projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Handle data block end_io through the async work queue
[linux-2.6-block.git] / fs / btrfs / volumes.c
CommitLineData
0b86a832
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/bio.h>
8a4b83cc 20#include <linux/buffer_head.h>
593060d7 21#include <asm/div64.h>
0b86a832
CM		 22#include "ctree.h"
23#include "extent_map.h"
24#include "disk-io.h"
25#include "transaction.h"
26#include "print-tree.h"
27#include "volumes.h"
28
593060d7
CM		 29struct map_lookup {
30 u64 type;
31 int io_align;
32 int io_width;
33 int stripe_len;
34 int sector_size;
35 int num_stripes;
cea9e445 36 struct btrfs_bio_stripe stripes[];
593060d7
CM		 37};
38
39#define map_lookup_size(n) (sizeof(struct map_lookup) + \
cea9e445 40 (sizeof(struct btrfs_bio_stripe) * (n)))
593060d7 41
8a4b83cc
CM		 42static DEFINE_MUTEX(uuid_mutex);
43static LIST_HEAD(fs_uuids);
44
45int btrfs_cleanup_fs_uuids(void)
46{
47 struct btrfs_fs_devices *fs_devices;
48 struct list_head *uuid_cur;
49 struct list_head *devices_cur;
50 struct btrfs_device *dev;
51
52 list_for_each(uuid_cur, &fs_uuids) {
53 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
54 list);
55 while(!list_empty(&fs_devices->devices)) {
56 devices_cur = fs_devices->devices.next;
57 dev = list_entry(devices_cur, struct btrfs_device,
58 dev_list);
59 printk("uuid cleanup finds %s\n", dev->name);
60 if (dev->bdev) {
61 printk("closing\n");
62 close_bdev_excl(dev->bdev);
63 }
64 list_del(&dev->dev_list);
65 kfree(dev);
66 }
67 }
68 return 0;
69}
70
71static struct btrfs_device *__find_device(struct list_head *head, u64 devid)
72{
73 struct btrfs_device *dev;
74 struct list_head *cur;
75
76 list_for_each(cur, head) {
77 dev = list_entry(cur, struct btrfs_device, dev_list);
78 if (dev->devid == devid)
79 return dev;
80 }
81 return NULL;
82}
83
84static struct btrfs_fs_devices *find_fsid(u8 *fsid)
85{
86 struct list_head *cur;
87 struct btrfs_fs_devices *fs_devices;
88
89 list_for_each(cur, &fs_uuids) {
90 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
91 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
92 return fs_devices;
93 }
94 return NULL;
95}
96
97static int device_list_add(const char *path,
98 struct btrfs_super_block *disk_super,
99 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
100{
101 struct btrfs_device *device;
102 struct btrfs_fs_devices *fs_devices;
103 u64 found_transid = btrfs_super_generation(disk_super);
104
105 fs_devices = find_fsid(disk_super->fsid);
106 if (!fs_devices) {
107 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
108 if (!fs_devices)
109 return -ENOMEM;
110 INIT_LIST_HEAD(&fs_devices->devices);
111 list_add(&fs_devices->list, &fs_uuids);
112 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
113 fs_devices->latest_devid = devid;
114 fs_devices->latest_trans = found_transid;
115 fs_devices->lowest_devid = (u64)-1;
116 fs_devices->num_devices = 0;
117 device = NULL;
118 } else {
119 device = __find_device(&fs_devices->devices, devid);
120 }
121 if (!device) {
122 device = kzalloc(sizeof(*device), GFP_NOFS);
123 if (!device) {
124 /* we can safely leave the fs_devices entry around */
125 return -ENOMEM;
126 }
127 device->devid = devid;
128 device->name = kstrdup(path, GFP_NOFS);
129 if (!device->name) {
130 kfree(device);
131 return -ENOMEM;
132 }
133 list_add(&device->dev_list, &fs_devices->devices);
134 fs_devices->num_devices++;
135 }
136
137 if (found_transid > fs_devices->latest_trans) {
138 fs_devices->latest_devid = devid;
139 fs_devices->latest_trans = found_transid;
140 }
141 if (fs_devices->lowest_devid > devid) {
142 fs_devices->lowest_devid = devid;
143 printk("lowest devid now %Lu\n", devid);
144 }
145 *fs_devices_ret = fs_devices;
146 return 0;
147}
148
149int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
150{
151 struct list_head *head = &fs_devices->devices;
152 struct list_head *cur;
153 struct btrfs_device *device;
154
155 mutex_lock(&uuid_mutex);
156 list_for_each(cur, head) {
157 device = list_entry(cur, struct btrfs_device, dev_list);
158 if (device->bdev) {
159 close_bdev_excl(device->bdev);
160 printk("close devices closes %s\n", device->name);
161 }
