blob: 198cff28766d494f3a919673775fe679d560722f [file] [log] [blame]
Chris Mason0b86a832008-03-24 15:01:56 -04001/*
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>
Chris Mason8a4b83c2008-03-24 15:02:07 -040020#include <linux/buffer_head.h>
Chris Masonf2d8d742008-04-21 10:03:05 -040021#include <linux/blkdev.h>
Chris Mason788f20e2008-04-28 15:29:42 -040022#include <linux/random.h>
Chris Masonb765ead2009-04-03 10:27:10 -040023#include <linux/iocontext.h>
Chris Mason593060d2008-03-25 16:50:33 -040024#include <asm/div64.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050025#include "compat.h"
Chris Mason0b86a832008-03-24 15:01:56 -040026#include "ctree.h"
27#include "extent_map.h"
28#include "disk-io.h"
29#include "transaction.h"
30#include "print-tree.h"
31#include "volumes.h"
Chris Mason8b712842008-06-11 16:50:36 -040032#include "async-thread.h"
Chris Mason0b86a832008-03-24 15:01:56 -040033
Chris Mason593060d2008-03-25 16:50:33 -040034struct map_lookup {
35 u64 type;
36 int io_align;
37 int io_width;
38 int stripe_len;
39 int sector_size;
40 int num_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -040041 int sub_stripes;
Chris Masoncea9e442008-04-09 16:28:12 -040042 struct btrfs_bio_stripe stripes[];
Chris Mason593060d2008-03-25 16:50:33 -040043};
44
Yan Zheng2b820322008-11-17 21:11:30 -050045static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49
Chris Mason593060d2008-03-25 16:50:33 -040050#define map_lookup_size(n) (sizeof(struct map_lookup) + \
Chris Masoncea9e442008-04-09 16:28:12 -040051 (sizeof(struct btrfs_bio_stripe) * (n)))
Chris Mason593060d2008-03-25 16:50:33 -040052
Chris Mason8a4b83c2008-03-24 15:02:07 -040053static DEFINE_MUTEX(uuid_mutex);
54static LIST_HEAD(fs_uuids);
55
Chris Masona061fc82008-05-07 11:43:44 -040056void btrfs_lock_volumes(void)
57{
58 mutex_lock(&uuid_mutex);
59}
60
61void btrfs_unlock_volumes(void)
62{
63 mutex_unlock(&uuid_mutex);
64}
65
Chris Mason7d9eb122008-07-08 14:19:17 -040066static void lock_chunks(struct btrfs_root *root)
67{
Chris Mason7d9eb122008-07-08 14:19:17 -040068 mutex_lock(&root->fs_info->chunk_mutex);
69}
70
71static void unlock_chunks(struct btrfs_root *root)
72{
Chris Mason7d9eb122008-07-08 14:19:17 -040073 mutex_unlock(&root->fs_info->chunk_mutex);
74}
75
Yan Zhenge4404d62008-12-12 10:03:26 -050076static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
77{
78 struct btrfs_device *device;
79 WARN_ON(fs_devices->opened);
80 while (!list_empty(&fs_devices->devices)) {
81 device = list_entry(fs_devices->devices.next,
82 struct btrfs_device, dev_list);
83 list_del(&device->dev_list);
84 kfree(device->name);
85 kfree(device);
86 }
87 kfree(fs_devices);
88}
89
Chris Mason8a4b83c2008-03-24 15:02:07 -040090int btrfs_cleanup_fs_uuids(void)
91{
92 struct btrfs_fs_devices *fs_devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -040093
Yan Zheng2b820322008-11-17 21:11:30 -050094 while (!list_empty(&fs_uuids)) {
95 fs_devices = list_entry(fs_uuids.next,
96 struct btrfs_fs_devices, list);
97 list_del(&fs_devices->list);
Yan Zhenge4404d62008-12-12 10:03:26 -050098 free_fs_devices(fs_devices);
Chris Mason8a4b83c2008-03-24 15:02:07 -040099 }
100 return 0;
101}
102
Chris Masona1b32a52008-09-05 16:09:51 -0400103static noinline struct btrfs_device *__find_device(struct list_head *head,
104 u64 devid, u8 *uuid)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400105{
106 struct btrfs_device *dev;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400107
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500108 list_for_each_entry(dev, head, dev_list) {
Chris Masona4437552008-04-18 10:29:38 -0400109 if (dev->devid == devid &&
Chris Mason8f18cf12008-04-25 16:53:30 -0400110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400111 return dev;
Chris Masona4437552008-04-18 10:29:38 -0400112 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400113 }
114 return NULL;
115}
116
Chris Masona1b32a52008-09-05 16:09:51 -0400117static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400118{
Chris Mason8a4b83c2008-03-24 15:02:07 -0400119 struct btrfs_fs_devices *fs_devices;
120
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500121 list_for_each_entry(fs_devices, &fs_uuids, list) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
123 return fs_devices;
124 }
125 return NULL;
126}
127
Chris Masonffbd5172009-04-20 15:50:09 -0400128static void requeue_list(struct btrfs_pending_bios *pending_bios,
129 struct bio *head, struct bio *tail)
130{
131
132 struct bio *old_head;
133
134 old_head = pending_bios->head;
135 pending_bios->head = head;
136 if (pending_bios->tail)
137 tail->bi_next = old_head;
138 else
139 pending_bios->tail = tail;
140}
141
Chris Mason8b712842008-06-11 16:50:36 -0400142/*
143 * we try to collect pending bios for a device so we don't get a large
144 * number of procs sending bios down to the same device. This greatly
145 * improves the schedulers ability to collect and merge the bios.
146 *
147 * But, it also turns into a long list of bios to process and that is sure
148 * to eventually make the worker thread block. The solution here is to
149 * make some progress and then put this work struct back at the end of
150 * the list if the block device is congested. This way, multiple devices
151 * can make progress from a single worker thread.
152 */
Chris Masond3977122009-01-05 21:25:51 -0500153static noinline int run_scheduled_bios(struct btrfs_device *device)
Chris Mason8b712842008-06-11 16:50:36 -0400154{
155 struct bio *pending;
156 struct backing_dev_info *bdi;
Chris Masonb64a2852008-08-20 13:39:41 -0400157 struct btrfs_fs_info *fs_info;
Chris Masonffbd5172009-04-20 15:50:09 -0400158 struct btrfs_pending_bios *pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -0400159 struct bio *tail;
160 struct bio *cur;
161 int again = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400162 unsigned long num_run;
163 unsigned long num_sync_run;
Chris Masond644d8a2009-06-09 15:59:22 -0400164 unsigned long batch_run = 0;
Chris Masonb64a2852008-08-20 13:39:41 -0400165 unsigned long limit;
Chris Masonb765ead2009-04-03 10:27:10 -0400166 unsigned long last_waited = 0;
Chris Masond84275c2009-06-09 15:39:08 -0400167 int force_reg = 0;
Chris Mason8b712842008-06-11 16:50:36 -0400168
Chris Masonbedf7622009-04-03 10:32:58 -0400169 bdi = blk_get_backing_dev_info(device->bdev);
Chris Masonb64a2852008-08-20 13:39:41 -0400170 fs_info = device->dev_root->fs_info;
171 limit = btrfs_async_submit_limit(fs_info);
172 limit = limit * 2 / 3;
173
Chris Masonffbd5172009-04-20 15:50:09 -0400174 /* we want to make sure that every time we switch from the sync
175 * list to the normal list, we unplug
176 */
177 num_sync_run = 0;
178
Chris Mason8b712842008-06-11 16:50:36 -0400179loop:
180 spin_lock(&device->io_lock);
181
Chris Masona6837052009-02-04 09:19:41 -0500182loop_lock:
Chris Masond84275c2009-06-09 15:39:08 -0400183 num_run = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400184
Chris Mason8b712842008-06-11 16:50:36 -0400185 /* take all the bios off the list at once and process them
186 * later on (without the lock held). But, remember the
187 * tail and other pointers so the bios can be properly reinserted
188 * into the list if we hit congestion
189 */
Chris Masond84275c2009-06-09 15:39:08 -0400190 if (!force_reg && device->pending_sync_bios.head) {
Chris Masonffbd5172009-04-20 15:50:09 -0400191 pending_bios = &device->pending_sync_bios;
Chris Masond84275c2009-06-09 15:39:08 -0400192 force_reg = 1;
193 } else {
Chris Masonffbd5172009-04-20 15:50:09 -0400194 pending_bios = &device->pending_bios;
Chris Masond84275c2009-06-09 15:39:08 -0400195 force_reg = 0;
196 }
Chris Masonffbd5172009-04-20 15:50:09 -0400197
198 pending = pending_bios->head;
199 tail = pending_bios->tail;
Chris Mason8b712842008-06-11 16:50:36 -0400200 WARN_ON(pending && !tail);
Chris Mason8b712842008-06-11 16:50:36 -0400201
202 /*
203 * if pending was null this time around, no bios need processing
204 * at all and we can stop. Otherwise it'll loop back up again
205 * and do an additional check so no bios are missed.
206 *
207 * device->running_pending is used to synchronize with the
208 * schedule_bio code.
209 */
Chris Masonffbd5172009-04-20 15:50:09 -0400210 if (device->pending_sync_bios.head == NULL &&
211 device->pending_bios.head == NULL) {
Chris Mason8b712842008-06-11 16:50:36 -0400212 again = 0;
213 device->running_pending = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400214 } else {
215 again = 1;
216 device->running_pending = 1;
Chris Mason8b712842008-06-11 16:50:36 -0400217 }
Chris Masonffbd5172009-04-20 15:50:09 -0400218
219 pending_bios->head = NULL;
220 pending_bios->tail = NULL;
221
Chris Mason8b712842008-06-11 16:50:36 -0400222 spin_unlock(&device->io_lock);
223
Chris Masonffbd5172009-04-20 15:50:09 -0400224 /*
225 * if we're doing the regular priority list, make sure we unplug
226 * for any high prio bios we've sent down
227 */
228 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
229 num_sync_run = 0;
230 blk_run_backing_dev(bdi, NULL);
231 }
232
Chris Masond3977122009-01-05 21:25:51 -0500233 while (pending) {
Chris Masonffbd5172009-04-20 15:50:09 -0400234
235 rmb();
Chris Masond84275c2009-06-09 15:39:08 -0400236 /* we want to work on both lists, but do more bios on the
237 * sync list than the regular list
238 */
239 if ((num_run > 32 &&
240 pending_bios != &device->pending_sync_bios &&
241 device->pending_sync_bios.head) ||
242 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
243 device->pending_bios.head)) {
Chris Masonffbd5172009-04-20 15:50:09 -0400244 spin_lock(&device->io_lock);
245 requeue_list(pending_bios, pending, tail);
246 goto loop_lock;
247 }
248
Chris Mason8b712842008-06-11 16:50:36 -0400249 cur = pending;
250 pending = pending->bi_next;
251 cur->bi_next = NULL;
Chris Masonb64a2852008-08-20 13:39:41 -0400252 atomic_dec(&fs_info->nr_async_bios);
253
254 if (atomic_read(&fs_info->nr_async_bios) < limit &&
255 waitqueue_active(&fs_info->async_submit_wait))
256 wake_up(&fs_info->async_submit_wait);
Chris Mason492bb6de2008-07-31 16:29:02 -0400257
258 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
Chris Mason8b712842008-06-11 16:50:36 -0400259 submit_bio(cur->bi_rw, cur);
260 num_run++;
Chris Masond644d8a2009-06-09 15:59:22 -0400261 batch_run++;
262
Jens Axboe1f98a132009-09-11 14:32:04 +0200263 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
Chris Masonffbd5172009-04-20 15:50:09 -0400264 num_sync_run++;
265
266 if (need_resched()) {
267 if (num_sync_run) {
268 blk_run_backing_dev(bdi, NULL);
269 num_sync_run = 0;
270 }
271 cond_resched();
272 }
Chris Mason8b712842008-06-11 16:50:36 -0400273
274 /*
275 * we made progress, there is more work to do and the bdi
276 * is now congested. Back off and let other work structs
277 * run instead
278 */
Chris Mason57fd5a52009-08-07 09:59:15 -0400279 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
Chris Mason5f2cc082008-11-07 18:22:45 -0500280 fs_info->fs_devices->open_devices > 1) {
Chris Masonb765ead2009-04-03 10:27:10 -0400281 struct io_context *ioc;
Chris Mason8b712842008-06-11 16:50:36 -0400282
Chris Masonb765ead2009-04-03 10:27:10 -0400283 ioc = current->io_context;
284
285 /*
286 * the main goal here is that we don't want to
287 * block if we're going to be able to submit
288 * more requests without blocking.
289 *
290 * This code does two great things, it pokes into
291 * the elevator code from a filesystem _and_
292 * it makes assumptions about how batching works.
293 */
294 if (ioc && ioc->nr_batch_requests > 0 &&
295 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
296 (last_waited == 0 ||
297 ioc->last_waited == last_waited)) {
298 /*
299 * we want to go through our batch of
300 * requests and stop. So, we copy out
301 * the ioc->last_waited time and test
302 * against it before looping
303 */
304 last_waited = ioc->last_waited;
Chris Masonffbd5172009-04-20 15:50:09 -0400305 if (need_resched()) {
306 if (num_sync_run) {
307 blk_run_backing_dev(bdi, NULL);
308 num_sync_run = 0;
309 }
310 cond_resched();
311 }
Chris Masonb765ead2009-04-03 10:27:10 -0400312 continue;
313 }
Chris Mason8b712842008-06-11 16:50:36 -0400314 spin_lock(&device->io_lock);
Chris Masonffbd5172009-04-20 15:50:09 -0400315 requeue_list(pending_bios, pending, tail);
Chris Masona6837052009-02-04 09:19:41 -0500316 device->running_pending = 1;
Chris Mason8b712842008-06-11 16:50:36 -0400317
318 spin_unlock(&device->io_lock);
319 btrfs_requeue_work(&device->work);
320 goto done;
321 }
322 }
Chris Masonffbd5172009-04-20 15:50:09 -0400323
324 if (num_sync_run) {
325 num_sync_run = 0;
326 blk_run_backing_dev(bdi, NULL);
327 }
328
329 cond_resched();
Chris Mason8b712842008-06-11 16:50:36 -0400330 if (again)
331 goto loop;
Chris Masona6837052009-02-04 09:19:41 -0500332
333 spin_lock(&device->io_lock);
Chris Masonffbd5172009-04-20 15:50:09 -0400334 if (device->pending_bios.head || device->pending_sync_bios.head)
Chris Masona6837052009-02-04 09:19:41 -0500335 goto loop_lock;
336 spin_unlock(&device->io_lock);
Chris Masonbedf7622009-04-03 10:32:58 -0400337
338 /*
339 * IO has already been through a long path to get here. Checksumming,
340 * async helper threads, perhaps compression. We've done a pretty
341 * good job of collecting a batch of IO and should just unplug
342 * the device right away.
343 *
344 * This will help anyone who is waiting on the IO, they might have
345 * already unplugged, but managed to do so before the bio they
346 * cared about found its way down here.
