 | ||
699 | if (zfs_nocacheflush) | |
700 | break; | |
701 | ||
702 | if (v->vdev_nowritecache) { | |
2e528b49 | 703 | zio->io_error = SET_ERROR(ENOTSUP); |
60101509 BB |
704 | break; |
705 | } | |
706 | ||
707 | error = vdev_disk_io_flush(vd->vd_bdev, zio); | |
708 | if (error == 0) | |
98b25418 | 709 | return; |
60101509 BB |
710 | |
711 | zio->io_error = error; | |
60101509 BB |
712 | |
713 | break; | |
714 | ||
715 | default: | |
2e528b49 | 716 | zio->io_error = SET_ERROR(ENOTSUP); |
60101509 BB |
717 | } |
718 | ||
98b25418 GW |
719 | zio_execute(zio); |
720 | return; | |
60101509 | 721 | case ZIO_TYPE_WRITE: |
3b86aeb2 | 722 | rw = WRITE; |
e6603b7c | 723 | #if defined(HAVE_BLK_QUEUE_HAVE_BIO_RW_UNPLUG) |
3b86aeb2 | 724 | flags = (1 << BIO_RW_UNPLUG); |
e6603b7c | 725 | #elif defined(REQ_UNPLUG) |
3b86aeb2 | 726 | flags = REQ_UNPLUG; |
e6603b7c | 727 | #else |
3b86aeb2 | 728 | flags = 0; |
e6603b7c | 729 | #endif |
60101509 BB |
730 | break; |
731 | ||
732 | case ZIO_TYPE_READ: | |
3b86aeb2 | 733 | rw = READ; |
e6603b7c | 734 | #if defined(HAVE_BLK_QUEUE_HAVE_BIO_RW_UNPLUG) |
3b86aeb2 | 735 | flags = (1 << BIO_RW_UNPLUG); |
e6603b7c | 736 | #elif defined(REQ_UNPLUG) |
3b86aeb2 | 737 | flags = REQ_UNPLUG; |
e6603b7c | 738 | #else |
3b86aeb2 | 739 | flags = 0; |
e6603b7c | 740 | #endif |
60101509 BB |
741 | break; |
742 | ||
743 | default: | |
2e528b49 | 744 | zio->io_error = SET_ERROR(ENOTSUP); |
98b25418 GW |
745 | zio_interrupt(zio); |
746 | return; | |
60101509 BB |
747 | } |
748 | ||
26ef0cc7 | 749 | zio->io_target_timestamp = zio_handle_io_delay(zio); |
b0be93e8 | 750 | error = __vdev_disk_physio(vd->vd_bdev, zio, |
3b86aeb2 | 751 | zio->io_size, zio->io_offset, rw, flags); |
60101509 BB |
752 | if (error) { |
753 | zio->io_error = error; | |
98b25418 GW |
754 | zio_interrupt(zio); |
755 | return; | |
60101509 | 756 | } |
60101509 BB |
757 | } |
758 | ||
759 | static void | |
760 | vdev_disk_io_done(zio_t *zio) | |
761 | { | |
762 | /* | |
763 | * If the device returned EIO, we revalidate the media. If it is | |
764 | * determined the media has changed this triggers the asynchronous | |
765 | * removal of the device from the configuration. | |
766 | */ | |
767 | if (zio->io_error == EIO) { | |
d1d7e268 | 768 | vdev_t *v = zio->io_vd; |
60101509 BB |
769 | vdev_disk_t *vd = v->vdev_tsd; |
770 | ||
771 | if (check_disk_change(vd->vd_bdev)) { | |
772 | vdev_bdev_invalidate(vd->vd_bdev); | |
773 | v->vdev_remove_wanted = B_TRUE; | |
774 | spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE); | |
775 | } | |
776 | } | |
777 | } | |
778 | ||
779 | static void | |
780 | vdev_disk_hold(vdev_t *vd) | |
781 | { | |
782 | ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); | |
783 | ||
784 | /* We must have a pathname, and it must be absolute. */ | |
785 | if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') | |
786 | return; | |
787 | ||
788 | /* | |
789 | * Only prefetch path and devid info if the device has | |
790 | * never been opened. | |
791 | */ | |
792 | if (vd->vdev_tsd != NULL) | |
793 | return; | |
794 | ||
795 | /* XXX: Implement me as a vnode lookup for the device */ | |
796 | vd->vdev_name_vp = NULL; | |
797 | vd->vdev_devid_vp = NULL; | |
798 | } | |
799 | ||
800 | static void | |
801 | vdev_disk_rele(vdev_t *vd) | |
802 | { | |
803 | ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); | |
804 | ||
805 | /* XXX: Implement me as a vnode rele for the device */ | |
806 | } | |
807 | ||
e771de53 BB |
808 | static int |
809 | param_set_vdev_scheduler(const char *val, zfs_kernel_param_t *kp) | |
810 | { | |
811 | spa_t *spa = NULL; | |
812 | char *p; | |
813 | ||
814 | if (val == NULL) | |
815 | return (SET_ERROR(-EINVAL)); | |
816 | ||
817 | if ((p = strchr(val, '\n')) != NULL) | |
818 | *p = '\0'; | |
819 | ||
bc5f51c5 | 820 | if (spa_mode_global != 0) { |
e771de53 | 821 | mutex_enter(&spa_namespace_lock); |
bc5f51c5 OF |
822 | while ((spa = spa_next(spa)) != NULL) { |
823 | if (spa_state(spa) != POOL_STATE_ACTIVE || | |
824 | !spa_writeable(spa) || spa_suspended(spa)) | |
825 | continue; | |
826 | ||
827 | spa_open_ref(spa, FTAG); | |
828 | mutex_exit(&spa_namespace_lock); | |
829 | vdev_elevator_switch(spa->spa_root_vdev, (char *)val); | |
830 | mutex_enter(&spa_namespace_lock); | |
831 | spa_close(spa, FTAG); | |
832 | } | |
833 | mutex_exit(&spa_namespace_lock); | |
e771de53 | 834 | } |
e771de53 BB |
835 | |
836 | return (param_set_charp(val, kp)); | |
837 | } | |
838 | ||
60101509 BB |
839 | vdev_ops_t vdev_disk_ops = { |
840 | vdev_disk_open, | |
841 | vdev_disk_close, | |
842 | vdev_default_asize, | |
843 | vdev_disk_io_start, | |
844 | vdev_disk_io_done, | |
845 | NULL, | |
3d6da72d | 846 | NULL, |
60101509 BB |
847 | vdev_disk_hold, |
848 | vdev_disk_rele, | |
a1d477c2 | 849 | NULL, |
60101509 BB |
850 | VDEV_TYPE_DISK, /* name of this vdev type */ |
851 | B_TRUE /* leaf vdev */ | |
852 | }; | |
853 | ||
e771de53 BB |
854 | module_param_call(zfs_vdev_scheduler, param_set_vdev_scheduler, |
855 | param_get_charp, &zfs_vdev_scheduler, 0644); | |
c409e464 | 856 | MODULE_PARM_DESC(zfs_vdev_scheduler, "I/O scheduler"); |