162 device->bdev = NULL;
163 }
164 mutex_unlock(&uuid_mutex);
165 return 0;
166}
167
168int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
169 int flags, void *holder)
170{
171 struct block_device *bdev;
172 struct list_head *head = &fs_devices->devices;
173 struct list_head *cur;
174 struct btrfs_device *device;
175 int ret;
176
177 mutex_lock(&uuid_mutex);
178 list_for_each(cur, head) {
179 device = list_entry(cur, struct btrfs_device, dev_list);
180 bdev = open_bdev_excl(device->name, flags, holder);
181printk("opening %s devid %Lu\n", device->name, device->devid);
182 if (IS_ERR(bdev)) {
183 printk("open %s failed\n", device->name);
184 ret = PTR_ERR(bdev);
185 goto fail;
186 }
187 if (device->devid == fs_devices->latest_devid)
188 fs_devices->latest_bdev = bdev;
189 if (device->devid == fs_devices->lowest_devid) {
190 fs_devices->lowest_bdev = bdev;
191printk("lowest bdev %s\n", device->name);
192 }
193 device->bdev = bdev;
194 }
195 mutex_unlock(&uuid_mutex);
196 return 0;
197fail:
198 mutex_unlock(&uuid_mutex);
199 btrfs_close_devices(fs_devices);
200 return ret;
201}
202
203int btrfs_scan_one_device(const char *path, int flags, void *holder,
204 struct btrfs_fs_devices **fs_devices_ret)
205{
206 struct btrfs_super_block *disk_super;
207 struct block_device *bdev;
208 struct buffer_head *bh;
209 int ret;
210 u64 devid;
211
212 mutex_lock(&uuid_mutex);
213
214 printk("scan one opens %s\n", path);
215 bdev = open_bdev_excl(path, flags, holder);
216
217 if (IS_ERR(bdev)) {
218 printk("open failed\n");
219 ret = PTR_ERR(bdev);
220 goto error;
221 }
222
223 ret = set_blocksize(bdev, 4096);
224 if (ret)
225 goto error_close;
226 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
227 if (!bh) {
228 ret = -EIO;
229 goto error_close;
230 }
231 disk_super = (struct btrfs_super_block *)bh->b_data;
232 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
233 sizeof(disk_super->magic))) {
234 printk("no btrfs found on %s\n", path);
e58ca020 235 ret = -EINVAL;
8a4b83cc
CM		 236 goto error_brelse;
237 }
238 devid = le64_to_cpu(disk_super->dev_item.devid);
239 printk("found device %Lu on %s\n", devid, path);
240 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
241
242error_brelse:
243 brelse(bh);
244error_close:
245 close_bdev_excl(bdev);
246 printk("scan one closes bdev %s\n", path);
247error:
248 mutex_unlock(&uuid_mutex);
249 return ret;
250}
0b86a832
CM		 251
252/*
253 * this uses a pretty simple search, the expectation is that it is
254 * called very infrequently and that a given device has a small number
255 * of extents
256 */
257static int find_free_dev_extent(struct btrfs_trans_handle *trans,
258 struct btrfs_device *device,
259 struct btrfs_path *path,
260 u64 num_bytes, u64 *start)
261{
262 struct btrfs_key key;
263 struct btrfs_root *root = device->dev_root;
264 struct btrfs_dev_extent *dev_extent = NULL;
265 u64 hole_size = 0;
266 u64 last_byte = 0;
267 u64 search_start = 0;
268 u64 search_end = device->total_bytes;
269 int ret;
270 int slot = 0;
271 int start_found;
272 struct extent_buffer *l;
273
274 start_found = 0;
275 path->reada = 2;
276
277 /* FIXME use last free of some kind */
278
8a4b83cc
CM		 279 /* we don't want to overwrite the superblock on the drive,
280 * so we make sure to start at an offset of at least 1MB
281 */
282 search_start = max((u64)1024 * 1024, search_start);
0b86a832
CM		 283 key.objectid = device->devid;
284 key.offset = search_start;
285 key.type = BTRFS_DEV_EXTENT_KEY;
286 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
287 if (ret < 0)
288 goto error;
289 ret = btrfs_previous_item(root, path, 0, key.type);
290 if (ret < 0)
291 goto error;
292 l = path->nodes[0];
293 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
294 while (1) {
295 l = path->nodes[0];
296 slot = path->slots[0];
297 if (slot >= btrfs_header_nritems(l)) {
298 ret = btrfs_next_leaf(root, path);
299 if (ret == 0)
300 continue;
301 if (ret < 0)
302 goto error;
303no_more_items:
304 if (!start_found) {
305 if (search_start >= search_end) {
306 ret = -ENOSPC;
307 goto error;
308 }
309 *start = search_start;
310 start_found = 1;
311 goto check_pending;
312 }
313 *start = last_byte > search_start ?