347 */
348 blk_run_backing_dev(bdi, NULL);
Chris Mason8b712842008-06-11 16:50:36 -0400349done:
350 return 0;
351}
352
Christoph Hellwigb2950862008-12-02 09:54:17 -0500353static void pending_bios_fn(struct btrfs_work *work)
Chris Mason8b712842008-06-11 16:50:36 -0400354{
355 struct btrfs_device *device;
356
357 device = container_of(work, struct btrfs_device, work);
358 run_scheduled_bios(device);
359}
360
Chris Masona1b32a52008-09-05 16:09:51 -0400361static noinline int device_list_add(const char *path,
Chris Mason8a4b83c2008-03-24 15:02:07 -0400362 struct btrfs_super_block *disk_super,
363 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
364{
365 struct btrfs_device *device;
366 struct btrfs_fs_devices *fs_devices;
367 u64 found_transid = btrfs_super_generation(disk_super);
368
369 fs_devices = find_fsid(disk_super->fsid);
370 if (!fs_devices) {
Chris Mason515dc322008-05-16 13:30:15 -0400371 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400372 if (!fs_devices)
373 return -ENOMEM;
374 INIT_LIST_HEAD(&fs_devices->devices);
Chris Masonb3075712008-04-22 09:22:07 -0400375 INIT_LIST_HEAD(&fs_devices->alloc_list);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400376 list_add(&fs_devices->list, &fs_uuids);
377 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
378 fs_devices->latest_devid = devid;
379 fs_devices->latest_trans = found_transid;
Chris Masone5e9a522009-06-10 15:17:02 -0400380 mutex_init(&fs_devices->device_list_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400381 device = NULL;
382 } else {
Chris Masona4437552008-04-18 10:29:38 -0400383 device = __find_device(&fs_devices->devices, devid,
384 disk_super->dev_item.uuid);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400385 }
386 if (!device) {
Yan Zheng2b820322008-11-17 21:11:30 -0500387 if (fs_devices->opened)
388 return -EBUSY;
389
Chris Mason8a4b83c2008-03-24 15:02:07 -0400390 device = kzalloc(sizeof(*device), GFP_NOFS);
391 if (!device) {
392 /* we can safely leave the fs_devices entry around */
393 return -ENOMEM;
394 }
395 device->devid = devid;
Chris Mason8b712842008-06-11 16:50:36 -0400396 device->work.func = pending_bios_fn;
Chris Masona4437552008-04-18 10:29:38 -0400397 memcpy(device->uuid, disk_super->dev_item.uuid,
398 BTRFS_UUID_SIZE);
Chris Masonf2984462008-04-10 16:19:33 -0400399 device->barriers = 1;
Chris Masonb248a412008-04-14 09:48:18 -0400400 spin_lock_init(&device->io_lock);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400401 device->name = kstrdup(path, GFP_NOFS);
402 if (!device->name) {
403 kfree(device);
404 return -ENOMEM;
405 }
Yan Zheng2b820322008-11-17 21:11:30 -0500406 INIT_LIST_HEAD(&device->dev_alloc_list);
Chris Masone5e9a522009-06-10 15:17:02 -0400407
408 mutex_lock(&fs_devices->device_list_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400409 list_add(&device->dev_list, &fs_devices->devices);
Chris Masone5e9a522009-06-10 15:17:02 -0400410 mutex_unlock(&fs_devices->device_list_mutex);
411
Yan Zheng2b820322008-11-17 21:11:30 -0500412 device->fs_devices = fs_devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400413 fs_devices->num_devices++;
414 }
415
416 if (found_transid > fs_devices->latest_trans) {
417 fs_devices->latest_devid = devid;
418 fs_devices->latest_trans = found_transid;
419 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400420 *fs_devices_ret = fs_devices;
421 return 0;
422}
423
Yan Zhenge4404d62008-12-12 10:03:26 -0500424static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
425{
426 struct btrfs_fs_devices *fs_devices;
427 struct btrfs_device *device;
428 struct btrfs_device *orig_dev;
429
430 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
431 if (!fs_devices)
432 return ERR_PTR(-ENOMEM);
433
434 INIT_LIST_HEAD(&fs_devices->devices);
435 INIT_LIST_HEAD(&fs_devices->alloc_list);
436 INIT_LIST_HEAD(&fs_devices->list);
Chris Masone5e9a522009-06-10 15:17:02 -0400437 mutex_init(&fs_devices->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500438 fs_devices->latest_devid = orig->latest_devid;
439 fs_devices->latest_trans = orig->latest_trans;
440 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
441
Chris Masone5e9a522009-06-10 15:17:02 -0400442 mutex_lock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500443 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
444 device = kzalloc(sizeof(*device), GFP_NOFS);
445 if (!device)
446 goto error;
447
448 device->name = kstrdup(orig_dev->name, GFP_NOFS);
Julia Lawallfd2696f2009-09-29 13:51:04 -0400449 if (!device->name) {
450 kfree(device);
Yan Zhenge4404d62008-12-12 10:03:26 -0500451 goto error;
Julia Lawallfd2696f2009-09-29 13:51:04 -0400452 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500453
454 device->devid = orig_dev->devid;
455 device->work.func = pending_bios_fn;
456 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
457 device->barriers = 1;
458 spin_lock_init(&device->io_lock);
459 INIT_LIST_HEAD(&device->dev_list);
460 INIT_LIST_HEAD(&device->dev_alloc_list);
461
462 list_add(&device->dev_list, &fs_devices->devices);
463 device->fs_devices = fs_devices;
464 fs_devices->num_devices++;
465 }
Chris Masone5e9a522009-06-10 15:17:02 -0400466 mutex_unlock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500467 return fs_devices;
468error:
Chris Masone5e9a522009-06-10 15:17:02 -0400469 mutex_unlock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500470 free_fs_devices(fs_devices);
471 return ERR_PTR(-ENOMEM);
472}
473
Chris Masondfe25022008-05-13 13:46:40 -0400474int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
475{
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500476 struct btrfs_device *device, *next;
Chris Masondfe25022008-05-13 13:46:40 -0400477
478 mutex_lock(&uuid_mutex);
479again:
Chris Masone5e9a522009-06-10 15:17:02 -0400480 mutex_lock(&fs_devices->device_list_mutex);
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500481 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
Yan Zheng2b820322008-11-17 21:11:30 -0500482 if (device->in_fs_metadata)
483 continue;
484
485 if (device->bdev) {
Chris Mason15916de2008-11-19 21:17:22 -0500486 close_bdev_exclusive(device->bdev, device->mode);
Yan Zheng2b820322008-11-17 21:11:30 -0500487 device->bdev = NULL;
488 fs_devices->open_devices--;
489 }
490 if (device->writeable) {
491 list_del_init(&device->dev_alloc_list);
492 device->writeable = 0;
493 fs_devices->rw_devices--;
494 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500495 list_del_init(&device->dev_list);
496 fs_devices->num_devices--;
497 kfree(device->name);
498 kfree(device);
Chris Masondfe25022008-05-13 13:46:40 -0400499 }
Chris Masone5e9a522009-06-10 15:17:02 -0400500 mutex_unlock(&fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -0500501
502 if (fs_devices->seed) {
503 fs_devices = fs_devices->seed;
Yan Zheng2b820322008-11-17 21:11:30 -0500504 goto again;
505 }
506
Chris Masondfe25022008-05-13 13:46:40 -0400507 mutex_unlock(&uuid_mutex);
508 return 0;
509}
Chris Masona0af4692008-05-13 16:03:06 -0400510
Yan Zheng2b820322008-11-17 21:11:30 -0500511static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400512{
Chris Mason8a4b83c2008-03-24 15:02:07 -0400513 struct btrfs_device *device;
Yan Zhenge4404d62008-12-12 10:03:26 -0500514
Yan Zheng2b820322008-11-17 21:11:30 -0500515 if (--fs_devices->opened > 0)
516 return 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400517
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500518 list_for_each_entry(device, &fs_devices->devices, dev_list) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400519 if (device->bdev) {
Chris Mason15916de2008-11-19 21:17:22 -0500520 close_bdev_exclusive(device->bdev, device->mode);
Chris Masona0af4692008-05-13 16:03:06 -0400521 fs_devices->open_devices--;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400522 }
Yan Zheng2b820322008-11-17 21:11:30 -0500523 if (device->writeable) {
524 list_del_init(&device->dev_alloc_list);
525 fs_devices->rw_devices--;
526 }
527
Chris Mason8a4b83c2008-03-24 15:02:07 -0400528 device->bdev = NULL;
Yan Zheng2b820322008-11-17 21:11:30 -0500529 device->writeable = 0;
Chris Masondfe25022008-05-13 13:46:40 -0400530 device->in_fs_metadata = 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400531 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500532 WARN_ON(fs_devices->open_devices);
533 WARN_ON(fs_devices->rw_devices);
Yan Zheng2b820322008-11-17 21:11:30 -0500534 fs_devices->opened = 0;
535 fs_devices->seeding = 0;
Yan Zheng2b820322008-11-17 21:11:30 -0500536
Chris Mason8a4b83c2008-03-24 15:02:07 -0400537 return 0;
538}
539
Yan Zheng2b820322008-11-17 21:11:30 -0500540int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
541{
Yan Zhenge4404d62008-12-12 10:03:26 -0500542 struct btrfs_fs_devices *seed_devices = NULL;
Yan Zheng2b820322008-11-17 21:11:30 -0500543 int ret;
544
545 mutex_lock(&uuid_mutex);
546 ret = __btrfs_close_devices(fs_devices);
Yan Zhenge4404d62008-12-12 10:03:26 -0500547 if (!fs_devices->opened) {
548 seed_devices = fs_devices->seed;
549 fs_devices->seed = NULL;
550 }
Yan Zheng2b820322008-11-17 21:11:30 -0500551 mutex_unlock(&uuid_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500552
553 while (seed_devices) {
554 fs_devices = seed_devices;
555 seed_devices = fs_devices->seed;
556 __btrfs_close_devices(fs_devices);
557 free_fs_devices(fs_devices);
558 }
Yan Zheng2b820322008-11-17 21:11:30 -0500559 return ret;
560}
561
Yan Zhenge4404d62008-12-12 10:03:26 -0500562static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
563 fmode_t flags, void *holder)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400564{
565 struct block_device *bdev;
566 struct list_head *head = &fs_devices->devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400567 struct btrfs_device *device;
Chris Masona0af4692008-05-13 16:03:06 -0400568 struct block_device *latest_bdev = NULL;
569 struct buffer_head *bh;
570 struct btrfs_super_block *disk_super;
571 u64 latest_devid = 0;
572 u64 latest_transid = 0;
Chris Masona0af4692008-05-13 16:03:06 -0400573 u64 devid;
Yan Zheng2b820322008-11-17 21:11:30 -0500574 int seeding = 1;
Chris Masona0af4692008-05-13 16:03:06 -0400575 int ret = 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400576
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500577 list_for_each_entry(device, head, dev_list) {
Chris Masonc1c4d912008-05-08 15:05:58 -0400578 if (device->bdev)
579 continue;
Chris Masondfe25022008-05-13 13:46:40 -0400580 if (!device->name)
581 continue;
582
Chris Mason15916de2008-11-19 21:17:22 -0500583 bdev = open_bdev_exclusive(device->name, flags, holder);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400584 if (IS_ERR(bdev)) {
Chris Masond3977122009-01-05 21:25:51 -0500585 printk(KERN_INFO "open %s failed\n", device->name);
Chris Masona0af4692008-05-13 16:03:06 -0400586 goto error;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400587 }
Chris Masona061fc82008-05-07 11:43:44 -0400588 set_blocksize(bdev, 4096);
Chris Masona0af4692008-05-13 16:03:06 -0400589
Yan Zhenga512bbf2008-12-08 16:46:26 -0500590 bh = btrfs_read_dev_super(bdev);
Chris Masona0af4692008-05-13 16:03:06 -0400591 if (!bh)
592 goto error_close;
593
594 disk_super = (struct btrfs_super_block *)bh->b_data;
Chris Masona0af4692008-05-13 16:03:06 -0400595 devid = le64_to_cpu(disk_super->dev_item.devid);
596 if (devid != device->devid)
597 goto error_brelse;
598
Yan Zheng2b820322008-11-17 21:11:30 -0500599 if (memcmp(device->uuid, disk_super->dev_item.uuid,
600 BTRFS_UUID_SIZE))
601 goto error_brelse;
602
603 device->generation = btrfs_super_generation(disk_super);
604 if (!latest_transid || device->generation > latest_transid) {
Chris Masona0af4692008-05-13 16:03:06 -0400605 latest_devid = devid;
Yan Zheng2b820322008-11-17 21:11:30 -0500606 latest_transid = device->generation;
Chris Masona0af4692008-05-13 16:03:06 -0400607 latest_bdev = bdev;
608 }
609
Yan Zheng2b820322008-11-17 21:11:30 -0500610 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
611 device->writeable = 0;
612 } else {
613 device->writeable = !bdev_read_only(bdev);
614 seeding = 0;
615 }
616
Chris Mason8a4b83c2008-03-24 15:02:07 -0400617 device->bdev = bdev;
Chris Masondfe25022008-05-13 13:46:40 -0400618 device->in_fs_metadata = 0;
Chris Mason15916de2008-11-19 21:17:22 -0500619 device->mode = flags;
620
Chris Masonc2898112009-06-10 09:51:32 -0400621 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
622 fs_devices->rotating = 1;
623
Chris Masona0af4692008-05-13 16:03:06 -0400624 fs_devices->open_devices++;
Yan Zheng2b820322008-11-17 21:11:30 -0500625 if (device->writeable) {
626 fs_devices->rw_devices++;
627 list_add(&device->dev_alloc_list,
628 &fs_devices->alloc_list);
629 }
Chris Masona0af4692008-05-13 16:03:06 -0400630 continue;
Chris Masona061fc82008-05-07 11:43:44 -0400631
Chris Masona0af4692008-05-13 16:03:06 -0400632error_brelse:
633 brelse(bh);
634error_close:
Christoph Hellwig97288f22008-12-02 06:36:09 -0500635 close_bdev_exclusive(bdev, FMODE_READ);
Chris Masona0af4692008-05-13 16:03:06 -0400636error:
637 continue;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400638 }
Chris Masona0af4692008-05-13 16:03:06 -0400639 if (fs_devices->open_devices == 0) {
640 ret = -EIO;
641 goto out;
642 }
Yan Zheng2b820322008-11-17 21:11:30 -0500643 fs_devices->seeding = seeding;
644 fs_devices->opened = 1;
Chris Masona0af4692008-05-13 16:03:06 -0400645 fs_devices->latest_bdev = latest_bdev;
646 fs_devices->latest_devid = latest_devid;
647 fs_devices->latest_trans = latest_transid;
Yan Zheng2b820322008-11-17 21:11:30 -0500648 fs_devices->total_rw_bytes = 0;
Chris Masona0af4692008-05-13 16:03:06 -0400649out:
Yan Zheng2b820322008-11-17 21:11:30 -0500650 return ret;
651}
652
653int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
Christoph Hellwig97288f22008-12-02 06:36:09 -0500654 fmode_t flags, void *holder)
Yan Zheng2b820322008-11-17 21:11:30 -0500655{
656 int ret;
657
658 mutex_lock(&uuid_mutex);
659 if (fs_devices->opened) {
Yan Zhenge4404d62008-12-12 10:03:26 -0500660 fs_devices->opened++;
661 ret = 0;
Yan Zheng2b820322008-11-17 21:11:30 -0500662 } else {
Chris Mason15916de2008-11-19 21:17:22 -0500663 ret = __btrfs_open_devices(fs_devices, flags, holder);
Yan Zheng2b820322008-11-17 21:11:30 -0500664 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400665 mutex_unlock(&uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400666 return ret;
667}
668
Christoph Hellwig97288f22008-12-02 06:36:09 -0500669int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
Chris Mason8a4b83c2008-03-24 15:02:07 -0400670 struct btrfs_fs_devices **fs_devices_ret)
671{
672 struct btrfs_super_block *disk_super;
673 struct block_device *bdev;
674 struct buffer_head *bh;
675 int ret;
676 u64 devid;
Chris Masonf2984462008-04-10 16:19:33 -0400677 u64 transid;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400678
679 mutex_lock(&uuid_mutex);
680
Chris Mason15916de2008-11-19 21:17:22 -0500681 bdev = open_bdev_exclusive(path, flags, holder);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400682
683 if (IS_ERR(bdev)) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400684 ret = PTR_ERR(bdev);
685 goto error;
686 }
687
688 ret = set_blocksize(bdev, 4096);
689 if (ret)
690 goto error_close;
Yan Zhenga512bbf2008-12-08 16:46:26 -0500691 bh = btrfs_read_dev_super(bdev);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400692 if (!bh) {
693 ret = -EIO;
694 goto error_close;
695 }
696 disk_super = (struct btrfs_super_block *)bh->b_data;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400697 devid = le64_to_cpu(disk_super->dev_item.devid);
Chris Masonf2984462008-04-10 16:19:33 -0400698 transid = btrfs_super_generation(disk_super);
Chris Mason7ae9c092008-04-18 10:29:49 -0400699 if (disk_super->label[0])
Chris Masond3977122009-01-05 21:25:51 -0500700 printk(KERN_INFO "device label %s ", disk_super->label);
Chris Mason7ae9c092008-04-18 10:29:49 -0400701 else {
702 /* FIXME, make a readl uuid parser */
Chris Masond3977122009-01-05 21:25:51 -0500703 printk(KERN_INFO "device fsid %llx-%llx ",
Chris Mason7ae9c092008-04-18 10:29:49 -0400704 *(unsigned long long *)disk_super->fsid,
705 *(unsigned long long *)(disk_super->fsid + 8));
706 }
Roland Dreier119e10c2009-01-21 10:49:16 -0500707 printk(KERN_CONT "devid %llu transid %llu %s\n",
Chris Masond3977122009-01-05 21:25:51 -0500708 (unsigned long long)devid, (unsigned long long)transid, path);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400709 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
710
Chris Mason8a4b83c2008-03-24 15:02:07 -0400711 brelse(bh);
712error_close:
Chris Mason15916de2008-11-19 21:17:22 -0500713 close_bdev_exclusive(bdev, flags);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400714error:
715 mutex_unlock(&uuid_mutex);
716 return ret;
717}
Chris Mason0b86a832008-03-24 15:01:56 -0400718
719/*
720 * this uses a pretty simple search, the expectation is that it is
721 * called very infrequently and that a given device has a small number
722 * of extents
723 */
Josef Bacikba1bf482009-09-11 16:11:19 -0400724int find_free_dev_extent(struct btrfs_trans_handle *trans,
725 struct btrfs_device *device, u64 num_bytes,
726 u64 *start, u64 *max_avail)
Chris Mason0b86a832008-03-24 15:01:56 -0400727{
728 struct btrfs_key key;
729 struct btrfs_root *root = device->dev_root;
730 struct btrfs_dev_extent *dev_extent = NULL;
Yan Zheng2b820322008-11-17 21:11:30 -0500731 struct btrfs_path *path;
Chris Mason0b86a832008-03-24 15:01:56 -0400732 u64 hole_size = 0;
733 u64 last_byte = 0;
734 u64 search_start = 0;
735 u64 search_end = device->total_bytes;
736 int ret;
737 int slot = 0;
738 int start_found;
739 struct extent_buffer *l;
740
Yan Zheng2b820322008-11-17 21:11:30 -0500741 path = btrfs_alloc_path();
742 if (!path)
743 return -ENOMEM;
Chris Mason0b86a832008-03-24 15:01:56 -0400744 path->reada = 2;
Yan Zheng2b820322008-11-17 21:11:30 -0500745 start_found = 0;
Chris Mason0b86a832008-03-24 15:01:56 -0400746
747 /* FIXME use last free of some kind */
748
Chris Mason8a4b83c2008-03-24 15:02:07 -0400749 /* we don't want to overwrite the superblock on the drive,
750 * so we make sure to start at an offset of at least 1MB
751 */
752 search_start = max((u64)1024 * 1024, search_start);
Chris Mason8f18cf12008-04-25 16:53:30 -0400753
754 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
755 search_start = max(root->fs_info->alloc_start, search_start);
756
Chris Mason0b86a832008-03-24 15:01:56 -0400757 key.objectid = device->devid;
758 key.offset = search_start;
759 key.type = BTRFS_DEV_EXTENT_KEY;
760 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
761 if (ret < 0)
762 goto error;
Yan Zheng1fcbac52009-07-24 11:06:53 -0400763 if (ret > 0) {
764 ret = btrfs_previous_item(root, path, key.objectid, key.type);
765 if (ret < 0)
766 goto error;
767 if (ret > 0)
768 start_found = 1;
769 }
Chris Mason0b86a832008-03-24 15:01:56 -0400770 l = path->nodes[0];
771 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
772 while (1) {
773 l = path->nodes[0];
774 slot = path->slots[0];
775 if (slot >= btrfs_header_nritems(l)) {
776 ret = btrfs_next_leaf(root, path);
777 if (ret == 0)
778 continue;
779 if (ret < 0)
780 goto error;
781no_more_items:
782 if (!start_found) {
783 if (search_start >= search_end) {
784 ret = -ENOSPC;
785 goto error;
786 }
787 *start = search_start;
788 start_found = 1;
789 goto check_pending;
790 }
791 *start = last_byte > search_start ?
