blob: 035efce603a9c2db3989c1057dd1e20f82d2eba2 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Chris Mason8a4b83c2008-03-24 15:02:07 -040021#include <linux/buffer_head.h>
Chris Masonf2d8d742008-04-21 10:03:05 -040022#include <linux/blkdev.h>
Chris Masonb765ead2009-04-03 10:27:10 -040023#include <linux/iocontext.h>
Ben Hutchings6f88a442010-12-29 14:55:03 +000024#include <linux/capability.h>
Stefan Behrens442a4f62012-05-25 16:06:08 +020025#include <linux/ratelimit.h>
Ilya Dryomov59641012012-01-16 22:04:48 +020026#include <linux/kthread.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050027#include <linux/raid/pq.h>
Stefan Behrens803b2f52013-08-15 17:11:21 +020028#include <linux/semaphore.h>
Andy Shevchenko8da4b8c2016-05-20 17:01:00 -070029#include <linux/uuid.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050030#include <asm/div64.h>
Chris Mason0b86a832008-03-24 15:01:56 -040031#include "ctree.h"
32#include "extent_map.h"
33#include "disk-io.h"
34#include "transaction.h"
35#include "print-tree.h"
36#include "volumes.h"
David Woodhouse53b381b2013-01-29 18:40:14 -050037#include "raid56.h"
Chris Mason8b712842008-06-11 16:50:36 -040038#include "async-thread.h"
Stefan Behrens21adbd52011-11-09 13:44:05 +010039#include "check-integrity.h"
Josef Bacik606686e2012-06-04 14:03:51 -040040#include "rcu-string.h"
Miao Xie3fed40c2012-09-13 04:51:36 -060041#include "math.h"
Stefan Behrens8dabb742012-11-06 13:15:27 +010042#include "dev-replace.h"
Anand Jain99994cd2014-06-03 11:36:00 +080043#include "sysfs.h"
Chris Mason0b86a832008-03-24 15:01:56 -040044
Zhao Leiaf902042015-09-15 21:08:06 +080045const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
47 .sub_stripes = 2,
48 .dev_stripes = 1,
49 .devs_max = 0, /* 0 == as many as possible */
50 .devs_min = 4,
Zhao Lei8789f4f2015-09-15 21:08:07 +080051 .tolerated_failures = 1,
Zhao Leiaf902042015-09-15 21:08:06 +080052 .devs_increment = 2,
53 .ncopies = 2,
54 },
55 [BTRFS_RAID_RAID1] = {
56 .sub_stripes = 1,
57 .dev_stripes = 1,
58 .devs_max = 2,
59 .devs_min = 2,
Zhao Lei8789f4f2015-09-15 21:08:07 +080060 .tolerated_failures = 1,
Zhao Leiaf902042015-09-15 21:08:06 +080061 .devs_increment = 2,
62 .ncopies = 2,
63 },
64 [BTRFS_RAID_DUP] = {
65 .sub_stripes = 1,
66 .dev_stripes = 2,
67 .devs_max = 1,
68 .devs_min = 1,
Zhao Lei8789f4f2015-09-15 21:08:07 +080069 .tolerated_failures = 0,
Zhao Leiaf902042015-09-15 21:08:06 +080070 .devs_increment = 1,
71 .ncopies = 2,
72 },
73 [BTRFS_RAID_RAID0] = {
74 .sub_stripes = 1,
75 .dev_stripes = 1,
76 .devs_max = 0,
77 .devs_min = 2,
Zhao Lei8789f4f2015-09-15 21:08:07 +080078 .tolerated_failures = 0,
Zhao Leiaf902042015-09-15 21:08:06 +080079 .devs_increment = 1,
80 .ncopies = 1,
81 },
82 [BTRFS_RAID_SINGLE] = {
83 .sub_stripes = 1,
84 .dev_stripes = 1,
85 .devs_max = 1,
86 .devs_min = 1,
Zhao Lei8789f4f2015-09-15 21:08:07 +080087 .tolerated_failures = 0,
Zhao Leiaf902042015-09-15 21:08:06 +080088 .devs_increment = 1,
89 .ncopies = 1,
90 },
91 [BTRFS_RAID_RAID5] = {
92 .sub_stripes = 1,
93 .dev_stripes = 1,
94 .devs_max = 0,
95 .devs_min = 2,
Zhao Lei8789f4f2015-09-15 21:08:07 +080096 .tolerated_failures = 1,
Zhao Leiaf902042015-09-15 21:08:06 +080097 .devs_increment = 1,
98 .ncopies = 2,
99 },
100 [BTRFS_RAID_RAID6] = {
101 .sub_stripes = 1,
102 .dev_stripes = 1,
103 .devs_max = 0,
104 .devs_min = 3,
Zhao Lei8789f4f2015-09-15 21:08:07 +0800105 .tolerated_failures = 2,
Zhao Leiaf902042015-09-15 21:08:06 +0800106 .devs_increment = 1,
107 .ncopies = 3,
108 },
109};
110
Colin Ian Kingfb75d852016-01-19 00:05:28 +0000111const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
Zhao Leiaf902042015-09-15 21:08:06 +0800112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
119};
120
David Sterba621292b2016-02-15 16:28:03 +0100121/*
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
125 */
126const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134};
135
Yan Zheng2b820322008-11-17 21:11:30 -0500136static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_device *device);
139static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200140static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
Eric Sandeen48a3b632013-04-25 20:41:01 +0000141static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
Stefan Behrens733f4fb2012-05-25 16:06:10 +0200142static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
Yan Zheng2b820322008-11-17 21:11:30 -0500143
Miao Xie67a2c452014-09-03 21:35:43 +0800144DEFINE_MUTEX(uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400145static LIST_HEAD(fs_uuids);
Anand Jainc73eccf2015-03-10 06:38:30 +0800146struct list_head *btrfs_get_fs_uuids(void)
147{
148 return &fs_uuids;
149}
Chris Mason8a4b83c2008-03-24 15:02:07 -0400150
Ilya Dryomov2208a372013-08-12 14:33:03 +0300151static struct btrfs_fs_devices *__alloc_fs_devices(void)
152{
153 struct btrfs_fs_devices *fs_devs;
154
David Sterba78f2c9e2016-02-11 14:25:38 +0100155 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
Ilya Dryomov2208a372013-08-12 14:33:03 +0300156 if (!fs_devs)
157 return ERR_PTR(-ENOMEM);
158
159 mutex_init(&fs_devs->device_list_mutex);
160
161 INIT_LIST_HEAD(&fs_devs->devices);
Miao Xie935e5cc2014-09-03 21:35:33 +0800162 INIT_LIST_HEAD(&fs_devs->resized_devices);
Ilya Dryomov2208a372013-08-12 14:33:03 +0300163 INIT_LIST_HEAD(&fs_devs->alloc_list);
164 INIT_LIST_HEAD(&fs_devs->list);
165
166 return fs_devs;
167}
168
169/**
170 * alloc_fs_devices - allocate struct btrfs_fs_devices
171 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
172 * generated.
173 *
174 * Return: a pointer to a new &struct btrfs_fs_devices on success;
175 * ERR_PTR() on error. Returned struct is not linked onto any lists and
176 * can be destroyed with kfree() right away.
177 */
178static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
179{
180 struct btrfs_fs_devices *fs_devs;
181
182 fs_devs = __alloc_fs_devices();
183 if (IS_ERR(fs_devs))
184 return fs_devs;
185
186 if (fsid)
187 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
188 else
189 generate_random_uuid(fs_devs->fsid);
190
191 return fs_devs;
192}
193
Yan Zhenge4404d62008-12-12 10:03:26 -0500194static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
195{
196 struct btrfs_device *device;
197 WARN_ON(fs_devices->opened);
198 while (!list_empty(&fs_devices->devices)) {
199 device = list_entry(fs_devices->devices.next,
200 struct btrfs_device, dev_list);
201 list_del(&device->dev_list);
Josef Bacik606686e2012-06-04 14:03:51 -0400202 rcu_string_free(device->name);
Yan Zhenge4404d62008-12-12 10:03:26 -0500203 kfree(device);
204 }
205 kfree(fs_devices);
206}
207
Lukas Czernerb8b8ff52012-12-06 19:25:48 +0000208static void btrfs_kobject_uevent(struct block_device *bdev,
209 enum kobject_action action)
210{
211 int ret;
212
213 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
214 if (ret)
Frank Holtonefe120a2013-12-20 11:37:06 -0500215 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
Lukas Czernerb8b8ff52012-12-06 19:25:48 +0000216 action,
217 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
218 &disk_to_dev(bdev->bd_disk)->kobj);
219}
220
Jeff Mahoney143bede2012-03-01 14:56:26 +0100221void btrfs_cleanup_fs_uuids(void)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400222{
223 struct btrfs_fs_devices *fs_devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400224
Yan Zheng2b820322008-11-17 21:11:30 -0500225 while (!list_empty(&fs_uuids)) {
226 fs_devices = list_entry(fs_uuids.next,
227 struct btrfs_fs_devices, list);
228 list_del(&fs_devices->list);
Yan Zhenge4404d62008-12-12 10:03:26 -0500229 free_fs_devices(fs_devices);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400230 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400231}
232
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300233static struct btrfs_device *__alloc_device(void)
234{
235 struct btrfs_device *dev;
236
David Sterba78f2c9e2016-02-11 14:25:38 +0100237 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300238 if (!dev)
239 return ERR_PTR(-ENOMEM);
240
241 INIT_LIST_HEAD(&dev->dev_list);
242 INIT_LIST_HEAD(&dev->dev_alloc_list);
Miao Xie935e5cc2014-09-03 21:35:33 +0800243 INIT_LIST_HEAD(&dev->resized_list);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300244
245 spin_lock_init(&dev->io_lock);
246
247 spin_lock_init(&dev->reada_lock);
248 atomic_set(&dev->reada_in_flight, 0);
Miao Xieaddc3fa2014-07-24 11:37:11 +0800249 atomic_set(&dev->dev_stats_ccnt, 0);
Sebastian Andrzej Siewior546bed62016-01-15 14:37:15 +0100250 btrfs_device_data_ordered_init(dev);
Mel Gormand0164ad2015-11-06 16:28:21 -0800251 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300253
254 return dev;
255}
256
Chris Masona1b32a52008-09-05 16:09:51 -0400257static noinline struct btrfs_device *__find_device(struct list_head *head,
258 u64 devid, u8 *uuid)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400259{
260 struct btrfs_device *dev;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400261
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500262 list_for_each_entry(dev, head, dev_list) {
Chris Masona4437552008-04-18 10:29:38 -0400263 if (dev->devid == devid &&
Chris Mason8f18cf12008-04-25 16:53:30 -0400264 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400265 return dev;
Chris Masona4437552008-04-18 10:29:38 -0400266 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400267 }
268 return NULL;
269}
270
Chris Masona1b32a52008-09-05 16:09:51 -0400271static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400272{
Chris Mason8a4b83c2008-03-24 15:02:07 -0400273 struct btrfs_fs_devices *fs_devices;
274
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500275 list_for_each_entry(fs_devices, &fs_uuids, list) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400276 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
277 return fs_devices;
278 }
279 return NULL;
280}
281
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100282static int
283btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
284 int flush, struct block_device **bdev,
285 struct buffer_head **bh)
286{
287 int ret;
288
289 *bdev = blkdev_get_by_path(device_path, flags, holder);
290
291 if (IS_ERR(*bdev)) {
292 ret = PTR_ERR(*bdev);
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100293 goto error;
294 }
295
296 if (flush)
297 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
298 ret = set_blocksize(*bdev, 4096);
299 if (ret) {
300 blkdev_put(*bdev, flags);
301 goto error;
302 }
303 invalidate_bdev(*bdev);
304 *bh = btrfs_read_dev_super(*bdev);
Anand Jain92fc03f2015-08-14 18:32:51 +0800305 if (IS_ERR(*bh)) {
306 ret = PTR_ERR(*bh);
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100307 blkdev_put(*bdev, flags);
308 goto error;
309 }
310
311 return 0;
312
313error:
314 *bdev = NULL;
315 *bh = NULL;
316 return ret;
317}
318
Chris Masonffbd5172009-04-20 15:50:09 -0400319static void requeue_list(struct btrfs_pending_bios *pending_bios,
320 struct bio *head, struct bio *tail)
321{
322
323 struct bio *old_head;
324
325 old_head = pending_bios->head;
326 pending_bios->head = head;
327 if (pending_bios->tail)
328 tail->bi_next = old_head;
329 else
330 pending_bios->tail = tail;
331}
332
Chris Mason8b712842008-06-11 16:50:36 -0400333/*
334 * we try to collect pending bios for a device so we don't get a large
335 * number of procs sending bios down to the same device. This greatly
336 * improves the schedulers ability to collect and merge the bios.
337 *
338 * But, it also turns into a long list of bios to process and that is sure
339 * to eventually make the worker thread block. The solution here is to
340 * make some progress and then put this work struct back at the end of
341 * the list if the block device is congested. This way, multiple devices
342 * can make progress from a single worker thread.
343 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100344static noinline void run_scheduled_bios(struct btrfs_device *device)
Chris Mason8b712842008-06-11 16:50:36 -0400345{
346 struct bio *pending;
347 struct backing_dev_info *bdi;
Chris Masonb64a2852008-08-20 13:39:41 -0400348 struct btrfs_fs_info *fs_info;
Chris Masonffbd5172009-04-20 15:50:09 -0400349 struct btrfs_pending_bios *pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -0400350 struct bio *tail;
351 struct bio *cur;
352 int again = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400353 unsigned long num_run;
Chris Masond644d8a2009-06-09 15:59:22 -0400354 unsigned long batch_run = 0;
Chris Masonb64a2852008-08-20 13:39:41 -0400355 unsigned long limit;
Chris Masonb765ead2009-04-03 10:27:10 -0400356 unsigned long last_waited = 0;
Chris Masond84275c2009-06-09 15:39:08 -0400357 int force_reg = 0;
Miao Xie0e588852011-08-05 09:32:37 +0000358 int sync_pending = 0;
Chris Mason211588a2011-04-19 20:12:40 -0400359 struct blk_plug plug;
360
361 /*
362 * this function runs all the bios we've collected for
363 * a particular device. We don't want to wander off to
364 * another device without first sending all of these down.
365 * So, setup a plug here and finish it off before we return
366 */
367 blk_start_plug(&plug);
Chris Mason8b712842008-06-11 16:50:36 -0400368
Chris Masonbedf7622009-04-03 10:32:58 -0400369 bdi = blk_get_backing_dev_info(device->bdev);
Chris Masonb64a2852008-08-20 13:39:41 -0400370 fs_info = device->dev_root->fs_info;
371 limit = btrfs_async_submit_limit(fs_info);
372 limit = limit * 2 / 3;
373
Chris Mason8b712842008-06-11 16:50:36 -0400374loop:
375 spin_lock(&device->io_lock);
376
Chris Masona6837052009-02-04 09:19:41 -0500377loop_lock:
Chris Masond84275c2009-06-09 15:39:08 -0400378 num_run = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400379
Chris Mason8b712842008-06-11 16:50:36 -0400380 /* take all the bios off the list at once and process them
381 * later on (without the lock held). But, remember the
382 * tail and other pointers so the bios can be properly reinserted
383 * into the list if we hit congestion
384 */
Chris Masond84275c2009-06-09 15:39:08 -0400385 if (!force_reg && device->pending_sync_bios.head) {
Chris Masonffbd5172009-04-20 15:50:09 -0400386 pending_bios = &device->pending_sync_bios;
Chris Masond84275c2009-06-09 15:39:08 -0400387 force_reg = 1;
388 } else {
Chris Masonffbd5172009-04-20 15:50:09 -0400389 pending_bios = &device->pending_bios;
Chris Masond84275c2009-06-09 15:39:08 -0400390 force_reg = 0;
391 }
Chris Masonffbd5172009-04-20 15:50:09 -0400392
393 pending = pending_bios->head;
394 tail = pending_bios->tail;
Chris Mason8b712842008-06-11 16:50:36 -0400395 WARN_ON(pending && !tail);
Chris Mason8b712842008-06-11 16:50:36 -0400396
397 /*
398 * if pending was null this time around, no bios need processing
399 * at all and we can stop. Otherwise it'll loop back up again
400 * and do an additional check so no bios are missed.
401 *
402 * device->running_pending is used to synchronize with the
403 * schedule_bio code.
404 */
Chris Masonffbd5172009-04-20 15:50:09 -0400405 if (device->pending_sync_bios.head == NULL &&
406 device->pending_bios.head == NULL) {
Chris Mason8b712842008-06-11 16:50:36 -0400407 again = 0;
408 device->running_pending = 0;
Chris Masonffbd5172009-04-20 15:50:09 -0400409 } else {
410 again = 1;
411 device->running_pending = 1;
Chris Mason8b712842008-06-11 16:50:36 -0400412 }
Chris Masonffbd5172009-04-20 15:50:09 -0400413
414 pending_bios->head = NULL;
415 pending_bios->tail = NULL;
416
Chris Mason8b712842008-06-11 16:50:36 -0400417 spin_unlock(&device->io_lock);
418
Chris Masond3977122009-01-05 21:25:51 -0500419 while (pending) {
Chris Masonffbd5172009-04-20 15:50:09 -0400420
421 rmb();
Chris Masond84275c2009-06-09 15:39:08 -0400422 /* we want to work on both lists, but do more bios on the
423 * sync list than the regular list
424 */
425 if ((num_run > 32 &&
426 pending_bios != &device->pending_sync_bios &&
427 device->pending_sync_bios.head) ||
428 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
429 device->pending_bios.head)) {
Chris Masonffbd5172009-04-20 15:50:09 -0400430 spin_lock(&device->io_lock);
431 requeue_list(pending_bios, pending, tail);
432 goto loop_lock;
433 }
434
Chris Mason8b712842008-06-11 16:50:36 -0400435 cur = pending;
436 pending = pending->bi_next;
437 cur->bi_next = NULL;
Chris Masonb64a2852008-08-20 13:39:41 -0400438
David Sterbaee863952015-02-16 19:41:40 +0100439 /*
440 * atomic_dec_return implies a barrier for waitqueue_active
441 */
Josef Bacik66657b32012-08-01 15:36:24 -0400442 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
Chris Masonb64a2852008-08-20 13:39:41 -0400443 waitqueue_active(&fs_info->async_submit_wait))
444 wake_up(&fs_info->async_submit_wait);
Chris Mason492bb6de2008-07-31 16:29:02 -0400445
Jens Axboedac56212015-04-17 16:23:59 -0600446 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
Chris Masond644d8a2009-06-09 15:59:22 -0400447
Chris Mason2ab1ba62011-08-04 14:28:36 -0400448 /*
449 * if we're doing the sync list, record that our
450 * plug has some sync requests on it
451 *
452 * If we're doing the regular list and there are
453 * sync requests sitting around, unplug before
454 * we add more
455 */
456 if (pending_bios == &device->pending_sync_bios) {
457 sync_pending = 1;
458 } else if (sync_pending) {
459 blk_finish_plug(&plug);
460 blk_start_plug(&plug);
461 sync_pending = 0;
462 }
463
Mike Christie4e49ea42016-06-05 14:31:41 -0500464 btrfsic_submit_bio(cur);
Chris Mason5ff7ba32010-03-15 10:21:30 -0400465 num_run++;
466 batch_run++;
David Sterba853d8ec2015-01-08 15:15:19 +0100467
468 cond_resched();
Chris Mason8b712842008-06-11 16:50:36 -0400469
470 /*
471 * we made progress, there is more work to do and the bdi
472 * is now congested. Back off and let other work structs
473 * run instead
474 */
Chris Mason57fd5a52009-08-07 09:59:15 -0400475 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
Chris Mason5f2cc082008-11-07 18:22:45 -0500476 fs_info->fs_devices->open_devices > 1) {
Chris Masonb765ead2009-04-03 10:27:10 -0400477 struct io_context *ioc;
Chris Mason8b712842008-06-11 16:50:36 -0400478
Chris Masonb765ead2009-04-03 10:27:10 -0400479 ioc = current->io_context;
480
481 /*
482 * the main goal here is that we don't want to
483 * block if we're going to be able to submit
484 * more requests without blocking.
485 *
486 * This code does two great things, it pokes into
487 * the elevator code from a filesystem _and_
488 * it makes assumptions about how batching works.
489 */
490 if (ioc && ioc->nr_batch_requests > 0 &&
491 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
492 (last_waited == 0 ||
493 ioc->last_waited == last_waited)) {
494 /*
495 * we want to go through our batch of
496 * requests and stop. So, we copy out
497 * the ioc->last_waited time and test
498 * against it before looping
499 */
500 last_waited = ioc->last_waited;
David Sterba853d8ec2015-01-08 15:15:19 +0100501 cond_resched();
Chris Masonb765ead2009-04-03 10:27:10 -0400502 continue;
503 }
Chris Mason8b712842008-06-11 16:50:36 -0400504 spin_lock(&device->io_lock);
Chris Masonffbd5172009-04-20 15:50:09 -0400505 requeue_list(pending_bios, pending, tail);
Chris Masona6837052009-02-04 09:19:41 -0500506 device->running_pending = 1;
Chris Mason8b712842008-06-11 16:50:36 -0400507
508 spin_unlock(&device->io_lock);
Qu Wenruoa8c93d42014-02-28 10:46:08 +0800509 btrfs_queue_work(fs_info->submit_workers,
510 &device->work);
Chris Mason8b712842008-06-11 16:50:36 -0400511 goto done;
512 }
Chris Masond85c8a6f2011-12-15 15:38:41 -0500513 /* unplug every 64 requests just for good measure */
514 if (batch_run % 64 == 0) {
515 blk_finish_plug(&plug);
516 blk_start_plug(&plug);
517 sync_pending = 0;
518 }
Chris Mason8b712842008-06-11 16:50:36 -0400519 }
Chris Masonffbd5172009-04-20 15:50:09 -0400520
Chris Mason51684082010-03-10 15:33:32 -0500521 cond_resched();
522 if (again)
523 goto loop;
524
525 spin_lock(&device->io_lock);
526 if (device->pending_bios.head || device->pending_sync_bios.head)
527 goto loop_lock;
528 spin_unlock(&device->io_lock);
529
Chris Mason8b712842008-06-11 16:50:36 -0400530done:
Chris Mason211588a2011-04-19 20:12:40 -0400531 blk_finish_plug(&plug);
Chris Mason8b712842008-06-11 16:50:36 -0400532}
533
Christoph Hellwigb2950862008-12-02 09:54:17 -0500534static void pending_bios_fn(struct btrfs_work *work)
Chris Mason8b712842008-06-11 16:50:36 -0400535{
536 struct btrfs_device *device;
537
538 device = container_of(work, struct btrfs_device, work);
539 run_scheduled_bios(device);
540}
541
Anand Jain4fde46f2015-06-17 21:10:48 +0800542
543void btrfs_free_stale_device(struct btrfs_device *cur_dev)
544{
545 struct btrfs_fs_devices *fs_devs;
546 struct btrfs_device *dev;
547
548 if (!cur_dev->name)
549 return;
550
551 list_for_each_entry(fs_devs, &fs_uuids, list) {
552 int del = 1;
553
554 if (fs_devs->opened)
555 continue;
556 if (fs_devs->seeding)
557 continue;
558
559 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
560
561 if (dev == cur_dev)
562 continue;
563 if (!dev->name)
564 continue;
565
566 /*
567 * Todo: This won't be enough. What if the same device
568 * comes back (with new uuid and) with its mapper path?
569 * But for now, this does help as mostly an admin will
570 * either use mapper or non mapper path throughout.