314 last_byte : search_start;
315 if (search_end <= *start) {
316 ret = -ENOSPC;
317 goto error;
318 }
319 goto check_pending;
320 }
321 btrfs_item_key_to_cpu(l, &key, slot);
322
323 if (key.objectid < device->devid)
324 goto next;
325
326 if (key.objectid > device->devid)
327 goto no_more_items;
328
329 if (key.offset >= search_start && key.offset > last_byte &&
330 start_found) {
331 if (last_byte < search_start)
332 last_byte = search_start;
333 hole_size = key.offset - last_byte;
334 if (key.offset > last_byte &&
335 hole_size >= num_bytes) {
336 *start = last_byte;
337 goto check_pending;
338 }
339 }
340 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
341 goto next;
342 }
343
344 start_found = 1;
345 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
346 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
347next:
348 path->slots[0]++;
349 cond_resched();
350 }
351check_pending:
352 /* we have to make sure we didn't find an extent that has already
353 * been allocated by the map tree or the original allocation
354 */
355 btrfs_release_path(root, path);
356 BUG_ON(*start < search_start);
357
6324fbf3 358 if (*start + num_bytes > search_end) {
0b86a832
CM		 359 ret = -ENOSPC;
360 goto error;
361 }
362 /* check for pending inserts here */
363 return 0;
364
365error:
366 btrfs_release_path(root, path);
367 return ret;
368}
369
370int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
371 struct btrfs_device *device,
372 u64 owner, u64 num_bytes, u64 *start)
373{
374 int ret;
375 struct btrfs_path *path;
376 struct btrfs_root *root = device->dev_root;
377 struct btrfs_dev_extent *extent;
378 struct extent_buffer *leaf;
379 struct btrfs_key key;
380
381 path = btrfs_alloc_path();
382 if (!path)
383 return -ENOMEM;
384
385 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
6324fbf3 386 if (ret) {
0b86a832 387 goto err;
6324fbf3 388 }
0b86a832
CM		 389
390 key.objectid = device->devid;
391 key.offset = *start;
392 key.type = BTRFS_DEV_EXTENT_KEY;
393 ret = btrfs_insert_empty_item(trans, root, path, &key,
394 sizeof(*extent));
395 BUG_ON(ret);
396
397 leaf = path->nodes[0];
398 extent = btrfs_item_ptr(leaf, path->slots[0],
399 struct btrfs_dev_extent);
400 btrfs_set_dev_extent_owner(leaf, extent, owner);
401 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
402 btrfs_mark_buffer_dirty(leaf);
403err:
404 btrfs_free_path(path);
405 return ret;
406}
407
408static int find_next_chunk(struct btrfs_root *root, u64 *objectid)
409{
410 struct btrfs_path *path;
411 int ret;
412 struct btrfs_key key;
413 struct btrfs_key found_key;
414
415 path = btrfs_alloc_path();
416 BUG_ON(!path);
417
418 key.objectid = (u64)-1;
419 key.offset = (u64)-1;
420 key.type = BTRFS_CHUNK_ITEM_KEY;
421
422 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
423 if (ret < 0)
424 goto error;
425
426 BUG_ON(ret == 0);
427
428 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
429 if (ret) {
430 *objectid = 0;
431 } else {
432 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
433 path->slots[0]);
434 *objectid = found_key.objectid + found_key.offset;
435 }
436 ret = 0;
437error:
438 btrfs_free_path(path);
439 return ret;
440}
441
0b86a832
CM		 442static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
443 u64 *objectid)
444{
445 int ret;
446 struct btrfs_key key;
447 struct btrfs_key found_key;
448
449 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
450 key.type = BTRFS_DEV_ITEM_KEY;
451 key.offset = (u64)-1;
452
453 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
454 if (ret < 0)
455 goto error;
456
457 BUG_ON(ret == 0);
458
459 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
460 BTRFS_DEV_ITEM_KEY);
461 if (ret) {
462 *objectid = 1;
463 } else {
464 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
465 path->slots[0]);
466 *objectid = found_key.offset + 1;
467 }
468 ret = 0;
469error:
470 btrfs_release_path(root, path);
471 return ret;
472}
473
474/*
475 * the device information is stored in the chunk root
476 * the btrfs_device struct should be fully filled in
477 */
478int btrfs_add_device(struct btrfs_trans_handle *trans,
479 struct btrfs_root *root,
480 struct btrfs_device *device)
481{
482 int ret;
483 struct btrfs_path *path;
484 struct btrfs_dev_item *dev_item;
485 struct extent_buffer *leaf;
486 struct btrfs_key key;
487 unsigned long ptr;
488 u64 free_devid;
489
490 root = root->fs_info->chunk_root;
491
492 path = btrfs_alloc_path();
493 if (!path)
494 return -ENOMEM;
495
496 ret = find_next_devid(root, path, &free_devid);
497 if (ret)
498 goto out;
499
500 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
501 key.type = BTRFS_DEV_ITEM_KEY;
502 key.offset = free_devid;