792 last_byte : search_start;
793 if (search_end <= *start) {
794 ret = -ENOSPC;
795 goto error;
796 }
797 goto check_pending;
798 }
799 btrfs_item_key_to_cpu(l, &key, slot);
800
801 if (key.objectid < device->devid)
802 goto next;
803
804 if (key.objectid > device->devid)
805 goto no_more_items;
806
807 if (key.offset >= search_start && key.offset > last_byte &&
808 start_found) {
809 if (last_byte < search_start)
810 last_byte = search_start;
811 hole_size = key.offset - last_byte;
Chris Mason9779b722009-07-24 16:41:41 -0400812
813 if (hole_size > *max_avail)
814 *max_avail = hole_size;
815
Chris Mason0b86a832008-03-24 15:01:56 -0400816 if (key.offset > last_byte &&
817 hole_size >= num_bytes) {
818 *start = last_byte;
819 goto check_pending;
820 }
821 }
Chris Masond3977122009-01-05 21:25:51 -0500822 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
Chris Mason0b86a832008-03-24 15:01:56 -0400823 goto next;
Chris Mason0b86a832008-03-24 15:01:56 -0400824
825 start_found = 1;
826 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
827 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
828next:
829 path->slots[0]++;
830 cond_resched();
831 }
832check_pending:
833 /* we have to make sure we didn't find an extent that has already
834 * been allocated by the map tree or the original allocation
835 */
Chris Mason0b86a832008-03-24 15:01:56 -0400836 BUG_ON(*start < search_start);
837
Chris Mason6324fbf2008-03-24 15:01:59 -0400838 if (*start + num_bytes > search_end) {
Chris Mason0b86a832008-03-24 15:01:56 -0400839 ret = -ENOSPC;
840 goto error;
841 }
842 /* check for pending inserts here */
Yan Zheng2b820322008-11-17 21:11:30 -0500843 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -0400844
845error:
Yan Zheng2b820322008-11-17 21:11:30 -0500846 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -0400847 return ret;
848}
849
Christoph Hellwigb2950862008-12-02 09:54:17 -0500850static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
Chris Mason8f18cf12008-04-25 16:53:30 -0400851 struct btrfs_device *device,
852 u64 start)
853{
854 int ret;
855 struct btrfs_path *path;
856 struct btrfs_root *root = device->dev_root;
857 struct btrfs_key key;
Chris Masona061fc82008-05-07 11:43:44 -0400858 struct btrfs_key found_key;
859 struct extent_buffer *leaf = NULL;
860 struct btrfs_dev_extent *extent = NULL;
Chris Mason8f18cf12008-04-25 16:53:30 -0400861
862 path = btrfs_alloc_path();
863 if (!path)
864 return -ENOMEM;
865
866 key.objectid = device->devid;
867 key.offset = start;
868 key.type = BTRFS_DEV_EXTENT_KEY;
869
870 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Chris Masona061fc82008-05-07 11:43:44 -0400871 if (ret > 0) {
872 ret = btrfs_previous_item(root, path, key.objectid,
873 BTRFS_DEV_EXTENT_KEY);
874 BUG_ON(ret);
875 leaf = path->nodes[0];
876 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
877 extent = btrfs_item_ptr(leaf, path->slots[0],
878 struct btrfs_dev_extent);
879 BUG_ON(found_key.offset > start || found_key.offset +
880 btrfs_dev_extent_length(leaf, extent) < start);
881 ret = 0;
882 } else if (ret == 0) {
883 leaf = path->nodes[0];
884 extent = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_dev_extent);
886 }
Chris Mason8f18cf12008-04-25 16:53:30 -0400887 BUG_ON(ret);
888
Chris Masondfe25022008-05-13 13:46:40 -0400889 if (device->bytes_used > 0)
890 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
Chris Mason8f18cf12008-04-25 16:53:30 -0400891 ret = btrfs_del_item(trans, root, path);
892 BUG_ON(ret);
893
894 btrfs_free_path(path);
895 return ret;
896}
897
Yan Zheng2b820322008-11-17 21:11:30 -0500898int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
Chris Mason0b86a832008-03-24 15:01:56 -0400899 struct btrfs_device *device,
Chris Masone17cade2008-04-15 15:41:47 -0400900 u64 chunk_tree, u64 chunk_objectid,
Yan Zheng2b820322008-11-17 21:11:30 -0500901 u64 chunk_offset, u64 start, u64 num_bytes)
Chris Mason0b86a832008-03-24 15:01:56 -0400902{
903 int ret;
904 struct btrfs_path *path;
905 struct btrfs_root *root = device->dev_root;
906 struct btrfs_dev_extent *extent;
907 struct extent_buffer *leaf;
908 struct btrfs_key key;
909
Chris Masondfe25022008-05-13 13:46:40 -0400910 WARN_ON(!device->in_fs_metadata);
Chris Mason0b86a832008-03-24 15:01:56 -0400911 path = btrfs_alloc_path();
912 if (!path)
913 return -ENOMEM;
914
Chris Mason0b86a832008-03-24 15:01:56 -0400915 key.objectid = device->devid;
Yan Zheng2b820322008-11-17 21:11:30 -0500916 key.offset = start;
Chris Mason0b86a832008-03-24 15:01:56 -0400917 key.type = BTRFS_DEV_EXTENT_KEY;
918 ret = btrfs_insert_empty_item(trans, root, path, &key,
919 sizeof(*extent));
920 BUG_ON(ret);
921
922 leaf = path->nodes[0];
923 extent = btrfs_item_ptr(leaf, path->slots[0],
924 struct btrfs_dev_extent);
Chris Masone17cade2008-04-15 15:41:47 -0400925 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
926 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
927 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
928
929 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
930 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
931 BTRFS_UUID_SIZE);
932
Chris Mason0b86a832008-03-24 15:01:56 -0400933 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
934 btrfs_mark_buffer_dirty(leaf);
Chris Mason0b86a832008-03-24 15:01:56 -0400935 btrfs_free_path(path);
936 return ret;
937}
938
Chris Masona1b32a52008-09-05 16:09:51 -0400939static noinline int find_next_chunk(struct btrfs_root *root,
940 u64 objectid, u64 *offset)
Chris Mason0b86a832008-03-24 15:01:56 -0400941{
942 struct btrfs_path *path;
943 int ret;
944 struct btrfs_key key;
Chris Masone17cade2008-04-15 15:41:47 -0400945 struct btrfs_chunk *chunk;
Chris Mason0b86a832008-03-24 15:01:56 -0400946 struct btrfs_key found_key;
947
948 path = btrfs_alloc_path();
949 BUG_ON(!path);
950
Chris Masone17cade2008-04-15 15:41:47 -0400951 key.objectid = objectid;
Chris Mason0b86a832008-03-24 15:01:56 -0400952 key.offset = (u64)-1;
953 key.type = BTRFS_CHUNK_ITEM_KEY;
954
955 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
956 if (ret < 0)
957 goto error;
958
959 BUG_ON(ret == 0);
960
961 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
962 if (ret) {
Chris Masone17cade2008-04-15 15:41:47 -0400963 *offset = 0;
Chris Mason0b86a832008-03-24 15:01:56 -0400964 } else {
965 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
966 path->slots[0]);
Chris Masone17cade2008-04-15 15:41:47 -0400967 if (found_key.objectid != objectid)
968 *offset = 0;
969 else {
970 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
971 struct btrfs_chunk);
972 *offset = found_key.offset +
973 btrfs_chunk_length(path->nodes[0], chunk);
974 }
Chris Mason0b86a832008-03-24 15:01:56 -0400975 }
976 ret = 0;
977error:
978 btrfs_free_path(path);
979 return ret;
980}
981
Yan Zheng2b820322008-11-17 21:11:30 -0500982static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
Chris Mason0b86a832008-03-24 15:01:56 -0400983{
984 int ret;
985 struct btrfs_key key;
986 struct btrfs_key found_key;
Yan Zheng2b820322008-11-17 21:11:30 -0500987 struct btrfs_path *path;
988
989 root = root->fs_info->chunk_root;
990
991 path = btrfs_alloc_path();
992 if (!path)
993 return -ENOMEM;
Chris Mason0b86a832008-03-24 15:01:56 -0400994
995 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
996 key.type = BTRFS_DEV_ITEM_KEY;
997 key.offset = (u64)-1;
998
999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1000 if (ret < 0)
1001 goto error;
1002
1003 BUG_ON(ret == 0);
1004
1005 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1006 BTRFS_DEV_ITEM_KEY);
1007 if (ret) {
1008 *objectid = 1;
1009 } else {
1010 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1011 path->slots[0]);
1012 *objectid = found_key.offset + 1;
1013 }
1014 ret = 0;
1015error:
Yan Zheng2b820322008-11-17 21:11:30 -05001016 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04001017 return ret;
1018}
1019
1020/*
1021 * the device information is stored in the chunk root
1022 * the btrfs_device struct should be fully filled in
1023 */
1024int btrfs_add_device(struct btrfs_trans_handle *trans,
1025 struct btrfs_root *root,
1026 struct btrfs_device *device)
1027{
1028 int ret;
1029 struct btrfs_path *path;
1030 struct btrfs_dev_item *dev_item;
1031 struct extent_buffer *leaf;
1032 struct btrfs_key key;
1033 unsigned long ptr;
Chris Mason0b86a832008-03-24 15:01:56 -04001034
1035 root = root->fs_info->chunk_root;
1036
1037 path = btrfs_alloc_path();
1038 if (!path)
1039 return -ENOMEM;
1040
Chris Mason0b86a832008-03-24 15:01:56 -04001041 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1042 key.type = BTRFS_DEV_ITEM_KEY;
Yan Zheng2b820322008-11-17 21:11:30 -05001043 key.offset = device->devid;
Chris Mason0b86a832008-03-24 15:01:56 -04001044
1045 ret = btrfs_insert_empty_item(trans, root, path, &key,
Chris Mason0d81ba52008-03-24 15:02:07 -04001046 sizeof(*dev_item));
Chris Mason0b86a832008-03-24 15:01:56 -04001047 if (ret)
1048 goto out;
1049
1050 leaf = path->nodes[0];
1051 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1052
1053 btrfs_set_device_id(leaf, dev_item, device->devid);
Yan Zheng2b820322008-11-17 21:11:30 -05001054 btrfs_set_device_generation(leaf, dev_item, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001055 btrfs_set_device_type(leaf, dev_item, device->type);
1056 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1057 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1058 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
Chris Mason0b86a832008-03-24 15:01:56 -04001059 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1060 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
Chris Masone17cade2008-04-15 15:41:47 -04001061 btrfs_set_device_group(leaf, dev_item, 0);
1062 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1063 btrfs_set_device_bandwidth(leaf, dev_item, 0);
Chris Masonc3027eb2008-12-08 16:40:21 -05001064 btrfs_set_device_start_offset(leaf, dev_item, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001065
Chris Mason0b86a832008-03-24 15:01:56 -04001066 ptr = (unsigned long)btrfs_device_uuid(dev_item);
Chris Masone17cade2008-04-15 15:41:47 -04001067 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
Yan Zheng2b820322008-11-17 21:11:30 -05001068 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1069 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04001070 btrfs_mark_buffer_dirty(leaf);
Chris Mason0b86a832008-03-24 15:01:56 -04001071
Yan Zheng2b820322008-11-17 21:11:30 -05001072 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04001073out:
1074 btrfs_free_path(path);
1075 return ret;
1076}
Chris Mason8f18cf12008-04-25 16:53:30 -04001077
Chris Masona061fc82008-05-07 11:43:44 -04001078static int btrfs_rm_dev_item(struct btrfs_root *root,
1079 struct btrfs_device *device)
1080{
1081 int ret;
1082 struct btrfs_path *path;
Chris Masona061fc82008-05-07 11:43:44 -04001083 struct btrfs_key key;
Chris Masona061fc82008-05-07 11:43:44 -04001084 struct btrfs_trans_handle *trans;
1085
1086 root = root->fs_info->chunk_root;
1087
1088 path = btrfs_alloc_path();
1089 if (!path)
1090 return -ENOMEM;
1091
1092 trans = btrfs_start_transaction(root, 1);
1093 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1094 key.type = BTRFS_DEV_ITEM_KEY;
1095 key.offset = device->devid;
Chris Mason7d9eb122008-07-08 14:19:17 -04001096 lock_chunks(root);
Chris Masona061fc82008-05-07 11:43:44 -04001097
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099 if (ret < 0)
1100 goto out;
1101
1102 if (ret > 0) {
1103 ret = -ENOENT;
1104 goto out;
1105 }
1106
1107 ret = btrfs_del_item(trans, root, path);
1108 if (ret)
1109 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001110out:
1111 btrfs_free_path(path);
Chris Mason7d9eb122008-07-08 14:19:17 -04001112 unlock_chunks(root);
Chris Masona061fc82008-05-07 11:43:44 -04001113 btrfs_commit_transaction(trans, root);
1114 return ret;
1115}
1116
1117int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1118{
1119 struct btrfs_device *device;
Yan Zheng2b820322008-11-17 21:11:30 -05001120 struct btrfs_device *next_device;
Chris Masona061fc82008-05-07 11:43:44 -04001121 struct block_device *bdev;
Chris Masondfe25022008-05-13 13:46:40 -04001122 struct buffer_head *bh = NULL;
Chris Masona061fc82008-05-07 11:43:44 -04001123 struct btrfs_super_block *disk_super;
1124 u64 all_avail;
1125 u64 devid;
Yan Zheng2b820322008-11-17 21:11:30 -05001126 u64 num_devices;
1127 u8 *dev_uuid;
Chris Masona061fc82008-05-07 11:43:44 -04001128 int ret = 0;
1129
Chris Masona061fc82008-05-07 11:43:44 -04001130 mutex_lock(&uuid_mutex);
Chris Mason7d9eb122008-07-08 14:19:17 -04001131 mutex_lock(&root->fs_info->volume_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001132
1133 all_avail = root->fs_info->avail_data_alloc_bits |
1134 root->fs_info->avail_system_alloc_bits |
1135 root->fs_info->avail_metadata_alloc_bits;
1136
1137 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
Yan Zheng2b820322008-11-17 21:11:30 -05001138 root->fs_info->fs_devices->rw_devices <= 4) {
Chris Masond3977122009-01-05 21:25:51 -05001139 printk(KERN_ERR "btrfs: unable to go below four devices "
1140 "on raid10\n");
Chris Masona061fc82008-05-07 11:43:44 -04001141 ret = -EINVAL;
1142 goto out;
1143 }
1144
1145 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
Yan Zheng2b820322008-11-17 21:11:30 -05001146 root->fs_info->fs_devices->rw_devices <= 2) {
Chris Masond3977122009-01-05 21:25:51 -05001147 printk(KERN_ERR "btrfs: unable to go below two "
1148 "devices on raid1\n");
Chris Masona061fc82008-05-07 11:43:44 -04001149 ret = -EINVAL;
1150 goto out;
1151 }
1152
Chris Masondfe25022008-05-13 13:46:40 -04001153 if (strcmp(device_path, "missing") == 0) {
Chris Masondfe25022008-05-13 13:46:40 -04001154 struct list_head *devices;
1155 struct btrfs_device *tmp;
Chris Masona061fc82008-05-07 11:43:44 -04001156
Chris Masondfe25022008-05-13 13:46:40 -04001157 device = NULL;
1158 devices = &root->fs_info->fs_devices->devices;
Chris Masone5e9a522009-06-10 15:17:02 -04001159 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001160 list_for_each_entry(tmp, devices, dev_list) {
Chris Masondfe25022008-05-13 13:46:40 -04001161 if (tmp->in_fs_metadata && !tmp->bdev) {
1162 device = tmp;
1163 break;
1164 }
1165 }
Chris Masone5e9a522009-06-10 15:17:02 -04001166 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Chris Masondfe25022008-05-13 13:46:40 -04001167 bdev = NULL;
1168 bh = NULL;
1169 disk_super = NULL;
1170 if (!device) {
Chris Masond3977122009-01-05 21:25:51 -05001171 printk(KERN_ERR "btrfs: no missing devices found to "
1172 "remove\n");
Chris Masondfe25022008-05-13 13:46:40 -04001173 goto out;
1174 }
Chris Masondfe25022008-05-13 13:46:40 -04001175 } else {
Christoph Hellwig97288f22008-12-02 06:36:09 -05001176 bdev = open_bdev_exclusive(device_path, FMODE_READ,
Chris Masondfe25022008-05-13 13:46:40 -04001177 root->fs_info->bdev_holder);
1178 if (IS_ERR(bdev)) {
1179 ret = PTR_ERR(bdev);
1180 goto out;
1181 }
1182
Yan Zheng2b820322008-11-17 21:11:30 -05001183 set_blocksize(bdev, 4096);
Yan Zhenga512bbf2008-12-08 16:46:26 -05001184 bh = btrfs_read_dev_super(bdev);
Chris Masondfe25022008-05-13 13:46:40 -04001185 if (!bh) {
1186 ret = -EIO;
1187 goto error_close;
1188 }
1189 disk_super = (struct btrfs_super_block *)bh->b_data;
Chris Masondfe25022008-05-13 13:46:40 -04001190 devid = le64_to_cpu(disk_super->dev_item.devid);
Yan Zheng2b820322008-11-17 21:11:30 -05001191 dev_uuid = disk_super->dev_item.uuid;
1192 device = btrfs_find_device(root, devid, dev_uuid,
1193 disk_super->fsid);
Chris Masondfe25022008-05-13 13:46:40 -04001194 if (!device) {
1195 ret = -ENOENT;
1196 goto error_brelse;
1197 }
Chris Masondfe25022008-05-13 13:46:40 -04001198 }
Yan Zheng2b820322008-11-17 21:11:30 -05001199
1200 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
Chris Masond3977122009-01-05 21:25:51 -05001201 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1202 "device\n");
Yan Zheng2b820322008-11-17 21:11:30 -05001203 ret = -EINVAL;
1204 goto error_brelse;
1205 }
1206
1207 if (device->writeable) {
1208 list_del_init(&device->dev_alloc_list);
1209 root->fs_info->fs_devices->rw_devices--;
1210 }
Chris Masona061fc82008-05-07 11:43:44 -04001211
1212 ret = btrfs_shrink_device(device, 0);
1213 if (ret)
1214 goto error_brelse;
1215
Chris Masona061fc82008-05-07 11:43:44 -04001216 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1217 if (ret)
1218 goto error_brelse;
1219
Yan Zheng2b820322008-11-17 21:11:30 -05001220 device->in_fs_metadata = 0;
Chris Masone5e9a522009-06-10 15:17:02 -04001221
1222 /*
1223 * the device list mutex makes sure that we don't change
1224 * the device list while someone else is writing out all
1225 * the device supers.
1226 */
1227 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -05001228 list_del_init(&device->dev_list);
Chris Masone5e9a522009-06-10 15:17:02 -04001229 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1230
Yan Zhenge4404d62008-12-12 10:03:26 -05001231 device->fs_devices->num_devices--;
Yan Zheng2b820322008-11-17 21:11:30 -05001232
1233 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1234 struct btrfs_device, dev_list);
1235 if (device->bdev == root->fs_info->sb->s_bdev)
1236 root->fs_info->sb->s_bdev = next_device->bdev;
1237 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1238 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1239
Yan Zhenge4404d62008-12-12 10:03:26 -05001240 if (device->bdev) {
1241 close_bdev_exclusive(device->bdev, device->mode);
1242 device->bdev = NULL;
1243 device->fs_devices->open_devices--;
1244 }
1245
Yan Zheng2b820322008-11-17 21:11:30 -05001246 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1247 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1248
Yan Zhenge4404d62008-12-12 10:03:26 -05001249 if (device->fs_devices->open_devices == 0) {
1250 struct btrfs_fs_devices *fs_devices;
1251 fs_devices = root->fs_info->fs_devices;
1252 while (fs_devices) {
1253 if (fs_devices->seed == device->fs_devices)
1254 break;
1255 fs_devices = fs_devices->seed;
Yan Zheng2b820322008-11-17 21:11:30 -05001256 }
Yan Zhenge4404d62008-12-12 10:03:26 -05001257 fs_devices->seed = device->fs_devices->seed;
1258 device->fs_devices->seed = NULL;
1259 __btrfs_close_devices(device->fs_devices);
1260 free_fs_devices(device->fs_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05001261 }
1262
1263 /*
1264 * at this point, the device is zero sized. We want to
1265 * remove it from the devices list and zero out the old super
1266 */
1267 if (device->writeable) {
Chris Masondfe25022008-05-13 13:46:40 -04001268 /* make sure this device isn't detected as part of
1269 * the FS anymore
1270 */
1271 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1272 set_buffer_dirty(bh);
1273 sync_dirty_buffer(bh);
Chris Masondfe25022008-05-13 13:46:40 -04001274 }
Chris Masona061fc82008-05-07 11:43:44 -04001275
Chris Masona061fc82008-05-07 11:43:44 -04001276 kfree(device->name);
1277 kfree(device);
1278 ret = 0;
Chris Masona061fc82008-05-07 11:43:44 -04001279
1280error_brelse:
1281 brelse(bh);
1282error_close:
Chris Masondfe25022008-05-13 13:46:40 -04001283 if (bdev)
Christoph Hellwig97288f22008-12-02 06:36:09 -05001284 close_bdev_exclusive(bdev, FMODE_READ);
Chris Masona061fc82008-05-07 11:43:44 -04001285out:
Chris Mason7d9eb122008-07-08 14:19:17 -04001286 mutex_unlock(&root->fs_info->volume_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001287 mutex_unlock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001288 return ret;
1289}
1290
Yan Zheng2b820322008-11-17 21:11:30 -05001291/*
1292 * does all the dirty work required for changing file system's UUID.