571 */
572 rcu_read_lock();
573 del = strcmp(rcu_str_deref(dev->name),
574 rcu_str_deref(cur_dev->name));
575 rcu_read_unlock();
576 if (!del)
577 break;
578 }
579
580 if (!del) {
581 /* delete the stale device */
582 if (fs_devs->num_devices == 1) {
583 btrfs_sysfs_remove_fsid(fs_devs);
584 list_del(&fs_devs->list);
585 free_fs_devices(fs_devs);
586 } else {
587 fs_devs->num_devices--;
588 list_del(&dev->dev_list);
589 rcu_string_free(dev->name);
590 kfree(dev);
591 }
592 break;
593 }
594 }
595}
596
David Sterba60999ca2014-03-26 18:26:36 +0100597/*
598 * Add new device to list of registered devices
599 *
600 * Returns:
601 * 1 - first time device is seen
602 * 0 - device already known
603 * < 0 - error
604 */
Chris Masona1b32a52008-09-05 16:09:51 -0400605static noinline int device_list_add(const char *path,
Chris Mason8a4b83c2008-03-24 15:02:07 -0400606 struct btrfs_super_block *disk_super,
607 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
608{
609 struct btrfs_device *device;
610 struct btrfs_fs_devices *fs_devices;
Josef Bacik606686e2012-06-04 14:03:51 -0400611 struct rcu_string *name;
David Sterba60999ca2014-03-26 18:26:36 +0100612 int ret = 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400613 u64 found_transid = btrfs_super_generation(disk_super);
614
615 fs_devices = find_fsid(disk_super->fsid);
616 if (!fs_devices) {
Ilya Dryomov2208a372013-08-12 14:33:03 +0300617 fs_devices = alloc_fs_devices(disk_super->fsid);
618 if (IS_ERR(fs_devices))
619 return PTR_ERR(fs_devices);
620
Chris Mason8a4b83c2008-03-24 15:02:07 -0400621 list_add(&fs_devices->list, &fs_uuids);
Ilya Dryomov2208a372013-08-12 14:33:03 +0300622
Chris Mason8a4b83c2008-03-24 15:02:07 -0400623 device = NULL;
624 } else {
Chris Masona4437552008-04-18 10:29:38 -0400625 device = __find_device(&fs_devices->devices, devid,
626 disk_super->dev_item.uuid);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400627 }
Miao Xie443f24f2014-07-24 11:37:15 +0800628
Chris Mason8a4b83c2008-03-24 15:02:07 -0400629 if (!device) {
Yan Zheng2b820322008-11-17 21:11:30 -0500630 if (fs_devices->opened)
631 return -EBUSY;
632
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300633 device = btrfs_alloc_device(NULL, &devid,
634 disk_super->dev_item.uuid);
635 if (IS_ERR(device)) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400636 /* we can safely leave the fs_devices entry around */
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300637 return PTR_ERR(device);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400638 }
Josef Bacik606686e2012-06-04 14:03:51 -0400639
640 name = rcu_string_strdup(path, GFP_NOFS);
641 if (!name) {
Chris Mason8a4b83c2008-03-24 15:02:07 -0400642 kfree(device);
643 return -ENOMEM;
644 }
Josef Bacik606686e2012-06-04 14:03:51 -0400645 rcu_assign_pointer(device->name, name);
Arne Jansen90519d62011-05-23 14:30:00 +0200646
Chris Masone5e9a522009-06-10 15:17:02 -0400647 mutex_lock(&fs_devices->device_list_mutex);
Xiao Guangrong1f781602011-04-20 10:09:16 +0000648 list_add_rcu(&device->dev_list, &fs_devices->devices);
Filipe David Borba Mananaf7171752013-08-12 20:56:58 +0100649 fs_devices->num_devices++;
Chris Masone5e9a522009-06-10 15:17:02 -0400650 mutex_unlock(&fs_devices->device_list_mutex);
651
David Sterba60999ca2014-03-26 18:26:36 +0100652 ret = 1;
Yan Zheng2b820322008-11-17 21:11:30 -0500653 device->fs_devices = fs_devices;
Josef Bacik606686e2012-06-04 14:03:51 -0400654 } else if (!device->name || strcmp(device->name->str, path)) {
Anand Jainb96de002014-07-03 18:22:05 +0800655 /*
656 * When FS is already mounted.
657 * 1. If you are here and if the device->name is NULL that
658 * means this device was missing at time of FS mount.
659 * 2. If you are here and if the device->name is different
660 * from 'path' that means either
661 * a. The same device disappeared and reappeared with
662 * different name. or
663 * b. The missing-disk-which-was-replaced, has
664 * reappeared now.
665 *
666 * We must allow 1 and 2a above. But 2b would be a spurious
667 * and unintentional.
668 *
669 * Further in case of 1 and 2a above, the disk at 'path'
670 * would have missed some transaction when it was away and
671 * in case of 2a the stale bdev has to be updated as well.
672 * 2b must not be allowed at all time.
673 */
674
675 /*
Chris Mason0f23ae72014-09-18 07:49:05 -0700676 * For now, we do allow update to btrfs_fs_device through the
677 * btrfs dev scan cli after FS has been mounted. We're still
678 * tracking a problem where systems fail mount by subvolume id
679 * when we reject replacement on a mounted FS.
Anand Jainb96de002014-07-03 18:22:05 +0800680 */
Chris Mason0f23ae72014-09-18 07:49:05 -0700681 if (!fs_devices->opened && found_transid < device->generation) {
Anand Jain77bdae42014-07-03 18:22:06 +0800682 /*
683 * That is if the FS is _not_ mounted and if you
684 * are here, that means there is more than one
685 * disk with same uuid and devid.We keep the one
686 * with larger generation number or the last-in if
687 * generation are equal.
688 */
Chris Mason0f23ae72014-09-18 07:49:05 -0700689 return -EEXIST;
Anand Jain77bdae42014-07-03 18:22:06 +0800690 }
Anand Jainb96de002014-07-03 18:22:05 +0800691
Josef Bacik606686e2012-06-04 14:03:51 -0400692 name = rcu_string_strdup(path, GFP_NOFS);
TARUISI Hiroaki3a0524d2010-02-09 06:36:45 +0000693 if (!name)
694 return -ENOMEM;
Josef Bacik606686e2012-06-04 14:03:51 -0400695 rcu_string_free(device->name);
696 rcu_assign_pointer(device->name, name);
Chris Masoncd02dca2010-12-13 14:56:23 -0500697 if (device->missing) {
698 fs_devices->missing_devices--;
699 device->missing = 0;
700 }
Chris Mason8a4b83c2008-03-24 15:02:07 -0400701 }
702
Anand Jain77bdae42014-07-03 18:22:06 +0800703 /*
704 * Unmount does not free the btrfs_device struct but would zero
705 * generation along with most of the other members. So just update
706 * it back. We need it to pick the disk with largest generation
707 * (as above).
708 */
709 if (!fs_devices->opened)
710 device->generation = found_transid;
711
Anand Jain4fde46f2015-06-17 21:10:48 +0800712 /*
713 * if there is new btrfs on an already registered device,
714 * then remove the stale device entry.
715 */
Anand Jain02feae32016-02-13 10:01:40 +0800716 if (ret > 0)
717 btrfs_free_stale_device(device);
Anand Jain4fde46f2015-06-17 21:10:48 +0800718
Chris Mason8a4b83c2008-03-24 15:02:07 -0400719 *fs_devices_ret = fs_devices;
David Sterba60999ca2014-03-26 18:26:36 +0100720
721 return ret;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400722}
723
Yan Zhenge4404d62008-12-12 10:03:26 -0500724static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
725{
726 struct btrfs_fs_devices *fs_devices;
727 struct btrfs_device *device;
728 struct btrfs_device *orig_dev;
729
Ilya Dryomov2208a372013-08-12 14:33:03 +0300730 fs_devices = alloc_fs_devices(orig->fsid);
731 if (IS_ERR(fs_devices))
732 return fs_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -0500733
Miao Xieadbbb862014-09-03 21:35:42 +0800734 mutex_lock(&orig->device_list_mutex);
Josef Bacik02db0842012-06-21 16:03:58 -0400735 fs_devices->total_devices = orig->total_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -0500736
Xiao Guangrong46224702011-04-20 10:08:47 +0000737 /* We have held the volume lock, it is safe to get the devices. */
Yan Zhenge4404d62008-12-12 10:03:26 -0500738 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
Josef Bacik606686e2012-06-04 14:03:51 -0400739 struct rcu_string *name;
740
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +0300741 device = btrfs_alloc_device(NULL, &orig_dev->devid,
742 orig_dev->uuid);
743 if (IS_ERR(device))
Yan Zhenge4404d62008-12-12 10:03:26 -0500744 goto error;
745
Josef Bacik606686e2012-06-04 14:03:51 -0400746 /*
747 * This is ok to do without rcu read locked because we hold the
748 * uuid mutex so nothing we touch in here is going to disappear.
749 */
Anand Jaine755f782014-06-30 17:12:47 +0800750 if (orig_dev->name) {
David Sterba78f2c9e2016-02-11 14:25:38 +0100751 name = rcu_string_strdup(orig_dev->name->str,
752 GFP_KERNEL);
Anand Jaine755f782014-06-30 17:12:47 +0800753 if (!name) {
754 kfree(device);
755 goto error;
756 }
757 rcu_assign_pointer(device->name, name);
Julia Lawallfd2696f2009-09-29 13:51:04 -0400758 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500759
Yan Zhenge4404d62008-12-12 10:03:26 -0500760 list_add(&device->dev_list, &fs_devices->devices);
761 device->fs_devices = fs_devices;
762 fs_devices->num_devices++;
763 }
Miao Xieadbbb862014-09-03 21:35:42 +0800764 mutex_unlock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500765 return fs_devices;
766error:
Miao Xieadbbb862014-09-03 21:35:42 +0800767 mutex_unlock(&orig->device_list_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500768 free_fs_devices(fs_devices);
769 return ERR_PTR(-ENOMEM);
770}
771
Eric Sandeen9eaed212014-08-01 18:12:35 -0500772void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
Chris Masondfe25022008-05-13 13:46:40 -0400773{
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500774 struct btrfs_device *device, *next;
Miao Xie443f24f2014-07-24 11:37:15 +0800775 struct btrfs_device *latest_dev = NULL;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500776
Chris Masondfe25022008-05-13 13:46:40 -0400777 mutex_lock(&uuid_mutex);
778again:
Xiao Guangrong46224702011-04-20 10:08:47 +0000779 /* This is the initialized path, it is safe to release the devices. */
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500780 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
Chris Masona6b0d5c2012-02-20 20:53:43 -0500781 if (device->in_fs_metadata) {
Stefan Behrens63a212a2012-11-05 18:29:28 +0100782 if (!device->is_tgtdev_for_dev_replace &&
Miao Xie443f24f2014-07-24 11:37:15 +0800783 (!latest_dev ||
784 device->generation > latest_dev->generation)) {
785 latest_dev = device;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500786 }
Yan Zheng2b820322008-11-17 21:11:30 -0500787 continue;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500788 }
Yan Zheng2b820322008-11-17 21:11:30 -0500789
Stefan Behrens8dabb742012-11-06 13:15:27 +0100790 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
791 /*
792 * In the first step, keep the device which has
793 * the correct fsid and the devid that is used
794 * for the dev_replace procedure.
795 * In the second step, the dev_replace state is
796 * read from the device tree and it is known
797 * whether the procedure is really active or
798 * not, which means whether this device is
799 * used or whether it should be removed.
800 */
801 if (step == 0 || device->is_tgtdev_for_dev_replace) {
802 continue;
803 }
804 }
Yan Zheng2b820322008-11-17 21:11:30 -0500805 if (device->bdev) {
Tejun Heod4d77622010-11-13 11:55:18 +0100806 blkdev_put(device->bdev, device->mode);
Yan Zheng2b820322008-11-17 21:11:30 -0500807 device->bdev = NULL;
808 fs_devices->open_devices--;
809 }
810 if (device->writeable) {
811 list_del_init(&device->dev_alloc_list);
812 device->writeable = 0;
Stefan Behrens8dabb742012-11-06 13:15:27 +0100813 if (!device->is_tgtdev_for_dev_replace)
814 fs_devices->rw_devices--;
Yan Zheng2b820322008-11-17 21:11:30 -0500815 }
Yan Zhenge4404d62008-12-12 10:03:26 -0500816 list_del_init(&device->dev_list);
817 fs_devices->num_devices--;
Josef Bacik606686e2012-06-04 14:03:51 -0400818 rcu_string_free(device->name);
Yan Zhenge4404d62008-12-12 10:03:26 -0500819 kfree(device);
Chris Masondfe25022008-05-13 13:46:40 -0400820 }
Yan Zheng2b820322008-11-17 21:11:30 -0500821
822 if (fs_devices->seed) {
823 fs_devices = fs_devices->seed;
Yan Zheng2b820322008-11-17 21:11:30 -0500824 goto again;
825 }
826
Miao Xie443f24f2014-07-24 11:37:15 +0800827 fs_devices->latest_bdev = latest_dev->bdev;
Chris Masona6b0d5c2012-02-20 20:53:43 -0500828
Chris Masondfe25022008-05-13 13:46:40 -0400829 mutex_unlock(&uuid_mutex);
Chris Masondfe25022008-05-13 13:46:40 -0400830}
Chris Masona0af4692008-05-13 16:03:06 -0400831
Xiao Guangrong1f781602011-04-20 10:09:16 +0000832static void __free_device(struct work_struct *work)
833{
834 struct btrfs_device *device;
835
836 device = container_of(work, struct btrfs_device, rcu_work);
Josef Bacik606686e2012-06-04 14:03:51 -0400837 rcu_string_free(device->name);
Xiao Guangrong1f781602011-04-20 10:09:16 +0000838 kfree(device);
839}
840
841static void free_device(struct rcu_head *head)
842{
843 struct btrfs_device *device;
844
845 device = container_of(head, struct btrfs_device, rcu);
846
847 INIT_WORK(&device->rcu_work, __free_device);
848 schedule_work(&device->rcu_work);
849}
850
Anand Jain14238812016-07-22 06:04:53 +0800851static void btrfs_close_bdev(struct btrfs_device *device)
852{
853 if (device->bdev && device->writeable) {
854 sync_blockdev(device->bdev);
855 invalidate_bdev(device->bdev);
856 }
857
858 if (device->bdev)
859 blkdev_put(device->bdev, device->mode);
860}
861
Anand Jainf4483412016-06-14 18:55:25 +0800862static void btrfs_close_one_device(struct btrfs_device *device)
863{
864 struct btrfs_fs_devices *fs_devices = device->fs_devices;
865 struct btrfs_device *new_device;
866 struct rcu_string *name;
867
868 if (device->bdev)
869 fs_devices->open_devices--;
870
871 if (device->writeable &&
872 device->devid != BTRFS_DEV_REPLACE_DEVID) {
873 list_del_init(&device->dev_alloc_list);
874 fs_devices->rw_devices--;
875 }
876
877 if (device->missing)
878 fs_devices->missing_devices--;
879
Anand Jain14238812016-07-22 06:04:53 +0800880 btrfs_close_bdev(device);
Anand Jaine2bf6e82016-06-23 20:54:07 +0800881
Anand Jainf4483412016-06-14 18:55:25 +0800882 new_device = btrfs_alloc_device(NULL, &device->devid,
883 device->uuid);
884 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
885
886 /* Safe because we are under uuid_mutex */
887 if (device->name) {
888 name = rcu_string_strdup(device->name->str, GFP_NOFS);
889 BUG_ON(!name); /* -ENOMEM */
890 rcu_assign_pointer(new_device->name, name);
891 }
892
893 list_replace_rcu(&device->dev_list, &new_device->dev_list);
894 new_device->fs_devices = device->fs_devices;
895
896 call_rcu(&device->rcu, free_device);
897}
898
Yan Zheng2b820322008-11-17 21:11:30 -0500899static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400900{
Sasha Levin2037a092015-05-12 19:31:37 -0400901 struct btrfs_device *device, *tmp;
Yan Zhenge4404d62008-12-12 10:03:26 -0500902
Yan Zheng2b820322008-11-17 21:11:30 -0500903 if (--fs_devices->opened > 0)
904 return 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400905
Xiao Guangrongc9513ed2011-04-20 10:07:30 +0000906 mutex_lock(&fs_devices->device_list_mutex);
Sasha Levin2037a092015-05-12 19:31:37 -0400907 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
Anand Jainf190aa42015-08-14 18:33:05 +0800908 btrfs_close_one_device(device);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400909 }
Xiao Guangrongc9513ed2011-04-20 10:07:30 +0000910 mutex_unlock(&fs_devices->device_list_mutex);
911
Yan Zhenge4404d62008-12-12 10:03:26 -0500912 WARN_ON(fs_devices->open_devices);
913 WARN_ON(fs_devices->rw_devices);
Yan Zheng2b820322008-11-17 21:11:30 -0500914 fs_devices->opened = 0;
915 fs_devices->seeding = 0;
Yan Zheng2b820322008-11-17 21:11:30 -0500916
Chris Mason8a4b83c2008-03-24 15:02:07 -0400917 return 0;
918}
919
Yan Zheng2b820322008-11-17 21:11:30 -0500920int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
921{
Yan Zhenge4404d62008-12-12 10:03:26 -0500922 struct btrfs_fs_devices *seed_devices = NULL;
Yan Zheng2b820322008-11-17 21:11:30 -0500923 int ret;
924
925 mutex_lock(&uuid_mutex);
926 ret = __btrfs_close_devices(fs_devices);
Yan Zhenge4404d62008-12-12 10:03:26 -0500927 if (!fs_devices->opened) {
928 seed_devices = fs_devices->seed;
929 fs_devices->seed = NULL;
930 }
Yan Zheng2b820322008-11-17 21:11:30 -0500931 mutex_unlock(&uuid_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -0500932
933 while (seed_devices) {
934 fs_devices = seed_devices;
935 seed_devices = fs_devices->seed;
936 __btrfs_close_devices(fs_devices);
937 free_fs_devices(fs_devices);
938 }
Eric Sandeenbc178622013-03-09 15:18:39 +0000939 /*
940 * Wait for rcu kworkers under __btrfs_close_devices
941 * to finish all blkdev_puts so device is really
942 * free when umount is done.
943 */
944 rcu_barrier();
Yan Zheng2b820322008-11-17 21:11:30 -0500945 return ret;
946}
947
Yan Zhenge4404d62008-12-12 10:03:26 -0500948static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
949 fmode_t flags, void *holder)
Chris Mason8a4b83c2008-03-24 15:02:07 -0400950{
Josef Bacikd5e20032011-08-04 14:52:27 +0000951 struct request_queue *q;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400952 struct block_device *bdev;
953 struct list_head *head = &fs_devices->devices;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400954 struct btrfs_device *device;
Miao Xie443f24f2014-07-24 11:37:15 +0800955 struct btrfs_device *latest_dev = NULL;
Chris Masona0af4692008-05-13 16:03:06 -0400956 struct buffer_head *bh;
957 struct btrfs_super_block *disk_super;
Chris Masona0af4692008-05-13 16:03:06 -0400958 u64 devid;
Yan Zheng2b820322008-11-17 21:11:30 -0500959 int seeding = 1;
Chris Masona0af4692008-05-13 16:03:06 -0400960 int ret = 0;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400961
Tejun Heod4d77622010-11-13 11:55:18 +0100962 flags |= FMODE_EXCL;
963
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500964 list_for_each_entry(device, head, dev_list) {
Chris Masonc1c4d912008-05-08 15:05:58 -0400965 if (device->bdev)
966 continue;
Chris Masondfe25022008-05-13 13:46:40 -0400967 if (!device->name)
968 continue;
969
Eric Sandeenf63e0cc2013-04-04 20:45:08 +0000970 /* Just open everything we can; ignore failures here */
971 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
972 &bdev, &bh))
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +0100973 continue;
Chris Masona0af4692008-05-13 16:03:06 -0400974
975 disk_super = (struct btrfs_super_block *)bh->b_data;
Xiao Guangronga3438322010-01-06 11:48:18 +0000976 devid = btrfs_stack_device_id(&disk_super->dev_item);
Chris Masona0af4692008-05-13 16:03:06 -0400977 if (devid != device->devid)
978 goto error_brelse;
979
Yan Zheng2b820322008-11-17 21:11:30 -0500980 if (memcmp(device->uuid, disk_super->dev_item.uuid,
981 BTRFS_UUID_SIZE))
982 goto error_brelse;
983
984 device->generation = btrfs_super_generation(disk_super);
Miao Xie443f24f2014-07-24 11:37:15 +0800985 if (!latest_dev ||
986 device->generation > latest_dev->generation)
987 latest_dev = device;
Chris Masona0af4692008-05-13 16:03:06 -0400988
Yan Zheng2b820322008-11-17 21:11:30 -0500989 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
990 device->writeable = 0;
991 } else {
992 device->writeable = !bdev_read_only(bdev);
993 seeding = 0;
994 }
995
Josef Bacikd5e20032011-08-04 14:52:27 +0000996 q = bdev_get_queue(bdev);
Miao Xie90180da2014-09-03 21:35:30 +0800997 if (blk_queue_discard(q))
Josef Bacikd5e20032011-08-04 14:52:27 +0000998 device->can_discard = 1;
Josef Bacikd5e20032011-08-04 14:52:27 +0000999
Chris Mason8a4b83c2008-03-24 15:02:07 -04001000 device->bdev = bdev;
Chris Masondfe25022008-05-13 13:46:40 -04001001 device->in_fs_metadata = 0;
Chris Mason15916de2008-11-19 21:17:22 -05001002 device->mode = flags;
1003
Chris Masonc2898112009-06-10 09:51:32 -04001004 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1005 fs_devices->rotating = 1;
1006
Chris Masona0af4692008-05-13 16:03:06 -04001007 fs_devices->open_devices++;
Ilya Dryomov55e50e42013-09-01 18:56:44 +03001008 if (device->writeable &&
1009 device->devid != BTRFS_DEV_REPLACE_DEVID) {
Yan Zheng2b820322008-11-17 21:11:30 -05001010 fs_devices->rw_devices++;
1011 list_add(&device->dev_alloc_list,
1012 &fs_devices->alloc_list);
1013 }
Xiao Guangrong4f6c9322011-04-20 10:06:40 +00001014 brelse(bh);
Chris Masona0af4692008-05-13 16:03:06 -04001015 continue;
Chris Masona061fc82008-05-07 11:43:44 -04001016
Chris Masona0af4692008-05-13 16:03:06 -04001017error_brelse:
1018 brelse(bh);
Tejun Heod4d77622010-11-13 11:55:18 +01001019 blkdev_put(bdev, flags);
Chris Masona0af4692008-05-13 16:03:06 -04001020 continue;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001021 }
Chris Masona0af4692008-05-13 16:03:06 -04001022 if (fs_devices->open_devices == 0) {
Ilya Dryomov20bcd642011-10-20 00:06:20 +03001023 ret = -EINVAL;
Chris Masona0af4692008-05-13 16:03:06 -04001024 goto out;
1025 }
Yan Zheng2b820322008-11-17 21:11:30 -05001026 fs_devices->seeding = seeding;
1027 fs_devices->opened = 1;
Miao Xie443f24f2014-07-24 11:37:15 +08001028 fs_devices->latest_bdev = latest_dev->bdev;
Yan Zheng2b820322008-11-17 21:11:30 -05001029 fs_devices->total_rw_bytes = 0;
Chris Masona0af4692008-05-13 16:03:06 -04001030out:
Yan Zheng2b820322008-11-17 21:11:30 -05001031 return ret;
1032}
1033
1034int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
Christoph Hellwig97288f22008-12-02 06:36:09 -05001035 fmode_t flags, void *holder)
Yan Zheng2b820322008-11-17 21:11:30 -05001036{
1037 int ret;
1038
1039 mutex_lock(&uuid_mutex);
1040 if (fs_devices->opened) {
Yan Zhenge4404d62008-12-12 10:03:26 -05001041 fs_devices->opened++;
1042 ret = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05001043 } else {
Chris Mason15916de2008-11-19 21:17:22 -05001044 ret = __btrfs_open_devices(fs_devices, flags, holder);
Yan Zheng2b820322008-11-17 21:11:30 -05001045 }
Chris Mason8a4b83c2008-03-24 15:02:07 -04001046 mutex_unlock(&uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -04001047 return ret;
1048}
1049
Anand Jain6cf86a002016-02-13 10:01:29 +08001050void btrfs_release_disk_super(struct page *page)
1051{
1052 kunmap(page);
1053 put_page(page);
1054}
1055
1056int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1057 struct page **page, struct btrfs_super_block **disk_super)
1058{
1059 void *p;
1060 pgoff_t index;
1061
1062 /* make sure our super fits in the device */
1063 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1064 return 1;
1065
1066 /* make sure our super fits in the page */
1067 if (sizeof(**disk_super) > PAGE_SIZE)
1068 return 1;
1069
1070 /* make sure our super doesn't straddle pages on disk */
1071 index = bytenr >> PAGE_SHIFT;
1072 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1073 return 1;
1074
1075 /* pull in the page with our super */
1076 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1077 index, GFP_KERNEL);
1078
1079 if (IS_ERR_OR_NULL(*page))
1080 return 1;
1081
1082 p = kmap(*page);
1083
1084 /* align our pointer to the offset of the super block */
1085 *disk_super = p + (bytenr & ~PAGE_MASK);
1086
1087 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1088 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1089 btrfs_release_disk_super(*page);
1090 return 1;
1091 }
1092
1093 if ((*disk_super)->label[0] &&
1094 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1095 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1096
1097 return 0;
1098}
1099
David Sterba6f60cbd2013-02-15 11:31:02 -07001100/*
1101 * Look for a btrfs signature on a device. This may be called out of the mount path
1102 * and we are not allowed to call set_blocksize during the scan. The superblock
1103 * is read via pagecache
1104 */
Christoph Hellwig97288f22008-12-02 06:36:09 -05001105int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
Chris Mason8a4b83c2008-03-24 15:02:07 -04001106 struct btrfs_fs_devices **fs_devices_ret)
1107{
1108 struct btrfs_super_block *disk_super;
1109 struct block_device *bdev;
David Sterba6f60cbd2013-02-15 11:31:02 -07001110 struct page *page;
David Sterba6f60cbd2013-02-15 11:31:02 -07001111 int ret = -EINVAL;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001112 u64 devid;
Chris Masonf2984462008-04-10 16:19:33 -04001113 u64 transid;
Josef Bacik02db0842012-06-21 16:03:58 -04001114 u64 total_devices;
David Sterba6f60cbd2013-02-15 11:31:02 -07001115 u64 bytenr;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001116
David Sterba6f60cbd2013-02-15 11:31:02 -07001117 /*
1118 * we would like to check all the supers, but that would make
1119 * a btrfs mount succeed after a mkfs from a different FS.