503
504 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 505 sizeof(*dev_item));
0b86a832
CM		 506 if (ret)
507 goto out;
508
509 leaf = path->nodes[0];
510 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
511
8a4b83cc 512 device->devid = free_devid;
0b86a832
CM		 513 btrfs_set_device_id(leaf, dev_item, device->devid);
514 btrfs_set_device_type(leaf, dev_item, device->type);
515 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
516 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
517 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 518 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
519 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
520
0b86a832
CM		 521 ptr = (unsigned long)btrfs_device_uuid(dev_item);
522 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
523 btrfs_mark_buffer_dirty(leaf);
524 ret = 0;
525
526out:
527 btrfs_free_path(path);
528 return ret;
529}
530int btrfs_update_device(struct btrfs_trans_handle *trans,
531 struct btrfs_device *device)
532{
533 int ret;
534 struct btrfs_path *path;
535 struct btrfs_root *root;
536 struct btrfs_dev_item *dev_item;
537 struct extent_buffer *leaf;
538 struct btrfs_key key;
539
540 root = device->dev_root->fs_info->chunk_root;
541
542 path = btrfs_alloc_path();
543 if (!path)
544 return -ENOMEM;
545
546 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
547 key.type = BTRFS_DEV_ITEM_KEY;
548 key.offset = device->devid;
549
550 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
551 if (ret < 0)
552 goto out;
553
554 if (ret > 0) {
555 ret = -ENOENT;
556 goto out;
557 }
558
559 leaf = path->nodes[0];
560 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
561
562 btrfs_set_device_id(leaf, dev_item, device->devid);
563 btrfs_set_device_type(leaf, dev_item, device->type);
564 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
565 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
566 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 567 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
568 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
569 btrfs_mark_buffer_dirty(leaf);
570
571out:
572 btrfs_free_path(path);
573 return ret;
574}
575
576int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
577 struct btrfs_root *root,
578 struct btrfs_key *key,
579 struct btrfs_chunk *chunk, int item_size)
580{
581 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
582 struct btrfs_disk_key disk_key;
583 u32 array_size;
584 u8 *ptr;
585
586 array_size = btrfs_super_sys_array_size(super_copy);
587 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
588 return -EFBIG;
589
590 ptr = super_copy->sys_chunk_array + array_size;
591 btrfs_cpu_key_to_disk(&disk_key, key);
592 memcpy(ptr, &disk_key, sizeof(disk_key));
593 ptr += sizeof(disk_key);
594 memcpy(ptr, chunk, item_size);
595 item_size += sizeof(disk_key);
596 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
597 return 0;
598}
599
600int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
601 struct btrfs_root *extent_root, u64 *start,
6324fbf3 602 u64 *num_bytes, u64 type)
0b86a832
CM		 603{
604 u64 dev_offset;
593060d7 605 struct btrfs_fs_info *info = extent_root->fs_info;
0b86a832
CM		 606 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
607 struct btrfs_stripe *stripes;
608 struct btrfs_device *device = NULL;
609 struct btrfs_chunk *chunk;
6324fbf3 610 struct list_head private_devs;
8a4b83cc 611 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
6324fbf3 612 struct list_head *cur;
0b86a832
CM		 613 struct extent_map_tree *em_tree;
614 struct map_lookup *map;
615 struct extent_map *em;
616 u64 physical;
617 u64 calc_size = 1024 * 1024 * 1024;
611f0e00 618 u64 min_free = calc_size;
6324fbf3
CM		 619 u64 avail;
620 u64 max_avail = 0;
621 int num_stripes = 1;
622 int looped = 0;
0b86a832 623 int ret;
6324fbf3 624 int index;
593060d7 625 int stripe_len = 64 * 1024;
0b86a832
CM		 626 struct btrfs_key key;
627
6324fbf3
CM		 628 if (list_empty(dev_list))
629 return -ENOSPC;
593060d7 630
8790d502 631 if (type & (BTRFS_BLOCK_GROUP_RAID0))
593060d7 632 num_stripes = btrfs_super_num_devices(&info->super_copy);
611f0e00
CM		 633 if (type & (BTRFS_BLOCK_GROUP_DUP))
634 num_stripes = 2;
8790d502
CM		 635 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
636 num_stripes = min_t(u64, 2,
637 btrfs_super_num_devices(&info->super_copy));
638 }
6324fbf3
CM		 639again:
640 INIT_LIST_HEAD(&private_devs);
641 cur = dev_list->next;
642 index = 0;
611f0e00
CM		 643
644 if (type & BTRFS_BLOCK_GROUP_DUP)
645 min_free = calc_size * 2;
646
6324fbf3
CM		 647 /* build a private list of devices we will allocate from */
648 while(index < num_stripes) {