1293 */
1294static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1295 struct btrfs_root *root)
1296{
1297 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1298 struct btrfs_fs_devices *old_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -05001299 struct btrfs_fs_devices *seed_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05001300 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1301 struct btrfs_device *device;
1302 u64 super_flags;
1303
1304 BUG_ON(!mutex_is_locked(&uuid_mutex));
Yan Zhenge4404d62008-12-12 10:03:26 -05001305 if (!fs_devices->seeding)
Yan Zheng2b820322008-11-17 21:11:30 -05001306 return -EINVAL;
1307
Yan Zhenge4404d62008-12-12 10:03:26 -05001308 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1309 if (!seed_devices)
Yan Zheng2b820322008-11-17 21:11:30 -05001310 return -ENOMEM;
1311
Yan Zhenge4404d62008-12-12 10:03:26 -05001312 old_devices = clone_fs_devices(fs_devices);
1313 if (IS_ERR(old_devices)) {
1314 kfree(seed_devices);
1315 return PTR_ERR(old_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05001316 }
Yan Zhenge4404d62008-12-12 10:03:26 -05001317
Yan Zheng2b820322008-11-17 21:11:30 -05001318 list_add(&old_devices->list, &fs_uuids);
1319
Yan Zhenge4404d62008-12-12 10:03:26 -05001320 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1321 seed_devices->opened = 1;
1322 INIT_LIST_HEAD(&seed_devices->devices);
1323 INIT_LIST_HEAD(&seed_devices->alloc_list);
Chris Masone5e9a522009-06-10 15:17:02 -04001324 mutex_init(&seed_devices->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -05001325 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1326 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1327 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1328 device->fs_devices = seed_devices;
1329 }
1330
Yan Zheng2b820322008-11-17 21:11:30 -05001331 fs_devices->seeding = 0;
1332 fs_devices->num_devices = 0;
1333 fs_devices->open_devices = 0;
Yan Zhenge4404d62008-12-12 10:03:26 -05001334 fs_devices->seed = seed_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05001335
1336 generate_random_uuid(fs_devices->fsid);
1337 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1338 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1339 super_flags = btrfs_super_flags(disk_super) &
1340 ~BTRFS_SUPER_FLAG_SEEDING;
1341 btrfs_set_super_flags(disk_super, super_flags);
1342
1343 return 0;
1344}
1345
1346/*
1347 * strore the expected generation for seed devices in device items.
1348 */
1349static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1350 struct btrfs_root *root)
1351{
1352 struct btrfs_path *path;
1353 struct extent_buffer *leaf;
1354 struct btrfs_dev_item *dev_item;
1355 struct btrfs_device *device;
1356 struct btrfs_key key;
1357 u8 fs_uuid[BTRFS_UUID_SIZE];
1358 u8 dev_uuid[BTRFS_UUID_SIZE];
1359 u64 devid;
1360 int ret;
1361
1362 path = btrfs_alloc_path();
1363 if (!path)
1364 return -ENOMEM;
1365
1366 root = root->fs_info->chunk_root;
1367 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1368 key.offset = 0;
1369 key.type = BTRFS_DEV_ITEM_KEY;
1370
1371 while (1) {
1372 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1373 if (ret < 0)
1374 goto error;
1375
1376 leaf = path->nodes[0];
1377next_slot:
1378 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1379 ret = btrfs_next_leaf(root, path);
1380 if (ret > 0)
1381 break;
1382 if (ret < 0)
1383 goto error;
1384 leaf = path->nodes[0];
1385 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1386 btrfs_release_path(root, path);
1387 continue;
1388 }
1389
1390 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1391 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1392 key.type != BTRFS_DEV_ITEM_KEY)
1393 break;
1394
1395 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1396 struct btrfs_dev_item);
1397 devid = btrfs_device_id(leaf, dev_item);
1398 read_extent_buffer(leaf, dev_uuid,
1399 (unsigned long)btrfs_device_uuid(dev_item),
1400 BTRFS_UUID_SIZE);
1401 read_extent_buffer(leaf, fs_uuid,
1402 (unsigned long)btrfs_device_fsid(dev_item),
1403 BTRFS_UUID_SIZE);
1404 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1405 BUG_ON(!device);
1406
1407 if (device->fs_devices->seeding) {
1408 btrfs_set_device_generation(leaf, dev_item,
1409 device->generation);
1410 btrfs_mark_buffer_dirty(leaf);
1411 }
1412
1413 path->slots[0]++;
1414 goto next_slot;
1415 }
1416 ret = 0;
1417error:
1418 btrfs_free_path(path);
1419 return ret;
1420}
1421
Chris Mason788f20e2008-04-28 15:29:42 -04001422int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1423{
1424 struct btrfs_trans_handle *trans;
1425 struct btrfs_device *device;
1426 struct block_device *bdev;
Chris Mason788f20e2008-04-28 15:29:42 -04001427 struct list_head *devices;
Yan Zheng2b820322008-11-17 21:11:30 -05001428 struct super_block *sb = root->fs_info->sb;
Chris Mason788f20e2008-04-28 15:29:42 -04001429 u64 total_bytes;
Yan Zheng2b820322008-11-17 21:11:30 -05001430 int seeding_dev = 0;
Chris Mason788f20e2008-04-28 15:29:42 -04001431 int ret = 0;
1432
Yan Zheng2b820322008-11-17 21:11:30 -05001433 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1434 return -EINVAL;
Chris Mason788f20e2008-04-28 15:29:42 -04001435
Chris Mason15916de2008-11-19 21:17:22 -05001436 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
Chris Masond3977122009-01-05 21:25:51 -05001437 if (!bdev)
Chris Mason788f20e2008-04-28 15:29:42 -04001438 return -EIO;
Chris Masona2135012008-06-25 16:01:30 -04001439
Yan Zheng2b820322008-11-17 21:11:30 -05001440 if (root->fs_info->fs_devices->seeding) {
1441 seeding_dev = 1;
1442 down_write(&sb->s_umount);
1443 mutex_lock(&uuid_mutex);
1444 }
1445
Chris Mason8c8bee12008-09-29 11:19:10 -04001446 filemap_write_and_wait(bdev->bd_inode->i_mapping);
Chris Mason7d9eb122008-07-08 14:19:17 -04001447 mutex_lock(&root->fs_info->volume_mutex);
Chris Masona2135012008-06-25 16:01:30 -04001448
Chris Mason788f20e2008-04-28 15:29:42 -04001449 devices = &root->fs_info->fs_devices->devices;
Chris Masone5e9a522009-06-10 15:17:02 -04001450 /*
1451 * we have the volume lock, so we don't need the extra
1452 * device list mutex while reading the list here.
1453 */
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001454 list_for_each_entry(device, devices, dev_list) {
Chris Mason788f20e2008-04-28 15:29:42 -04001455 if (device->bdev == bdev) {
1456 ret = -EEXIST;
Yan Zheng2b820322008-11-17 21:11:30 -05001457 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04001458 }
1459 }
1460
1461 device = kzalloc(sizeof(*device), GFP_NOFS);
1462 if (!device) {
1463 /* we can safely leave the fs_devices entry around */
1464 ret = -ENOMEM;
Yan Zheng2b820322008-11-17 21:11:30 -05001465 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04001466 }
1467
Chris Mason788f20e2008-04-28 15:29:42 -04001468 device->name = kstrdup(device_path, GFP_NOFS);
1469 if (!device->name) {
1470 kfree(device);
Yan Zheng2b820322008-11-17 21:11:30 -05001471 ret = -ENOMEM;
1472 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04001473 }
Yan Zheng2b820322008-11-17 21:11:30 -05001474
1475 ret = find_next_devid(root, &device->devid);
1476 if (ret) {
1477 kfree(device);
1478 goto error;
1479 }
1480
1481 trans = btrfs_start_transaction(root, 1);
1482 lock_chunks(root);
1483
1484 device->barriers = 1;
1485 device->writeable = 1;
1486 device->work.func = pending_bios_fn;
1487 generate_random_uuid(device->uuid);
1488 spin_lock_init(&device->io_lock);
1489 device->generation = trans->transid;
Chris Mason788f20e2008-04-28 15:29:42 -04001490 device->io_width = root->sectorsize;
1491 device->io_align = root->sectorsize;
1492 device->sector_size = root->sectorsize;
1493 device->total_bytes = i_size_read(bdev->bd_inode);
Yan Zheng2cc3c552009-06-04 09:23:50 -04001494 device->disk_total_bytes = device->total_bytes;
Chris Mason788f20e2008-04-28 15:29:42 -04001495 device->dev_root = root->fs_info->dev_root;
1496 device->bdev = bdev;
Chris Masondfe25022008-05-13 13:46:40 -04001497 device->in_fs_metadata = 1;
Chris Mason15916de2008-11-19 21:17:22 -05001498 device->mode = 0;
Zheng Yan325cd4b2008-09-05 16:43:54 -04001499 set_blocksize(device->bdev, 4096);
1500
Yan Zheng2b820322008-11-17 21:11:30 -05001501 if (seeding_dev) {
1502 sb->s_flags &= ~MS_RDONLY;
1503 ret = btrfs_prepare_sprout(trans, root);
1504 BUG_ON(ret);
1505 }
1506
1507 device->fs_devices = root->fs_info->fs_devices;
Chris Masone5e9a522009-06-10 15:17:02 -04001508
1509 /*
1510 * we don't want write_supers to jump in here with our device
1511 * half setup
1512 */
1513 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05001514 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1515 list_add(&device->dev_alloc_list,
1516 &root->fs_info->fs_devices->alloc_list);
1517 root->fs_info->fs_devices->num_devices++;
1518 root->fs_info->fs_devices->open_devices++;
1519 root->fs_info->fs_devices->rw_devices++;
1520 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1521
Chris Masonc2898112009-06-10 09:51:32 -04001522 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1523 root->fs_info->fs_devices->rotating = 1;
1524
Chris Mason788f20e2008-04-28 15:29:42 -04001525 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1526 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1527 total_bytes + device->total_bytes);
1528
1529 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1530 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1531 total_bytes + 1);
Chris Masone5e9a522009-06-10 15:17:02 -04001532 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Chris Mason788f20e2008-04-28 15:29:42 -04001533
Yan Zheng2b820322008-11-17 21:11:30 -05001534 if (seeding_dev) {
1535 ret = init_first_rw_device(trans, root, device);
1536 BUG_ON(ret);
1537 ret = btrfs_finish_sprout(trans, root);
1538 BUG_ON(ret);
1539 } else {
1540 ret = btrfs_add_device(trans, root, device);
1541 }
1542
Chris Mason913d9522009-03-10 13:17:18 -04001543 /*
1544 * we've got more storage, clear any full flags on the space
1545 * infos
1546 */
1547 btrfs_clear_space_info_full(root->fs_info);
1548
Chris Mason7d9eb122008-07-08 14:19:17 -04001549 unlock_chunks(root);
Yan Zheng2b820322008-11-17 21:11:30 -05001550 btrfs_commit_transaction(trans, root);
1551
1552 if (seeding_dev) {
1553 mutex_unlock(&uuid_mutex);
1554 up_write(&sb->s_umount);
1555
1556 ret = btrfs_relocate_sys_chunks(root);
1557 BUG_ON(ret);
1558 }
1559out:
Chris Mason7d9eb122008-07-08 14:19:17 -04001560 mutex_unlock(&root->fs_info->volume_mutex);
Chris Mason788f20e2008-04-28 15:29:42 -04001561 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05001562error:
Chris Mason15916de2008-11-19 21:17:22 -05001563 close_bdev_exclusive(bdev, 0);
Yan Zheng2b820322008-11-17 21:11:30 -05001564 if (seeding_dev) {
1565 mutex_unlock(&uuid_mutex);
1566 up_write(&sb->s_umount);
1567 }
Chris Mason788f20e2008-04-28 15:29:42 -04001568 goto out;
1569}
1570
Chris Masond3977122009-01-05 21:25:51 -05001571static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1572 struct btrfs_device *device)
Chris Mason0b86a832008-03-24 15:01:56 -04001573{
1574 int ret;
1575 struct btrfs_path *path;
1576 struct btrfs_root *root;
1577 struct btrfs_dev_item *dev_item;
1578 struct extent_buffer *leaf;
1579 struct btrfs_key key;
1580
1581 root = device->dev_root->fs_info->chunk_root;
1582
1583 path = btrfs_alloc_path();
1584 if (!path)
1585 return -ENOMEM;
1586
1587 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1588 key.type = BTRFS_DEV_ITEM_KEY;
1589 key.offset = device->devid;
1590
1591 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1592 if (ret < 0)
1593 goto out;
1594
1595 if (ret > 0) {
1596 ret = -ENOENT;
1597 goto out;
1598 }
1599
1600 leaf = path->nodes[0];
1601 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1602
1603 btrfs_set_device_id(leaf, dev_item, device->devid);
1604 btrfs_set_device_type(leaf, dev_item, device->type);
1605 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1606 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1607 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
Chris Balld6397ba2009-04-27 07:29:03 -04001608 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
Chris Mason0b86a832008-03-24 15:01:56 -04001609 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1610 btrfs_mark_buffer_dirty(leaf);
1611
1612out:
1613 btrfs_free_path(path);
1614 return ret;
1615}
1616
Chris Mason7d9eb122008-07-08 14:19:17 -04001617static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
Chris Mason8f18cf12008-04-25 16:53:30 -04001618 struct btrfs_device *device, u64 new_size)
1619{
1620 struct btrfs_super_block *super_copy =
1621 &device->dev_root->fs_info->super_copy;
1622 u64 old_total = btrfs_super_total_bytes(super_copy);
1623 u64 diff = new_size - device->total_bytes;
1624
Yan Zheng2b820322008-11-17 21:11:30 -05001625 if (!device->writeable)
1626 return -EACCES;
1627 if (new_size <= device->total_bytes)
1628 return -EINVAL;
1629
Chris Mason8f18cf12008-04-25 16:53:30 -04001630 btrfs_set_super_total_bytes(super_copy, old_total + diff);
Yan Zheng2b820322008-11-17 21:11:30 -05001631 device->fs_devices->total_rw_bytes += diff;
1632
1633 device->total_bytes = new_size;
Chris Mason9779b722009-07-24 16:41:41 -04001634 device->disk_total_bytes = new_size;
Chris Mason4184ea72009-03-10 12:39:20 -04001635 btrfs_clear_space_info_full(device->dev_root->fs_info);
1636
Chris Mason8f18cf12008-04-25 16:53:30 -04001637 return btrfs_update_device(trans, device);
1638}
1639
Chris Mason7d9eb122008-07-08 14:19:17 -04001640int btrfs_grow_device(struct btrfs_trans_handle *trans,
1641 struct btrfs_device *device, u64 new_size)
1642{
1643 int ret;
1644 lock_chunks(device->dev_root);
1645 ret = __btrfs_grow_device(trans, device, new_size);
1646 unlock_chunks(device->dev_root);
1647 return ret;
1648}
1649
Chris Mason8f18cf12008-04-25 16:53:30 -04001650static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1651 struct btrfs_root *root,
1652 u64 chunk_tree, u64 chunk_objectid,
1653 u64 chunk_offset)
1654{
1655 int ret;
1656 struct btrfs_path *path;
1657 struct btrfs_key key;
1658
1659 root = root->fs_info->chunk_root;
1660 path = btrfs_alloc_path();
1661 if (!path)
1662 return -ENOMEM;
1663
1664 key.objectid = chunk_objectid;
1665 key.offset = chunk_offset;
1666 key.type = BTRFS_CHUNK_ITEM_KEY;
1667
1668 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1669 BUG_ON(ret);
1670
1671 ret = btrfs_del_item(trans, root, path);
1672 BUG_ON(ret);
1673
1674 btrfs_free_path(path);
1675 return 0;
1676}
1677
Christoph Hellwigb2950862008-12-02 09:54:17 -05001678static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
Chris Mason8f18cf12008-04-25 16:53:30 -04001679 chunk_offset)
1680{
1681 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1682 struct btrfs_disk_key *disk_key;
1683 struct btrfs_chunk *chunk;
1684 u8 *ptr;
1685 int ret = 0;
1686 u32 num_stripes;
1687 u32 array_size;
1688 u32 len = 0;
1689 u32 cur;
1690 struct btrfs_key key;
1691
1692 array_size = btrfs_super_sys_array_size(super_copy);
1693
1694 ptr = super_copy->sys_chunk_array;
1695 cur = 0;
1696
1697 while (cur < array_size) {
1698 disk_key = (struct btrfs_disk_key *)ptr;
1699 btrfs_disk_key_to_cpu(&key, disk_key);
1700
1701 len = sizeof(*disk_key);
1702
1703 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1704 chunk = (struct btrfs_chunk *)(ptr + len);
1705 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1706 len += btrfs_chunk_item_size(num_stripes);
1707 } else {
1708 ret = -EIO;
1709 break;
1710 }
1711 if (key.objectid == chunk_objectid &&
1712 key.offset == chunk_offset) {
1713 memmove(ptr, ptr + len, array_size - (cur + len));
1714 array_size -= len;
1715 btrfs_set_super_sys_array_size(super_copy, array_size);
1716 } else {
1717 ptr += len;
1718 cur += len;
1719 }
1720 }
1721 return ret;
1722}
1723
Christoph Hellwigb2950862008-12-02 09:54:17 -05001724static int btrfs_relocate_chunk(struct btrfs_root *root,