1120 * So, we need to add a special mount option to scan for
1121 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1122 */
1123 bytenr = btrfs_sb_offset(0);
Tejun Heod4d77622010-11-13 11:55:18 +01001124 flags |= FMODE_EXCL;
Al Viro10f63272011-11-17 15:05:22 -05001125 mutex_lock(&uuid_mutex);
David Sterba6f60cbd2013-02-15 11:31:02 -07001126
1127 bdev = blkdev_get_by_path(path, flags, holder);
David Sterba6f60cbd2013-02-15 11:31:02 -07001128 if (IS_ERR(bdev)) {
1129 ret = PTR_ERR(bdev);
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +01001130 goto error;
David Sterba6f60cbd2013-02-15 11:31:02 -07001131 }
1132
Anand Jain6cf86a002016-02-13 10:01:29 +08001133 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
David Sterba6f60cbd2013-02-15 11:31:02 -07001134 goto error_bdev_put;
1135
Xiao Guangronga3438322010-01-06 11:48:18 +00001136 devid = btrfs_stack_device_id(&disk_super->dev_item);
Chris Masonf2984462008-04-10 16:19:33 -04001137 transid = btrfs_super_generation(disk_super);
Josef Bacik02db0842012-06-21 16:03:58 -04001138 total_devices = btrfs_super_num_devices(disk_super);
David Sterba6f60cbd2013-02-15 11:31:02 -07001139
Chris Mason8a4b83c2008-03-24 15:02:07 -04001140 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
David Sterba60999ca2014-03-26 18:26:36 +01001141 if (ret > 0) {
1142 if (disk_super->label[0]) {
David Sterba60999ca2014-03-26 18:26:36 +01001143 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1144 } else {
1145 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1146 }
1147
1148 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1149 ret = 0;
1150 }
Josef Bacik02db0842012-06-21 16:03:58 -04001151 if (!ret && fs_devices_ret)
1152 (*fs_devices_ret)->total_devices = total_devices;
David Sterba6f60cbd2013-02-15 11:31:02 -07001153
Anand Jain6cf86a002016-02-13 10:01:29 +08001154 btrfs_release_disk_super(page);
David Sterba6f60cbd2013-02-15 11:31:02 -07001155
1156error_bdev_put:
Tejun Heod4d77622010-11-13 11:55:18 +01001157 blkdev_put(bdev, flags);
Chris Mason8a4b83c2008-03-24 15:02:07 -04001158error:
Stefan Behrensbeaf8ab2012-11-12 14:03:45 +01001159 mutex_unlock(&uuid_mutex);
Chris Mason8a4b83c2008-03-24 15:02:07 -04001160 return ret;
1161}
Chris Mason0b86a832008-03-24 15:01:56 -04001162
Miao Xie6d07bce2011-01-05 10:07:31 +00001163/* helper to account the used device space in the range */
1164int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1165 u64 end, u64 *length)
Chris Mason0b86a832008-03-24 15:01:56 -04001166{
1167 struct btrfs_key key;
1168 struct btrfs_root *root = device->dev_root;
Miao Xie6d07bce2011-01-05 10:07:31 +00001169 struct btrfs_dev_extent *dev_extent;
Yan Zheng2b820322008-11-17 21:11:30 -05001170 struct btrfs_path *path;
Miao Xie6d07bce2011-01-05 10:07:31 +00001171 u64 extent_end;
Chris Mason0b86a832008-03-24 15:01:56 -04001172 int ret;
Miao Xie6d07bce2011-01-05 10:07:31 +00001173 int slot;
Chris Mason0b86a832008-03-24 15:01:56 -04001174 struct extent_buffer *l;
1175
Miao Xie6d07bce2011-01-05 10:07:31 +00001176 *length = 0;
1177
Stefan Behrens63a212a2012-11-05 18:29:28 +01001178 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
Miao Xie6d07bce2011-01-05 10:07:31 +00001179 return 0;
1180
Yan Zheng2b820322008-11-17 21:11:30 -05001181 path = btrfs_alloc_path();
1182 if (!path)
1183 return -ENOMEM;
David Sterbae4058b52015-11-27 16:31:35 +01001184 path->reada = READA_FORWARD;
Chris Mason8f18cf12008-04-25 16:53:30 -04001185
Chris Mason0b86a832008-03-24 15:01:56 -04001186 key.objectid = device->devid;
Miao Xie6d07bce2011-01-05 10:07:31 +00001187 key.offset = start;
Chris Mason0b86a832008-03-24 15:01:56 -04001188 key.type = BTRFS_DEV_EXTENT_KEY;
Miao Xie6d07bce2011-01-05 10:07:31 +00001189
1190 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001191 if (ret < 0)
Miao Xie6d07bce2011-01-05 10:07:31 +00001192 goto out;
Yan Zheng1fcbac52009-07-24 11:06:53 -04001193 if (ret > 0) {
1194 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1195 if (ret < 0)
Miao Xie6d07bce2011-01-05 10:07:31 +00001196 goto out;
Yan Zheng1fcbac52009-07-24 11:06:53 -04001197 }
Miao Xie6d07bce2011-01-05 10:07:31 +00001198
Chris Mason0b86a832008-03-24 15:01:56 -04001199 while (1) {
1200 l = path->nodes[0];
1201 slot = path->slots[0];
1202 if (slot >= btrfs_header_nritems(l)) {
1203 ret = btrfs_next_leaf(root, path);
1204 if (ret == 0)
1205 continue;
1206 if (ret < 0)
Miao Xie6d07bce2011-01-05 10:07:31 +00001207 goto out;
1208
1209 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001210 }
1211 btrfs_item_key_to_cpu(l, &key, slot);
1212
1213 if (key.objectid < device->devid)
1214 goto next;
1215
1216 if (key.objectid > device->devid)
Miao Xie6d07bce2011-01-05 10:07:31 +00001217 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001218
David Sterba962a2982014-06-04 18:41:45 +02001219 if (key.type != BTRFS_DEV_EXTENT_KEY)
Chris Mason0b86a832008-03-24 15:01:56 -04001220 goto next;
Chris Mason0b86a832008-03-24 15:01:56 -04001221
Chris Mason0b86a832008-03-24 15:01:56 -04001222 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
Miao Xie6d07bce2011-01-05 10:07:31 +00001223 extent_end = key.offset + btrfs_dev_extent_length(l,
1224 dev_extent);
1225 if (key.offset <= start && extent_end > end) {
1226 *length = end - start + 1;
1227 break;
1228 } else if (key.offset <= start && extent_end > start)
1229 *length += extent_end - start;
1230 else if (key.offset > start && extent_end <= end)
1231 *length += extent_end - key.offset;
1232 else if (key.offset > start && key.offset <= end) {
1233 *length += end - key.offset + 1;
1234 break;
1235 } else if (key.offset > end)
1236 break;
1237
1238next:
1239 path->slots[0]++;
1240 }
1241 ret = 0;
1242out:
1243 btrfs_free_path(path);
1244 return ret;
1245}
1246
Jeff Mahoney499f3772015-06-15 09:41:17 -04001247static int contains_pending_extent(struct btrfs_transaction *transaction,
Josef Bacik6df9a952013-06-27 13:22:46 -04001248 struct btrfs_device *device,
1249 u64 *start, u64 len)
1250{
Jeff Mahoney499f3772015-06-15 09:41:17 -04001251 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
Josef Bacik6df9a952013-06-27 13:22:46 -04001252 struct extent_map *em;
Jeff Mahoney499f3772015-06-15 09:41:17 -04001253 struct list_head *search_list = &fs_info->pinned_chunks;
Josef Bacik6df9a952013-06-27 13:22:46 -04001254 int ret = 0;
Forrest Liu1b984502015-02-09 17:30:47 +08001255 u64 physical_start = *start;
Josef Bacik6df9a952013-06-27 13:22:46 -04001256
Jeff Mahoney499f3772015-06-15 09:41:17 -04001257 if (transaction)
1258 search_list = &transaction->pending_chunks;
Filipe Manana04216822014-11-27 21:14:15 +00001259again:
1260 list_for_each_entry(em, search_list, list) {
Josef Bacik6df9a952013-06-27 13:22:46 -04001261 struct map_lookup *map;
1262 int i;
1263
Jeff Mahoney95617d62015-06-03 10:55:48 -04001264 map = em->map_lookup;
Josef Bacik6df9a952013-06-27 13:22:46 -04001265 for (i = 0; i < map->num_stripes; i++) {
Filipe Mananac152b632015-05-14 10:46:03 +01001266 u64 end;
1267
Josef Bacik6df9a952013-06-27 13:22:46 -04001268 if (map->stripes[i].dev != device)
1269 continue;
Forrest Liu1b984502015-02-09 17:30:47 +08001270 if (map->stripes[i].physical >= physical_start + len ||
Josef Bacik6df9a952013-06-27 13:22:46 -04001271 map->stripes[i].physical + em->orig_block_len <=
Forrest Liu1b984502015-02-09 17:30:47 +08001272 physical_start)
Josef Bacik6df9a952013-06-27 13:22:46 -04001273 continue;
Filipe Mananac152b632015-05-14 10:46:03 +01001274 /*
1275 * Make sure that while processing the pinned list we do
1276 * not override our *start with a lower value, because
1277 * we can have pinned chunks that fall within this
1278 * device hole and that have lower physical addresses
1279 * than the pending chunks we processed before. If we
1280 * do not take this special care we can end up getting
1281 * 2 pending chunks that start at the same physical
1282 * device offsets because the end offset of a pinned
1283 * chunk can be equal to the start offset of some
1284 * pending chunk.
1285 */
1286 end = map->stripes[i].physical + em->orig_block_len;
1287 if (end > *start) {
1288 *start = end;
1289 ret = 1;
1290 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001291 }
1292 }
Jeff Mahoney499f3772015-06-15 09:41:17 -04001293 if (search_list != &fs_info->pinned_chunks) {
1294 search_list = &fs_info->pinned_chunks;
Filipe Manana04216822014-11-27 21:14:15 +00001295 goto again;
1296 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001297
1298 return ret;
1299}
1300
1301
Chris Mason0b86a832008-03-24 15:01:56 -04001302/*
Jeff Mahoney499f3772015-06-15 09:41:17 -04001303 * find_free_dev_extent_start - find free space in the specified device
1304 * @device: the device which we search the free space in
1305 * @num_bytes: the size of the free space that we need
1306 * @search_start: the position from which to begin the search
1307 * @start: store the start of the free space.
1308 * @len: the size of the free space. that we find, or the size
1309 * of the max free space if we don't find suitable free space
Miao Xie7bfc8372011-01-05 10:07:26 +00001310 *
Chris Mason0b86a832008-03-24 15:01:56 -04001311 * this uses a pretty simple search, the expectation is that it is
1312 * called very infrequently and that a given device has a small number
1313 * of extents
Miao Xie7bfc8372011-01-05 10:07:26 +00001314 *
1315 * @start is used to store the start of the free space if we find. But if we
1316 * don't find suitable free space, it will be used to store the start position
1317 * of the max free space.
1318 *
1319 * @len is used to store the size of the free space that we find.
1320 * But if we don't find suitable free space, it is used to store the size of
1321 * the max free space.
Chris Mason0b86a832008-03-24 15:01:56 -04001322 */
Jeff Mahoney499f3772015-06-15 09:41:17 -04001323int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1324 struct btrfs_device *device, u64 num_bytes,
1325 u64 search_start, u64 *start, u64 *len)
Chris Mason0b86a832008-03-24 15:01:56 -04001326{
1327 struct btrfs_key key;
1328 struct btrfs_root *root = device->dev_root;
Miao Xie7bfc8372011-01-05 10:07:26 +00001329 struct btrfs_dev_extent *dev_extent;
Chris Mason0b86a832008-03-24 15:01:56 -04001330 struct btrfs_path *path;
Miao Xie7bfc8372011-01-05 10:07:26 +00001331 u64 hole_size;
1332 u64 max_hole_start;
1333 u64 max_hole_size;
1334 u64 extent_end;
Chris Mason0b86a832008-03-24 15:01:56 -04001335 u64 search_end = device->total_bytes;
1336 int ret;
Miao Xie7bfc8372011-01-05 10:07:26 +00001337 int slot;
Chris Mason0b86a832008-03-24 15:01:56 -04001338 struct extent_buffer *l;
Filipe Manana8cdc7c52016-01-06 22:42:35 +00001339 u64 min_search_start;
1340
1341 /*
1342 * We don't want to overwrite the superblock on the drive nor any area
1343 * used by the boot loader (grub for example), so we make sure to start
1344 * at an offset of at least 1MB.
1345 */
1346 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1347 search_start = max(search_start, min_search_start);
Chris Mason0b86a832008-03-24 15:01:56 -04001348
Josef Bacik6df9a952013-06-27 13:22:46 -04001349 path = btrfs_alloc_path();
1350 if (!path)
1351 return -ENOMEM;
Zhao Leif2ab7612015-02-16 18:52:17 +08001352
Miao Xie7bfc8372011-01-05 10:07:26 +00001353 max_hole_start = search_start;
1354 max_hole_size = 0;
1355
Zhao Leif2ab7612015-02-16 18:52:17 +08001356again:
Stefan Behrens63a212a2012-11-05 18:29:28 +01001357 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
Miao Xie7bfc8372011-01-05 10:07:26 +00001358 ret = -ENOSPC;
Josef Bacik6df9a952013-06-27 13:22:46 -04001359 goto out;
Miao Xie7bfc8372011-01-05 10:07:26 +00001360 }
1361
David Sterbae4058b52015-11-27 16:31:35 +01001362 path->reada = READA_FORWARD;
Josef Bacik6df9a952013-06-27 13:22:46 -04001363 path->search_commit_root = 1;
1364 path->skip_locking = 1;
Miao Xie7bfc8372011-01-05 10:07:26 +00001365
Chris Mason0b86a832008-03-24 15:01:56 -04001366 key.objectid = device->devid;
1367 key.offset = search_start;
1368 key.type = BTRFS_DEV_EXTENT_KEY;
Miao Xie7bfc8372011-01-05 10:07:26 +00001369
Li Zefan125ccb02011-12-08 15:07:24 +08001370 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001371 if (ret < 0)
Miao Xie7bfc8372011-01-05 10:07:26 +00001372 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04001373 if (ret > 0) {
1374 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1375 if (ret < 0)
Miao Xie7bfc8372011-01-05 10:07:26 +00001376 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04001377 }
Miao Xie7bfc8372011-01-05 10:07:26 +00001378
Chris Mason0b86a832008-03-24 15:01:56 -04001379 while (1) {
1380 l = path->nodes[0];
1381 slot = path->slots[0];
1382 if (slot >= btrfs_header_nritems(l)) {
1383 ret = btrfs_next_leaf(root, path);
1384 if (ret == 0)
1385 continue;
1386 if (ret < 0)
Miao Xie7bfc8372011-01-05 10:07:26 +00001387 goto out;
1388
1389 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001390 }
1391 btrfs_item_key_to_cpu(l, &key, slot);
1392
1393 if (key.objectid < device->devid)
1394 goto next;
1395
1396 if (key.objectid > device->devid)
Miao Xie7bfc8372011-01-05 10:07:26 +00001397 break;
Chris Mason0b86a832008-03-24 15:01:56 -04001398
David Sterba962a2982014-06-04 18:41:45 +02001399 if (key.type != BTRFS_DEV_EXTENT_KEY)
Chris Mason0b86a832008-03-24 15:01:56 -04001400 goto next;
1401
Miao Xie7bfc8372011-01-05 10:07:26 +00001402 if (key.offset > search_start) {
1403 hole_size = key.offset - search_start;
1404
Josef Bacik6df9a952013-06-27 13:22:46 -04001405 /*
1406 * Have to check before we set max_hole_start, otherwise
1407 * we could end up sending back this offset anyway.
1408 */
Jeff Mahoney499f3772015-06-15 09:41:17 -04001409 if (contains_pending_extent(transaction, device,
Josef Bacik6df9a952013-06-27 13:22:46 -04001410 &search_start,
Forrest Liu1b984502015-02-09 17:30:47 +08001411 hole_size)) {
1412 if (key.offset >= search_start) {
1413 hole_size = key.offset - search_start;
1414 } else {
1415 WARN_ON_ONCE(1);
1416 hole_size = 0;
1417 }
1418 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001419
Miao Xie7bfc8372011-01-05 10:07:26 +00001420 if (hole_size > max_hole_size) {
1421 max_hole_start = search_start;
1422 max_hole_size = hole_size;
1423 }
1424
1425 /*
1426 * If this free space is greater than which we need,
1427 * it must be the max free space that we have found
1428 * until now, so max_hole_start must point to the start
1429 * of this free space and the length of this free space
1430 * is stored in max_hole_size. Thus, we return
1431 * max_hole_start and max_hole_size and go back to the
1432 * caller.
1433 */
1434 if (hole_size >= num_bytes) {
1435 ret = 0;
1436 goto out;
1437 }
1438 }
1439
Chris Mason0b86a832008-03-24 15:01:56 -04001440 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
Miao Xie7bfc8372011-01-05 10:07:26 +00001441 extent_end = key.offset + btrfs_dev_extent_length(l,
1442 dev_extent);
1443 if (extent_end > search_start)
1444 search_start = extent_end;
Chris Mason0b86a832008-03-24 15:01:56 -04001445next:
1446 path->slots[0]++;
1447 cond_resched();
1448 }
Chris Mason0b86a832008-03-24 15:01:56 -04001449
liubo38c01b92011-08-02 02:39:03 +00001450 /*
1451 * At this point, search_start should be the end of
1452 * allocated dev extents, and when shrinking the device,
1453 * search_end may be smaller than search_start.