649 device = list_entry(cur, struct btrfs_device, dev_list);
611f0e00 650
6324fbf3
CM		 651 avail = device->total_bytes - device->bytes_used;
652 cur = cur->next;
653 if (avail > max_avail)
654 max_avail = avail;
611f0e00 655 if (avail >= min_free) {
6324fbf3
CM		 656 list_move_tail(&device->dev_list, &private_devs);
657 index++;
611f0e00
CM		 658 if (type & BTRFS_BLOCK_GROUP_DUP)
659 index++;
6324fbf3
CM		 660 }
661 if (cur == dev_list)
662 break;
663 }
664 if (index < num_stripes) {
665 list_splice(&private_devs, dev_list);
666 if (!looped && max_avail > 0) {
667 looped = 1;
668 calc_size = max_avail;
669 goto again;
670 }
671 return -ENOSPC;
672 }
0b86a832
CM		 673
674 ret = find_next_chunk(chunk_root, &key.objectid);
675 if (ret)
676 return ret;
677
0b86a832
CM		 678 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
679 if (!chunk)
680 return -ENOMEM;
681
593060d7
CM		 682 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
683 if (!map) {
684 kfree(chunk);
685 return -ENOMEM;
686 }
687
0b86a832
CM		 688 stripes = &chunk->stripe;
689
611f0e00 690 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
8790d502
CM		 691 *num_bytes = calc_size;
692 else
693 *num_bytes = calc_size * num_stripes;
694
6324fbf3 695 index = 0;
611f0e00 696printk("new chunk type %Lu start %Lu size %Lu\n", type, key.objectid, *num_bytes);
0b86a832 697 while(index < num_stripes) {
6324fbf3
CM		 698 BUG_ON(list_empty(&private_devs));
699 cur = private_devs.next;
700 device = list_entry(cur, struct btrfs_device, dev_list);
611f0e00
CM		 701
702 /* loop over this device again if we're doing a dup group */
703 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
704 (index == num_stripes - 1))
705 list_move_tail(&device->dev_list, dev_list);
0b86a832
CM		 706
707 ret = btrfs_alloc_dev_extent(trans, device,
708 key.objectid,
709 calc_size, &dev_offset);
710 BUG_ON(ret);
8790d502 711printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.objectid, calc_size, device->devid, type);
0b86a832
CM		 712 device->bytes_used += calc_size;
713 ret = btrfs_update_device(trans, device);
714 BUG_ON(ret);
715
593060d7
CM		 716 map->stripes[index].dev = device;
717 map->stripes[index].physical = dev_offset;
0b86a832
CM		 718 btrfs_set_stack_stripe_devid(stripes + index, device->devid);
719 btrfs_set_stack_stripe_offset(stripes + index, dev_offset);
720 physical = dev_offset;
721 index++;
722 }
6324fbf3 723 BUG_ON(!list_empty(&private_devs));
0b86a832
CM		 724
725 /* key.objectid was set above */
726 key.offset = *num_bytes;
727 key.type = BTRFS_CHUNK_ITEM_KEY;
728 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
593060d7 729 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
0b86a832
CM		 730 btrfs_set_stack_chunk_type(chunk, type);
731 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
593060d7
CM		 732 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
733 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
0b86a832 734 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
593060d7
CM		 735 map->sector_size = extent_root->sectorsize;
736 map->stripe_len = stripe_len;
737 map->io_align = stripe_len;
738 map->io_width = stripe_len;
739 map->type = type;
740 map->num_stripes = num_stripes;
0b86a832
CM		 741
742 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
743 btrfs_chunk_item_size(num_stripes));
744 BUG_ON(ret);
745 *start = key.objectid;
746
747 em = alloc_extent_map(GFP_NOFS);
748 if (!em)
749 return -ENOMEM;
0b86a832
CM		 750 em->bdev = (struct block_device *)map;
751 em->start = key.objectid;
752 em->len = key.offset;
753 em->block_start = 0;
754
0b86a832
CM		 755 kfree(chunk);
756
757 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
758 spin_lock(&em_tree->lock);
759 ret = add_extent_mapping(em_tree, em);
760 BUG_ON(ret);
761 spin_unlock(&em_tree->lock);
762 free_extent_map(em);
763 return ret;
764}
765
766void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
767{
768 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
769}
770
771void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
772{
773 struct extent_map *em;
774
775 while(1) {
776 spin_lock(&tree->map_tree.lock);
777 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
778 if (em)
779 remove_extent_mapping(&tree->map_tree, em);
780 spin_unlock(&tree->map_tree.lock);
781 if (!em)
782 break;
783 kfree(em->bdev);
784 /* once for us */
785 free_extent_map(em);
786 /* once for the tree */
787 free_extent_map(em);
788 }
789}
790
8790d502 791int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
cea9e445
CM		 792 u64 logical, u64 *length,
793 struct btrfs_multi_bio **multi_ret)
0b86a832
CM		 794{
795 struct extent_map *em;