Chris Mason8f18cf12008-04-25 16:53:30 -04001725 u64 chunk_tree, u64 chunk_objectid,
1726 u64 chunk_offset)
1727{
1728 struct extent_map_tree *em_tree;
1729 struct btrfs_root *extent_root;
1730 struct btrfs_trans_handle *trans;
1731 struct extent_map *em;
1732 struct map_lookup *map;
1733 int ret;
1734 int i;
1735
1736 root = root->fs_info->chunk_root;
1737 extent_root = root->fs_info->extent_root;
1738 em_tree = &root->fs_info->mapping_tree.map_tree;
1739
Josef Bacikba1bf482009-09-11 16:11:19 -04001740 ret = btrfs_can_relocate(extent_root, chunk_offset);
1741 if (ret)
1742 return -ENOSPC;
1743
Chris Mason8f18cf12008-04-25 16:53:30 -04001744 /* step one, relocate all the extents inside this chunk */
Zheng Yan1a40e232008-09-26 10:09:34 -04001745 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
Chris Mason8f18cf12008-04-25 16:53:30 -04001746 BUG_ON(ret);
1747
1748 trans = btrfs_start_transaction(root, 1);
1749 BUG_ON(!trans);
1750
Chris Mason7d9eb122008-07-08 14:19:17 -04001751 lock_chunks(root);
1752
Chris Mason8f18cf12008-04-25 16:53:30 -04001753 /*
1754 * step two, delete the device extents and the
1755 * chunk tree entries
1756 */
Chris Mason890871b2009-09-02 16:24:52 -04001757 read_lock(&em_tree->lock);
Chris Mason8f18cf12008-04-25 16:53:30 -04001758 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
Chris Mason890871b2009-09-02 16:24:52 -04001759 read_unlock(&em_tree->lock);
Chris Mason8f18cf12008-04-25 16:53:30 -04001760
Chris Masona061fc82008-05-07 11:43:44 -04001761 BUG_ON(em->start > chunk_offset ||
1762 em->start + em->len < chunk_offset);
Chris Mason8f18cf12008-04-25 16:53:30 -04001763 map = (struct map_lookup *)em->bdev;
1764
1765 for (i = 0; i < map->num_stripes; i++) {
1766 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1767 map->stripes[i].physical);
1768 BUG_ON(ret);
Chris Masona061fc82008-05-07 11:43:44 -04001769
Chris Masondfe25022008-05-13 13:46:40 -04001770 if (map->stripes[i].dev) {
1771 ret = btrfs_update_device(trans, map->stripes[i].dev);
1772 BUG_ON(ret);
1773 }
Chris Mason8f18cf12008-04-25 16:53:30 -04001774 }
1775 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1776 chunk_offset);
1777
1778 BUG_ON(ret);
1779
1780 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1781 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1782 BUG_ON(ret);
Chris Mason8f18cf12008-04-25 16:53:30 -04001783 }
1784
Zheng Yan1a40e232008-09-26 10:09:34 -04001785 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1786 BUG_ON(ret);
1787
Chris Mason890871b2009-09-02 16:24:52 -04001788 write_lock(&em_tree->lock);
Chris Mason8f18cf12008-04-25 16:53:30 -04001789 remove_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04001790 write_unlock(&em_tree->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04001791
Chris Mason8f18cf12008-04-25 16:53:30 -04001792 kfree(map);
1793 em->bdev = NULL;
1794
1795 /* once for the tree */
1796 free_extent_map(em);
Chris Mason8f18cf12008-04-25 16:53:30 -04001797 /* once for us */
1798 free_extent_map(em);
1799
Chris Mason7d9eb122008-07-08 14:19:17 -04001800 unlock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04001801 btrfs_end_transaction(trans, root);
1802 return 0;
1803}
1804
Yan Zheng2b820322008-11-17 21:11:30 -05001805static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1806{
1807 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1808 struct btrfs_path *path;
1809 struct extent_buffer *leaf;
1810 struct btrfs_chunk *chunk;
1811 struct btrfs_key key;
1812 struct btrfs_key found_key;
1813 u64 chunk_tree = chunk_root->root_key.objectid;
1814 u64 chunk_type;
Josef Bacikba1bf482009-09-11 16:11:19 -04001815 bool retried = false;
1816 int failed = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05001817 int ret;
1818
1819 path = btrfs_alloc_path();
1820 if (!path)
1821 return -ENOMEM;
1822
Josef Bacikba1bf482009-09-11 16:11:19 -04001823again:
Yan Zheng2b820322008-11-17 21:11:30 -05001824 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1825 key.offset = (u64)-1;
1826 key.type = BTRFS_CHUNK_ITEM_KEY;
1827
1828 while (1) {
1829 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1830 if (ret < 0)
1831 goto error;
1832 BUG_ON(ret == 0);
1833
1834 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1835 key.type);
1836 if (ret < 0)
1837 goto error;
1838 if (ret > 0)
1839 break;
1840
1841 leaf = path->nodes[0];
1842 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1843
1844 chunk = btrfs_item_ptr(leaf, path->slots[0],
1845 struct btrfs_chunk);
1846 chunk_type = btrfs_chunk_type(leaf, chunk);
1847 btrfs_release_path(chunk_root, path);
1848
1849 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1850 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1851 found_key.objectid,
1852 found_key.offset);
Josef Bacikba1bf482009-09-11 16:11:19 -04001853 if (ret == -ENOSPC)
1854 failed++;
1855 else if (ret)
1856 BUG();
Yan Zheng2b820322008-11-17 21:11:30 -05001857 }
1858
1859 if (found_key.offset == 0)
1860 break;
1861 key.offset = found_key.offset - 1;
1862 }
1863 ret = 0;
Josef Bacikba1bf482009-09-11 16:11:19 -04001864 if (failed && !retried) {
1865 failed = 0;
1866 retried = true;
1867 goto again;
1868 } else if (failed && retried) {
1869 WARN_ON(1);
1870 ret = -ENOSPC;
1871 }
Yan Zheng2b820322008-11-17 21:11:30 -05001872error:
1873 btrfs_free_path(path);
1874 return ret;
1875}
1876
Chris Masonec44a352008-04-28 15:29:52 -04001877static u64 div_factor(u64 num, int factor)
1878{
1879 if (factor == 10)
1880 return num;
1881 num *= factor;
1882 do_div(num, 10);
1883 return num;
1884}
1885
Chris Masonec44a352008-04-28 15:29:52 -04001886int btrfs_balance(struct btrfs_root *dev_root)
1887{
1888 int ret;
Chris Masonec44a352008-04-28 15:29:52 -04001889 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1890 struct btrfs_device *device;
1891 u64 old_size;
1892 u64 size_to_free;
1893 struct btrfs_path *path;
1894 struct btrfs_key key;
1895 struct btrfs_chunk *chunk;
1896 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1897 struct btrfs_trans_handle *trans;
1898 struct btrfs_key found_key;
1899
Yan Zheng2b820322008-11-17 21:11:30 -05001900 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1901 return -EROFS;
Chris Masonec44a352008-04-28 15:29:52 -04001902
Chris Mason7d9eb122008-07-08 14:19:17 -04001903 mutex_lock(&dev_root->fs_info->volume_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04001904 dev_root = dev_root->fs_info->dev_root;
1905
Chris Masonec44a352008-04-28 15:29:52 -04001906 /* step one make some room on all the devices */
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001907 list_for_each_entry(device, devices, dev_list) {
Chris Masonec44a352008-04-28 15:29:52 -04001908 old_size = device->total_bytes;
1909 size_to_free = div_factor(old_size, 1);
1910 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
Yan Zheng2b820322008-11-17 21:11:30 -05001911 if (!device->writeable ||
1912 device->total_bytes - device->bytes_used > size_to_free)
Chris Masonec44a352008-04-28 15:29:52 -04001913 continue;
1914
1915 ret = btrfs_shrink_device(device, old_size - size_to_free);
Josef Bacikba1bf482009-09-11 16:11:19 -04001916 if (ret == -ENOSPC)
1917 break;
Chris Masonec44a352008-04-28 15:29:52 -04001918 BUG_ON(ret);
1919
1920 trans = btrfs_start_transaction(dev_root, 1);
1921 BUG_ON(!trans);
1922
1923 ret = btrfs_grow_device(trans, device, old_size);
1924 BUG_ON(ret);
1925
1926 btrfs_end_transaction(trans, dev_root);
1927 }
1928
1929 /* step two, relocate all the chunks */
1930 path = btrfs_alloc_path();
1931 BUG_ON(!path);
1932
1933 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1934 key.offset = (u64)-1;
1935 key.type = BTRFS_CHUNK_ITEM_KEY;
1936
Chris Masond3977122009-01-05 21:25:51 -05001937 while (1) {
Chris Masonec44a352008-04-28 15:29:52 -04001938 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1939 if (ret < 0)
1940 goto error;
1941
1942 /*
1943 * this shouldn't happen, it means the last relocate
1944 * failed
1945 */
1946 if (ret == 0)
1947 break;
1948
1949 ret = btrfs_previous_item(chunk_root, path, 0,
1950 BTRFS_CHUNK_ITEM_KEY);
Chris Mason7d9eb122008-07-08 14:19:17 -04001951 if (ret)
Chris Masonec44a352008-04-28 15:29:52 -04001952 break;
Chris Mason7d9eb122008-07-08 14:19:17 -04001953
Chris Masonec44a352008-04-28 15:29:52 -04001954 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1955 path->slots[0]);
1956 if (found_key.objectid != key.objectid)
1957 break;
Chris Mason7d9eb122008-07-08 14:19:17 -04001958
Chris Masonec44a352008-04-28 15:29:52 -04001959 chunk = btrfs_item_ptr(path->nodes[0],
1960 path->slots[0],
1961 struct btrfs_chunk);
Chris Masonec44a352008-04-28 15:29:52 -04001962 /* chunk zero is special */
Josef Bacikba1bf482009-09-11 16:11:19 -04001963 if (found_key.offset == 0)
Chris Masonec44a352008-04-28 15:29:52 -04001964 break;
1965
Chris Mason7d9eb122008-07-08 14:19:17 -04001966 btrfs_release_path(chunk_root, path);
Chris Masonec44a352008-04-28 15:29:52 -04001967 ret = btrfs_relocate_chunk(chunk_root,
1968 chunk_root->root_key.objectid,
1969 found_key.objectid,
1970 found_key.offset);
Josef Bacikba1bf482009-09-11 16:11:19 -04001971 BUG_ON(ret && ret != -ENOSPC);
1972 key.offset = found_key.offset - 1;
Chris Masonec44a352008-04-28 15:29:52 -04001973 }
1974 ret = 0;
1975error:
1976 btrfs_free_path(path);
Chris Mason7d9eb122008-07-08 14:19:17 -04001977 mutex_unlock(&dev_root->fs_info->volume_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04001978 return ret;
1979}
1980
Chris Mason8f18cf12008-04-25 16:53:30 -04001981/*
1982 * shrinking a device means finding all of the device extents past
1983 * the new size, and then following the back refs to the chunks.
1984 * The chunk relocation code actually frees the device extent
1985 */
1986int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1987{
1988 struct btrfs_trans_handle *trans;
1989 struct btrfs_root *root = device->dev_root;
1990 struct btrfs_dev_extent *dev_extent = NULL;
1991 struct btrfs_path *path;
1992 u64 length;
1993 u64 chunk_tree;
1994 u64 chunk_objectid;
1995 u64 chunk_offset;
1996 int ret;
1997 int slot;
Josef Bacikba1bf482009-09-11 16:11:19 -04001998 int failed = 0;
1999 bool retried = false;
Chris Mason8f18cf12008-04-25 16:53:30 -04002000 struct extent_buffer *l;
2001 struct btrfs_key key;
2002 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2003 u64 old_total = btrfs_super_total_bytes(super_copy);
Josef Bacikba1bf482009-09-11 16:11:19 -04002004 u64 old_size = device->total_bytes;
Chris Mason8f18cf12008-04-25 16:53:30 -04002005 u64 diff = device->total_bytes - new_size;
2006
Yan Zheng2b820322008-11-17 21:11:30 -05002007 if (new_size >= device->total_bytes)
2008 return -EINVAL;
Chris Mason8f18cf12008-04-25 16:53:30 -04002009
2010 path = btrfs_alloc_path();
2011 if (!path)
2012 return -ENOMEM;
2013
Chris Mason8f18cf12008-04-25 16:53:30 -04002014 path->reada = 2;
2015
Chris Mason7d9eb122008-07-08 14:19:17 -04002016 lock_chunks(root);
2017
Chris Mason8f18cf12008-04-25 16:53:30 -04002018 device->total_bytes = new_size;
Yan Zheng2b820322008-11-17 21:11:30 -05002019 if (device->writeable)
2020 device->fs_devices->total_rw_bytes -= diff;
Chris Mason7d9eb122008-07-08 14:19:17 -04002021 unlock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04002022
Josef Bacikba1bf482009-09-11 16:11:19 -04002023again:
Chris Mason8f18cf12008-04-25 16:53:30 -04002024 key.objectid = device->devid;
2025 key.offset = (u64)-1;
2026 key.type = BTRFS_DEV_EXTENT_KEY;
2027
2028 while (1) {
2029 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2030 if (ret < 0)
2031 goto done;
2032
2033 ret = btrfs_previous_item(root, path, 0, key.type);
2034 if (ret < 0)
2035 goto done;
2036 if (ret) {
2037 ret = 0;
Josef Bacikba1bf482009-09-11 16:11:19 -04002038 btrfs_release_path(root, path);
Yan Zhengbf1fb512009-07-22 09:59:00 -04002039 break;
Chris Mason8f18cf12008-04-25 16:53:30 -04002040 }
2041
2042 l = path->nodes[0];
2043 slot = path->slots[0];
2044 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2045
Josef Bacikba1bf482009-09-11 16:11:19 -04002046 if (key.objectid != device->devid) {
2047 btrfs_release_path(root, path);
Yan Zhengbf1fb512009-07-22 09:59:00 -04002048 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04002049 }
Chris Mason8f18cf12008-04-25 16:53:30 -04002050
2051 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2052 length = btrfs_dev_extent_length(l, dev_extent);
2053
Josef Bacikba1bf482009-09-11 16:11:19 -04002054 if (key.offset + length <= new_size) {
2055 btrfs_release_path(root, path);
Chris Balld6397ba2009-04-27 07:29:03 -04002056 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04002057 }
Chris Mason8f18cf12008-04-25 16:53:30 -04002058
2059 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2060 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2061 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2062 btrfs_release_path(root, path);
2063
2064 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2065 chunk_offset);
Josef Bacikba1bf482009-09-11 16:11:19 -04002066 if (ret && ret != -ENOSPC)
Chris Mason8f18cf12008-04-25 16:53:30 -04002067 goto done;
Josef Bacikba1bf482009-09-11 16:11:19 -04002068 if (ret == -ENOSPC)
2069 failed++;
2070 key.offset -= 1;
2071 }
2072
2073 if (failed && !retried) {
2074 failed = 0;
2075 retried = true;
2076 goto again;
2077 } else if (failed && retried) {
2078 ret = -ENOSPC;
2079 lock_chunks(root);
2080
2081 device->total_bytes = old_size;
2082 if (device->writeable)
2083 device->fs_devices->total_rw_bytes += diff;
2084 unlock_chunks(root);
2085 goto done;
Chris Mason8f18cf12008-04-25 16:53:30 -04002086 }
2087
Chris Balld6397ba2009-04-27 07:29:03 -04002088 /* Shrinking succeeded, else we would be at "done". */
2089 trans = btrfs_start_transaction(root, 1);
2090 if (!trans) {
2091 ret = -ENOMEM;
2092 goto done;
2093 }
2094 lock_chunks(root);
2095
2096 device->disk_total_bytes = new_size;
2097 /* Now btrfs_update_device() will change the on-disk size. */
2098 ret = btrfs_update_device(trans, device);
2099 if (ret) {
2100 unlock_chunks(root);
2101 btrfs_end_transaction(trans, root);
2102 goto done;
2103 }
2104 WARN_ON(diff > old_total);
2105 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2106 unlock_chunks(root);
2107 btrfs_end_transaction(trans, root);
Chris Mason8f18cf12008-04-25 16:53:30 -04002108done:
2109 btrfs_free_path(path);
2110 return ret;
2111}
2112
Christoph Hellwigb2950862008-12-02 09:54:17 -05002113static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
Chris Mason0b86a832008-03-24 15:01:56 -04002114 struct btrfs_root *root,
2115 struct btrfs_key *key,
2116 struct btrfs_chunk *chunk, int item_size)
2117{
2118 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2119 struct btrfs_disk_key disk_key;
2120 u32 array_size;
2121 u8 *ptr;
2122
2123 array_size = btrfs_super_sys_array_size(super_copy);
2124 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2125 return -EFBIG;
2126
2127 ptr = super_copy->sys_chunk_array + array_size;
2128 btrfs_cpu_key_to_disk(&disk_key, key);
2129 memcpy(ptr, &disk_key, sizeof(disk_key));
2130 ptr += sizeof(disk_key);
2131 memcpy(ptr, chunk, item_size);
2132 item_size += sizeof(disk_key);
2133 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2134 return 0;
2135}
2136
Chris Masond3977122009-01-05 21:25:51 -05002137static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
Chris Masona1b32a52008-09-05 16:09:51 -04002138 int num_stripes, int sub_stripes)
Chris Mason9b3f68b2008-04-18 10:29:51 -04002139{
2140 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2141 return calc_size;
2142 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2143 return calc_size * (num_stripes / sub_stripes);
2144 else
2145 return calc_size * num_stripes;
2146}
2147
Yan Zheng2b820322008-11-17 21:11:30 -05002148static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *extent_root,
2150 struct map_lookup **map_ret,