1454 */
Zhao Leif2ab7612015-02-16 18:52:17 +08001455 if (search_end > search_start) {
liubo38c01b92011-08-02 02:39:03 +00001456 hole_size = search_end - search_start;
1457
Jeff Mahoney499f3772015-06-15 09:41:17 -04001458 if (contains_pending_extent(transaction, device, &search_start,
Zhao Leif2ab7612015-02-16 18:52:17 +08001459 hole_size)) {
1460 btrfs_release_path(path);
1461 goto again;
1462 }
Chris Mason0b86a832008-03-24 15:01:56 -04001463
Zhao Leif2ab7612015-02-16 18:52:17 +08001464 if (hole_size > max_hole_size) {
1465 max_hole_start = search_start;
1466 max_hole_size = hole_size;
1467 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001468 }
1469
Miao Xie7bfc8372011-01-05 10:07:26 +00001470 /* See above. */
Zhao Leif2ab7612015-02-16 18:52:17 +08001471 if (max_hole_size < num_bytes)
Miao Xie7bfc8372011-01-05 10:07:26 +00001472 ret = -ENOSPC;
1473 else
1474 ret = 0;
1475
1476out:
Yan Zheng2b820322008-11-17 21:11:30 -05001477 btrfs_free_path(path);
Miao Xie7bfc8372011-01-05 10:07:26 +00001478 *start = max_hole_start;
Miao Xieb2117a32011-01-05 10:07:28 +00001479 if (len)
Miao Xie7bfc8372011-01-05 10:07:26 +00001480 *len = max_hole_size;
Chris Mason0b86a832008-03-24 15:01:56 -04001481 return ret;
1482}
1483
Jeff Mahoney499f3772015-06-15 09:41:17 -04001484int find_free_dev_extent(struct btrfs_trans_handle *trans,
1485 struct btrfs_device *device, u64 num_bytes,
1486 u64 *start, u64 *len)
1487{
Jeff Mahoney499f3772015-06-15 09:41:17 -04001488 /* FIXME use last free of some kind */
Jeff Mahoney499f3772015-06-15 09:41:17 -04001489 return find_free_dev_extent_start(trans->transaction, device,
Filipe Manana8cdc7c52016-01-06 22:42:35 +00001490 num_bytes, 0, start, len);
Jeff Mahoney499f3772015-06-15 09:41:17 -04001491}
1492
Christoph Hellwigb2950862008-12-02 09:54:17 -05001493static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
Chris Mason8f18cf12008-04-25 16:53:30 -04001494 struct btrfs_device *device,
Miao Xie2196d6e2014-09-03 21:35:41 +08001495 u64 start, u64 *dev_extent_len)
Chris Mason8f18cf12008-04-25 16:53:30 -04001496{
1497 int ret;
1498 struct btrfs_path *path;
1499 struct btrfs_root *root = device->dev_root;
1500 struct btrfs_key key;
Chris Masona061fc82008-05-07 11:43:44 -04001501 struct btrfs_key found_key;
1502 struct extent_buffer *leaf = NULL;
1503 struct btrfs_dev_extent *extent = NULL;
Chris Mason8f18cf12008-04-25 16:53:30 -04001504
1505 path = btrfs_alloc_path();
1506 if (!path)
1507 return -ENOMEM;
1508
1509 key.objectid = device->devid;
1510 key.offset = start;
1511 key.type = BTRFS_DEV_EXTENT_KEY;
Miao Xie924cd8f2011-11-10 20:45:04 -05001512again:
Chris Mason8f18cf12008-04-25 16:53:30 -04001513 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Chris Masona061fc82008-05-07 11:43:44 -04001514 if (ret > 0) {
1515 ret = btrfs_previous_item(root, path, key.objectid,
1516 BTRFS_DEV_EXTENT_KEY);
Tsutomu Itohb0b802d2011-05-19 07:03:42 +00001517 if (ret)
1518 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001519 leaf = path->nodes[0];
1520 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1521 extent = btrfs_item_ptr(leaf, path->slots[0],
1522 struct btrfs_dev_extent);
1523 BUG_ON(found_key.offset > start || found_key.offset +
1524 btrfs_dev_extent_length(leaf, extent) < start);
Miao Xie924cd8f2011-11-10 20:45:04 -05001525 key = found_key;
1526 btrfs_release_path(path);
1527 goto again;
Chris Masona061fc82008-05-07 11:43:44 -04001528 } else if (ret == 0) {
1529 leaf = path->nodes[0];
1530 extent = btrfs_item_ptr(leaf, path->slots[0],
1531 struct btrfs_dev_extent);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001532 } else {
Anand Jain34d97002016-03-16 16:43:06 +08001533 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001534 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001535 }
Chris Mason8f18cf12008-04-25 16:53:30 -04001536
Miao Xie2196d6e2014-09-03 21:35:41 +08001537 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1538
Chris Mason8f18cf12008-04-25 16:53:30 -04001539 ret = btrfs_del_item(trans, root, path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001540 if (ret) {
Anand Jain34d97002016-03-16 16:43:06 +08001541 btrfs_handle_fs_error(root->fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001542 "Failed to remove dev extent item");
Zhao Lei13212b52015-02-12 14:18:17 +08001543 } else {
Josef Bacik3204d332015-09-24 10:46:10 -04001544 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001545 }
Tsutomu Itohb0b802d2011-05-19 07:03:42 +00001546out:
Chris Mason8f18cf12008-04-25 16:53:30 -04001547 btrfs_free_path(path);
1548 return ret;
1549}
1550
Eric Sandeen48a3b632013-04-25 20:41:01 +00001551static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1552 struct btrfs_device *device,
1553 u64 chunk_tree, u64 chunk_objectid,
1554 u64 chunk_offset, u64 start, u64 num_bytes)
Chris Mason0b86a832008-03-24 15:01:56 -04001555{
1556 int ret;
1557 struct btrfs_path *path;
1558 struct btrfs_root *root = device->dev_root;
1559 struct btrfs_dev_extent *extent;
1560 struct extent_buffer *leaf;
1561 struct btrfs_key key;
1562
Chris Masondfe25022008-05-13 13:46:40 -04001563 WARN_ON(!device->in_fs_metadata);
Stefan Behrens63a212a2012-11-05 18:29:28 +01001564 WARN_ON(device->is_tgtdev_for_dev_replace);
Chris Mason0b86a832008-03-24 15:01:56 -04001565 path = btrfs_alloc_path();
1566 if (!path)
1567 return -ENOMEM;
1568
Chris Mason0b86a832008-03-24 15:01:56 -04001569 key.objectid = device->devid;
Yan Zheng2b820322008-11-17 21:11:30 -05001570 key.offset = start;
Chris Mason0b86a832008-03-24 15:01:56 -04001571 key.type = BTRFS_DEV_EXTENT_KEY;
1572 ret = btrfs_insert_empty_item(trans, root, path, &key,
1573 sizeof(*extent));
Mark Fasheh2cdcecb2011-09-08 17:14:32 -07001574 if (ret)
1575 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04001576
1577 leaf = path->nodes[0];
1578 extent = btrfs_item_ptr(leaf, path->slots[0],
1579 struct btrfs_dev_extent);
Chris Masone17cade2008-04-15 15:41:47 -04001580 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1581 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1582 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1583
1584 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
Geert Uytterhoeven231e88f2013-08-20 13:20:13 +02001585 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
Chris Masone17cade2008-04-15 15:41:47 -04001586
Chris Mason0b86a832008-03-24 15:01:56 -04001587 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1588 btrfs_mark_buffer_dirty(leaf);
Mark Fasheh2cdcecb2011-09-08 17:14:32 -07001589out:
Chris Mason0b86a832008-03-24 15:01:56 -04001590 btrfs_free_path(path);
1591 return ret;
1592}
1593
Josef Bacik6df9a952013-06-27 13:22:46 -04001594static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
Chris Mason0b86a832008-03-24 15:01:56 -04001595{
Josef Bacik6df9a952013-06-27 13:22:46 -04001596 struct extent_map_tree *em_tree;
1597 struct extent_map *em;
1598 struct rb_node *n;
1599 u64 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04001600
Josef Bacik6df9a952013-06-27 13:22:46 -04001601 em_tree = &fs_info->mapping_tree.map_tree;
1602 read_lock(&em_tree->lock);
1603 n = rb_last(&em_tree->map);
1604 if (n) {
1605 em = rb_entry(n, struct extent_map, rb_node);
1606 ret = em->start + em->len;
Chris Mason0b86a832008-03-24 15:01:56 -04001607 }
Josef Bacik6df9a952013-06-27 13:22:46 -04001608 read_unlock(&em_tree->lock);
1609
Chris Mason0b86a832008-03-24 15:01:56 -04001610 return ret;
1611}
1612
Ilya Dryomov53f10652013-08-12 14:33:01 +03001613static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1614 u64 *devid_ret)
Chris Mason0b86a832008-03-24 15:01:56 -04001615{
1616 int ret;
1617 struct btrfs_key key;
1618 struct btrfs_key found_key;
Yan Zheng2b820322008-11-17 21:11:30 -05001619 struct btrfs_path *path;
1620
Yan Zheng2b820322008-11-17 21:11:30 -05001621 path = btrfs_alloc_path();
1622 if (!path)
1623 return -ENOMEM;
Chris Mason0b86a832008-03-24 15:01:56 -04001624
1625 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1626 key.type = BTRFS_DEV_ITEM_KEY;
1627 key.offset = (u64)-1;
1628
Ilya Dryomov53f10652013-08-12 14:33:01 +03001629 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001630 if (ret < 0)
1631 goto error;
1632
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001633 BUG_ON(ret == 0); /* Corruption */
Chris Mason0b86a832008-03-24 15:01:56 -04001634
Ilya Dryomov53f10652013-08-12 14:33:01 +03001635 ret = btrfs_previous_item(fs_info->chunk_root, path,
1636 BTRFS_DEV_ITEMS_OBJECTID,
Chris Mason0b86a832008-03-24 15:01:56 -04001637 BTRFS_DEV_ITEM_KEY);
1638 if (ret) {
Ilya Dryomov53f10652013-08-12 14:33:01 +03001639 *devid_ret = 1;
Chris Mason0b86a832008-03-24 15:01:56 -04001640 } else {
1641 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1642 path->slots[0]);
Ilya Dryomov53f10652013-08-12 14:33:01 +03001643 *devid_ret = found_key.offset + 1;
Chris Mason0b86a832008-03-24 15:01:56 -04001644 }
1645 ret = 0;
1646error:
Yan Zheng2b820322008-11-17 21:11:30 -05001647 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04001648 return ret;
1649}
1650
1651/*
1652 * the device information is stored in the chunk root
1653 * the btrfs_device struct should be fully filled in
1654 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00001655static int btrfs_add_device(struct btrfs_trans_handle *trans,
1656 struct btrfs_root *root,
1657 struct btrfs_device *device)
Chris Mason0b86a832008-03-24 15:01:56 -04001658{
1659 int ret;
1660 struct btrfs_path *path;
1661 struct btrfs_dev_item *dev_item;
1662 struct extent_buffer *leaf;
1663 struct btrfs_key key;
1664 unsigned long ptr;
Chris Mason0b86a832008-03-24 15:01:56 -04001665
1666 root = root->fs_info->chunk_root;
1667
1668 path = btrfs_alloc_path();
1669 if (!path)
1670 return -ENOMEM;
1671
Chris Mason0b86a832008-03-24 15:01:56 -04001672 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1673 key.type = BTRFS_DEV_ITEM_KEY;
Yan Zheng2b820322008-11-17 21:11:30 -05001674 key.offset = device->devid;
Chris Mason0b86a832008-03-24 15:01:56 -04001675
1676 ret = btrfs_insert_empty_item(trans, root, path, &key,
Chris Mason0d81ba52008-03-24 15:02:07 -04001677 sizeof(*dev_item));
Chris Mason0b86a832008-03-24 15:01:56 -04001678 if (ret)
1679 goto out;
1680
1681 leaf = path->nodes[0];
1682 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1683
1684 btrfs_set_device_id(leaf, dev_item, device->devid);
Yan Zheng2b820322008-11-17 21:11:30 -05001685 btrfs_set_device_generation(leaf, dev_item, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001686 btrfs_set_device_type(leaf, dev_item, device->type);
1687 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1688 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1689 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
Miao Xie7cc8e582014-09-03 21:35:38 +08001690 btrfs_set_device_total_bytes(leaf, dev_item,
1691 btrfs_device_get_disk_total_bytes(device));
1692 btrfs_set_device_bytes_used(leaf, dev_item,
1693 btrfs_device_get_bytes_used(device));
Chris Masone17cade2008-04-15 15:41:47 -04001694 btrfs_set_device_group(leaf, dev_item, 0);
1695 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1696 btrfs_set_device_bandwidth(leaf, dev_item, 0);
Chris Masonc3027eb2008-12-08 16:40:21 -05001697 btrfs_set_device_start_offset(leaf, dev_item, 0);
Chris Mason0b86a832008-03-24 15:01:56 -04001698
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02001699 ptr = btrfs_device_uuid(dev_item);
Chris Masone17cade2008-04-15 15:41:47 -04001700 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
Geert Uytterhoeven1473b242013-08-20 13:20:12 +02001701 ptr = btrfs_device_fsid(dev_item);
Yan Zheng2b820322008-11-17 21:11:30 -05001702 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04001703 btrfs_mark_buffer_dirty(leaf);
Chris Mason0b86a832008-03-24 15:01:56 -04001704
Yan Zheng2b820322008-11-17 21:11:30 -05001705 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04001706out:
1707 btrfs_free_path(path);
1708 return ret;
1709}
Chris Mason8f18cf12008-04-25 16:53:30 -04001710
Qu Wenruo5a1972b2014-04-16 17:02:32 +08001711/*
1712 * Function to update ctime/mtime for a given device path.
1713 * Mainly used for ctime/mtime based probe like libblkid.
1714 */
1715static void update_dev_time(char *path_name)
1716{
1717 struct file *filp;
1718
1719 filp = filp_open(path_name, O_RDWR, 0);
Al Viro98af5922014-12-14 02:59:17 -05001720 if (IS_ERR(filp))
Qu Wenruo5a1972b2014-04-16 17:02:32 +08001721 return;
1722 file_update_time(filp);
1723 filp_close(filp, NULL);
Qu Wenruo5a1972b2014-04-16 17:02:32 +08001724}
1725
Chris Masona061fc82008-05-07 11:43:44 -04001726static int btrfs_rm_dev_item(struct btrfs_root *root,
1727 struct btrfs_device *device)
1728{
1729 int ret;
1730 struct btrfs_path *path;
Chris Masona061fc82008-05-07 11:43:44 -04001731 struct btrfs_key key;
Chris Masona061fc82008-05-07 11:43:44 -04001732 struct btrfs_trans_handle *trans;
1733
1734 root = root->fs_info->chunk_root;
1735
1736 path = btrfs_alloc_path();
1737 if (!path)
1738 return -ENOMEM;
1739
Yan, Zhenga22285a2010-05-16 10:48:46 -04001740 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00001741 if (IS_ERR(trans)) {
1742 btrfs_free_path(path);
1743 return PTR_ERR(trans);
1744 }
Chris Masona061fc82008-05-07 11:43:44 -04001745 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1746 key.type = BTRFS_DEV_ITEM_KEY;
1747 key.offset = device->devid;
1748
1749 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1750 if (ret < 0)
1751 goto out;
1752
1753 if (ret > 0) {
1754 ret = -ENOENT;
1755 goto out;
1756 }
1757
1758 ret = btrfs_del_item(trans, root, path);
1759 if (ret)
1760 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001761out:
1762 btrfs_free_path(path);
1763 btrfs_commit_transaction(trans, root);
1764 return ret;
1765}
1766
David Sterba3cc31a02016-02-15 16:00:26 +01001767/*
1768 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1769 * filesystem. It's up to the caller to adjust that number regarding eg. device
1770 * replace.
1771 */
1772static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1773 u64 num_devices)
Chris Masona061fc82008-05-07 11:43:44 -04001774{
Chris Masona061fc82008-05-07 11:43:44 -04001775 u64 all_avail;
Miao Xiede98ced2013-01-29 10:13:12 +00001776 unsigned seq;
David Sterba418775a2016-02-15 16:28:14 +01001777 int i;
Chris Masona061fc82008-05-07 11:43:44 -04001778
Miao Xiede98ced2013-01-29 10:13:12 +00001779 do {
Anand Jainbd45ffb2016-02-13 10:01:34 +08001780 seq = read_seqbegin(&fs_info->profiles_lock);
Miao Xiede98ced2013-01-29 10:13:12 +00001781
Anand Jainbd45ffb2016-02-13 10:01:34 +08001782 all_avail = fs_info->avail_data_alloc_bits |
1783 fs_info->avail_system_alloc_bits |
1784 fs_info->avail_metadata_alloc_bits;
1785 } while (read_seqretry(&fs_info->profiles_lock, seq));
Anand Jainf1fa7f22016-02-13 10:01:33 +08001786
David Sterba418775a2016-02-15 16:28:14 +01001787 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1788 if (!(all_avail & btrfs_raid_group[i]))
1789 continue;
Chris Masona061fc82008-05-07 11:43:44 -04001790
David Sterba418775a2016-02-15 16:28:14 +01001791 if (num_devices < btrfs_raid_array[i].devs_min) {
1792 int ret = btrfs_raid_mindev_error[i];
Chris Masona061fc82008-05-07 11:43:44 -04001793
David Sterba418775a2016-02-15 16:28:14 +01001794 if (ret)
1795 return ret;
1796 }
Anand Jainbd45ffb2016-02-13 10:01:34 +08001797 }
1798
1799 return 0;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001800}
1801
Anand Jain88acff62016-05-03 17:44:43 +08001802struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1803 struct btrfs_device *device)
1804{
1805 struct btrfs_device *next_device;
1806
1807 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1808 if (next_device != device &&
1809 !next_device->missing && next_device->bdev)
1810 return next_device;
1811 }
1812
1813 return NULL;
1814}
1815
1816/*
1817 * Helper function to check if the given device is part of s_bdev / latest_bdev
1818 * and replace it with the provided or the next active device, in the context
1819 * where this function called, there should be always be another device (or
1820 * this_dev) which is active.
1821 */
1822void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1823 struct btrfs_device *device, struct btrfs_device *this_dev)
1824{
1825 struct btrfs_device *next_device;
1826
1827 if (this_dev)
1828 next_device = this_dev;
1829 else
1830 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1831 device);
1832 ASSERT(next_device);
1833
1834 if (fs_info->sb->s_bdev &&
1835 (fs_info->sb->s_bdev == device->bdev))
1836 fs_info->sb->s_bdev = next_device->bdev;
1837
1838 if (fs_info->fs_devices->latest_bdev == device->bdev)
1839 fs_info->fs_devices->latest_bdev = next_device->bdev;
1840}
1841
Anand Jain6b526ed2016-02-13 10:01:39 +08001842int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
Anand Jainf1fa7f22016-02-13 10:01:33 +08001843{
1844 struct btrfs_device *device;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001845 struct btrfs_fs_devices *cur_devices;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001846 u64 num_devices;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001847 int ret = 0;
1848 bool clear_super = false;
Anand Jain42b67422016-02-13 10:01:38 +08001849 char *dev_name = NULL;
Anand Jainf1fa7f22016-02-13 10:01:33 +08001850
1851 mutex_lock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001852
1853 num_devices = root->fs_info->fs_devices->num_devices;
1854 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1855 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1856 WARN_ON(num_devices < 1);
1857 num_devices--;
1858 }
1859 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1860
David Sterba3cc31a02016-02-15 16:00:26 +01001861 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
Anand Jainf1fa7f22016-02-13 10:01:33 +08001862 if (ret)
Chris Masona061fc82008-05-07 11:43:44 -04001863 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001864
David Sterba5c5c0df2016-02-15 16:39:55 +01001865 ret = btrfs_find_device_by_devspec(root, devid, device_path,
Anand Jain24fc5722016-02-13 10:01:36 +08001866 &device);
1867 if (ret)
Chris Masona061fc82008-05-07 11:43:44 -04001868 goto out;
Yan Zheng2b820322008-11-17 21:11:30 -05001869
Stefan Behrens63a212a2012-11-05 18:29:28 +01001870 if (device->is_tgtdev_for_dev_replace) {
Anand Jain183860f2013-05-17 10:52:45 +00001871 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
Anand Jain24fc5722016-02-13 10:01:36 +08001872 goto out;
Stefan Behrens63a212a2012-11-05 18:29:28 +01001873 }
1874
Yan Zheng2b820322008-11-17 21:11:30 -05001875 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
Anand Jain183860f2013-05-17 10:52:45 +00001876 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
Anand Jain24fc5722016-02-13 10:01:36 +08001877 goto out;
Yan Zheng2b820322008-11-17 21:11:30 -05001878 }
1879
1880 if (device->writeable) {
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001881 lock_chunks(root);
Yan Zheng2b820322008-11-17 21:11:30 -05001882 list_del_init(&device->dev_alloc_list);
Miao Xiec3929c32014-09-03 21:35:47 +08001883 device->fs_devices->rw_devices--;
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001884 unlock_chunks(root);
Anand Jain42b67422016-02-13 10:01:38 +08001885 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1886 if (!dev_name) {
1887 ret = -ENOMEM;
1888 goto error_undo;
1889 }
Xiao Guangrong1f781602011-04-20 10:09:16 +00001890 clear_super = true;
Yan Zheng2b820322008-11-17 21:11:30 -05001891 }
Chris Masona061fc82008-05-07 11:43:44 -04001892
Carey Underwoodd7901552013-03-04 16:37:06 -06001893 mutex_unlock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001894 ret = btrfs_shrink_device(device, 0);
Carey Underwoodd7901552013-03-04 16:37:06 -06001895 mutex_lock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001896 if (ret)
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001897 goto error_undo;
Chris Masona061fc82008-05-07 11:43:44 -04001898
Stefan Behrens63a212a2012-11-05 18:29:28 +01001899 /*
1900 * TODO: the superblock still includes this device in its num_devices
1901 * counter although write_all_supers() is not locked out. This
1902 * could give a filesystem state which requires a degraded mount.
1903 */
Chris Masona061fc82008-05-07 11:43:44 -04001904 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1905 if (ret)
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001906 goto error_undo;
Chris Masona061fc82008-05-07 11:43:44 -04001907
Yan Zheng2b820322008-11-17 21:11:30 -05001908 device->in_fs_metadata = 0;
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01001909 btrfs_scrub_cancel_dev(root->fs_info, device);
Chris Masone5e9a522009-06-10 15:17:02 -04001910
1911 /*
1912 * the device list mutex makes sure that we don't change
1913 * the device list while someone else is writing out all
Filipe David Borba Mananad7306802013-08-09 15:41:36 +01001914 * the device supers. Whoever is writing all supers, should
1915 * lock the device list mutex before getting the number of
1916 * devices in the super block (super_copy). Conversely,
1917 * whoever updates the number of devices in the super block
1918 * (super_copy) should hold the device list mutex.
Chris Masone5e9a522009-06-10 15:17:02 -04001919 */
Xiao Guangrong1f781602011-04-20 10:09:16 +00001920
1921 cur_devices = device->fs_devices;
Chris Masone5e9a522009-06-10 15:17:02 -04001922 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Xiao Guangrong1f781602011-04-20 10:09:16 +00001923 list_del_rcu(&device->dev_list);
Chris Masone5e9a522009-06-10 15:17:02 -04001924
Yan Zhenge4404d62008-12-12 10:03:26 -05001925 device->fs_devices->num_devices--;
Josef Bacik02db0842012-06-21 16:03:58 -04001926 device->fs_devices->total_devices--;
Yan Zheng2b820322008-11-17 21:11:30 -05001927
Chris Masoncd02dca2010-12-13 14:56:23 -05001928 if (device->missing)
Miao Xie3a7d55c2014-07-16 18:38:01 +08001929 device->fs_devices->missing_devices--;
Chris Masoncd02dca2010-12-13 14:56:23 -05001930
Anand Jain88acff62016-05-03 17:44:43 +08001931 btrfs_assign_next_active_device(root->fs_info, device, NULL);
Yan Zheng2b820322008-11-17 21:11:30 -05001932
Eric Sandeen0bfaa9c2014-07-07 12:34:49 -05001933 if (device->bdev) {
Yan Zhenge4404d62008-12-12 10:03:26 -05001934 device->fs_devices->open_devices--;
Eric Sandeen0bfaa9c2014-07-07 12:34:49 -05001935 /* remove sysfs entry */
Anand Jain32576042015-08-14 18:32:49 +08001936 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
Eric Sandeen0bfaa9c2014-07-07 12:34:49 -05001937 }
Anand Jain99994cd2014-06-03 11:36:00 +08001938
Anand Jain14238812016-07-22 06:04:53 +08001939 btrfs_close_bdev(device);
1940
Xiao Guangrong1f781602011-04-20 10:09:16 +00001941 call_rcu(&device->rcu, free_device);
Yan Zhenge4404d62008-12-12 10:03:26 -05001942
David Sterba6c417612011-04-13 15:41:04 +02001943 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1944 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
Filipe David Borba Mananad7306802013-08-09 15:41:36 +01001945 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05001946
Xiao Guangrong1f781602011-04-20 10:09:16 +00001947 if (cur_devices->open_devices == 0) {
Yan Zhenge4404d62008-12-12 10:03:26 -05001948 struct btrfs_fs_devices *fs_devices;
1949 fs_devices = root->fs_info->fs_devices;
1950 while (fs_devices) {
Rickard Strandqvist8321cf22014-05-22 22:43:43 +02001951 if (fs_devices->seed == cur_devices) {
1952 fs_devices->seed = cur_devices->seed;
Yan Zhenge4404d62008-12-12 10:03:26 -05001953 break;
Rickard Strandqvist8321cf22014-05-22 22:43:43 +02001954 }
Yan Zhenge4404d62008-12-12 10:03:26 -05001955 fs_devices = fs_devices->seed;
Yan Zheng2b820322008-11-17 21:11:30 -05001956 }
Xiao Guangrong1f781602011-04-20 10:09:16 +00001957 cur_devices->seed = NULL;
Xiao Guangrong1f781602011-04-20 10:09:16 +00001958 __btrfs_close_devices(cur_devices);
Xiao Guangrong1f781602011-04-20 10:09:16 +00001959 free_fs_devices(cur_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05001960 }
1961
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02001962 root->fs_info->num_tolerated_disk_barrier_failures =
1963 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1964
Yan Zheng2b820322008-11-17 21:11:30 -05001965 /*
1966 * at this point, the device is zero sized. We want to
1967 * remove it from the devices list and zero out the old super
1968 */
Anand Jain24fc5722016-02-13 10:01:36 +08001969 if (clear_super) {
Anand Jain42b67422016-02-13 10:01:38 +08001970 struct block_device *bdev;
Anand Jain4d90d282014-04-06 12:59:07 +08001971
Anand Jain42b67422016-02-13 10:01:38 +08001972 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1973 root->fs_info->bdev_holder);
1974 if (!IS_ERR(bdev)) {
1975 btrfs_scratch_superblocks(bdev, dev_name);
Anand Jain24fc5722016-02-13 10:01:36 +08001976 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
Anand Jain4d90d282014-04-06 12:59:07 +08001977 }
Chris Masondfe25022008-05-13 13:46:40 -04001978 }
Chris Masona061fc82008-05-07 11:43:44 -04001979
Chris Masona061fc82008-05-07 11:43:44 -04001980out:
Anand Jain42b67422016-02-13 10:01:38 +08001981 kfree(dev_name);
1982
Chris Masona061fc82008-05-07 11:43:44 -04001983 mutex_unlock(&uuid_mutex);
Chris Masona061fc82008-05-07 11:43:44 -04001984 return ret;
Anand Jain24fc5722016-02-13 10:01:36 +08001985
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001986error_undo:
1987 if (device->writeable) {
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001988 lock_chunks(root);
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001989 list_add(&device->dev_alloc_list,
1990 &root->fs_info->fs_devices->alloc_list);
Miao Xiec3929c32014-09-03 21:35:47 +08001991 device->fs_devices->rw_devices++;
Xiao Guangrong0c1daee2011-04-20 10:08:16 +00001992 unlock_chunks(root);
Ilya Dryomov9b3517e2011-02-15 18:14:25 +00001993 }
Anand Jain24fc5722016-02-13 10:01:36 +08001994 goto out;
Chris Masona061fc82008-05-07 11:43:44 -04001995}
1996
Qu Wenruo084b6e7c2014-10-30 16:52:31 +08001997void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1998 struct btrfs_device *srcdev)
Stefan Behrense93c89c2012-11-05 17:33:06 +01001999{
Anand Jaind51908c2014-08-13 14:24:19 +08002000 struct btrfs_fs_devices *fs_devices;
2001
Stefan Behrense93c89c2012-11-05 17:33:06 +01002002 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
Ilya Dryomov13572722013-10-02 20:41:01 +03002003
Anand Jain25e8e912014-08-20 10:56:56 +08002004 /*
2005 * in case of fs with no seed, srcdev->fs_devices will point
2006 * to fs_devices of fs_info. However when the dev being replaced is
2007 * a seed dev it will point to the seed's local fs_devices. In short
2008 * srcdev will have its correct fs_devices in both the cases.