796 struct map_lookup *map;
797 struct extent_map_tree *em_tree = &map_tree->map_tree;
798 u64 offset;
593060d7
CM		 799 u64 stripe_offset;
800 u64 stripe_nr;
cea9e445 801 int stripes_allocated = 8;
593060d7 802 int stripe_index;
cea9e445
CM		 803 int i;
804 struct btrfs_multi_bio *multi = NULL;
0b86a832 805
cea9e445
CM		 806 if (multi_ret && !(rw & (1 << BIO_RW))) {
807 stripes_allocated = 1;
808 }
809again:
810 if (multi_ret) {
811 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
812 GFP_NOFS);
813 if (!multi)
814 return -ENOMEM;
815 }
0b86a832
CM		 816
817 spin_lock(&em_tree->lock);
818 em = lookup_extent_mapping(em_tree, logical, *length);
819 BUG_ON(!em);
820
821 BUG_ON(em->start > logical || em->start + em->len < logical);
822 map = (struct map_lookup *)em->bdev;
823 offset = logical - em->start;
593060d7 824
cea9e445
CM		 825 /* if our multi bio struct is too small, back off and try again */
826 if (multi_ret && (rw & (1 << BIO_RW)) &&
827 stripes_allocated < map->num_stripes &&
828 ((map->type & BTRFS_BLOCK_GROUP_RAID1) ||
829 (map->type & BTRFS_BLOCK_GROUP_DUP))) {
830 stripes_allocated = map->num_stripes;
831 spin_unlock(&em_tree->lock);
832 free_extent_map(em);
833 kfree(multi);
834 goto again;
835 }
593060d7
CM		 836 stripe_nr = offset;
837 /*
838 * stripe_nr counts the total number of stripes we have to stride
839 * to get to this block
840 */
841 do_div(stripe_nr, map->stripe_len);
842
843 stripe_offset = stripe_nr * map->stripe_len;
844 BUG_ON(offset < stripe_offset);
845
846 /* stripe_offset is the offset of this block in its stripe*/
847 stripe_offset = offset - stripe_offset;
848
cea9e445
CM		 849 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
850 BTRFS_BLOCK_GROUP_DUP)) {
851 /* we limit the length of each bio to what fits in a stripe */
852 *length = min_t(u64, em->len - offset,
853 map->stripe_len - stripe_offset);
854 } else {
855 *length = em->len - offset;
856 }
857 if (!multi_ret)
858 goto out;
859
860 multi->num_stripes = 1;
861 stripe_index = 0;
8790d502 862 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
8790d502 863 if (rw & (1 << BIO_RW))
cea9e445 864 multi->num_stripes = map->num_stripes;
8790d502
CM		 865 else {
866 int i;
867 u64 least = (u64)-1;
868 struct btrfs_device *cur;
869
870 for (i = 0; i < map->num_stripes; i++) {
871 cur = map->stripes[i].dev;
872 spin_lock(&cur->io_lock);
873 if (cur->total_ios < least) {
874 least = cur->total_ios;
875 stripe_index = i;
876 }
877 spin_unlock(&cur->io_lock);
878 }
8790d502 879 }
611f0e00 880 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
cea9e445
CM		 881 if (rw & (1 << BIO_RW))
882 multi->num_stripes = map->num_stripes;
8790d502
CM		 883 } else {
884 /*
885 * after this do_div call, stripe_nr is the number of stripes
886 * on this device we have to walk to find the data, and
887 * stripe_index is the number of our device in the stripe array
888 */
889 stripe_index = do_div(stripe_nr, map->num_stripes);
890 }
593060d7 891 BUG_ON(stripe_index >= map->num_stripes);
cea9e445
CM		 892 BUG_ON(stripe_index != 0 && multi->num_stripes > 1);
893
894 for (i = 0; i < multi->num_stripes; i++) {
895 multi->stripes[i].physical =
896 map->stripes[stripe_index].physical + stripe_offset +
897 stripe_nr * map->stripe_len;
898 multi->stripes[i].dev = map->stripes[stripe_index].dev;
899 stripe_index++;
593060d7 900 }
cea9e445
CM		 901 *multi_ret = multi;
902out:
0b86a832
CM		 903 free_extent_map(em);
904 spin_unlock(&em_tree->lock);
905 return 0;
906}
907
8790d502
CM		 908#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
909static void end_bio_multi_stripe(struct bio *bio, int err)
910#else
911static int end_bio_multi_stripe(struct bio *bio,
912 unsigned int bytes_done, int err)
913#endif
914{
cea9e445 915 struct btrfs_multi_bio *multi = bio->bi_private;
8790d502
CM		 916
917#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
918 if (bio->bi_size)
919 return 1;
920#endif
921 if (err)
922 multi->error = err;
923
cea9e445 924 if (atomic_dec_and_test(&multi->stripes_pending)) {
8790d502
CM		 925 bio->bi_private = multi->private;
926 bio->bi_end_io = multi->end_io;
927
928 if (!err && multi->error)
929 err = multi->error;
930 kfree(multi);
931
932 bio_endio(bio, err);
933 } else {
934 bio_put(bio);
935 }
936#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
937 return 0;
938#endif
939}
940
0b86a832
CM		 941int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio)
942{
943 struct btrfs_mapping_tree *map_tree;
944 struct btrfs_device *dev;
8790d502 945 struct bio *first_bio = bio;
0b86a832 946 u64 logical = bio->bi_sector << 9;
0b86a832
CM		 947 u64 length = 0;
948 u64 map_length;
949 struct bio_vec *bvec;