2151 u64 *num_bytes, u64 *stripe_size,
2152 u64 start, u64 type)
Chris Mason0b86a832008-03-24 15:01:56 -04002153{
Chris Mason593060d2008-03-25 16:50:33 -04002154 struct btrfs_fs_info *info = extent_root->fs_info;
Chris Mason0b86a832008-03-24 15:01:56 -04002155 struct btrfs_device *device = NULL;
Yan Zheng2b820322008-11-17 21:11:30 -05002156 struct btrfs_fs_devices *fs_devices = info->fs_devices;
Chris Mason6324fbf2008-03-24 15:01:59 -04002157 struct list_head *cur;
Yan Zheng2b820322008-11-17 21:11:30 -05002158 struct map_lookup *map = NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04002159 struct extent_map_tree *em_tree;
Chris Mason0b86a832008-03-24 15:01:56 -04002160 struct extent_map *em;
Yan Zheng2b820322008-11-17 21:11:30 -05002161 struct list_head private_devs;
Chris Masona40a90a2008-04-18 11:55:51 -04002162 int min_stripe_size = 1 * 1024 * 1024;
Chris Mason0b86a832008-03-24 15:01:56 -04002163 u64 calc_size = 1024 * 1024 * 1024;
Chris Mason9b3f68b2008-04-18 10:29:51 -04002164 u64 max_chunk_size = calc_size;
2165 u64 min_free;
Chris Mason6324fbf2008-03-24 15:01:59 -04002166 u64 avail;
2167 u64 max_avail = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05002168 u64 dev_offset;
Chris Mason6324fbf2008-03-24 15:01:59 -04002169 int num_stripes = 1;
Chris Masona40a90a2008-04-18 11:55:51 -04002170 int min_stripes = 1;
Chris Mason321aecc2008-04-16 10:49:51 -04002171 int sub_stripes = 0;
Chris Mason6324fbf2008-03-24 15:01:59 -04002172 int looped = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04002173 int ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04002174 int index;
Chris Mason593060d2008-03-25 16:50:33 -04002175 int stripe_len = 64 * 1024;
Chris Mason0b86a832008-03-24 15:01:56 -04002176
Chris Masonec44a352008-04-28 15:29:52 -04002177 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2178 (type & BTRFS_BLOCK_GROUP_DUP)) {
2179 WARN_ON(1);
2180 type &= ~BTRFS_BLOCK_GROUP_DUP;
2181 }
Yan Zheng2b820322008-11-17 21:11:30 -05002182 if (list_empty(&fs_devices->alloc_list))
Chris Mason6324fbf2008-03-24 15:01:59 -04002183 return -ENOSPC;
Chris Mason593060d2008-03-25 16:50:33 -04002184
Chris Masona40a90a2008-04-18 11:55:51 -04002185 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
Yan Zheng2b820322008-11-17 21:11:30 -05002186 num_stripes = fs_devices->rw_devices;
Chris Masona40a90a2008-04-18 11:55:51 -04002187 min_stripes = 2;
2188 }
2189 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
Chris Mason611f0e02008-04-03 16:29:03 -04002190 num_stripes = 2;
Chris Masona40a90a2008-04-18 11:55:51 -04002191 min_stripes = 2;
2192 }
Chris Mason8790d502008-04-03 16:29:03 -04002193 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
Yan Zheng2b820322008-11-17 21:11:30 -05002194 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
Chris Mason9b3f68b2008-04-18 10:29:51 -04002195 if (num_stripes < 2)
2196 return -ENOSPC;
Chris Masona40a90a2008-04-18 11:55:51 -04002197 min_stripes = 2;
Chris Mason8790d502008-04-03 16:29:03 -04002198 }
Chris Mason321aecc2008-04-16 10:49:51 -04002199 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
Yan Zheng2b820322008-11-17 21:11:30 -05002200 num_stripes = fs_devices->rw_devices;
Chris Mason321aecc2008-04-16 10:49:51 -04002201 if (num_stripes < 4)
2202 return -ENOSPC;
2203 num_stripes &= ~(u32)1;
2204 sub_stripes = 2;
Chris Masona40a90a2008-04-18 11:55:51 -04002205 min_stripes = 4;
Chris Mason321aecc2008-04-16 10:49:51 -04002206 }
Chris Mason9b3f68b2008-04-18 10:29:51 -04002207
2208 if (type & BTRFS_BLOCK_GROUP_DATA) {
2209 max_chunk_size = 10 * calc_size;
Chris Masona40a90a2008-04-18 11:55:51 -04002210 min_stripe_size = 64 * 1024 * 1024;
Chris Mason9b3f68b2008-04-18 10:29:51 -04002211 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
Josef Bacik83d3c962009-12-07 21:45:59 +00002212 max_chunk_size = 256 * 1024 * 1024;
Chris Masona40a90a2008-04-18 11:55:51 -04002213 min_stripe_size = 32 * 1024 * 1024;
2214 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2215 calc_size = 8 * 1024 * 1024;
2216 max_chunk_size = calc_size * 2;
2217 min_stripe_size = 1 * 1024 * 1024;
Chris Mason9b3f68b2008-04-18 10:29:51 -04002218 }
2219
Yan Zheng2b820322008-11-17 21:11:30 -05002220 /* we don't want a chunk larger than 10% of writeable space */
2221 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2222 max_chunk_size);
Chris Mason9b3f68b2008-04-18 10:29:51 -04002223
Chris Masona40a90a2008-04-18 11:55:51 -04002224again:
Chris Mason9779b722009-07-24 16:41:41 -04002225 max_avail = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05002226 if (!map || map->num_stripes != num_stripes) {
2227 kfree(map);
2228 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2229 if (!map)
2230 return -ENOMEM;
2231 map->num_stripes = num_stripes;
2232 }
2233
Chris Mason9b3f68b2008-04-18 10:29:51 -04002234 if (calc_size * num_stripes > max_chunk_size) {
2235 calc_size = max_chunk_size;
2236 do_div(calc_size, num_stripes);
2237 do_div(calc_size, stripe_len);
2238 calc_size *= stripe_len;
2239 }
2240 /* we don't want tiny stripes */
Chris Masona40a90a2008-04-18 11:55:51 -04002241 calc_size = max_t(u64, min_stripe_size, calc_size);
Chris Mason9b3f68b2008-04-18 10:29:51 -04002242
Chris Mason9b3f68b2008-04-18 10:29:51 -04002243 do_div(calc_size, stripe_len);
2244 calc_size *= stripe_len;
2245
Yan Zheng2b820322008-11-17 21:11:30 -05002246 cur = fs_devices->alloc_list.next;
Chris Mason6324fbf2008-03-24 15:01:59 -04002247 index = 0;
Chris Mason611f0e02008-04-03 16:29:03 -04002248
2249 if (type & BTRFS_BLOCK_GROUP_DUP)
2250 min_free = calc_size * 2;
Chris Mason9b3f68b2008-04-18 10:29:51 -04002251 else
2252 min_free = calc_size;
Chris Mason611f0e02008-04-03 16:29:03 -04002253
Josef Bacik0f9dd462008-09-23 13:14:11 -04002254 /*
2255 * we add 1MB because we never use the first 1MB of the device, unless
2256 * we've looped, then we are likely allocating the maximum amount of
2257 * space left already
2258 */
2259 if (!looped)
2260 min_free += 1024 * 1024;
Chris Masonad5bd912008-04-21 08:28:10 -04002261
Yan Zheng2b820322008-11-17 21:11:30 -05002262 INIT_LIST_HEAD(&private_devs);
Chris Masond3977122009-01-05 21:25:51 -05002263 while (index < num_stripes) {
Chris Masonb3075712008-04-22 09:22:07 -04002264 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
Yan Zheng2b820322008-11-17 21:11:30 -05002265 BUG_ON(!device->writeable);
Chris Masondfe25022008-05-13 13:46:40 -04002266 if (device->total_bytes > device->bytes_used)
2267 avail = device->total_bytes - device->bytes_used;
2268 else
2269 avail = 0;
Chris Mason6324fbf2008-03-24 15:01:59 -04002270 cur = cur->next;
Chris Mason8f18cf12008-04-25 16:53:30 -04002271
Chris Masondfe25022008-05-13 13:46:40 -04002272 if (device->in_fs_metadata && avail >= min_free) {
Yan Zheng2b820322008-11-17 21:11:30 -05002273 ret = find_free_dev_extent(trans, device,
Chris Mason9779b722009-07-24 16:41:41 -04002274 min_free, &dev_offset,
2275 &max_avail);
Chris Mason8f18cf12008-04-25 16:53:30 -04002276 if (ret == 0) {
2277 list_move_tail(&device->dev_alloc_list,
2278 &private_devs);
Yan Zheng2b820322008-11-17 21:11:30 -05002279 map->stripes[index].dev = device;
2280 map->stripes[index].physical = dev_offset;
Chris Mason611f0e02008-04-03 16:29:03 -04002281 index++;
Yan Zheng2b820322008-11-17 21:11:30 -05002282 if (type & BTRFS_BLOCK_GROUP_DUP) {
2283 map->stripes[index].dev = device;
2284 map->stripes[index].physical =
2285 dev_offset + calc_size;
Chris Mason8f18cf12008-04-25 16:53:30 -04002286 index++;
Yan Zheng2b820322008-11-17 21:11:30 -05002287 }
Chris Mason8f18cf12008-04-25 16:53:30 -04002288 }
Chris Masondfe25022008-05-13 13:46:40 -04002289 } else if (device->in_fs_metadata && avail > max_avail)
Chris Masona40a90a2008-04-18 11:55:51 -04002290 max_avail = avail;
Yan Zheng2b820322008-11-17 21:11:30 -05002291 if (cur == &fs_devices->alloc_list)
Chris Mason6324fbf2008-03-24 15:01:59 -04002292 break;
2293 }
Yan Zheng2b820322008-11-17 21:11:30 -05002294 list_splice(&private_devs, &fs_devices->alloc_list);
Chris Mason6324fbf2008-03-24 15:01:59 -04002295 if (index < num_stripes) {
Chris Masona40a90a2008-04-18 11:55:51 -04002296 if (index >= min_stripes) {
2297 num_stripes = index;
2298 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2299 num_stripes /= sub_stripes;
2300 num_stripes *= sub_stripes;
2301 }
2302 looped = 1;
2303 goto again;
2304 }
Chris Mason6324fbf2008-03-24 15:01:59 -04002305 if (!looped && max_avail > 0) {
2306 looped = 1;
2307 calc_size = max_avail;
2308 goto again;
2309 }
Yan Zheng2b820322008-11-17 21:11:30 -05002310 kfree(map);
Chris Mason6324fbf2008-03-24 15:01:59 -04002311 return -ENOSPC;
2312 }
Chris Mason593060d2008-03-25 16:50:33 -04002313 map->sector_size = extent_root->sectorsize;
2314 map->stripe_len = stripe_len;
2315 map->io_align = stripe_len;
2316 map->io_width = stripe_len;
2317 map->type = type;
2318 map->num_stripes = num_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04002319 map->sub_stripes = sub_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04002320
Yan Zheng2b820322008-11-17 21:11:30 -05002321 *map_ret = map;
2322 *stripe_size = calc_size;
2323 *num_bytes = chunk_bytes_by_type(type, calc_size,
2324 num_stripes, sub_stripes);
Chris Mason0b86a832008-03-24 15:01:56 -04002325
2326 em = alloc_extent_map(GFP_NOFS);
Yan Zheng2b820322008-11-17 21:11:30 -05002327 if (!em) {
2328 kfree(map);
Chris Mason0b86a832008-03-24 15:01:56 -04002329 return -ENOMEM;
Yan Zheng2b820322008-11-17 21:11:30 -05002330 }
Chris Mason0b86a832008-03-24 15:01:56 -04002331 em->bdev = (struct block_device *)map;
Yan Zheng2b820322008-11-17 21:11:30 -05002332 em->start = start;
Chris Masone17cade2008-04-15 15:41:47 -04002333 em->len = *num_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04002334 em->block_start = 0;
Chris Masonc8b97812008-10-29 14:49:59 -04002335 em->block_len = em->len;
Chris Mason0b86a832008-03-24 15:01:56 -04002336
Chris Mason0b86a832008-03-24 15:01:56 -04002337 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
Chris Mason890871b2009-09-02 16:24:52 -04002338 write_lock(&em_tree->lock);
Chris Mason0b86a832008-03-24 15:01:56 -04002339 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04002340 write_unlock(&em_tree->lock);
Chris Masonb248a412008-04-14 09:48:18 -04002341 BUG_ON(ret);
Chris Mason0b86a832008-03-24 15:01:56 -04002342 free_extent_map(em);
Yan Zheng2b820322008-11-17 21:11:30 -05002343
2344 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2345 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2346 start, *num_bytes);
2347 BUG_ON(ret);
2348
2349 index = 0;
2350 while (index < map->num_stripes) {
2351 device = map->stripes[index].dev;
2352 dev_offset = map->stripes[index].physical;
2353
2354 ret = btrfs_alloc_dev_extent(trans, device,
2355 info->chunk_root->root_key.objectid,
2356 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2357 start, dev_offset, calc_size);
2358 BUG_ON(ret);
2359 index++;
2360 }
2361
2362 return 0;
2363}
2364
2365static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2366 struct btrfs_root *extent_root,
2367 struct map_lookup *map, u64 chunk_offset,
2368 u64 chunk_size, u64 stripe_size)
2369{
2370 u64 dev_offset;
2371 struct btrfs_key key;
2372 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2373 struct btrfs_device *device;
2374 struct btrfs_chunk *chunk;
2375 struct btrfs_stripe *stripe;
2376 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2377 int index = 0;
2378 int ret;
2379
2380 chunk = kzalloc(item_size, GFP_NOFS);
2381 if (!chunk)
2382 return -ENOMEM;
2383
2384 index = 0;
2385 while (index < map->num_stripes) {
2386 device = map->stripes[index].dev;
2387 device->bytes_used += stripe_size;
2388 ret = btrfs_update_device(trans, device);
2389 BUG_ON(ret);
2390 index++;
2391 }
2392
2393 index = 0;
2394 stripe = &chunk->stripe;
2395 while (index < map->num_stripes) {
2396 device = map->stripes[index].dev;
2397 dev_offset = map->stripes[index].physical;
2398
2399 btrfs_set_stack_stripe_devid(stripe, device->devid);
2400 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2401 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2402 stripe++;
2403 index++;
2404 }
2405
2406 btrfs_set_stack_chunk_length(chunk, chunk_size);
2407 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2408 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2409 btrfs_set_stack_chunk_type(chunk, map->type);
2410 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2411 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2412 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2413 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2414 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2415
2416 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2417 key.type = BTRFS_CHUNK_ITEM_KEY;
2418 key.offset = chunk_offset;
2419
2420 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2421 BUG_ON(ret);
2422
2423 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2424 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2425 item_size);
2426 BUG_ON(ret);
2427 }
2428 kfree(chunk);
2429 return 0;
2430}
2431
2432/*
2433 * Chunk allocation falls into two parts. The first part does works
2434 * that make the new allocated chunk useable, but not do any operation
2435 * that modifies the chunk tree. The second part does the works that
2436 * require modifying the chunk tree. This division is important for the
2437 * bootstrap process of adding storage to a seed btrfs.
2438 */
2439int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2440 struct btrfs_root *extent_root, u64 type)
2441{
2442 u64 chunk_offset;
2443 u64 chunk_size;
2444 u64 stripe_size;
2445 struct map_lookup *map;
2446 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2447 int ret;
2448
2449 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2450 &chunk_offset);
2451 if (ret)
2452 return ret;
2453
2454 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2455 &stripe_size, chunk_offset, type);
2456 if (ret)
2457 return ret;
2458
2459 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2460 chunk_size, stripe_size);
2461 BUG_ON(ret);
2462 return 0;
2463}
2464
Chris Masond3977122009-01-05 21:25:51 -05002465static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
Yan Zheng2b820322008-11-17 21:11:30 -05002466 struct btrfs_root *root,
2467 struct btrfs_device *device)
2468{
2469 u64 chunk_offset;
2470 u64 sys_chunk_offset;
2471 u64 chunk_size;
2472 u64 sys_chunk_size;
2473 u64 stripe_size;
2474 u64 sys_stripe_size;
2475 u64 alloc_profile;
2476 struct map_lookup *map;
2477 struct map_lookup *sys_map;
2478 struct btrfs_fs_info *fs_info = root->fs_info;
2479 struct btrfs_root *extent_root = fs_info->extent_root;
2480 int ret;
2481
2482 ret = find_next_chunk(fs_info->chunk_root,
2483 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2484 BUG_ON(ret);
2485
2486 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2487 (fs_info->metadata_alloc_profile &
2488 fs_info->avail_metadata_alloc_bits);
2489 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2490
2491 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2492 &stripe_size, chunk_offset, alloc_profile);
2493 BUG_ON(ret);
2494
2495 sys_chunk_offset = chunk_offset + chunk_size;
2496
2497 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2498 (fs_info->system_alloc_profile &
2499 fs_info->avail_system_alloc_bits);
2500 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2501
2502 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2503 &sys_chunk_size, &sys_stripe_size,
2504 sys_chunk_offset, alloc_profile);
2505 BUG_ON(ret);
2506
2507 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2508 BUG_ON(ret);
2509
2510 /*
2511 * Modifying chunk tree needs allocating new blocks from both
2512 * system block group and metadata block group. So we only can
2513 * do operations require modifying the chunk tree after both
2514 * block groups were created.