2009 */
2010 fs_devices = srcdev->fs_devices;
Anand Jaind51908c2014-08-13 14:24:19 +08002011
Stefan Behrense93c89c2012-11-05 17:33:06 +01002012 list_del_rcu(&srcdev->dev_list);
2013 list_del_rcu(&srcdev->dev_alloc_list);
Anand Jaind51908c2014-08-13 14:24:19 +08002014 fs_devices->num_devices--;
Miao Xie82372bc2014-09-03 21:35:44 +08002015 if (srcdev->missing)
Anand Jaind51908c2014-08-13 14:24:19 +08002016 fs_devices->missing_devices--;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002017
Anand Jain48b3b9d2016-04-12 21:36:16 +08002018 if (srcdev->writeable)
Miao Xie82372bc2014-09-03 21:35:44 +08002019 fs_devices->rw_devices--;
Ilya Dryomov13572722013-10-02 20:41:01 +03002020
Miao Xie82372bc2014-09-03 21:35:44 +08002021 if (srcdev->bdev)
Anand Jaind51908c2014-08-13 14:24:19 +08002022 fs_devices->open_devices--;
Qu Wenruo084b6e7c2014-10-30 16:52:31 +08002023}
2024
2025void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2026 struct btrfs_device *srcdev)
2027{
2028 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002029
Anand Jain48b3b9d2016-04-12 21:36:16 +08002030 if (srcdev->writeable) {
2031 /* zero out the old super if it is writable */
2032 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2033 }
Anand Jain14238812016-07-22 06:04:53 +08002034
2035 btrfs_close_bdev(srcdev);
2036
Stefan Behrense93c89c2012-11-05 17:33:06 +01002037 call_rcu(&srcdev->rcu, free_device);
Anand Jain94d5f0c2014-08-13 14:24:22 +08002038
2039 /*
2040 * unless fs_devices is seed fs, num_devices shouldn't go
2041 * zero
2042 */
2043 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2044
2045 /* if this is no devs we rather delete the fs_devices */
2046 if (!fs_devices->num_devices) {
2047 struct btrfs_fs_devices *tmp_fs_devices;
2048
2049 tmp_fs_devices = fs_info->fs_devices;
2050 while (tmp_fs_devices) {
2051 if (tmp_fs_devices->seed == fs_devices) {
2052 tmp_fs_devices->seed = fs_devices->seed;
2053 break;
2054 }
2055 tmp_fs_devices = tmp_fs_devices->seed;
2056 }
2057 fs_devices->seed = NULL;
Anand Jain8bef8402014-08-13 14:24:23 +08002058 __btrfs_close_devices(fs_devices);
2059 free_fs_devices(fs_devices);
Anand Jain94d5f0c2014-08-13 14:24:22 +08002060 }
Stefan Behrense93c89c2012-11-05 17:33:06 +01002061}
2062
2063void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2064 struct btrfs_device *tgtdev)
2065{
Miao Xie67a2c452014-09-03 21:35:43 +08002066 mutex_lock(&uuid_mutex);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002067 WARN_ON(!tgtdev);
2068 mutex_lock(&fs_info->fs_devices->device_list_mutex);
Anand Jaind2ff1b22015-03-10 06:38:42 +08002069
Anand Jain32576042015-08-14 18:32:49 +08002070 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
Anand Jaind2ff1b22015-03-10 06:38:42 +08002071
Anand Jain779bf3f2016-04-18 16:51:23 +08002072 if (tgtdev->bdev)
Stefan Behrense93c89c2012-11-05 17:33:06 +01002073 fs_info->fs_devices->open_devices--;
Anand Jain779bf3f2016-04-18 16:51:23 +08002074
Stefan Behrense93c89c2012-11-05 17:33:06 +01002075 fs_info->fs_devices->num_devices--;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002076
Anand Jain88acff62016-05-03 17:44:43 +08002077 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002078
Stefan Behrense93c89c2012-11-05 17:33:06 +01002079 list_del_rcu(&tgtdev->dev_list);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002080
2081 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Miao Xie67a2c452014-09-03 21:35:43 +08002082 mutex_unlock(&uuid_mutex);
Anand Jain779bf3f2016-04-18 16:51:23 +08002083
2084 /*
2085 * The update_dev_time() with in btrfs_scratch_superblocks()
2086 * may lead to a call to btrfs_show_devname() which will try
2087 * to hold device_list_mutex. And here this device
2088 * is already out of device list, so we don't have to hold
2089 * the device_list_mutex lock.
2090 */
2091 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
Anand Jain14238812016-07-22 06:04:53 +08002092
2093 btrfs_close_bdev(tgtdev);
Anand Jain779bf3f2016-04-18 16:51:23 +08002094 call_rcu(&tgtdev->rcu, free_device);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002095}
2096
Eric Sandeen48a3b632013-04-25 20:41:01 +00002097static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2098 struct btrfs_device **device)
Stefan Behrens7ba15b72012-11-05 14:42:30 +01002099{
2100 int ret = 0;
2101 struct btrfs_super_block *disk_super;
2102 u64 devid;
2103 u8 *dev_uuid;
2104 struct block_device *bdev;
2105 struct buffer_head *bh;
2106
2107 *device = NULL;
2108 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2109 root->fs_info->bdev_holder, 0, &bdev, &bh);
2110 if (ret)
2111 return ret;
2112 disk_super = (struct btrfs_super_block *)bh->b_data;
2113 devid = btrfs_stack_device_id(&disk_super->dev_item);
2114 dev_uuid = disk_super->dev_item.uuid;
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002115 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
Stefan Behrens7ba15b72012-11-05 14:42:30 +01002116 disk_super->fsid);
2117 brelse(bh);
2118 if (!*device)
2119 ret = -ENOENT;
2120 blkdev_put(bdev, FMODE_READ);
2121 return ret;
2122}
2123
2124int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2125 char *device_path,
2126 struct btrfs_device **device)
2127{
2128 *device = NULL;
2129 if (strcmp(device_path, "missing") == 0) {
2130 struct list_head *devices;
2131 struct btrfs_device *tmp;
2132
2133 devices = &root->fs_info->fs_devices->devices;
2134 /*
2135 * It is safe to read the devices since the volume_mutex
2136 * is held by the caller.
2137 */
2138 list_for_each_entry(tmp, devices, dev_list) {
2139 if (tmp->in_fs_metadata && !tmp->bdev) {
2140 *device = tmp;
2141 break;
2142 }
2143 }
2144
Anand Jaind74a6252015-08-14 18:32:56 +08002145 if (!*device)
2146 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
Stefan Behrens7ba15b72012-11-05 14:42:30 +01002147
2148 return 0;
2149 } else {
2150 return btrfs_find_device_by_path(root, device_path, device);
2151 }
2152}
2153
Yan Zheng2b820322008-11-17 21:11:30 -05002154/*
David Sterba5c5c0df2016-02-15 16:39:55 +01002155 * Lookup a device given by device id, or the path if the id is 0.
2156 */
2157int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2158 char *devpath,
Anand Jain24e04742016-02-13 10:01:35 +08002159 struct btrfs_device **device)
2160{
2161 int ret;
2162
David Sterba5c5c0df2016-02-15 16:39:55 +01002163 if (devid) {
Anand Jain24e04742016-02-13 10:01:35 +08002164 ret = 0;
David Sterba5c5c0df2016-02-15 16:39:55 +01002165 *device = btrfs_find_device(root->fs_info, devid, NULL,
Anand Jain24e04742016-02-13 10:01:35 +08002166 NULL);
2167 if (!*device)
2168 ret = -ENOENT;
2169 } else {
David Sterba5c5c0df2016-02-15 16:39:55 +01002170 if (!devpath || !devpath[0])
Anand Jainb3d1b152016-02-13 10:01:37 +08002171 return -EINVAL;
2172
David Sterba5c5c0df2016-02-15 16:39:55 +01002173 ret = btrfs_find_device_missing_or_by_path(root, devpath,
Anand Jain24e04742016-02-13 10:01:35 +08002174 device);
2175 }
2176 return ret;
2177}
2178
Yan Zheng2b820322008-11-17 21:11:30 -05002179/*
2180 * does all the dirty work required for changing file system's UUID.
2181 */
Li Zefan125ccb02011-12-08 15:07:24 +08002182static int btrfs_prepare_sprout(struct btrfs_root *root)
Yan Zheng2b820322008-11-17 21:11:30 -05002183{
2184 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2185 struct btrfs_fs_devices *old_devices;
Yan Zhenge4404d62008-12-12 10:03:26 -05002186 struct btrfs_fs_devices *seed_devices;
David Sterba6c417612011-04-13 15:41:04 +02002187 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
Yan Zheng2b820322008-11-17 21:11:30 -05002188 struct btrfs_device *device;
2189 u64 super_flags;
2190
2191 BUG_ON(!mutex_is_locked(&uuid_mutex));
Yan Zhenge4404d62008-12-12 10:03:26 -05002192 if (!fs_devices->seeding)
Yan Zheng2b820322008-11-17 21:11:30 -05002193 return -EINVAL;
2194
Ilya Dryomov2208a372013-08-12 14:33:03 +03002195 seed_devices = __alloc_fs_devices();
2196 if (IS_ERR(seed_devices))
2197 return PTR_ERR(seed_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05002198
Yan Zhenge4404d62008-12-12 10:03:26 -05002199 old_devices = clone_fs_devices(fs_devices);
2200 if (IS_ERR(old_devices)) {
2201 kfree(seed_devices);
2202 return PTR_ERR(old_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05002203 }
Yan Zhenge4404d62008-12-12 10:03:26 -05002204
Yan Zheng2b820322008-11-17 21:11:30 -05002205 list_add(&old_devices->list, &fs_uuids);
2206
Yan Zhenge4404d62008-12-12 10:03:26 -05002207 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2208 seed_devices->opened = 1;
2209 INIT_LIST_HEAD(&seed_devices->devices);
2210 INIT_LIST_HEAD(&seed_devices->alloc_list);
Chris Masone5e9a522009-06-10 15:17:02 -04002211 mutex_init(&seed_devices->device_list_mutex);
Xiao Guangrongc9513ed2011-04-20 10:07:30 +00002212
2213 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Xiao Guangrong1f781602011-04-20 10:09:16 +00002214 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2215 synchronize_rcu);
Miao Xie2196d6e2014-09-03 21:35:41 +08002216 list_for_each_entry(device, &seed_devices->devices, dev_list)
Yan Zhenge4404d62008-12-12 10:03:26 -05002217 device->fs_devices = seed_devices;
Miao Xie2196d6e2014-09-03 21:35:41 +08002218
2219 lock_chunks(root);
2220 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2221 unlock_chunks(root);
Yan Zhenge4404d62008-12-12 10:03:26 -05002222
Yan Zheng2b820322008-11-17 21:11:30 -05002223 fs_devices->seeding = 0;
2224 fs_devices->num_devices = 0;
2225 fs_devices->open_devices = 0;
Miao Xie69611ac2014-07-03 18:22:12 +08002226 fs_devices->missing_devices = 0;
Miao Xie69611ac2014-07-03 18:22:12 +08002227 fs_devices->rotating = 0;
Yan Zhenge4404d62008-12-12 10:03:26 -05002228 fs_devices->seed = seed_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05002229
2230 generate_random_uuid(fs_devices->fsid);
2231 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2232 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
Filipe David Borba Mananaf7171752013-08-12 20:56:58 +01002233 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2234
Yan Zheng2b820322008-11-17 21:11:30 -05002235 super_flags = btrfs_super_flags(disk_super) &
2236 ~BTRFS_SUPER_FLAG_SEEDING;
2237 btrfs_set_super_flags(disk_super, super_flags);
2238
2239 return 0;
2240}
2241
2242/*
Nicholas D Steeves01327612016-05-19 21:18:45 -04002243 * Store the expected generation for seed devices in device items.
Yan Zheng2b820322008-11-17 21:11:30 -05002244 */
2245static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2246 struct btrfs_root *root)
2247{
2248 struct btrfs_path *path;
2249 struct extent_buffer *leaf;
2250 struct btrfs_dev_item *dev_item;
2251 struct btrfs_device *device;
2252 struct btrfs_key key;
2253 u8 fs_uuid[BTRFS_UUID_SIZE];
2254 u8 dev_uuid[BTRFS_UUID_SIZE];
2255 u64 devid;
2256 int ret;
2257
2258 path = btrfs_alloc_path();
2259 if (!path)
2260 return -ENOMEM;
2261
2262 root = root->fs_info->chunk_root;
2263 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2264 key.offset = 0;
2265 key.type = BTRFS_DEV_ITEM_KEY;
2266
2267 while (1) {
2268 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2269 if (ret < 0)
2270 goto error;
2271
2272 leaf = path->nodes[0];
2273next_slot:
2274 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2275 ret = btrfs_next_leaf(root, path);
2276 if (ret > 0)
2277 break;
2278 if (ret < 0)
2279 goto error;
2280 leaf = path->nodes[0];
2281 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002282 btrfs_release_path(path);
Yan Zheng2b820322008-11-17 21:11:30 -05002283 continue;
2284 }
2285
2286 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2288 key.type != BTRFS_DEV_ITEM_KEY)
2289 break;
2290
2291 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2292 struct btrfs_dev_item);
2293 devid = btrfs_device_id(leaf, dev_item);
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02002294 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
Yan Zheng2b820322008-11-17 21:11:30 -05002295 BTRFS_UUID_SIZE);
Geert Uytterhoeven1473b242013-08-20 13:20:12 +02002296 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
Yan Zheng2b820322008-11-17 21:11:30 -05002297 BTRFS_UUID_SIZE);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002298 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2299 fs_uuid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002300 BUG_ON(!device); /* Logic error */
Yan Zheng2b820322008-11-17 21:11:30 -05002301
2302 if (device->fs_devices->seeding) {
2303 btrfs_set_device_generation(leaf, dev_item,
2304 device->generation);
2305 btrfs_mark_buffer_dirty(leaf);
2306 }
2307
2308 path->slots[0]++;
2309 goto next_slot;
2310 }
2311 ret = 0;
2312error:
2313 btrfs_free_path(path);
2314 return ret;
2315}
2316
Chris Mason788f20e2008-04-28 15:29:42 -04002317int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2318{
Josef Bacikd5e20032011-08-04 14:52:27 +00002319 struct request_queue *q;
Chris Mason788f20e2008-04-28 15:29:42 -04002320 struct btrfs_trans_handle *trans;
2321 struct btrfs_device *device;
2322 struct block_device *bdev;
Chris Mason788f20e2008-04-28 15:29:42 -04002323 struct list_head *devices;
Yan Zheng2b820322008-11-17 21:11:30 -05002324 struct super_block *sb = root->fs_info->sb;
Josef Bacik606686e2012-06-04 14:03:51 -04002325 struct rcu_string *name;
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002326 u64 tmp;
Yan Zheng2b820322008-11-17 21:11:30 -05002327 int seeding_dev = 0;
Chris Mason788f20e2008-04-28 15:29:42 -04002328 int ret = 0;
2329
Yan Zheng2b820322008-11-17 21:11:30 -05002330 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
Liu Bof8c5d0b2012-05-10 18:10:38 +08002331 return -EROFS;
Chris Mason788f20e2008-04-28 15:29:42 -04002332
Li Zefana5d16332011-12-07 20:08:40 -05002333 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
Tejun Heod4d77622010-11-13 11:55:18 +01002334 root->fs_info->bdev_holder);
Josef Bacik7f592032010-01-27 02:09:00 +00002335 if (IS_ERR(bdev))
2336 return PTR_ERR(bdev);
Chris Masona2135012008-06-25 16:01:30 -04002337
Yan Zheng2b820322008-11-17 21:11:30 -05002338 if (root->fs_info->fs_devices->seeding) {
2339 seeding_dev = 1;
2340 down_write(&sb->s_umount);
2341 mutex_lock(&uuid_mutex);
2342 }
2343
Chris Mason8c8bee12008-09-29 11:19:10 -04002344 filemap_write_and_wait(bdev->bd_inode->i_mapping);
Chris Masona2135012008-06-25 16:01:30 -04002345
Chris Mason788f20e2008-04-28 15:29:42 -04002346 devices = &root->fs_info->fs_devices->devices;
Liu Bod25628b2012-11-14 14:35:30 +00002347
2348 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05002349 list_for_each_entry(device, devices, dev_list) {
Chris Mason788f20e2008-04-28 15:29:42 -04002350 if (device->bdev == bdev) {
2351 ret = -EEXIST;
Liu Bod25628b2012-11-14 14:35:30 +00002352 mutex_unlock(
2353 &root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002354 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04002355 }
2356 }
Liu Bod25628b2012-11-14 14:35:30 +00002357 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Chris Mason788f20e2008-04-28 15:29:42 -04002358
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002359 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2360 if (IS_ERR(device)) {
Chris Mason788f20e2008-04-28 15:29:42 -04002361 /* we can safely leave the fs_devices entry around */
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002362 ret = PTR_ERR(device);
Yan Zheng2b820322008-11-17 21:11:30 -05002363 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04002364 }
2365
David Sterba78f2c9e2016-02-11 14:25:38 +01002366 name = rcu_string_strdup(device_path, GFP_KERNEL);
Josef Bacik606686e2012-06-04 14:03:51 -04002367 if (!name) {
Chris Mason788f20e2008-04-28 15:29:42 -04002368 kfree(device);
Yan Zheng2b820322008-11-17 21:11:30 -05002369 ret = -ENOMEM;
2370 goto error;
Chris Mason788f20e2008-04-28 15:29:42 -04002371 }
Josef Bacik606686e2012-06-04 14:03:51 -04002372 rcu_assign_pointer(device->name, name);
Yan Zheng2b820322008-11-17 21:11:30 -05002373
Yan, Zhenga22285a2010-05-16 10:48:46 -04002374 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00002375 if (IS_ERR(trans)) {
Josef Bacik606686e2012-06-04 14:03:51 -04002376 rcu_string_free(device->name);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00002377 kfree(device);
2378 ret = PTR_ERR(trans);
2379 goto error;
2380 }
2381
Josef Bacikd5e20032011-08-04 14:52:27 +00002382 q = bdev_get_queue(bdev);
2383 if (blk_queue_discard(q))
2384 device->can_discard = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05002385 device->writeable = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05002386 device->generation = trans->transid;
Chris Mason788f20e2008-04-28 15:29:42 -04002387 device->io_width = root->sectorsize;
2388 device->io_align = root->sectorsize;
2389 device->sector_size = root->sectorsize;
2390 device->total_bytes = i_size_read(bdev->bd_inode);
Yan Zheng2cc3c552009-06-04 09:23:50 -04002391 device->disk_total_bytes = device->total_bytes;
Miao Xie935e5cc2014-09-03 21:35:33 +08002392 device->commit_total_bytes = device->total_bytes;
Chris Mason788f20e2008-04-28 15:29:42 -04002393 device->dev_root = root->fs_info->dev_root;
2394 device->bdev = bdev;
Chris Masondfe25022008-05-13 13:46:40 -04002395 device->in_fs_metadata = 1;
Stefan Behrens63a212a2012-11-05 18:29:28 +01002396 device->is_tgtdev_for_dev_replace = 0;
Ilya Dryomovfb01aa82011-02-15 18:12:57 +00002397 device->mode = FMODE_EXCL;
Stefan Behrens27087f32013-10-11 15:20:42 +02002398 device->dev_stats_valid = 1;
Zheng Yan325cd4b2008-09-05 16:43:54 -04002399 set_blocksize(device->bdev, 4096);
2400
Yan Zheng2b820322008-11-17 21:11:30 -05002401 if (seeding_dev) {
2402 sb->s_flags &= ~MS_RDONLY;
Li Zefan125ccb02011-12-08 15:07:24 +08002403 ret = btrfs_prepare_sprout(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002404 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2b820322008-11-17 21:11:30 -05002405 }
2406
2407 device->fs_devices = root->fs_info->fs_devices;
Chris Masone5e9a522009-06-10 15:17:02 -04002408
Chris Masone5e9a522009-06-10 15:17:02 -04002409 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Miao Xie2196d6e2014-09-03 21:35:41 +08002410 lock_chunks(root);
Xiao Guangrong1f781602011-04-20 10:09:16 +00002411 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
Yan Zheng2b820322008-11-17 21:11:30 -05002412 list_add(&device->dev_alloc_list,
2413 &root->fs_info->fs_devices->alloc_list);
2414 root->fs_info->fs_devices->num_devices++;
2415 root->fs_info->fs_devices->open_devices++;
2416 root->fs_info->fs_devices->rw_devices++;
Josef Bacik02db0842012-06-21 16:03:58 -04002417 root->fs_info->fs_devices->total_devices++;
Yan Zheng2b820322008-11-17 21:11:30 -05002418 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2419
Josef Bacik2bf64752011-09-26 17:12:22 -04002420 spin_lock(&root->fs_info->free_chunk_lock);
2421 root->fs_info->free_chunk_space += device->total_bytes;
2422 spin_unlock(&root->fs_info->free_chunk_lock);
2423
Chris Masonc2898112009-06-10 09:51:32 -04002424 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2425 root->fs_info->fs_devices->rotating = 1;
2426
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002427 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
David Sterba6c417612011-04-13 15:41:04 +02002428 btrfs_set_super_total_bytes(root->fs_info->super_copy,
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002429 tmp + device->total_bytes);
Chris Mason788f20e2008-04-28 15:29:42 -04002430
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002431 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
David Sterba6c417612011-04-13 15:41:04 +02002432 btrfs_set_super_num_devices(root->fs_info->super_copy,
Anand Jain3c1dbdf2014-08-20 10:54:17 +08002433 tmp + 1);
Anand Jain0d393762014-06-03 11:36:01 +08002434
2435 /* add sysfs device entry */
Anand Jaine3bd6972015-08-14 18:32:48 +08002436 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
Anand Jain0d393762014-06-03 11:36:01 +08002437
Miao Xie2196d6e2014-09-03 21:35:41 +08002438 /*
2439 * we've got more storage, clear any full flags on the space
2440 * infos
2441 */
2442 btrfs_clear_space_info_full(root->fs_info);
2443
2444 unlock_chunks(root);
Chris Masone5e9a522009-06-10 15:17:02 -04002445 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
Chris Mason788f20e2008-04-28 15:29:42 -04002446
Yan Zheng2b820322008-11-17 21:11:30 -05002447 if (seeding_dev) {
Miao Xie2196d6e2014-09-03 21:35:41 +08002448 lock_chunks(root);
Yan Zheng2b820322008-11-17 21:11:30 -05002449 ret = init_first_rw_device(trans, root, device);
Miao Xie2196d6e2014-09-03 21:35:41 +08002450 unlock_chunks(root);