cea9e445 950 struct btrfs_multi_bio *multi = NULL;
0b86a832
CM		 951 int i;
952 int ret;
8790d502
CM		 953 int dev_nr = 0;
954 int total_devs = 1;
0b86a832
CM		 955
956 bio_for_each_segment(bvec, bio, i) {
957 length += bvec->bv_len;
958 }
8790d502 959
0b86a832
CM		 960 map_tree = &root->fs_info->mapping_tree;
961 map_length = length;
cea9e445
CM		 962
963 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi);
964 BUG_ON(ret);
965
966 total_devs = multi->num_stripes;
967 if (map_length < length) {
968 printk("mapping failed logical %Lu bio len %Lu "
969 "len %Lu\n", logical, length, map_length);
970 BUG();
971 }
972 multi->end_io = first_bio->bi_end_io;
973 multi->private = first_bio->bi_private;
974 atomic_set(&multi->stripes_pending, multi->num_stripes);
975
8790d502 976 while(dev_nr < total_devs) {
8790d502 977 if (total_devs > 1) {
8790d502
CM		 978 if (dev_nr < total_devs - 1) {
979 bio = bio_clone(first_bio, GFP_NOFS);
980 BUG_ON(!bio);
981 } else {
982 bio = first_bio;
983 }
984 bio->bi_private = multi;
985 bio->bi_end_io = end_bio_multi_stripe;
986 }
cea9e445
CM		 987 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
988 dev = multi->stripes[dev_nr].dev;
8790d502
CM		 989 bio->bi_bdev = dev->bdev;
990 spin_lock(&dev->io_lock);
991 dev->total_ios++;
992 spin_unlock(&dev->io_lock);
993 submit_bio(rw, bio);
994 dev_nr++;
995 }
cea9e445
CM		 996 if (total_devs == 1)
997 kfree(multi);
0b86a832
CM		 998 return 0;
999}
1000
1001struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid)
1002{
8a4b83cc 1003 struct list_head *head = &root->fs_info->fs_devices->devices;
0b86a832 1004
8a4b83cc 1005 return __find_device(head, devid);
0b86a832
CM		 1006}
1007
1008static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1009 struct extent_buffer *leaf,
1010 struct btrfs_chunk *chunk)
1011{
1012 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1013 struct map_lookup *map;
1014 struct extent_map *em;
1015 u64 logical;
1016 u64 length;
1017 u64 devid;
593060d7 1018 int num_stripes;
0b86a832 1019 int ret;
593060d7 1020 int i;
0b86a832
CM		 1021
1022 logical = key->objectid;
1023 length = key->offset;
1024 spin_lock(&map_tree->map_tree.lock);
1025 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1026
1027 /* already mapped? */
1028 if (em && em->start <= logical && em->start + em->len > logical) {
1029 free_extent_map(em);
1030 spin_unlock(&map_tree->map_tree.lock);
1031 return 0;
1032 } else if (em) {
1033 free_extent_map(em);
1034 }
1035 spin_unlock(&map_tree->map_tree.lock);
1036
1037 map = kzalloc(sizeof(*map), GFP_NOFS);
1038 if (!map)
1039 return -ENOMEM;
1040
1041 em = alloc_extent_map(GFP_NOFS);
1042 if (!em)
1043 return -ENOMEM;
593060d7
CM		 1044 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1045 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 1046 if (!map) {
1047 free_extent_map(em);
1048 return -ENOMEM;
1049 }
1050
1051 em->bdev = (struct block_device *)map;
1052 em->start = logical;
1053 em->len = length;
1054 em->block_start = 0;
1055
593060d7
CM		 1056 map->num_stripes = num_stripes;
1057 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1058 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1059 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1060 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1061 map->type = btrfs_chunk_type(leaf, chunk);
1062 for (i = 0; i < num_stripes; i++) {
1063 map->stripes[i].physical =
1064 btrfs_stripe_offset_nr(leaf, chunk, i);
1065 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1066 map->stripes[i].dev = btrfs_find_device(root, devid);
1067 if (!map->stripes[i].dev) {
1068 kfree(map);
1069 free_extent_map(em);
1070 return -EIO;
1071 }
0b86a832
CM		 1072 }
1073
1074 spin_lock(&map_tree->map_tree.lock);
1075 ret = add_extent_mapping(&map_tree->map_tree, em);
1076 BUG_ON(ret);
1077 spin_unlock(&map_tree->map_tree.lock);
1078 free_extent_map(em);
1079
1080 return 0;
1081}
1082
1083static int fill_device_from_item(struct extent_buffer *leaf,
1084 struct btrfs_dev_item *dev_item,
1085 struct btrfs_device *device)
1086{
1087 unsigned long ptr;
0b86a832
CM		 1088
1089 device->devid = btrfs_device_id(leaf, dev_item);
1090 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1091 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1092 device->type = btrfs_device_type(leaf, dev_item);
1093 device->io_align = btrfs_device_io_align(leaf, dev_item);
1094 device->io_width = btrfs_device_io_width(leaf, dev_item);
1095 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 1096
1097 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1098 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
1099
0b86a832