2515 */
2516 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2517 chunk_size, stripe_size);
2518 BUG_ON(ret);
2519
2520 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2521 sys_chunk_offset, sys_chunk_size,
2522 sys_stripe_size);
2523 BUG_ON(ret);
2524 return 0;
2525}
2526
2527int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2528{
2529 struct extent_map *em;
2530 struct map_lookup *map;
2531 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2532 int readonly = 0;
2533 int i;
2534
Chris Mason890871b2009-09-02 16:24:52 -04002535 read_lock(&map_tree->map_tree.lock);
Yan Zheng2b820322008-11-17 21:11:30 -05002536 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
Chris Mason890871b2009-09-02 16:24:52 -04002537 read_unlock(&map_tree->map_tree.lock);
Yan Zheng2b820322008-11-17 21:11:30 -05002538 if (!em)
2539 return 1;
2540
2541 map = (struct map_lookup *)em->bdev;
2542 for (i = 0; i < map->num_stripes; i++) {
2543 if (!map->stripes[i].dev->writeable) {
2544 readonly = 1;
2545 break;
2546 }
2547 }
2548 free_extent_map(em);
2549 return readonly;
Chris Mason0b86a832008-03-24 15:01:56 -04002550}
2551
2552void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2553{
2554 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2555}
2556
2557void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2558{
2559 struct extent_map *em;
2560
Chris Masond3977122009-01-05 21:25:51 -05002561 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04002562 write_lock(&tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04002563 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2564 if (em)
2565 remove_extent_mapping(&tree->map_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04002566 write_unlock(&tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04002567 if (!em)
2568 break;
2569 kfree(em->bdev);
2570 /* once for us */
2571 free_extent_map(em);
2572 /* once for the tree */
2573 free_extent_map(em);
2574 }
2575}
2576
Chris Masonf1885912008-04-09 16:28:12 -04002577int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2578{
2579 struct extent_map *em;
2580 struct map_lookup *map;
2581 struct extent_map_tree *em_tree = &map_tree->map_tree;
2582 int ret;
2583
Chris Mason890871b2009-09-02 16:24:52 -04002584 read_lock(&em_tree->lock);
Chris Masonf1885912008-04-09 16:28:12 -04002585 em = lookup_extent_mapping(em_tree, logical, len);
Chris Mason890871b2009-09-02 16:24:52 -04002586 read_unlock(&em_tree->lock);
Chris Masonf1885912008-04-09 16:28:12 -04002587 BUG_ON(!em);
2588
2589 BUG_ON(em->start > logical || em->start + em->len < logical);
2590 map = (struct map_lookup *)em->bdev;
2591 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2592 ret = map->num_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04002593 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2594 ret = map->sub_stripes;
Chris Masonf1885912008-04-09 16:28:12 -04002595 else
2596 ret = 1;
2597 free_extent_map(em);
Chris Masonf1885912008-04-09 16:28:12 -04002598 return ret;
2599}
2600
Chris Masondfe25022008-05-13 13:46:40 -04002601static int find_live_mirror(struct map_lookup *map, int first, int num,
2602 int optimal)
2603{
2604 int i;
2605 if (map->stripes[optimal].dev->bdev)
2606 return optimal;
2607 for (i = first; i < first + num; i++) {
2608 if (map->stripes[i].dev->bdev)
2609 return i;
2610 }
2611 /* we couldn't find one that doesn't fail. Just return something
2612 * and the io error handling code will clean up eventually
2613 */
2614 return optimal;
2615}
2616
Chris Masonf2d8d742008-04-21 10:03:05 -04002617static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2618 u64 logical, u64 *length,
2619 struct btrfs_multi_bio **multi_ret,
2620 int mirror_num, struct page *unplug_page)
Chris Mason0b86a832008-03-24 15:01:56 -04002621{
2622 struct extent_map *em;
2623 struct map_lookup *map;
2624 struct extent_map_tree *em_tree = &map_tree->map_tree;
2625 u64 offset;
Chris Mason593060d2008-03-25 16:50:33 -04002626 u64 stripe_offset;
2627 u64 stripe_nr;
Chris Masoncea9e442008-04-09 16:28:12 -04002628 int stripes_allocated = 8;
Chris Mason321aecc2008-04-16 10:49:51 -04002629 int stripes_required = 1;
Chris Mason593060d2008-03-25 16:50:33 -04002630 int stripe_index;
Chris Masoncea9e442008-04-09 16:28:12 -04002631 int i;
Chris Masonf2d8d742008-04-21 10:03:05 -04002632 int num_stripes;
Chris Masona236aed2008-04-29 09:38:00 -04002633 int max_errors = 0;
Chris Masoncea9e442008-04-09 16:28:12 -04002634 struct btrfs_multi_bio *multi = NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04002635
Chris Masond3977122009-01-05 21:25:51 -05002636 if (multi_ret && !(rw & (1 << BIO_RW)))
Chris Masoncea9e442008-04-09 16:28:12 -04002637 stripes_allocated = 1;
Chris Masoncea9e442008-04-09 16:28:12 -04002638again:
2639 if (multi_ret) {
2640 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2641 GFP_NOFS);
2642 if (!multi)
2643 return -ENOMEM;
Chris Masona236aed2008-04-29 09:38:00 -04002644
2645 atomic_set(&multi->error, 0);
Chris Masoncea9e442008-04-09 16:28:12 -04002646 }
Chris Mason0b86a832008-03-24 15:01:56 -04002647
Chris Mason890871b2009-09-02 16:24:52 -04002648 read_lock(&em_tree->lock);
Chris Mason0b86a832008-03-24 15:01:56 -04002649 em = lookup_extent_mapping(em_tree, logical, *length);
Chris Mason890871b2009-09-02 16:24:52 -04002650 read_unlock(&em_tree->lock);
Chris Masonf2d8d742008-04-21 10:03:05 -04002651
2652 if (!em && unplug_page)
2653 return 0;
2654
Chris Mason3b951512008-04-17 11:29:12 -04002655 if (!em) {
Chris Masond3977122009-01-05 21:25:51 -05002656 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2657 (unsigned long long)logical,
2658 (unsigned long long)*length);
Chris Masonf2d8d742008-04-21 10:03:05 -04002659 BUG();
Chris Mason3b951512008-04-17 11:29:12 -04002660 }
Chris Mason0b86a832008-03-24 15:01:56 -04002661
2662 BUG_ON(em->start > logical || em->start + em->len < logical);
2663 map = (struct map_lookup *)em->bdev;
2664 offset = logical - em->start;
Chris Mason593060d2008-03-25 16:50:33 -04002665
Chris Masonf1885912008-04-09 16:28:12 -04002666 if (mirror_num > map->num_stripes)
2667 mirror_num = 0;
2668
Chris Masoncea9e442008-04-09 16:28:12 -04002669 /* if our multi bio struct is too small, back off and try again */
Chris Mason321aecc2008-04-16 10:49:51 -04002670 if (rw & (1 << BIO_RW)) {
2671 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2672 BTRFS_BLOCK_GROUP_DUP)) {
2673 stripes_required = map->num_stripes;
Chris Masona236aed2008-04-29 09:38:00 -04002674 max_errors = 1;
Chris Mason321aecc2008-04-16 10:49:51 -04002675 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2676 stripes_required = map->sub_stripes;
Chris Masona236aed2008-04-29 09:38:00 -04002677 max_errors = 1;
Chris Mason321aecc2008-04-16 10:49:51 -04002678 }
2679 }
Chris Masonffbd5172009-04-20 15:50:09 -04002680 if (multi_ret && (rw & (1 << BIO_RW)) &&
Chris Mason321aecc2008-04-16 10:49:51 -04002681 stripes_allocated < stripes_required) {
Chris Masoncea9e442008-04-09 16:28:12 -04002682 stripes_allocated = map->num_stripes;
Chris Masoncea9e442008-04-09 16:28:12 -04002683 free_extent_map(em);
2684 kfree(multi);
2685 goto again;
2686 }
Chris Mason593060d2008-03-25 16:50:33 -04002687 stripe_nr = offset;
2688 /*
2689 * stripe_nr counts the total number of stripes we have to stride
2690 * to get to this block
2691 */
2692 do_div(stripe_nr, map->stripe_len);
2693
2694 stripe_offset = stripe_nr * map->stripe_len;
2695 BUG_ON(offset < stripe_offset);
2696
2697 /* stripe_offset is the offset of this block in its stripe*/
2698 stripe_offset = offset - stripe_offset;
2699
Chris Masoncea9e442008-04-09 16:28:12 -04002700 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
Chris Mason321aecc2008-04-16 10:49:51 -04002701 BTRFS_BLOCK_GROUP_RAID10 |
Chris Masoncea9e442008-04-09 16:28:12 -04002702 BTRFS_BLOCK_GROUP_DUP)) {
2703 /* we limit the length of each bio to what fits in a stripe */
2704 *length = min_t(u64, em->len - offset,
2705 map->stripe_len - stripe_offset);
2706 } else {
2707 *length = em->len - offset;
2708 }
Chris Masonf2d8d742008-04-21 10:03:05 -04002709
2710 if (!multi_ret && !unplug_page)
Chris Masoncea9e442008-04-09 16:28:12 -04002711 goto out;
2712
Chris Masonf2d8d742008-04-21 10:03:05 -04002713 num_stripes = 1;
Chris Masoncea9e442008-04-09 16:28:12 -04002714 stripe_index = 0;
Chris Mason8790d502008-04-03 16:29:03 -04002715 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
Chris Masonf2d8d742008-04-21 10:03:05 -04002716 if (unplug_page || (rw & (1 << BIO_RW)))
2717 num_stripes = map->num_stripes;
Chris Mason2fff7342008-04-29 14:12:09 -04002718 else if (mirror_num)
Chris Masonf1885912008-04-09 16:28:12 -04002719 stripe_index = mirror_num - 1;
Chris Masondfe25022008-05-13 13:46:40 -04002720 else {
2721 stripe_index = find_live_mirror(map, 0,
2722 map->num_stripes,
2723 current->pid % map->num_stripes);
2724 }
Chris Mason2fff7342008-04-29 14:12:09 -04002725
Chris Mason611f0e02008-04-03 16:29:03 -04002726 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
Chris Masoncea9e442008-04-09 16:28:12 -04002727 if (rw & (1 << BIO_RW))
Chris Masonf2d8d742008-04-21 10:03:05 -04002728 num_stripes = map->num_stripes;
Chris Masonf1885912008-04-09 16:28:12 -04002729 else if (mirror_num)
2730 stripe_index = mirror_num - 1;
Chris Mason2fff7342008-04-29 14:12:09 -04002731
Chris Mason321aecc2008-04-16 10:49:51 -04002732 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2733 int factor = map->num_stripes / map->sub_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04002734
2735 stripe_index = do_div(stripe_nr, factor);
2736 stripe_index *= map->sub_stripes;
2737
Chris Masonf2d8d742008-04-21 10:03:05 -04002738 if (unplug_page || (rw & (1 << BIO_RW)))
2739 num_stripes = map->sub_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04002740 else if (mirror_num)
2741 stripe_index += mirror_num - 1;
Chris Masondfe25022008-05-13 13:46:40 -04002742 else {
2743 stripe_index = find_live_mirror(map, stripe_index,
2744 map->sub_stripes, stripe_index +
2745 current->pid % map->sub_stripes);
2746 }
Chris Mason8790d502008-04-03 16:29:03 -04002747 } else {
2748 /*
2749 * after this do_div call, stripe_nr is the number of stripes
2750 * on this device we have to walk to find the data, and
2751 * stripe_index is the number of our device in the stripe array
2752 */
2753 stripe_index = do_div(stripe_nr, map->num_stripes);
2754 }
Chris Mason593060d2008-03-25 16:50:33 -04002755 BUG_ON(stripe_index >= map->num_stripes);
Chris Mason593060d2008-03-25 16:50:33 -04002756
Chris Masonf2d8d742008-04-21 10:03:05 -04002757 for (i = 0; i < num_stripes; i++) {
2758 if (unplug_page) {
2759 struct btrfs_device *device;
2760 struct backing_dev_info *bdi;
2761
2762 device = map->stripes[stripe_index].dev;
Chris Masondfe25022008-05-13 13:46:40 -04002763 if (device->bdev) {
2764 bdi = blk_get_backing_dev_info(device->bdev);
Chris Masond3977122009-01-05 21:25:51 -05002765 if (bdi->unplug_io_fn)
Chris Masondfe25022008-05-13 13:46:40 -04002766 bdi->unplug_io_fn(bdi, unplug_page);
Chris Masonf2d8d742008-04-21 10:03:05 -04002767 }
2768 } else {
2769 multi->stripes[i].physical =
2770 map->stripes[stripe_index].physical +
2771 stripe_offset + stripe_nr * map->stripe_len;
2772 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2773 }
Chris Masoncea9e442008-04-09 16:28:12 -04002774 stripe_index++;
Chris Mason593060d2008-03-25 16:50:33 -04002775 }
Chris Masonf2d8d742008-04-21 10:03:05 -04002776 if (multi_ret) {
2777 *multi_ret = multi;
2778 multi->num_stripes = num_stripes;
Chris Masona236aed2008-04-29 09:38:00 -04002779 multi->max_errors = max_errors;
Chris Masonf2d8d742008-04-21 10:03:05 -04002780 }
Chris Masoncea9e442008-04-09 16:28:12 -04002781out:
Chris Mason0b86a832008-03-24 15:01:56 -04002782 free_extent_map(em);
Chris Mason0b86a832008-03-24 15:01:56 -04002783 return 0;
2784}
2785
Chris Masonf2d8d742008-04-21 10:03:05 -04002786int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2787 u64 logical, u64 *length,
2788 struct btrfs_multi_bio **multi_ret, int mirror_num)
2789{
2790 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2791 mirror_num, NULL);
2792}
2793
Yan Zhenga512bbf2008-12-08 16:46:26 -05002794int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2795 u64 chunk_start, u64 physical, u64 devid,
2796 u64 **logical, int *naddrs, int *stripe_len)
2797{
2798 struct extent_map_tree *em_tree = &map_tree->map_tree;
2799 struct extent_map *em;
2800 struct map_lookup *map;
2801 u64 *buf;
2802 u64 bytenr;
2803 u64 length;
2804 u64 stripe_nr;
2805 int i, j, nr = 0;
2806
Chris Mason890871b2009-09-02 16:24:52 -04002807 read_lock(&em_tree->lock);
Yan Zhenga512bbf2008-12-08 16:46:26 -05002808 em = lookup_extent_mapping(em_tree, chunk_start, 1);
Chris Mason890871b2009-09-02 16:24:52 -04002809 read_unlock(&em_tree->lock);
Yan Zhenga512bbf2008-12-08 16:46:26 -05002810
2811 BUG_ON(!em || em->start != chunk_start);
2812 map = (struct map_lookup *)em->bdev;
2813
2814 length = em->len;
2815 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2816 do_div(length, map->num_stripes / map->sub_stripes);
2817 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2818 do_div(length, map->num_stripes);
2819
2820 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2821 BUG_ON(!buf);
2822
2823 for (i = 0; i < map->num_stripes; i++) {
2824 if (devid && map->stripes[i].dev->devid != devid)
2825 continue;
2826 if (map->stripes[i].physical > physical ||
2827 map->stripes[i].physical + length <= physical)
2828 continue;
2829
2830 stripe_nr = physical - map->stripes[i].physical;
2831 do_div(stripe_nr, map->stripe_len);
2832
2833 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2834 stripe_nr = stripe_nr * map->num_stripes + i;
2835 do_div(stripe_nr, map->sub_stripes);
2836 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2837 stripe_nr = stripe_nr * map->num_stripes + i;
2838 }
2839 bytenr = chunk_start + stripe_nr * map->stripe_len;
Chris Mason934d3752008-12-08 16:43:10 -05002840 WARN_ON(nr >= map->num_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05002841 for (j = 0; j < nr; j++) {
2842 if (buf[j] == bytenr)
2843 break;
2844 }
Chris Mason934d3752008-12-08 16:43:10 -05002845 if (j == nr) {
2846 WARN_ON(nr >= map->num_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05002847 buf[nr++] = bytenr;
Chris Mason934d3752008-12-08 16:43:10 -05002848 }
Yan Zhenga512bbf2008-12-08 16:46:26 -05002849 }
2850
Yan Zhenga512bbf2008-12-08 16:46:26 -05002851 *logical = buf;
2852 *naddrs = nr;
2853 *stripe_len = map->stripe_len;
2854
2855 free_extent_map(em);
2856 return 0;
2857}
2858
Chris Masonf2d8d742008-04-21 10:03:05 -04002859int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2860 u64 logical, struct page *page)
2861{
2862 u64 length = PAGE_CACHE_SIZE;
2863 return __btrfs_map_block(map_tree, READ, logical, &length,
2864 NULL, 0, page);
2865}
2866
Chris Mason8790d502008-04-03 16:29:03 -04002867static void end_bio_multi_stripe(struct bio *bio, int err)
Chris Mason8790d502008-04-03 16:29:03 -04002868{
Chris Masoncea9e442008-04-09 16:28:12 -04002869 struct btrfs_multi_bio *multi = bio->bi_private;
Chris Mason7d2b4da2008-08-05 10:13:57 -04002870 int is_orig_bio = 0;
Chris Mason8790d502008-04-03 16:29:03 -04002871
Chris Mason8790d502008-04-03 16:29:03 -04002872 if (err)
Chris Masona236aed2008-04-29 09:38:00 -04002873 atomic_inc(&multi->error);
Chris Mason8790d502008-04-03 16:29:03 -04002874
Chris Mason7d2b4da2008-08-05 10:13:57 -04002875 if (bio == multi->orig_bio)
2876 is_orig_bio = 1;
2877
Chris Masoncea9e442008-04-09 16:28:12 -04002878 if (atomic_dec_and_test(&multi->stripes_pending)) {
Chris Mason7d2b4da2008-08-05 10:13:57 -04002879 if (!is_orig_bio) {
2880 bio_put(bio);
2881 bio = multi->orig_bio;
2882 }
Chris Mason8790d502008-04-03 16:29:03 -04002883 bio->bi_private = multi->private;
2884 bio->bi_end_io = multi->end_io;
Chris Masona236aed2008-04-29 09:38:00 -04002885 /* only send an error to the higher layers if it is
2886 * beyond the tolerance of the multi-bio
2887 */
Chris Mason1259ab72008-05-12 13:39:03 -04002888 if (atomic_read(&multi->error) > multi->max_errors) {
Chris Masona236aed2008-04-29 09:38:00 -04002889 err = -EIO;
Chris Mason1259ab72008-05-12 13:39:03 -04002890 } else if (err) {
2891 /*
2892 * this bio is actually up to date, we didn't
2893 * go over the max number of errors
2894 */
2895 set_bit(BIO_UPTODATE, &bio->bi_flags);
Chris Masona236aed2008-04-29 09:38:00 -04002896 err = 0;
Chris Mason1259ab72008-05-12 13:39:03 -04002897 }
Chris Mason8790d502008-04-03 16:29:03 -04002898 kfree(multi);
2899
2900 bio_endio(bio, err);
Chris Mason7d2b4da2008-08-05 10:13:57 -04002901 } else if (!is_orig_bio) {
Chris Mason8790d502008-04-03 16:29:03 -04002902 bio_put(bio);
2903 }
Chris Mason8790d502008-04-03 16:29:03 -04002904}
2905
Chris Mason8b712842008-06-11 16:50:36 -04002906struct async_sched {
2907 struct bio *bio;
2908 int rw;
2909 struct btrfs_fs_info *info;
2910 struct btrfs_work work;
2911};
2912
2913/*
2914 * see run_scheduled_bios for a description of why bios are collected for
2915 * async submit.
2916 *
2917 * This will add one bio to the pending list for a device and make sure
2918 * the work struct is scheduled.