David Sterba005d6422012-09-18 07:52:32 -06002451 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002452 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002453 goto error_trans;
David Sterba005d6422012-09-18 07:52:32 -06002454 }
Miao Xie2196d6e2014-09-03 21:35:41 +08002455 }
2456
2457 ret = btrfs_add_device(trans, root, device);
2458 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002459 btrfs_abort_transaction(trans, ret);
Miao Xie2196d6e2014-09-03 21:35:41 +08002460 goto error_trans;
2461 }
2462
2463 if (seeding_dev) {
2464 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2465
Yan Zheng2b820322008-11-17 21:11:30 -05002466 ret = btrfs_finish_sprout(trans, root);
David Sterba005d6422012-09-18 07:52:32 -06002467 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002468 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002469 goto error_trans;
David Sterba005d6422012-09-18 07:52:32 -06002470 }
Anand Jainb2373f22014-06-03 11:36:03 +08002471
2472 /* Sprouting would change fsid of the mounted root,
2473 * so rename the fsid on the sysfs
2474 */
2475 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2476 root->fs_info->fsid);
Anand Jainc1b7e472015-08-14 18:32:50 +08002477 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
Anand Jain2e7910d2015-03-10 06:38:29 +08002478 fsid_buf))
David Sterbaf14d1042015-10-08 11:37:06 +02002479 btrfs_warn(root->fs_info,
2480 "sysfs: failed to create fsid for sprout");
Yan Zheng2b820322008-11-17 21:11:30 -05002481 }
2482
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02002483 root->fs_info->num_tolerated_disk_barrier_failures =
2484 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002485 ret = btrfs_commit_transaction(trans, root);
Yan Zheng2b820322008-11-17 21:11:30 -05002486
2487 if (seeding_dev) {
2488 mutex_unlock(&uuid_mutex);
2489 up_write(&sb->s_umount);
2490
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002491 if (ret) /* transaction commit */
2492 return ret;
2493
Yan Zheng2b820322008-11-17 21:11:30 -05002494 ret = btrfs_relocate_sys_chunks(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002495 if (ret < 0)
Anand Jain34d97002016-03-16 16:43:06 +08002496 btrfs_handle_fs_error(root->fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002497 "Failed to relocate sys chunks after "
2498 "device initialization. This can be fixed "
2499 "using the \"btrfs balance\" command.");
Miao Xie671415b2012-10-16 11:26:46 +00002500 trans = btrfs_attach_transaction(root);
2501 if (IS_ERR(trans)) {
2502 if (PTR_ERR(trans) == -ENOENT)
2503 return 0;
2504 return PTR_ERR(trans);
2505 }
2506 ret = btrfs_commit_transaction(trans, root);
Yan Zheng2b820322008-11-17 21:11:30 -05002507 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02002508
Qu Wenruo5a1972b2014-04-16 17:02:32 +08002509 /* Update ctime/mtime for libblkid */
2510 update_dev_time(device_path);
Chris Mason788f20e2008-04-28 15:29:42 -04002511 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002512
2513error_trans:
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002514 btrfs_end_transaction(trans, root);
Josef Bacik606686e2012-06-04 14:03:51 -04002515 rcu_string_free(device->name);
Anand Jain32576042015-08-14 18:32:49 +08002516 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002517 kfree(device);
Yan Zheng2b820322008-11-17 21:11:30 -05002518error:
Tejun Heoe525fd82010-11-13 11:55:17 +01002519 blkdev_put(bdev, FMODE_EXCL);
Yan Zheng2b820322008-11-17 21:11:30 -05002520 if (seeding_dev) {
2521 mutex_unlock(&uuid_mutex);
2522 up_write(&sb->s_umount);
2523 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02002524 return ret;
Chris Mason788f20e2008-04-28 15:29:42 -04002525}
2526
Stefan Behrense93c89c2012-11-05 17:33:06 +01002527int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
Miao Xie1c433662014-09-03 21:35:32 +08002528 struct btrfs_device *srcdev,
Stefan Behrense93c89c2012-11-05 17:33:06 +01002529 struct btrfs_device **device_out)
2530{
2531 struct request_queue *q;
2532 struct btrfs_device *device;
2533 struct block_device *bdev;
2534 struct btrfs_fs_info *fs_info = root->fs_info;
2535 struct list_head *devices;
2536 struct rcu_string *name;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002537 u64 devid = BTRFS_DEV_REPLACE_DEVID;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002538 int ret = 0;
2539
2540 *device_out = NULL;
Miao Xie1c433662014-09-03 21:35:32 +08002541 if (fs_info->fs_devices->seeding) {
2542 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
Stefan Behrense93c89c2012-11-05 17:33:06 +01002543 return -EINVAL;
Miao Xie1c433662014-09-03 21:35:32 +08002544 }
Stefan Behrense93c89c2012-11-05 17:33:06 +01002545
2546 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2547 fs_info->bdev_holder);
Miao Xie1c433662014-09-03 21:35:32 +08002548 if (IS_ERR(bdev)) {
2549 btrfs_err(fs_info, "target device %s is invalid!", device_path);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002550 return PTR_ERR(bdev);
Miao Xie1c433662014-09-03 21:35:32 +08002551 }
Stefan Behrense93c89c2012-11-05 17:33:06 +01002552
2553 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2554
2555 devices = &fs_info->fs_devices->devices;
2556 list_for_each_entry(device, devices, dev_list) {
2557 if (device->bdev == bdev) {
Miao Xie1c433662014-09-03 21:35:32 +08002558 btrfs_err(fs_info, "target device is in the filesystem!");
Stefan Behrense93c89c2012-11-05 17:33:06 +01002559 ret = -EEXIST;
2560 goto error;
2561 }
2562 }
2563
Miao Xie1c433662014-09-03 21:35:32 +08002564
Miao Xie7cc8e582014-09-03 21:35:38 +08002565 if (i_size_read(bdev->bd_inode) <
2566 btrfs_device_get_total_bytes(srcdev)) {
Miao Xie1c433662014-09-03 21:35:32 +08002567 btrfs_err(fs_info, "target device is smaller than source device!");
2568 ret = -EINVAL;
2569 goto error;
2570 }
2571
2572
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03002573 device = btrfs_alloc_device(NULL, &devid, NULL);
2574 if (IS_ERR(device)) {
2575 ret = PTR_ERR(device);
Stefan Behrense93c89c2012-11-05 17:33:06 +01002576 goto error;
2577 }
2578
2579 name = rcu_string_strdup(device_path, GFP_NOFS);
2580 if (!name) {
2581 kfree(device);
2582 ret = -ENOMEM;
2583 goto error;
2584 }
2585 rcu_assign_pointer(device->name, name);
2586
2587 q = bdev_get_queue(bdev);
2588 if (blk_queue_discard(q))
2589 device->can_discard = 1;
2590 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2591 device->writeable = 1;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002592 device->generation = 0;
2593 device->io_width = root->sectorsize;
2594 device->io_align = root->sectorsize;
2595 device->sector_size = root->sectorsize;
Miao Xie7cc8e582014-09-03 21:35:38 +08002596 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2597 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2598 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
Miao Xie935e5cc2014-09-03 21:35:33 +08002599 ASSERT(list_empty(&srcdev->resized_list));
2600 device->commit_total_bytes = srcdev->commit_total_bytes;
Miao Xiece7213c2014-09-03 21:35:34 +08002601 device->commit_bytes_used = device->bytes_used;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002602 device->dev_root = fs_info->dev_root;
2603 device->bdev = bdev;
2604 device->in_fs_metadata = 1;
2605 device->is_tgtdev_for_dev_replace = 1;
2606 device->mode = FMODE_EXCL;
Stefan Behrens27087f32013-10-11 15:20:42 +02002607 device->dev_stats_valid = 1;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002608 set_blocksize(device->bdev, 4096);
2609 device->fs_devices = fs_info->fs_devices;
2610 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2611 fs_info->fs_devices->num_devices++;
2612 fs_info->fs_devices->open_devices++;
Stefan Behrense93c89c2012-11-05 17:33:06 +01002613 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2614
2615 *device_out = device;
2616 return ret;
2617
2618error:
2619 blkdev_put(bdev, FMODE_EXCL);
2620 return ret;
2621}
2622
2623void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2624 struct btrfs_device *tgtdev)
2625{
2626 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2627 tgtdev->io_width = fs_info->dev_root->sectorsize;
2628 tgtdev->io_align = fs_info->dev_root->sectorsize;
2629 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2630 tgtdev->dev_root = fs_info->dev_root;
2631 tgtdev->in_fs_metadata = 1;
2632}
2633
Chris Masond3977122009-01-05 21:25:51 -05002634static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2635 struct btrfs_device *device)
Chris Mason0b86a832008-03-24 15:01:56 -04002636{
2637 int ret;
2638 struct btrfs_path *path;
2639 struct btrfs_root *root;
2640 struct btrfs_dev_item *dev_item;
2641 struct extent_buffer *leaf;
2642 struct btrfs_key key;
2643
2644 root = device->dev_root->fs_info->chunk_root;
2645
2646 path = btrfs_alloc_path();
2647 if (!path)
2648 return -ENOMEM;
2649
2650 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2651 key.type = BTRFS_DEV_ITEM_KEY;
2652 key.offset = device->devid;
2653
2654 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2655 if (ret < 0)
2656 goto out;
2657
2658 if (ret > 0) {
2659 ret = -ENOENT;
2660 goto out;
2661 }
2662
2663 leaf = path->nodes[0];
2664 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2665
2666 btrfs_set_device_id(leaf, dev_item, device->devid);
2667 btrfs_set_device_type(leaf, dev_item, device->type);
2668 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2669 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2670 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
Miao Xie7cc8e582014-09-03 21:35:38 +08002671 btrfs_set_device_total_bytes(leaf, dev_item,
2672 btrfs_device_get_disk_total_bytes(device));
2673 btrfs_set_device_bytes_used(leaf, dev_item,
2674 btrfs_device_get_bytes_used(device));
Chris Mason0b86a832008-03-24 15:01:56 -04002675 btrfs_mark_buffer_dirty(leaf);
2676
2677out:
2678 btrfs_free_path(path);
2679 return ret;
2680}
2681
Miao Xie2196d6e2014-09-03 21:35:41 +08002682int btrfs_grow_device(struct btrfs_trans_handle *trans,
Chris Mason8f18cf12008-04-25 16:53:30 -04002683 struct btrfs_device *device, u64 new_size)
2684{
2685 struct btrfs_super_block *super_copy =
David Sterba6c417612011-04-13 15:41:04 +02002686 device->dev_root->fs_info->super_copy;
Miao Xie935e5cc2014-09-03 21:35:33 +08002687 struct btrfs_fs_devices *fs_devices;
Miao Xie2196d6e2014-09-03 21:35:41 +08002688 u64 old_total;
2689 u64 diff;
Chris Mason8f18cf12008-04-25 16:53:30 -04002690
Yan Zheng2b820322008-11-17 21:11:30 -05002691 if (!device->writeable)
2692 return -EACCES;
Miao Xie2196d6e2014-09-03 21:35:41 +08002693
2694 lock_chunks(device->dev_root);
2695 old_total = btrfs_super_total_bytes(super_copy);
2696 diff = new_size - device->total_bytes;
2697
Stefan Behrens63a212a2012-11-05 18:29:28 +01002698 if (new_size <= device->total_bytes ||
Miao Xie2196d6e2014-09-03 21:35:41 +08002699 device->is_tgtdev_for_dev_replace) {
2700 unlock_chunks(device->dev_root);
Yan Zheng2b820322008-11-17 21:11:30 -05002701 return -EINVAL;
Miao Xie2196d6e2014-09-03 21:35:41 +08002702 }
Yan Zheng2b820322008-11-17 21:11:30 -05002703
Miao Xie935e5cc2014-09-03 21:35:33 +08002704 fs_devices = device->dev_root->fs_info->fs_devices;
Chris Mason8f18cf12008-04-25 16:53:30 -04002705
2706 btrfs_set_super_total_bytes(super_copy, old_total + diff);
Yan Zheng2b820322008-11-17 21:11:30 -05002707 device->fs_devices->total_rw_bytes += diff;
2708
Miao Xie7cc8e582014-09-03 21:35:38 +08002709 btrfs_device_set_total_bytes(device, new_size);
2710 btrfs_device_set_disk_total_bytes(device, new_size);
Chris Mason4184ea72009-03-10 12:39:20 -04002711 btrfs_clear_space_info_full(device->dev_root->fs_info);
Miao Xie935e5cc2014-09-03 21:35:33 +08002712 if (list_empty(&device->resized_list))
2713 list_add_tail(&device->resized_list,
2714 &fs_devices->resized_devices);
Miao Xie2196d6e2014-09-03 21:35:41 +08002715 unlock_chunks(device->dev_root);
Chris Mason4184ea72009-03-10 12:39:20 -04002716
Chris Mason8f18cf12008-04-25 16:53:30 -04002717 return btrfs_update_device(trans, device);
2718}
2719
2720static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
Zhao Leia688a042015-02-09 20:31:44 +08002721 struct btrfs_root *root, u64 chunk_objectid,
Chris Mason8f18cf12008-04-25 16:53:30 -04002722 u64 chunk_offset)
2723{
2724 int ret;
2725 struct btrfs_path *path;
2726 struct btrfs_key key;
2727
2728 root = root->fs_info->chunk_root;
2729 path = btrfs_alloc_path();
2730 if (!path)
2731 return -ENOMEM;
2732
2733 key.objectid = chunk_objectid;
2734 key.offset = chunk_offset;
2735 key.type = BTRFS_CHUNK_ITEM_KEY;
2736
2737 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002738 if (ret < 0)
2739 goto out;
2740 else if (ret > 0) { /* Logic error or corruption */
Anand Jain34d97002016-03-16 16:43:06 +08002741 btrfs_handle_fs_error(root->fs_info, -ENOENT,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002742 "Failed lookup while freeing chunk.");
2743 ret = -ENOENT;
2744 goto out;
2745 }
Chris Mason8f18cf12008-04-25 16:53:30 -04002746
2747 ret = btrfs_del_item(trans, root, path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002748 if (ret < 0)
Anand Jain34d97002016-03-16 16:43:06 +08002749 btrfs_handle_fs_error(root->fs_info, ret,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002750 "Failed to delete chunk item.");
2751out:
Chris Mason8f18cf12008-04-25 16:53:30 -04002752 btrfs_free_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00002753 return ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002754}
2755
Christoph Hellwigb2950862008-12-02 09:54:17 -05002756static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
Chris Mason8f18cf12008-04-25 16:53:30 -04002757 chunk_offset)
2758{
David Sterba6c417612011-04-13 15:41:04 +02002759 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Mason8f18cf12008-04-25 16:53:30 -04002760 struct btrfs_disk_key *disk_key;
2761 struct btrfs_chunk *chunk;
2762 u8 *ptr;
2763 int ret = 0;
2764 u32 num_stripes;
2765 u32 array_size;
2766 u32 len = 0;
2767 u32 cur;
2768 struct btrfs_key key;
2769
Miao Xie2196d6e2014-09-03 21:35:41 +08002770 lock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04002771 array_size = btrfs_super_sys_array_size(super_copy);
2772
2773 ptr = super_copy->sys_chunk_array;
2774 cur = 0;
2775
2776 while (cur < array_size) {
2777 disk_key = (struct btrfs_disk_key *)ptr;
2778 btrfs_disk_key_to_cpu(&key, disk_key);
2779
2780 len = sizeof(*disk_key);
2781
2782 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2783 chunk = (struct btrfs_chunk *)(ptr + len);
2784 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2785 len += btrfs_chunk_item_size(num_stripes);
2786 } else {
2787 ret = -EIO;
2788 break;
2789 }
2790 if (key.objectid == chunk_objectid &&
2791 key.offset == chunk_offset) {
2792 memmove(ptr, ptr + len, array_size - (cur + len));
2793 array_size -= len;
2794 btrfs_set_super_sys_array_size(super_copy, array_size);
2795 } else {
2796 ptr += len;
2797 cur += len;
2798 }
2799 }
Miao Xie2196d6e2014-09-03 21:35:41 +08002800 unlock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04002801 return ret;
2802}
2803
Josef Bacik47ab2a62014-09-18 11:20:02 -04002804int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2805 struct btrfs_root *root, u64 chunk_offset)
2806{
2807 struct extent_map_tree *em_tree;
2808 struct extent_map *em;
2809 struct btrfs_root *extent_root = root->fs_info->extent_root;
2810 struct map_lookup *map;
2811 u64 dev_extent_len = 0;
2812 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
Josef Bacik47ab2a62014-09-18 11:20:02 -04002813 int i, ret = 0;
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002814 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
Josef Bacik47ab2a62014-09-18 11:20:02 -04002815
2816 /* Just in case */
2817 root = root->fs_info->chunk_root;
2818 em_tree = &root->fs_info->mapping_tree.map_tree;
2819
2820 read_lock(&em_tree->lock);
2821 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2822 read_unlock(&em_tree->lock);
2823
2824 if (!em || em->start > chunk_offset ||
2825 em->start + em->len < chunk_offset) {
2826 /*
2827 * This is a logic error, but we don't want to just rely on the
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08002828 * user having built with ASSERT enabled, so if ASSERT doesn't
Josef Bacik47ab2a62014-09-18 11:20:02 -04002829 * do anything we still error out.
2830 */
2831 ASSERT(0);
2832 if (em)
2833 free_extent_map(em);
2834 return -EINVAL;
2835 }
Jeff Mahoney95617d62015-06-03 10:55:48 -04002836 map = em->map_lookup;
Filipe Manana39c2d7f2015-05-20 14:01:55 +01002837 lock_chunks(root->fs_info->chunk_root);
Filipe Manana4617ea32015-06-09 17:48:21 +01002838 check_system_chunk(trans, extent_root, map->type);
Filipe Manana39c2d7f2015-05-20 14:01:55 +01002839 unlock_chunks(root->fs_info->chunk_root);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002840
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002841 /*
2842 * Take the device list mutex to prevent races with the final phase of
2843 * a device replace operation that replaces the device object associated
2844 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2845 */
2846 mutex_lock(&fs_devices->device_list_mutex);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002847 for (i = 0; i < map->num_stripes; i++) {
2848 struct btrfs_device *device = map->stripes[i].dev;
2849 ret = btrfs_free_dev_extent(trans, device,
2850 map->stripes[i].physical,
2851 &dev_extent_len);
2852 if (ret) {
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002853 mutex_unlock(&fs_devices->device_list_mutex);
Jeff Mahoney66642832016-06-10 18:19:25 -04002854 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002855 goto out;
2856 }
2857
2858 if (device->bytes_used > 0) {
2859 lock_chunks(root);
2860 btrfs_device_set_bytes_used(device,
2861 device->bytes_used - dev_extent_len);
2862 spin_lock(&root->fs_info->free_chunk_lock);
2863 root->fs_info->free_chunk_space += dev_extent_len;
2864 spin_unlock(&root->fs_info->free_chunk_lock);
2865 btrfs_clear_space_info_full(root->fs_info);
2866 unlock_chunks(root);
2867 }
2868
2869 if (map->stripes[i].dev) {
2870 ret = btrfs_update_device(trans, map->stripes[i].dev);
2871 if (ret) {
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002872 mutex_unlock(&fs_devices->device_list_mutex);
Jeff Mahoney66642832016-06-10 18:19:25 -04002873 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002874 goto out;
2875 }
2876 }
2877 }
Filipe Manana57ba4cb2016-05-20 04:34:23 +01002878 mutex_unlock(&fs_devices->device_list_mutex);
2879
Zhao Leia688a042015-02-09 20:31:44 +08002880 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002881 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002882 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002883 goto out;
2884 }
2885
2886 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2887
2888 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2889 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2890 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002891 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002892 goto out;
2893 }
2894 }
2895
Filipe Manana04216822014-11-27 21:14:15 +00002896 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002897 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002898 btrfs_abort_transaction(trans, ret);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002899 goto out;
2900 }
2901
Josef Bacik47ab2a62014-09-18 11:20:02 -04002902out:
2903 /* once for us */
2904 free_extent_map(em);
Chris Mason8f18cf12008-04-25 16:53:30 -04002905 return ret;
2906}
2907
Zhaoleidc2ee4e2015-08-05 18:00:04 +08002908static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
Chris Mason8f18cf12008-04-25 16:53:30 -04002909{
Chris Mason8f18cf12008-04-25 16:53:30 -04002910 struct btrfs_root *extent_root;
2911 struct btrfs_trans_handle *trans;
Chris Mason8f18cf12008-04-25 16:53:30 -04002912 int ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002913
2914 root = root->fs_info->chunk_root;
2915 extent_root = root->fs_info->extent_root;
Chris Mason8f18cf12008-04-25 16:53:30 -04002916
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002917 /*
2918 * Prevent races with automatic removal of unused block groups.
2919 * After we relocate and before we remove the chunk with offset
2920 * chunk_offset, automatic removal of the block group can kick in,
2921 * resulting in a failure when calling btrfs_remove_chunk() below.
2922 *
2923 * Make sure to acquire this mutex before doing a tree search (dev
2924 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2925 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2926 * we release the path used to search the chunk/dev tree and before
2927 * the current task acquires this mutex and calls us.