CM		 1100 return 0;
1101}
1102
0d81ba5d 1103static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 1104 struct extent_buffer *leaf,
1105 struct btrfs_dev_item *dev_item)
1106{
1107 struct btrfs_device *device;
1108 u64 devid;
1109 int ret;
1110
1111 devid = btrfs_device_id(leaf, dev_item);
6324fbf3
CM		 1112 device = btrfs_find_device(root, devid);
1113 if (!device) {
8a4b83cc 1114 printk("warning devid %Lu not found already\n", devid);
6324fbf3
CM		 1115 device = kmalloc(sizeof(*device), GFP_NOFS);
1116 if (!device)
1117 return -ENOMEM;
8a4b83cc
CM		 1118 list_add(&device->dev_list,
1119 &root->fs_info->fs_devices->devices);
8790d502
CM		 1120 device->total_ios = 0;
1121 spin_lock_init(&device->io_lock);
6324fbf3 1122 }
0b86a832
CM		 1123
1124 fill_device_from_item(leaf, dev_item, device);
1125 device->dev_root = root->fs_info->dev_root;
0b86a832
CM		 1126 ret = 0;
1127#if 0
1128 ret = btrfs_open_device(device);
1129 if (ret) {
1130 kfree(device);
1131 }
1132#endif
1133 return ret;
1134}
1135
0d81ba5d
CM		 1136int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1137{
1138 struct btrfs_dev_item *dev_item;
1139
1140 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1141 dev_item);
1142 return read_one_dev(root, buf, dev_item);
1143}
1144
0b86a832
CM		 1145int btrfs_read_sys_array(struct btrfs_root *root)
1146{
1147 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1148 struct extent_buffer *sb = root->fs_info->sb_buffer;
1149 struct btrfs_disk_key *disk_key;
0b86a832
CM		 1150 struct btrfs_chunk *chunk;
1151 struct btrfs_key key;
1152 u32 num_stripes;
1153 u32 array_size;
1154 u32 len = 0;
1155 u8 *ptr;
1156 unsigned long sb_ptr;
1157 u32 cur;
1158 int ret;
0b86a832
CM		 1159
1160 array_size = btrfs_super_sys_array_size(super_copy);
1161
1162 /*
1163 * we do this loop twice, once for the device items and
1164 * once for all of the chunks. This way there are device
1165 * structs filled in for every chunk
1166 */
0b86a832
CM		 1167 ptr = super_copy->sys_chunk_array;
1168 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1169 cur = 0;
1170
1171 while (cur < array_size) {
1172 disk_key = (struct btrfs_disk_key *)ptr;
1173 btrfs_disk_key_to_cpu(&key, disk_key);
1174
1175 len = sizeof(*disk_key);
1176 ptr += len;
1177 sb_ptr += len;
1178 cur += len;
1179
0d81ba5d 1180 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 1181 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d
CM		 1182 ret = read_one_chunk(root, &key, sb, chunk);
1183 BUG_ON(ret);
0b86a832
CM		 1184 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1185 len = btrfs_chunk_item_size(num_stripes);
1186 } else {
1187 BUG();
1188 }
1189 ptr += len;
1190 sb_ptr += len;
1191 cur += len;
1192 }
0b86a832
CM		 1193 return 0;
1194}
1195
1196int btrfs_read_chunk_tree(struct btrfs_root *root)
1197{
1198 struct btrfs_path *path;
1199 struct extent_buffer *leaf;
1200 struct btrfs_key key;
1201 struct btrfs_key found_key;
1202 int ret;
1203 int slot;
1204
1205 root = root->fs_info->chunk_root;
1206
1207 path = btrfs_alloc_path();
1208 if (!path)
1209 return -ENOMEM;
1210
1211 /* first we search for all of the device items, and then we
1212 * read in all of the chunk items. This way we can create chunk
1213 * mappings that reference all of the devices that are afound
1214 */
1215 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1216 key.offset = 0;
1217 key.type = 0;
1218again:
1219 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1220 while(1) {
1221 leaf = path->nodes[0];
1222 slot = path->slots[0];
1223 if (slot >= btrfs_header_nritems(leaf)) {
1224 ret = btrfs_next_leaf(root, path);
1225 if (ret == 0)
1226 continue;
1227 if (ret < 0)
1228 goto error;
1229 break;
1230 }
1231 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1232 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1233 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1234 break;
1235 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1236 struct btrfs_dev_item *dev_item;
1237 dev_item = btrfs_item_ptr(leaf, slot,
1238 struct btrfs_dev_item);
0d81ba5d 1239 ret = read_one_dev(root, leaf, dev_item);
0b86a832
CM		 1240 BUG_ON(ret);
1241 }
1242 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1243 struct btrfs_chunk *chunk;
1244 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1245 ret = read_one_chunk(root, &found_key, leaf, chunk);
1246 }
1247 path->slots[0]++;
1248 }
1249 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1250 key.objectid = 0;
1251 btrfs_release_path(root, path);
1252 goto again;
1253 }
1254
1255 btrfs_free_path(path);
1256 ret = 0;
1257error:
1258 return ret;
1259}
1260
Linux 6.x block layer and io_uring tree(s)