2919 */
Chris Masond3977122009-01-05 21:25:51 -05002920static noinline int schedule_bio(struct btrfs_root *root,
Chris Masona1b32a52008-09-05 16:09:51 -04002921 struct btrfs_device *device,
2922 int rw, struct bio *bio)
Chris Mason8b712842008-06-11 16:50:36 -04002923{
2924 int should_queue = 1;
Chris Masonffbd5172009-04-20 15:50:09 -04002925 struct btrfs_pending_bios *pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -04002926
2927 /* don't bother with additional async steps for reads, right now */
2928 if (!(rw & (1 << BIO_RW))) {
Chris Mason492bb6de2008-07-31 16:29:02 -04002929 bio_get(bio);
Chris Mason8b712842008-06-11 16:50:36 -04002930 submit_bio(rw, bio);
Chris Mason492bb6de2008-07-31 16:29:02 -04002931 bio_put(bio);
Chris Mason8b712842008-06-11 16:50:36 -04002932 return 0;
2933 }
2934
2935 /*
Chris Mason0986fe9e2008-08-15 15:34:15 -04002936 * nr_async_bios allows us to reliably return congestion to the
Chris Mason8b712842008-06-11 16:50:36 -04002937 * higher layers. Otherwise, the async bio makes it appear we have
2938 * made progress against dirty pages when we've really just put it
2939 * on a queue for later
2940 */
Chris Mason0986fe9e2008-08-15 15:34:15 -04002941 atomic_inc(&root->fs_info->nr_async_bios);
Chris Mason492bb6de2008-07-31 16:29:02 -04002942 WARN_ON(bio->bi_next);
Chris Mason8b712842008-06-11 16:50:36 -04002943 bio->bi_next = NULL;
2944 bio->bi_rw |= rw;
2945
2946 spin_lock(&device->io_lock);
Jens Axboe1f98a132009-09-11 14:32:04 +02002947 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
Chris Masonffbd5172009-04-20 15:50:09 -04002948 pending_bios = &device->pending_sync_bios;
2949 else
2950 pending_bios = &device->pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -04002951
Chris Masonffbd5172009-04-20 15:50:09 -04002952 if (pending_bios->tail)
2953 pending_bios->tail->bi_next = bio;
Chris Mason8b712842008-06-11 16:50:36 -04002954
Chris Masonffbd5172009-04-20 15:50:09 -04002955 pending_bios->tail = bio;
2956 if (!pending_bios->head)
2957 pending_bios->head = bio;
Chris Mason8b712842008-06-11 16:50:36 -04002958 if (device->running_pending)
2959 should_queue = 0;
2960
2961 spin_unlock(&device->io_lock);
2962
2963 if (should_queue)
Chris Mason1cc127b2008-06-12 14:46:17 -04002964 btrfs_queue_worker(&root->fs_info->submit_workers,
2965 &device->work);
Chris Mason8b712842008-06-11 16:50:36 -04002966 return 0;
2967}
2968
Chris Masonf1885912008-04-09 16:28:12 -04002969int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
Chris Mason8b712842008-06-11 16:50:36 -04002970 int mirror_num, int async_submit)
Chris Mason0b86a832008-03-24 15:01:56 -04002971{
2972 struct btrfs_mapping_tree *map_tree;
2973 struct btrfs_device *dev;
Chris Mason8790d502008-04-03 16:29:03 -04002974 struct bio *first_bio = bio;
Chris Masona62b9402008-10-03 16:31:08 -04002975 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason0b86a832008-03-24 15:01:56 -04002976 u64 length = 0;
2977 u64 map_length;
Chris Masoncea9e442008-04-09 16:28:12 -04002978 struct btrfs_multi_bio *multi = NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04002979 int ret;
Chris Mason8790d502008-04-03 16:29:03 -04002980 int dev_nr = 0;
2981 int total_devs = 1;
Chris Mason0b86a832008-03-24 15:01:56 -04002982
Chris Masonf2d8d742008-04-21 10:03:05 -04002983 length = bio->bi_size;
Chris Mason0b86a832008-03-24 15:01:56 -04002984 map_tree = &root->fs_info->mapping_tree;
2985 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04002986
Chris Masonf1885912008-04-09 16:28:12 -04002987 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2988 mirror_num);
Chris Masoncea9e442008-04-09 16:28:12 -04002989 BUG_ON(ret);
2990
2991 total_devs = multi->num_stripes;
2992 if (map_length < length) {
Chris Masond3977122009-01-05 21:25:51 -05002993 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2994 "len %llu\n", (unsigned long long)logical,
2995 (unsigned long long)length,
2996 (unsigned long long)map_length);
Chris Masoncea9e442008-04-09 16:28:12 -04002997 BUG();
2998 }
2999 multi->end_io = first_bio->bi_end_io;
3000 multi->private = first_bio->bi_private;
Chris Mason7d2b4da2008-08-05 10:13:57 -04003001 multi->orig_bio = first_bio;
Chris Masoncea9e442008-04-09 16:28:12 -04003002 atomic_set(&multi->stripes_pending, multi->num_stripes);
3003
Chris Masond3977122009-01-05 21:25:51 -05003004 while (dev_nr < total_devs) {
Chris Mason8790d502008-04-03 16:29:03 -04003005 if (total_devs > 1) {
Chris Mason8790d502008-04-03 16:29:03 -04003006 if (dev_nr < total_devs - 1) {
3007 bio = bio_clone(first_bio, GFP_NOFS);
3008 BUG_ON(!bio);
3009 } else {
3010 bio = first_bio;
3011 }
3012 bio->bi_private = multi;
3013 bio->bi_end_io = end_bio_multi_stripe;
3014 }
Chris Masoncea9e442008-04-09 16:28:12 -04003015 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3016 dev = multi->stripes[dev_nr].dev;
Yan Zheng2b820322008-11-17 21:11:30 -05003017 BUG_ON(rw == WRITE && !dev->writeable);
Chris Masondfe25022008-05-13 13:46:40 -04003018 if (dev && dev->bdev) {
3019 bio->bi_bdev = dev->bdev;
Chris Mason8b712842008-06-11 16:50:36 -04003020 if (async_submit)
3021 schedule_bio(root, dev, rw, bio);
3022 else
3023 submit_bio(rw, bio);
Chris Masondfe25022008-05-13 13:46:40 -04003024 } else {
3025 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3026 bio->bi_sector = logical >> 9;
Chris Masondfe25022008-05-13 13:46:40 -04003027 bio_endio(bio, -EIO);
Chris Masondfe25022008-05-13 13:46:40 -04003028 }
Chris Mason8790d502008-04-03 16:29:03 -04003029 dev_nr++;
Chris Mason239b14b2008-03-24 15:02:07 -04003030 }
Chris Masoncea9e442008-04-09 16:28:12 -04003031 if (total_devs == 1)
3032 kfree(multi);
Chris Mason0b86a832008-03-24 15:01:56 -04003033 return 0;
3034}
3035
Chris Masona4437552008-04-18 10:29:38 -04003036struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
Yan Zheng2b820322008-11-17 21:11:30 -05003037 u8 *uuid, u8 *fsid)
Chris Mason0b86a832008-03-24 15:01:56 -04003038{
Yan Zheng2b820322008-11-17 21:11:30 -05003039 struct btrfs_device *device;
3040 struct btrfs_fs_devices *cur_devices;
Chris Mason0b86a832008-03-24 15:01:56 -04003041
Yan Zheng2b820322008-11-17 21:11:30 -05003042 cur_devices = root->fs_info->fs_devices;
3043 while (cur_devices) {
3044 if (!fsid ||
3045 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3046 device = __find_device(&cur_devices->devices,
3047 devid, uuid);
3048 if (device)
3049 return device;
3050 }
3051 cur_devices = cur_devices->seed;
3052 }
3053 return NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04003054}
3055
Chris Masondfe25022008-05-13 13:46:40 -04003056static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3057 u64 devid, u8 *dev_uuid)
3058{
3059 struct btrfs_device *device;
3060 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3061
3062 device = kzalloc(sizeof(*device), GFP_NOFS);
yanhai zhu7cbd8a82008-11-12 14:38:54 -05003063 if (!device)
3064 return NULL;
Chris Masondfe25022008-05-13 13:46:40 -04003065 list_add(&device->dev_list,
3066 &fs_devices->devices);
Chris Masondfe25022008-05-13 13:46:40 -04003067 device->barriers = 1;
3068 device->dev_root = root->fs_info->dev_root;
3069 device->devid = devid;
Chris Mason8b712842008-06-11 16:50:36 -04003070 device->work.func = pending_bios_fn;
Yan Zhenge4404d62008-12-12 10:03:26 -05003071 device->fs_devices = fs_devices;
Chris Masondfe25022008-05-13 13:46:40 -04003072 fs_devices->num_devices++;
3073 spin_lock_init(&device->io_lock);
Chris Masond20f7042008-12-08 16:58:54 -05003074 INIT_LIST_HEAD(&device->dev_alloc_list);
Chris Masondfe25022008-05-13 13:46:40 -04003075 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3076 return device;
3077}
3078
Chris Mason0b86a832008-03-24 15:01:56 -04003079static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3080 struct extent_buffer *leaf,
3081 struct btrfs_chunk *chunk)
3082{
3083 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3084 struct map_lookup *map;
3085 struct extent_map *em;
3086 u64 logical;
3087 u64 length;
3088 u64 devid;
Chris Masona4437552008-04-18 10:29:38 -04003089 u8 uuid[BTRFS_UUID_SIZE];
Chris Mason593060d2008-03-25 16:50:33 -04003090 int num_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04003091 int ret;
Chris Mason593060d2008-03-25 16:50:33 -04003092 int i;
Chris Mason0b86a832008-03-24 15:01:56 -04003093
Chris Masone17cade2008-04-15 15:41:47 -04003094 logical = key->offset;
3095 length = btrfs_chunk_length(leaf, chunk);
Chris Masona061fc82008-05-07 11:43:44 -04003096
Chris Mason890871b2009-09-02 16:24:52 -04003097 read_lock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04003098 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
Chris Mason890871b2009-09-02 16:24:52 -04003099 read_unlock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04003100
3101 /* already mapped? */
3102 if (em && em->start <= logical && em->start + em->len > logical) {
3103 free_extent_map(em);
Chris Mason0b86a832008-03-24 15:01:56 -04003104 return 0;
3105 } else if (em) {
3106 free_extent_map(em);
3107 }
Chris Mason0b86a832008-03-24 15:01:56 -04003108
Chris Mason0b86a832008-03-24 15:01:56 -04003109 em = alloc_extent_map(GFP_NOFS);
3110 if (!em)
3111 return -ENOMEM;
Chris Mason593060d2008-03-25 16:50:33 -04003112 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3113 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
Chris Mason0b86a832008-03-24 15:01:56 -04003114 if (!map) {
3115 free_extent_map(em);
3116 return -ENOMEM;
3117 }
3118
3119 em->bdev = (struct block_device *)map;
3120 em->start = logical;
3121 em->len = length;
3122 em->block_start = 0;
Chris Masonc8b97812008-10-29 14:49:59 -04003123 em->block_len = em->len;
Chris Mason0b86a832008-03-24 15:01:56 -04003124
Chris Mason593060d2008-03-25 16:50:33 -04003125 map->num_stripes = num_stripes;
3126 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3127 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3128 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3129 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3130 map->type = btrfs_chunk_type(leaf, chunk);
Chris Mason321aecc2008-04-16 10:49:51 -04003131 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
Chris Mason593060d2008-03-25 16:50:33 -04003132 for (i = 0; i < num_stripes; i++) {
3133 map->stripes[i].physical =
3134 btrfs_stripe_offset_nr(leaf, chunk, i);
3135 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
Chris Masona4437552008-04-18 10:29:38 -04003136 read_extent_buffer(leaf, uuid, (unsigned long)
3137 btrfs_stripe_dev_uuid_nr(chunk, i),
3138 BTRFS_UUID_SIZE);
Yan Zheng2b820322008-11-17 21:11:30 -05003139 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3140 NULL);
Chris Masondfe25022008-05-13 13:46:40 -04003141 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
Chris Mason593060d2008-03-25 16:50:33 -04003142 kfree(map);
3143 free_extent_map(em);
3144 return -EIO;
3145 }
Chris Masondfe25022008-05-13 13:46:40 -04003146 if (!map->stripes[i].dev) {
3147 map->stripes[i].dev =
3148 add_missing_dev(root, devid, uuid);
3149 if (!map->stripes[i].dev) {
3150 kfree(map);
3151 free_extent_map(em);
3152 return -EIO;
3153 }
3154 }
3155 map->stripes[i].dev->in_fs_metadata = 1;
Chris Mason0b86a832008-03-24 15:01:56 -04003156 }
3157
Chris Mason890871b2009-09-02 16:24:52 -04003158 write_lock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04003159 ret = add_extent_mapping(&map_tree->map_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04003160 write_unlock(&map_tree->map_tree.lock);
Chris Masonb248a412008-04-14 09:48:18 -04003161 BUG_ON(ret);
Chris Mason0b86a832008-03-24 15:01:56 -04003162 free_extent_map(em);
3163
3164 return 0;
3165}
3166
3167static int fill_device_from_item(struct extent_buffer *leaf,
3168 struct btrfs_dev_item *dev_item,
3169 struct btrfs_device *device)
3170{
3171 unsigned long ptr;
Chris Mason0b86a832008-03-24 15:01:56 -04003172
3173 device->devid = btrfs_device_id(leaf, dev_item);
Chris Balld6397ba2009-04-27 07:29:03 -04003174 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3175 device->total_bytes = device->disk_total_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04003176 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3177 device->type = btrfs_device_type(leaf, dev_item);
3178 device->io_align = btrfs_device_io_align(leaf, dev_item);
3179 device->io_width = btrfs_device_io_width(leaf, dev_item);
3180 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
Chris Mason0b86a832008-03-24 15:01:56 -04003181
3182 ptr = (unsigned long)btrfs_device_uuid(dev_item);
Chris Masone17cade2008-04-15 15:41:47 -04003183 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04003184
Chris Mason0b86a832008-03-24 15:01:56 -04003185 return 0;
3186}
3187
Yan Zheng2b820322008-11-17 21:11:30 -05003188static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3189{
3190 struct btrfs_fs_devices *fs_devices;
3191 int ret;
3192
3193 mutex_lock(&uuid_mutex);
3194
3195 fs_devices = root->fs_info->fs_devices->seed;
3196 while (fs_devices) {
3197 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3198 ret = 0;
3199 goto out;
3200 }
3201 fs_devices = fs_devices->seed;
3202 }
3203
3204 fs_devices = find_fsid(fsid);
3205 if (!fs_devices) {
3206 ret = -ENOENT;
3207 goto out;
3208 }
Yan Zhenge4404d62008-12-12 10:03:26 -05003209
3210 fs_devices = clone_fs_devices(fs_devices);
3211 if (IS_ERR(fs_devices)) {
3212 ret = PTR_ERR(fs_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05003213 goto out;
3214 }
3215
Christoph Hellwig97288f22008-12-02 06:36:09 -05003216 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
Chris Mason15916de2008-11-19 21:17:22 -05003217 root->fs_info->bdev_holder);
Yan Zheng2b820322008-11-17 21:11:30 -05003218 if (ret)
3219 goto out;
3220
3221 if (!fs_devices->seeding) {
3222 __btrfs_close_devices(fs_devices);
Yan Zhenge4404d62008-12-12 10:03:26 -05003223 free_fs_devices(fs_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05003224 ret = -EINVAL;
3225 goto out;
3226 }
3227
3228 fs_devices->seed = root->fs_info->fs_devices->seed;
3229 root->fs_info->fs_devices->seed = fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05003230out:
3231 mutex_unlock(&uuid_mutex);
3232 return ret;
3233}
3234
Chris Mason0d81ba52008-03-24 15:02:07 -04003235static int read_one_dev(struct btrfs_root *root,
Chris Mason0b86a832008-03-24 15:01:56 -04003236 struct extent_buffer *leaf,
3237 struct btrfs_dev_item *dev_item)
3238{
3239 struct btrfs_device *device;
3240 u64 devid;
3241 int ret;
Yan Zheng2b820322008-11-17 21:11:30 -05003242 u8 fs_uuid[BTRFS_UUID_SIZE];
Chris Masona4437552008-04-18 10:29:38 -04003243 u8 dev_uuid[BTRFS_UUID_SIZE];
3244
Chris Mason0b86a832008-03-24 15:01:56 -04003245 devid = btrfs_device_id(leaf, dev_item);
Chris Masona4437552008-04-18 10:29:38 -04003246 read_extent_buffer(leaf, dev_uuid,
3247 (unsigned long)btrfs_device_uuid(dev_item),
3248 BTRFS_UUID_SIZE);
Yan Zheng2b820322008-11-17 21:11:30 -05003249 read_extent_buffer(leaf, fs_uuid,
3250 (unsigned long)btrfs_device_fsid(dev_item),
3251 BTRFS_UUID_SIZE);
3252
3253 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3254 ret = open_seed_devices(root, fs_uuid);
Yan Zhenge4404d62008-12-12 10:03:26 -05003255 if (ret && !btrfs_test_opt(root, DEGRADED))
Yan Zheng2b820322008-11-17 21:11:30 -05003256 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05003257 }
3258
3259 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3260 if (!device || !device->bdev) {
Yan Zhenge4404d62008-12-12 10:03:26 -05003261 if (!btrfs_test_opt(root, DEGRADED))
Yan Zheng2b820322008-11-17 21:11:30 -05003262 return -EIO;
3263
3264 if (!device) {
Chris Masond3977122009-01-05 21:25:51 -05003265 printk(KERN_WARNING "warning devid %llu missing\n",
3266 (unsigned long long)devid);
Yan Zheng2b820322008-11-17 21:11:30 -05003267 device = add_missing_dev(root, devid, dev_uuid);
3268 if (!device)
3269 return -ENOMEM;
3270 }
3271 }
3272
3273 if (device->fs_devices != root->fs_info->fs_devices) {
3274 BUG_ON(device->writeable);
3275 if (device->generation !=
3276 btrfs_device_generation(leaf, dev_item))
3277 return -EINVAL;
Chris Mason6324fbf2008-03-24 15:01:59 -04003278 }
Chris Mason0b86a832008-03-24 15:01:56 -04003279
3280 fill_device_from_item(leaf, dev_item, device);
3281 device->dev_root = root->fs_info->dev_root;
Chris Masondfe25022008-05-13 13:46:40 -04003282 device->in_fs_metadata = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05003283 if (device->writeable)
3284 device->fs_devices->total_rw_bytes += device->total_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04003285 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04003286 return ret;
3287}
3288
Chris Mason0d81ba52008-03-24 15:02:07 -04003289int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3290{
3291 struct btrfs_dev_item *dev_item;
3292
3293 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3294 dev_item);
3295 return read_one_dev(root, buf, dev_item);
3296}
3297
Yan Zhenge4404d62008-12-12 10:03:26 -05003298int btrfs_read_sys_array(struct btrfs_root *root)
Chris Mason0b86a832008-03-24 15:01:56 -04003299{
3300 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
Chris Masona061fc82008-05-07 11:43:44 -04003301 struct extent_buffer *sb;
Chris Mason0b86a832008-03-24 15:01:56 -04003302 struct btrfs_disk_key *disk_key;
Chris Mason0b86a832008-03-24 15:01:56 -04003303 struct btrfs_chunk *chunk;
Chris Mason84eed902008-04-25 09:04:37 -04003304 u8 *ptr;
3305 unsigned long sb_ptr;
3306 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04003307 u32 num_stripes;
3308 u32 array_size;
3309 u32 len = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04003310 u32 cur;
Chris Mason84eed902008-04-25 09:04:37 -04003311 struct btrfs_key key;
Chris Mason0b86a832008-03-24 15:01:56 -04003312
Yan Zhenge4404d62008-12-12 10:03:26 -05003313 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
Chris Masona061fc82008-05-07 11:43:44 -04003314 BTRFS_SUPER_INFO_SIZE);
3315 if (!sb)
3316 return -ENOMEM;
3317 btrfs_set_buffer_uptodate(sb);
Chris Mason4008c042009-02-12 14:09:45 -05003318 btrfs_set_buffer_lockdep_class(sb, 0);
3319
Chris Masona061fc82008-05-07 11:43:44 -04003320 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04003321 array_size = btrfs_super_sys_array_size(super_copy);
3322
Chris Mason0b86a832008-03-24 15:01:56 -04003323 ptr = super_copy->sys_chunk_array;
3324 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3325 cur = 0;
3326
3327 while (cur < array_size) {
3328 disk_key = (struct btrfs_disk_key *)ptr;
3329 btrfs_disk_key_to_cpu(&key, disk_key);
3330
Chris Masona061fc82008-05-07 11:43:44 -04003331 len = sizeof(*disk_key); ptr += len;
Chris Mason0b86a832008-03-24 15:01:56 -04003332 sb_ptr += len;
3333 cur += len;
3334
Chris Mason0d81ba52008-03-24 15:02:07 -04003335 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
Chris Mason0b86a832008-03-24 15:01:56 -04003336 chunk = (struct btrfs_chunk *)sb_ptr;
Chris Mason0d81ba52008-03-24 15:02:07 -04003337 ret = read_one_chunk(root, &key, sb, chunk);
Chris Mason84eed902008-04-25 09:04:37 -04003338 if (ret)
3339 break;
Chris Mason0b86a832008-03-24 15:01:56 -04003340 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3341 len = btrfs_chunk_item_size(num_stripes);
3342 } else {
Chris Mason84eed902008-04-25 09:04:37 -04003343 ret = -EIO;
3344 break;
Chris Mason0b86a832008-03-24 15:01:56 -04003345 }
3346 ptr += len;
3347 sb_ptr += len;
3348 cur += len;
3349 }
Chris Masona061fc82008-05-07 11:43:44 -04003350 free_extent_buffer(sb);
Chris Mason84eed902008-04-25 09:04:37 -04003351 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -04003352}
3353
3354int btrfs_read_chunk_tree(struct btrfs_root *root)
3355{
3356 struct btrfs_path *path;
3357 struct extent_buffer *leaf;
3358 struct btrfs_key key;
3359 struct btrfs_key found_key;
3360 int ret;
3361 int slot;
3362
3363 root = root->fs_info->chunk_root;
3364
3365 path = btrfs_alloc_path();
3366 if (!path)
3367 return -ENOMEM;
3368
3369 /* first we search for all of the device items, and then we
3370 * read in all of the chunk items. This way we can create chunk
3371 * mappings that reference all of the devices that are afound
3372 */
3373 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3374 key.offset = 0;
3375 key.type = 0;
3376again:
3377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Chris Masond3977122009-01-05 21:25:51 -05003378 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04003379 leaf = path->nodes[0];
3380 slot = path->slots[0];
3381 if (slot >= btrfs_header_nritems(leaf)) {
3382 ret = btrfs_next_leaf(root, path);
3383 if (ret == 0)
3384 continue;
3385 if (ret < 0)
3386 goto error;
3387 break;
3388 }
3389 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3390 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3391 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3392 break;
3393 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3394 struct btrfs_dev_item *dev_item;
3395 dev_item = btrfs_item_ptr(leaf, slot,
3396 struct btrfs_dev_item);
Chris Mason0d81ba52008-03-24 15:02:07 -04003397 ret = read_one_dev(root, leaf, dev_item);
Yan Zheng2b820322008-11-17 21:11:30 -05003398 if (ret)
3399 goto error;
Chris Mason0b86a832008-03-24 15:01:56 -04003400 }
3401 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3402 struct btrfs_chunk *chunk;
3403 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3404 ret = read_one_chunk(root, &found_key, leaf, chunk);
Yan Zheng2b820322008-11-17 21:11:30 -05003405 if (ret)
3406 goto error;
Chris Mason0b86a832008-03-24 15:01:56 -04003407 }
3408 path->slots[0]++;
3409 }
3410 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3411 key.objectid = 0;
3412 btrfs_release_path(root, path);
3413 goto again;
3414 }
Chris Mason0b86a832008-03-24 15:01:56 -04003415 ret = 0;
3416error:
Yan Zheng2b820322008-11-17 21:11:30 -05003417 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04003418 return ret;
3419}