2928 */
2929 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2930
Josef Bacikba1bf482009-09-11 16:11:19 -04002931 ret = btrfs_can_relocate(extent_root, chunk_offset);
2932 if (ret)
2933 return -ENOSPC;
2934
Chris Mason8f18cf12008-04-25 16:53:30 -04002935 /* step one, relocate all the extents inside this chunk */
Zhaolei55e3a602015-08-05 16:43:30 +08002936 btrfs_scrub_pause(root);
Zheng Yan1a40e232008-09-26 10:09:34 -04002937 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
Zhaolei55e3a602015-08-05 16:43:30 +08002938 btrfs_scrub_continue(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002939 if (ret)
2940 return ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002941
Filipe Manana7fd01182015-11-13 23:57:17 +00002942 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2943 chunk_offset);
Liu Bo0f788c52013-03-04 16:25:40 +00002944 if (IS_ERR(trans)) {
2945 ret = PTR_ERR(trans);
Anand Jain34d97002016-03-16 16:43:06 +08002946 btrfs_handle_fs_error(root->fs_info, ret, NULL);
Liu Bo0f788c52013-03-04 16:25:40 +00002947 return ret;
2948 }
Chris Mason8f18cf12008-04-25 16:53:30 -04002949
2950 /*
2951 * step two, delete the device extents and the
2952 * chunk tree entries
2953 */
Josef Bacik47ab2a62014-09-18 11:20:02 -04002954 ret = btrfs_remove_chunk(trans, root, chunk_offset);
Jeff Mahoney05f9a782016-06-24 16:37:38 -04002955 btrfs_end_transaction(trans, extent_root);
Josef Bacik47ab2a62014-09-18 11:20:02 -04002956 return ret;
Chris Mason8f18cf12008-04-25 16:53:30 -04002957}
2958
Yan Zheng2b820322008-11-17 21:11:30 -05002959static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2960{
2961 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2962 struct btrfs_path *path;
2963 struct extent_buffer *leaf;
2964 struct btrfs_chunk *chunk;
2965 struct btrfs_key key;
2966 struct btrfs_key found_key;
Yan Zheng2b820322008-11-17 21:11:30 -05002967 u64 chunk_type;
Josef Bacikba1bf482009-09-11 16:11:19 -04002968 bool retried = false;
2969 int failed = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05002970 int ret;
2971
2972 path = btrfs_alloc_path();
2973 if (!path)
2974 return -ENOMEM;
2975
Josef Bacikba1bf482009-09-11 16:11:19 -04002976again:
Yan Zheng2b820322008-11-17 21:11:30 -05002977 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2978 key.offset = (u64)-1;
2979 key.type = BTRFS_CHUNK_ITEM_KEY;
2980
2981 while (1) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002982 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002983 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002984 if (ret < 0) {
2985 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002986 goto error;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002987 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002988 BUG_ON(ret == 0); /* Corruption */
Yan Zheng2b820322008-11-17 21:11:30 -05002989
2990 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2991 key.type);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01002992 if (ret)
2993 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05002994 if (ret < 0)
2995 goto error;
2996 if (ret > 0)
2997 break;
2998
2999 leaf = path->nodes[0];
3000 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3001
3002 chunk = btrfs_item_ptr(leaf, path->slots[0],
3003 struct btrfs_chunk);
3004 chunk_type = btrfs_chunk_type(leaf, chunk);
David Sterbab3b4aa72011-04-21 01:20:15 +02003005 btrfs_release_path(path);
Yan Zheng2b820322008-11-17 21:11:30 -05003006
3007 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
Zhao Leia688a042015-02-09 20:31:44 +08003008 ret = btrfs_relocate_chunk(chunk_root,
Yan Zheng2b820322008-11-17 21:11:30 -05003009 found_key.offset);
Josef Bacikba1bf482009-09-11 16:11:19 -04003010 if (ret == -ENOSPC)
3011 failed++;
HIMANGI SARAOGI14586652014-07-09 03:51:41 +05303012 else
3013 BUG_ON(ret);
Yan Zheng2b820322008-11-17 21:11:30 -05003014 }
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003015 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05003016
3017 if (found_key.offset == 0)
3018 break;
3019 key.offset = found_key.offset - 1;
3020 }
3021 ret = 0;
Josef Bacikba1bf482009-09-11 16:11:19 -04003022 if (failed && !retried) {
3023 failed = 0;
3024 retried = true;
3025 goto again;
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303026 } else if (WARN_ON(failed && retried)) {
Josef Bacikba1bf482009-09-11 16:11:19 -04003027 ret = -ENOSPC;
3028 }
Yan Zheng2b820322008-11-17 21:11:30 -05003029error:
3030 btrfs_free_path(path);
3031 return ret;
3032}
3033
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003034static int insert_balance_item(struct btrfs_root *root,
3035 struct btrfs_balance_control *bctl)
3036{
3037 struct btrfs_trans_handle *trans;
3038 struct btrfs_balance_item *item;
3039 struct btrfs_disk_balance_args disk_bargs;
3040 struct btrfs_path *path;
3041 struct extent_buffer *leaf;
3042 struct btrfs_key key;
3043 int ret, err;
3044
3045 path = btrfs_alloc_path();
3046 if (!path)
3047 return -ENOMEM;
3048
3049 trans = btrfs_start_transaction(root, 0);
3050 if (IS_ERR(trans)) {
3051 btrfs_free_path(path);
3052 return PTR_ERR(trans);
3053 }
3054
3055 key.objectid = BTRFS_BALANCE_OBJECTID;
David Sterbac479cb42016-01-25 17:51:31 +01003056 key.type = BTRFS_TEMPORARY_ITEM_KEY;
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003057 key.offset = 0;
3058
3059 ret = btrfs_insert_empty_item(trans, root, path, &key,
3060 sizeof(*item));
3061 if (ret)
3062 goto out;
3063
3064 leaf = path->nodes[0];
3065 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3066
3067 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3068
3069 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3070 btrfs_set_balance_data(leaf, item, &disk_bargs);
3071 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3072 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3073 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3074 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3075
3076 btrfs_set_balance_flags(leaf, item, bctl->flags);
3077
3078 btrfs_mark_buffer_dirty(leaf);
3079out:
3080 btrfs_free_path(path);
3081 err = btrfs_commit_transaction(trans, root);
3082 if (err && !ret)
3083 ret = err;
3084 return ret;
3085}
3086
3087static int del_balance_item(struct btrfs_root *root)
3088{
3089 struct btrfs_trans_handle *trans;
3090 struct btrfs_path *path;
3091 struct btrfs_key key;
3092 int ret, err;
3093
3094 path = btrfs_alloc_path();
3095 if (!path)
3096 return -ENOMEM;
3097
3098 trans = btrfs_start_transaction(root, 0);
3099 if (IS_ERR(trans)) {
3100 btrfs_free_path(path);
3101 return PTR_ERR(trans);
3102 }
3103
3104 key.objectid = BTRFS_BALANCE_OBJECTID;
David Sterbac479cb42016-01-25 17:51:31 +01003105 key.type = BTRFS_TEMPORARY_ITEM_KEY;
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003106 key.offset = 0;
3107
3108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3109 if (ret < 0)
3110 goto out;
3111 if (ret > 0) {
3112 ret = -ENOENT;
3113 goto out;
3114 }
3115
3116 ret = btrfs_del_item(trans, root, path);
3117out:
3118 btrfs_free_path(path);
3119 err = btrfs_commit_transaction(trans, root);
3120 if (err && !ret)
3121 ret = err;
3122 return ret;
3123}
3124
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003125/*
Ilya Dryomov59641012012-01-16 22:04:48 +02003126 * This is a heuristic used to reduce the number of chunks balanced on
3127 * resume after balance was interrupted.
3128 */
3129static void update_balance_args(struct btrfs_balance_control *bctl)
3130{
3131 /*
3132 * Turn on soft mode for chunk types that were being converted.
3133 */
3134 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3139 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3140
3141 /*
3142 * Turn on usage filter if is not already used. The idea is
3143 * that chunks that we have already balanced should be
3144 * reasonably full. Don't do it for chunks that are being
3145 * converted - that will keep us from relocating unconverted
3146 * (albeit full) chunks.
3147 */
3148 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
David Sterbabc309462015-10-20 18:22:13 +02003149 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
Ilya Dryomov59641012012-01-16 22:04:48 +02003150 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3151 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3152 bctl->data.usage = 90;
3153 }
3154 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
David Sterbabc309462015-10-20 18:22:13 +02003155 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
Ilya Dryomov59641012012-01-16 22:04:48 +02003156 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3157 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3158 bctl->sys.usage = 90;
3159 }
3160 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
David Sterbabc309462015-10-20 18:22:13 +02003161 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
Ilya Dryomov59641012012-01-16 22:04:48 +02003162 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3163 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3164 bctl->meta.usage = 90;
3165 }
3166}
3167
3168/*
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003169 * Should be called with both balance and volume mutexes held to
3170 * serialize other volume operations (add_dev/rm_dev/resize) with
3171 * restriper. Same goes for unset_balance_control.
3172 */
3173static void set_balance_control(struct btrfs_balance_control *bctl)
3174{
3175 struct btrfs_fs_info *fs_info = bctl->fs_info;
3176
3177 BUG_ON(fs_info->balance_ctl);
3178
3179 spin_lock(&fs_info->balance_lock);
3180 fs_info->balance_ctl = bctl;
3181 spin_unlock(&fs_info->balance_lock);
3182}
3183
3184static void unset_balance_control(struct btrfs_fs_info *fs_info)
3185{
3186 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3187
3188 BUG_ON(!fs_info->balance_ctl);
3189
3190 spin_lock(&fs_info->balance_lock);
3191 fs_info->balance_ctl = NULL;
3192 spin_unlock(&fs_info->balance_lock);
3193
3194 kfree(bctl);
3195}
3196
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003197/*
3198 * Balance filters. Return 1 if chunk should be filtered out
3199 * (should not be balanced).
3200 */
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003201static int chunk_profiles_filter(u64 chunk_type,
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003202 struct btrfs_balance_args *bargs)
3203{
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003204 chunk_type = chunk_to_extended(chunk_type) &
3205 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003206
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003207 if (bargs->profiles & chunk_type)
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003208 return 0;
3209
3210 return 1;
3211}
3212
Holger Hoffstättedba72cb2015-11-17 12:29:32 +01003213static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003214 struct btrfs_balance_args *bargs)
3215{
3216 struct btrfs_block_group_cache *cache;
David Sterbabc309462015-10-20 18:22:13 +02003217 u64 chunk_used;
3218 u64 user_thresh_min;
3219 u64 user_thresh_max;
3220 int ret = 1;
3221
3222 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3223 chunk_used = btrfs_block_group_used(&cache->item);
3224
3225 if (bargs->usage_min == 0)
3226 user_thresh_min = 0;
3227 else
3228 user_thresh_min = div_factor_fine(cache->key.offset,
3229 bargs->usage_min);
3230
3231 if (bargs->usage_max == 0)
3232 user_thresh_max = 1;
3233 else if (bargs->usage_max > 100)
3234 user_thresh_max = cache->key.offset;
3235 else
3236 user_thresh_max = div_factor_fine(cache->key.offset,
3237 bargs->usage_max);
3238
3239 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3240 ret = 0;
3241
3242 btrfs_put_block_group(cache);
3243 return ret;
3244}
3245
Holger Hoffstättedba72cb2015-11-17 12:29:32 +01003246static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
David Sterbabc309462015-10-20 18:22:13 +02003247 u64 chunk_offset, struct btrfs_balance_args *bargs)
3248{
3249 struct btrfs_block_group_cache *cache;
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003250 u64 chunk_used, user_thresh;
3251 int ret = 1;
3252
3253 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3254 chunk_used = btrfs_block_group_used(&cache->item);
3255
David Sterbabc309462015-10-20 18:22:13 +02003256 if (bargs->usage_min == 0)
Ilya Dryomov3e39cea2013-02-12 16:28:59 +00003257 user_thresh = 1;
Ilya Dryomova105bb82013-01-21 15:15:56 +02003258 else if (bargs->usage > 100)
3259 user_thresh = cache->key.offset;
3260 else
3261 user_thresh = div_factor_fine(cache->key.offset,
3262 bargs->usage);
3263
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003264 if (chunk_used < user_thresh)
3265 ret = 0;
3266
3267 btrfs_put_block_group(cache);
3268 return ret;
3269}
3270
Ilya Dryomov409d4042012-01-16 22:04:47 +02003271static int chunk_devid_filter(struct extent_buffer *leaf,
3272 struct btrfs_chunk *chunk,
3273 struct btrfs_balance_args *bargs)
3274{
3275 struct btrfs_stripe *stripe;
3276 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3277 int i;
3278
3279 for (i = 0; i < num_stripes; i++) {
3280 stripe = btrfs_stripe_nr(chunk, i);
3281 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3282 return 0;
3283 }
3284
3285 return 1;
3286}
3287
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003288/* [pstart, pend) */
3289static int chunk_drange_filter(struct extent_buffer *leaf,
3290 struct btrfs_chunk *chunk,
3291 u64 chunk_offset,
3292 struct btrfs_balance_args *bargs)
3293{
3294 struct btrfs_stripe *stripe;
3295 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3296 u64 stripe_offset;
3297 u64 stripe_length;
3298 int factor;
3299 int i;
3300
3301 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3302 return 0;
3303
3304 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
David Woodhouse53b381b2013-01-29 18:40:14 -05003305 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3306 factor = num_stripes / 2;
3307 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3308 factor = num_stripes - 1;
3309 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3310 factor = num_stripes - 2;
3311 } else {
3312 factor = num_stripes;
3313 }
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003314
3315 for (i = 0; i < num_stripes; i++) {
3316 stripe = btrfs_stripe_nr(chunk, i);
3317 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3318 continue;
3319
3320 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3321 stripe_length = btrfs_chunk_length(leaf, chunk);
David Sterbab8b93ad2015-01-16 17:26:13 +01003322 stripe_length = div_u64(stripe_length, factor);
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003323
3324 if (stripe_offset < bargs->pend &&
3325 stripe_offset + stripe_length > bargs->pstart)
3326 return 0;
3327 }
3328
3329 return 1;
3330}
3331
Ilya Dryomovea671762012-01-16 22:04:48 +02003332/* [vstart, vend) */
3333static int chunk_vrange_filter(struct extent_buffer *leaf,
3334 struct btrfs_chunk *chunk,
3335 u64 chunk_offset,
3336 struct btrfs_balance_args *bargs)
3337{
3338 if (chunk_offset < bargs->vend &&
3339 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3340 /* at least part of the chunk is inside this vrange */
3341 return 0;
3342
3343 return 1;
3344}
3345
Gabríel Arthúr Péturssondee32d02015-09-28 22:32:41 +00003346static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3347 struct btrfs_chunk *chunk,
3348 struct btrfs_balance_args *bargs)
3349{
3350 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3351
3352 if (bargs->stripes_min <= num_stripes
3353 && num_stripes <= bargs->stripes_max)
3354 return 0;
3355
3356 return 1;
3357}
3358
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003359static int chunk_soft_convert_filter(u64 chunk_type,
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003360 struct btrfs_balance_args *bargs)
3361{
3362 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3363 return 0;
3364
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003365 chunk_type = chunk_to_extended(chunk_type) &
3366 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003367
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003368 if (bargs->target == chunk_type)
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003369 return 1;
3370
3371 return 0;
3372}
3373
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003374static int should_balance_chunk(struct btrfs_root *root,
3375 struct extent_buffer *leaf,
3376 struct btrfs_chunk *chunk, u64 chunk_offset)
3377{
3378 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3379 struct btrfs_balance_args *bargs = NULL;
3380 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3381
3382 /* type filter */
3383 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3384 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3385 return 0;
3386 }
3387
3388 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3389 bargs = &bctl->data;
3390 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3391 bargs = &bctl->sys;
3392 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3393 bargs = &bctl->meta;
3394
Ilya Dryomoved25e9b2012-01-16 22:04:47 +02003395 /* profiles filter */
3396 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3397 chunk_profiles_filter(chunk_type, bargs)) {
3398 return 0;
3399 }
3400
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003401 /* usage filter */
3402 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3403 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3404 return 0;
David Sterbabc309462015-10-20 18:22:13 +02003405 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3406 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3407 return 0;
Ilya Dryomov5ce5b3c2012-01-16 22:04:47 +02003408 }
3409
Ilya Dryomov409d4042012-01-16 22:04:47 +02003410 /* devid filter */
3411 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3412 chunk_devid_filter(leaf, chunk, bargs)) {
3413 return 0;
3414 }
3415
Ilya Dryomov94e60d52012-01-16 22:04:48 +02003416 /* drange filter, makes sense only with devid filter */
3417 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3418 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3419 return 0;
3420 }
3421
Ilya Dryomovea671762012-01-16 22:04:48 +02003422 /* vrange filter */
3423 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3424 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3425 return 0;
3426 }
3427
Gabríel Arthúr Péturssondee32d02015-09-28 22:32:41 +00003428 /* stripes filter */
3429 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3430 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3431 return 0;
3432 }
3433
Ilya Dryomovcfa4c962012-01-16 22:04:48 +02003434 /* soft profile changing mode */
3435 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3436 chunk_soft_convert_filter(chunk_type, bargs)) {
3437 return 0;
3438 }
3439
David Sterba7d824b62014-05-07 17:37:51 +02003440 /*
3441 * limited by count, must be the last filter
3442 */
3443 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3444 if (bargs->limit == 0)
3445 return 0;
3446 else
3447 bargs->limit--;
David Sterba12907fc2015-10-10 17:16:50 +02003448 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3449 /*
3450 * Same logic as the 'limit' filter; the minimum cannot be
Nicholas D Steeves01327612016-05-19 21:18:45 -04003451 * determined here because we do not have the global information
David Sterba12907fc2015-10-10 17:16:50 +02003452 * about the count of all chunks that satisfy the filters.
3453 */
3454 if (bargs->limit_max == 0)
3455 return 0;
3456 else
3457 bargs->limit_max--;
David Sterba7d824b62014-05-07 17:37:51 +02003458 }
3459
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003460 return 1;
3461}
3462
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003463static int __btrfs_balance(struct btrfs_fs_info *fs_info)
Chris Masonec44a352008-04-28 15:29:52 -04003464{
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003465 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003466 struct btrfs_root *chunk_root = fs_info->chunk_root;
3467 struct btrfs_root *dev_root = fs_info->dev_root;
3468 struct list_head *devices;
Chris Masonec44a352008-04-28 15:29:52 -04003469 struct btrfs_device *device;
3470 u64 old_size;
3471 u64 size_to_free;
David Sterba12907fc2015-10-10 17:16:50 +02003472 u64 chunk_type;
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003473 struct btrfs_chunk *chunk;
Liu Bo5a488b92016-07-12 11:24:21 -07003474 struct btrfs_path *path = NULL;
Chris Masonec44a352008-04-28 15:29:52 -04003475 struct btrfs_key key;
Chris Masonec44a352008-04-28 15:29:52 -04003476 struct btrfs_key found_key;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003477 struct btrfs_trans_handle *trans;
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003478 struct extent_buffer *leaf;
3479 int slot;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003480 int ret;
3481 int enospc_errors = 0;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003482 bool counting = true;
David Sterba12907fc2015-10-10 17:16:50 +02003483 /* The single value limit and min/max limits use the same bytes in the */
David Sterba7d824b62014-05-07 17:37:51 +02003484 u64 limit_data = bctl->data.limit;
3485 u64 limit_meta = bctl->meta.limit;
3486 u64 limit_sys = bctl->sys.limit;
David Sterba12907fc2015-10-10 17:16:50 +02003487 u32 count_data = 0;
3488 u32 count_meta = 0;
3489 u32 count_sys = 0;
Zhao Lei2c9fe832015-11-09 11:51:32 +08003490 int chunk_reserved = 0;
Liu Bocf25ce52015-12-14 18:29:32 -08003491 u64 bytes_used = 0;
Chris Masonec44a352008-04-28 15:29:52 -04003492
Chris Masonec44a352008-04-28 15:29:52 -04003493 /* step one make some room on all the devices */
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003494 devices = &fs_info->fs_devices->devices;
Qinghuang Fengc6e30872009-01-21 10:59:08 -05003495 list_for_each_entry(device, devices, dev_list) {
Miao Xie7cc8e582014-09-03 21:35:38 +08003496 old_size = btrfs_device_get_total_bytes(device);
Chris Masonec44a352008-04-28 15:29:52 -04003497 size_to_free = div_factor(old_size, 1);
Byongho Leeee221842015-12-15 01:42:10 +09003498 size_to_free = min_t(u64, size_to_free, SZ_1M);
Yan Zheng2b820322008-11-17 21:11:30 -05003499 if (!device->writeable ||
Miao Xie7cc8e582014-09-03 21:35:38 +08003500 btrfs_device_get_total_bytes(device) -
3501 btrfs_device_get_bytes_used(device) > size_to_free ||
Stefan Behrens63a212a2012-11-05 18:29:28 +01003502 device->is_tgtdev_for_dev_replace)
Chris Masonec44a352008-04-28 15:29:52 -04003503 continue;
3504
3505 ret = btrfs_shrink_device(device, old_size - size_to_free);
Josef Bacikba1bf482009-09-11 16:11:19 -04003506 if (ret == -ENOSPC)
3507 break;
Liu Bo5a488b92016-07-12 11:24:21 -07003508 if (ret) {
3509 /* btrfs_shrink_device never returns ret > 0 */
3510 WARN_ON(ret > 0);
3511 goto error;
3512 }
Chris Masonec44a352008-04-28 15:29:52 -04003513
Yan, Zhenga22285a2010-05-16 10:48:46 -04003514 trans = btrfs_start_transaction(dev_root, 0);
Liu Bo5a488b92016-07-12 11:24:21 -07003515 if (IS_ERR(trans)) {
3516 ret = PTR_ERR(trans);
3517 btrfs_info_in_rcu(fs_info,
3518 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3519 rcu_str_deref(device->name), ret,
3520 old_size, old_size - size_to_free);
3521 goto error;
3522 }
Chris Masonec44a352008-04-28 15:29:52 -04003523
3524 ret = btrfs_grow_device(trans, device, old_size);
Liu Bo5a488b92016-07-12 11:24:21 -07003525 if (ret) {
3526 btrfs_end_transaction(trans, dev_root);
3527 /* btrfs_grow_device never returns ret > 0 */
3528 WARN_ON(ret > 0);
3529 btrfs_info_in_rcu(fs_info,
3530 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3531 rcu_str_deref(device->name), ret,
3532 old_size, old_size - size_to_free);
3533 goto error;
3534 }
Chris Masonec44a352008-04-28 15:29:52 -04003535
3536 btrfs_end_transaction(trans, dev_root);
3537 }
3538
3539 /* step two, relocate all the chunks */
3540 path = btrfs_alloc_path();
Mark Fasheh17e9f792011-07-12 11:10:23 -07003541 if (!path) {
3542 ret = -ENOMEM;
3543 goto error;
3544 }
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003545
3546 /* zero out stat counters */
3547 spin_lock(&fs_info->balance_lock);
3548 memset(&bctl->stat, 0, sizeof(bctl->stat));
3549 spin_unlock(&fs_info->balance_lock);
3550again:
David Sterba7d824b62014-05-07 17:37:51 +02003551 if (!counting) {
David Sterba12907fc2015-10-10 17:16:50 +02003552 /*
3553 * The single value limit and min/max limits use the same bytes
3554 * in the
3555 */
David Sterba7d824b62014-05-07 17:37:51 +02003556 bctl->data.limit = limit_data;
3557 bctl->meta.limit = limit_meta;
3558 bctl->sys.limit = limit_sys;
3559 }
Chris Masonec44a352008-04-28 15:29:52 -04003560 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3561 key.offset = (u64)-1;
3562 key.type = BTRFS_CHUNK_ITEM_KEY;
3563
Chris Masond3977122009-01-05 21:25:51 -05003564 while (1) {
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003565 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
Ilya Dryomova7e99c62012-01-16 22:04:49 +02003566 atomic_read(&fs_info->balance_cancel_req)) {
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003567 ret = -ECANCELED;
3568 goto error;
3569 }
3570
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003571 mutex_lock(&fs_info->delete_unused_bgs_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04003572 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003573 if (ret < 0) {
3574 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04003575 goto error;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003576 }
Chris Masonec44a352008-04-28 15:29:52 -04003577
3578 /*
3579 * this shouldn't happen, it means the last relocate
3580 * failed
3581 */
3582 if (ret == 0)
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003583 BUG(); /* FIXME break ? */
Chris Masonec44a352008-04-28 15:29:52 -04003584
3585 ret = btrfs_previous_item(chunk_root, path, 0,
3586 BTRFS_CHUNK_ITEM_KEY);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003587 if (ret) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003588 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003589 ret = 0;
Chris Masonec44a352008-04-28 15:29:52 -04003590 break;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003591 }
Chris Mason7d9eb122008-07-08 14:19:17 -04003592
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003593 leaf = path->nodes[0];
3594 slot = path->slots[0];
3595 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3596
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003597 if (found_key.objectid != key.objectid) {
3598 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Chris Masonec44a352008-04-28 15:29:52 -04003599 break;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003600 }
Chris Mason7d9eb122008-07-08 14:19:17 -04003601
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003602 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
David Sterba12907fc2015-10-10 17:16:50 +02003603 chunk_type = btrfs_chunk_type(leaf, chunk);
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003604
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003605 if (!counting) {
3606 spin_lock(&fs_info->balance_lock);
3607 bctl->stat.considered++;
3608 spin_unlock(&fs_info->balance_lock);
3609 }
3610
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003611 ret = should_balance_chunk(chunk_root, leaf, chunk,
3612 found_key.offset);
Zhao Lei2c9fe832015-11-09 11:51:32 +08003613
David Sterbab3b4aa72011-04-21 01:20:15 +02003614 btrfs_release_path(path);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003615 if (!ret) {
3616 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003617 goto loop;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003618 }
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003619
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003620 if (counting) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003621 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003622 spin_lock(&fs_info->balance_lock);
3623 bctl->stat.expected++;
3624 spin_unlock(&fs_info->balance_lock);
David Sterba12907fc2015-10-10 17:16:50 +02003625
3626 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3627 count_data++;
3628 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3629 count_sys++;
3630 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3631 count_meta++;
3632
3633 goto loop;
3634 }
3635
3636 /*
3637 * Apply limit_min filter, no need to check if the LIMITS
3638 * filter is used, limit_min is 0 by default
3639 */
3640 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3641 count_data < bctl->data.limit_min)
3642 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3643 count_meta < bctl->meta.limit_min)
3644 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3645 count_sys < bctl->sys.limit_min)) {
3646 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003647 goto loop;
3648 }
3649
Liu Bocf25ce52015-12-14 18:29:32 -08003650 ASSERT(fs_info->data_sinfo);
3651 spin_lock(&fs_info->data_sinfo->lock);
3652 bytes_used = fs_info->data_sinfo->bytes_used;
3653 spin_unlock(&fs_info->data_sinfo->lock);
3654
3655 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3656 !chunk_reserved && !bytes_used) {
Zhao Lei2c9fe832015-11-09 11:51:32 +08003657 trans = btrfs_start_transaction(chunk_root, 0);
3658 if (IS_ERR(trans)) {
3659 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660 ret = PTR_ERR(trans);
3661 goto error;
3662 }
3663
3664 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3665 BTRFS_BLOCK_GROUP_DATA);
Filipe Manana8a7d6562015-12-10 11:08:59 +00003666 btrfs_end_transaction(trans, chunk_root);
Zhao Lei2c9fe832015-11-09 11:51:32 +08003667 if (ret < 0) {
3668 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3669 goto error;
3670 }
Zhao Lei2c9fe832015-11-09 11:51:32 +08003671 chunk_reserved = 1;
3672 }
3673
Chris Masonec44a352008-04-28 15:29:52 -04003674 ret = btrfs_relocate_chunk(chunk_root,
Chris Masonec44a352008-04-28 15:29:52 -04003675 found_key.offset);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01003676 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
Josef Bacik508794e2011-07-02 21:24:41 +00003677 if (ret && ret != -ENOSPC)
3678 goto error;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003679 if (ret == -ENOSPC) {
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003680 enospc_errors++;
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003681 } else {
3682 spin_lock(&fs_info->balance_lock);
3683 bctl->stat.completed++;
3684 spin_unlock(&fs_info->balance_lock);
3685 }
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003686loop:
Ilya Dryomov795a3322013-08-27 13:50:44 +03003687 if (found_key.offset == 0)
3688 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04003689 key.offset = found_key.offset - 1;
Chris Masonec44a352008-04-28 15:29:52 -04003690 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003691
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003692 if (counting) {
3693 btrfs_release_path(path);
3694 counting = false;
3695 goto again;
3696 }
Chris Masonec44a352008-04-28 15:29:52 -04003697error:
3698 btrfs_free_path(path);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003699 if (enospc_errors) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003700 btrfs_info(fs_info, "%d enospc errors during balance",
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003701 enospc_errors);
3702 if (!ret)
3703 ret = -ENOSPC;
3704 }
3705
Chris Masonec44a352008-04-28 15:29:52 -04003706 return ret;
3707}
3708
Ilya Dryomov0c460c02012-03-27 17:09:17 +03003709/**
3710 * alloc_profile_is_valid - see if a given profile is valid and reduced
3711 * @flags: profile to validate
3712 * @extended: if true @flags is treated as an extended profile
3713 */
3714static int alloc_profile_is_valid(u64 flags, int extended)
3715{
3716 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3717 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3718
3719 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3720
3721 /* 1) check that all other bits are zeroed */
3722 if (flags & ~mask)
3723 return 0;
3724
3725 /* 2) see if profile is reduced */
3726 if (flags == 0)
3727 return !extended; /* "0" is valid for usual profiles */
3728
3729 /* true if exactly one bit set */
3730 return (flags & (flags - 1)) == 0;
3731}
3732
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003733static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3734{
Ilya Dryomova7e99c62012-01-16 22:04:49 +02003735 /* cancel requested || normal exit path */
3736 return atomic_read(&fs_info->balance_cancel_req) ||
3737 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3738 atomic_read(&fs_info->balance_cancel_req) == 0);
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003739}
3740
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003741static void __cancel_balance(struct btrfs_fs_info *fs_info)
3742{
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003743 int ret;
3744
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003745 unset_balance_control(fs_info);
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003746 ret = del_balance_item(fs_info->tree_root);
Liu Bo0f788c52013-03-04 16:25:40 +00003747 if (ret)
Anand Jain34d97002016-03-16 16:43:06 +08003748 btrfs_handle_fs_error(fs_info, ret, NULL);
Ilya Dryomoved0fb782013-01-20 15:57:57 +02003749
3750 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003751}
3752
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003753/* Non-zero return value signifies invalidity */
3754static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3755 u64 allowed)
3756{
3757 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3758 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3759 (bctl_arg->target & ~allowed)));
3760}
3761
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003762/*
3763 * Should be called with both balance and volume mutexes held
3764 */
3765int btrfs_balance(struct btrfs_balance_control *bctl,
3766 struct btrfs_ioctl_balance_args *bargs)
3767{
3768 struct btrfs_fs_info *fs_info = bctl->fs_info;
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003769 u64 allowed;
Ilya Dryomove4837f82012-03-27 17:09:17 +03003770 int mixed = 0;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003771 int ret;
Stefan Behrens8dabb742012-11-06 13:15:27 +01003772 u64 num_devices;
Miao Xiede98ced2013-01-29 10:13:12 +00003773 unsigned seq;
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003774
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003775 if (btrfs_fs_closing(fs_info) ||
Ilya Dryomova7e99c62012-01-16 22:04:49 +02003776 atomic_read(&fs_info->balance_pause_req) ||
3777 atomic_read(&fs_info->balance_cancel_req)) {
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003778 ret = -EINVAL;
3779 goto out;
3780 }
3781
Ilya Dryomove4837f82012-03-27 17:09:17 +03003782 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3783 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3784 mixed = 1;
3785
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003786 /*
3787 * In case of mixed groups both data and meta should be picked,
3788 * and identical options should be given for both of them.
3789 */
Ilya Dryomove4837f82012-03-27 17:09:17 +03003790 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3791 if (mixed && (bctl->flags & allowed)) {
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003792 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3793 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3794 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003795 btrfs_err(fs_info, "with mixed groups data and "
3796 "metadata balance options must be the same");
Ilya Dryomovf43ffb62012-01-16 22:04:47 +02003797 ret = -EINVAL;
3798 goto out;
3799 }
3800 }
3801
Stefan Behrens8dabb742012-11-06 13:15:27 +01003802 num_devices = fs_info->fs_devices->num_devices;
Liu Bo73beece2015-07-17 16:49:19 +08003803 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
Stefan Behrens8dabb742012-11-06 13:15:27 +01003804 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3805 BUG_ON(num_devices < 1);
3806 num_devices--;
3807 }
Liu Bo73beece2015-07-17 16:49:19 +08003808 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
Austin S. Hemmelgarn88be1592016-03-23 14:22:59 -04003809 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3810 if (num_devices > 1)
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003811 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
Andreas Philipp8250dab2013-05-11 11:13:03 +00003812 if (num_devices > 2)
3813 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3814 if (num_devices > 3)
3815 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3816 BTRFS_BLOCK_GROUP_RAID6);
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003817 if (validate_convert_profile(&bctl->data, allowed)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003818 btrfs_err(fs_info, "unable to start balance with target "
3819 "data profile %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003820 bctl->data.target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003821 ret = -EINVAL;
3822 goto out;
3823 }
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003824 if (validate_convert_profile(&bctl->meta, allowed)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003825 btrfs_err(fs_info,
3826 "unable to start balance with target metadata profile %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003827 bctl->meta.target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003828 ret = -EINVAL;
3829 goto out;
3830 }
Alexandru Moisebdcd3c92015-09-22 20:02:25 +00003831 if (validate_convert_profile(&bctl->sys, allowed)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003832 btrfs_err(fs_info,
3833 "unable to start balance with target system profile %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003834 bctl->sys.target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003835 ret = -EINVAL;
3836 goto out;
3837 }
3838
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003839 /* allow to reduce meta or sys integrity only if force set */
3840 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
David Woodhouse53b381b2013-01-29 18:40:14 -05003841 BTRFS_BLOCK_GROUP_RAID10 |
3842 BTRFS_BLOCK_GROUP_RAID5 |
3843 BTRFS_BLOCK_GROUP_RAID6;
Miao Xiede98ced2013-01-29 10:13:12 +00003844 do {
3845 seq = read_seqbegin(&fs_info->profiles_lock);
3846
3847 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3848 (fs_info->avail_system_alloc_bits & allowed) &&
3849 !(bctl->sys.target & allowed)) ||
3850 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3851 (fs_info->avail_metadata_alloc_bits & allowed) &&
3852 !(bctl->meta.target & allowed))) {
3853 if (bctl->flags & BTRFS_BALANCE_FORCE) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003854 btrfs_info(fs_info, "force reducing metadata integrity");
Miao Xiede98ced2013-01-29 10:13:12 +00003855 } else {
Frank Holtonefe120a2013-12-20 11:37:06 -05003856 btrfs_err(fs_info, "balance will reduce metadata "
3857 "integrity, use force if you want this");
Miao Xiede98ced2013-01-29 10:13:12 +00003858 ret = -EINVAL;
3859 goto out;
3860 }
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003861 }
Miao Xiede98ced2013-01-29 10:13:12 +00003862 } while (read_seqretry(&fs_info->profiles_lock, seq));
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003863
Sam Tygieree592d02016-01-06 08:46:12 +00003864 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3865 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3866 btrfs_warn(fs_info,
Filipe Mananafedc0042016-01-15 10:56:15 +00003867 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
Sam Tygieree592d02016-01-06 08:46:12 +00003868 bctl->meta.target, bctl->data.target);
3869 }
3870
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02003871 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
Zhao Lei943c6e92015-08-19 15:54:15 +08003872 fs_info->num_tolerated_disk_barrier_failures = min(
3873 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3874 btrfs_get_num_tolerated_disk_barrier_failures(
3875 bctl->sys.target));
Stefan Behrens5af3e8c2012-08-01 18:56:49 +02003876 }
3877
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003878 ret = insert_balance_item(fs_info->tree_root, bctl);
Ilya Dryomov59641012012-01-16 22:04:48 +02003879 if (ret && ret != -EEXIST)
Ilya Dryomov0940ebf2012-01-16 22:04:48 +02003880 goto out;
3881
Ilya Dryomov59641012012-01-16 22:04:48 +02003882 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3883 BUG_ON(ret == -EEXIST);
3884 set_balance_control(bctl);
3885 } else {
3886 BUG_ON(ret != -EEXIST);
3887 spin_lock(&fs_info->balance_lock);
3888 update_balance_args(bctl);
3889 spin_unlock(&fs_info->balance_lock);
3890 }
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003891
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003892 atomic_inc(&fs_info->balance_running);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003893 mutex_unlock(&fs_info->balance_mutex);
3894
3895 ret = __btrfs_balance(fs_info);
3896
3897 mutex_lock(&fs_info->balance_mutex);
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003898 atomic_dec(&fs_info->balance_running);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003899
Ilya Dryomovbf023ec2013-02-12 16:27:46 +00003900 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3901 fs_info->num_tolerated_disk_barrier_failures =
3902 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3903 }
3904
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003905 if (bargs) {
3906 memset(bargs, 0, sizeof(*bargs));
Ilya Dryomov19a39dc2012-01-16 22:04:49 +02003907 update_ioctl_balance_args(fs_info, 0, bargs);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003908 }
3909
Ilya Dryomov3a01aa72013-03-06 01:57:55 -07003910 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3911 balance_need_close(fs_info)) {
3912 __cancel_balance(fs_info);
3913 }
3914
Ilya Dryomov837d5b62012-01-16 22:04:49 +02003915 wake_up(&fs_info->balance_wait_q);
Ilya Dryomovc9e9f972012-01-16 22:04:47 +02003916
3917 return ret;
3918out:
Ilya Dryomov59641012012-01-16 22:04:48 +02003919 if (bctl->flags & BTRFS_BALANCE_RESUME)
3920 __cancel_balance(fs_info);
Ilya Dryomoved0fb782013-01-20 15:57:57 +02003921 else {
Ilya Dryomov59641012012-01-16 22:04:48 +02003922 kfree(bctl);
Ilya Dryomoved0fb782013-01-20 15:57:57 +02003923 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3924 }
Ilya Dryomov59641012012-01-16 22:04:48 +02003925 return ret;
3926}
3927
3928static int balance_kthread(void *data)
3929{
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003930 struct btrfs_fs_info *fs_info = data;
Ilya Dryomov9555c6c2012-01-16 22:04:48 +02003931 int ret = 0;
Ilya Dryomov59641012012-01-16 22:04:48 +02003932
3933 mutex_lock(&fs_info->volume_mutex);
3934 mutex_lock(&fs_info->balance_mutex);
3935
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003936 if (fs_info->balance_ctl) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003937 btrfs_info(fs_info, "continuing balance");
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003938 ret = btrfs_balance(fs_info->balance_ctl, NULL);
Ilya Dryomov9555c6c2012-01-16 22:04:48 +02003939 }
Ilya Dryomov59641012012-01-16 22:04:48 +02003940
3941 mutex_unlock(&fs_info->balance_mutex);
3942 mutex_unlock(&fs_info->volume_mutex);
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003943
Ilya Dryomov59641012012-01-16 22:04:48 +02003944 return ret;
3945}
3946
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003947int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3948{
3949 struct task_struct *tsk;
3950
3951 spin_lock(&fs_info->balance_lock);
3952 if (!fs_info->balance_ctl) {
3953 spin_unlock(&fs_info->balance_lock);
3954 return 0;
3955 }
3956 spin_unlock(&fs_info->balance_lock);
3957
Jeff Mahoney3cdde222016-06-09 21:38:35 -04003958 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003959 btrfs_info(fs_info, "force skipping balance");
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003960 return 0;
3961 }
3962
3963 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
Sachin Kamatcd633972013-07-15 16:52:18 +05303964 return PTR_ERR_OR_ZERO(tsk);
Ilya Dryomov2b6ba622012-06-22 12:24:13 -06003965}
3966
Ilya Dryomov68310a52012-06-22 12:24:12 -06003967int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
Ilya Dryomov59641012012-01-16 22:04:48 +02003968{
Ilya Dryomov59641012012-01-16 22:04:48 +02003969 struct btrfs_balance_control *bctl;
3970 struct btrfs_balance_item *item;
3971 struct btrfs_disk_balance_args disk_bargs;
3972 struct btrfs_path *path;
3973 struct extent_buffer *leaf;
3974 struct btrfs_key key;
3975 int ret;
3976
3977 path = btrfs_alloc_path();
3978 if (!path)
3979 return -ENOMEM;
3980
Ilya Dryomov68310a52012-06-22 12:24:12 -06003981 key.objectid = BTRFS_BALANCE_OBJECTID;
David Sterbac479cb42016-01-25 17:51:31 +01003982 key.type = BTRFS_TEMPORARY_ITEM_KEY;
Ilya Dryomov68310a52012-06-22 12:24:12 -06003983 key.offset = 0;
3984
3985 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3986 if (ret < 0)
3987 goto out;
3988 if (ret > 0) { /* ret = -ENOENT; */
3989 ret = 0;
3990 goto out;
3991 }
3992
Ilya Dryomov59641012012-01-16 22:04:48 +02003993 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3994 if (!bctl) {
3995 ret = -ENOMEM;
3996 goto out;
3997 }
3998
Ilya Dryomov59641012012-01-16 22:04:48 +02003999 leaf = path->nodes[0];
4000 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4001
Ilya Dryomov68310a52012-06-22 12:24:12 -06004002 bctl->fs_info = fs_info;
4003 bctl->flags = btrfs_balance_flags(leaf, item);
4004 bctl->flags |= BTRFS_BALANCE_RESUME;
Ilya Dryomov59641012012-01-16 22:04:48 +02004005
4006 btrfs_balance_data(leaf, item, &disk_bargs);
4007 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4008 btrfs_balance_meta(leaf, item, &disk_bargs);
4009 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4010 btrfs_balance_sys(leaf, item, &disk_bargs);
4011 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4012
Ilya Dryomoved0fb782013-01-20 15:57:57 +02004013 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4014
Ilya Dryomov68310a52012-06-22 12:24:12 -06004015 mutex_lock(&fs_info->volume_mutex);
4016 mutex_lock(&fs_info->balance_mutex);
Ilya Dryomov59641012012-01-16 22:04:48 +02004017
Ilya Dryomov68310a52012-06-22 12:24:12 -06004018 set_balance_control(bctl);
4019
4020 mutex_unlock(&fs_info->balance_mutex);
4021 mutex_unlock(&fs_info->volume_mutex);
Ilya Dryomov59641012012-01-16 22:04:48 +02004022out:
4023 btrfs_free_path(path);
Chris Mason8f18cf12008-04-25 16:53:30 -04004024 return ret;
4025}
4026
Ilya Dryomov837d5b62012-01-16 22:04:49 +02004027int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4028{
4029 int ret = 0;
4030
4031 mutex_lock(&fs_info->balance_mutex);
4032 if (!fs_info->balance_ctl) {
4033 mutex_unlock(&fs_info->balance_mutex);
4034 return -ENOTCONN;
4035 }
4036
4037 if (atomic_read(&fs_info->balance_running)) {
4038 atomic_inc(&fs_info->balance_pause_req);
4039 mutex_unlock(&fs_info->balance_mutex);
4040
4041 wait_event(fs_info->balance_wait_q,
4042 atomic_read(&fs_info->balance_running) == 0);
4043
4044 mutex_lock(&fs_info->balance_mutex);
4045 /* we are good with balance_ctl ripped off from under us */
4046 BUG_ON(atomic_read(&fs_info->balance_running));
4047 atomic_dec(&fs_info->balance_pause_req);
4048 } else {
4049 ret = -ENOTCONN;
4050 }
4051
4052 mutex_unlock(&fs_info->balance_mutex);
4053 return ret;
4054}
4055
Ilya Dryomova7e99c62012-01-16 22:04:49 +02004056int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4057{
Ilya Dryomove649e582013-10-10 20:40:21 +03004058 if (fs_info->sb->s_flags & MS_RDONLY)
4059 return -EROFS;
4060
Ilya Dryomova7e99c62012-01-16 22:04:49 +02004061 mutex_lock(&fs_info->balance_mutex);
4062 if (!fs_info->balance_ctl) {
4063 mutex_unlock(&fs_info->balance_mutex);
4064 return -ENOTCONN;
4065 }
4066
4067 atomic_inc(&fs_info->balance_cancel_req);
4068 /*
4069 * if we are running just wait and return, balance item is
4070 * deleted in btrfs_balance in this case
4071 */
4072 if (atomic_read(&fs_info->balance_running)) {
4073 mutex_unlock(&fs_info->balance_mutex);
4074 wait_event(fs_info->balance_wait_q,
4075 atomic_read(&fs_info->balance_running) == 0);
4076 mutex_lock(&fs_info->balance_mutex);
4077 } else {
4078 /* __cancel_balance needs volume_mutex */
4079 mutex_unlock(&fs_info->balance_mutex);
4080 mutex_lock(&fs_info->volume_mutex);
4081 mutex_lock(&fs_info->balance_mutex);
4082
4083 if (fs_info->balance_ctl)
4084 __cancel_balance(fs_info);
4085
4086 mutex_unlock(&fs_info->volume_mutex);
4087 }
4088
4089 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4090 atomic_dec(&fs_info->balance_cancel_req);
4091 mutex_unlock(&fs_info->balance_mutex);
4092 return 0;
4093}
4094
Stefan Behrens803b2f52013-08-15 17:11:21 +02004095static int btrfs_uuid_scan_kthread(void *data)
4096{
4097 struct btrfs_fs_info *fs_info = data;
4098 struct btrfs_root *root = fs_info->tree_root;
4099 struct btrfs_key key;
4100 struct btrfs_key max_key;
4101 struct btrfs_path *path = NULL;
4102 int ret = 0;
4103 struct extent_buffer *eb;
4104 int slot;
4105 struct btrfs_root_item root_item;
4106 u32 item_size;
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004107 struct btrfs_trans_handle *trans = NULL;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004108
4109 path = btrfs_alloc_path();
4110 if (!path) {
4111 ret = -ENOMEM;
4112 goto out;
4113 }
4114
4115 key.objectid = 0;
4116 key.type = BTRFS_ROOT_ITEM_KEY;
4117 key.offset = 0;
4118
4119 max_key.objectid = (u64)-1;
4120 max_key.type = BTRFS_ROOT_ITEM_KEY;
4121 max_key.offset = (u64)-1;
4122
Stefan Behrens803b2f52013-08-15 17:11:21 +02004123 while (1) {
Filipe David Borba Manana6174d3c2013-10-01 16:13:42 +01004124 ret = btrfs_search_forward(root, &key, path, 0);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004125 if (ret) {
4126 if (ret > 0)
4127 ret = 0;
4128 break;
4129 }
4130
4131 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4132 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4133 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4134 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4135 goto skip;
4136
4137 eb = path->nodes[0];
4138 slot = path->slots[0];
4139 item_size = btrfs_item_size_nr(eb, slot);
4140 if (item_size < sizeof(root_item))
4141 goto skip;
4142
Stefan Behrens803b2f52013-08-15 17:11:21 +02004143 read_extent_buffer(eb, &root_item,
4144 btrfs_item_ptr_offset(eb, slot),
4145 (int)sizeof(root_item));
4146 if (btrfs_root_refs(&root_item) == 0)
4147 goto skip;
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004148
4149 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4150 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4151 if (trans)
4152 goto update_tree;
4153
4154 btrfs_release_path(path);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004155 /*
4156 * 1 - subvol uuid item
4157 * 1 - received_subvol uuid item
4158 */
4159 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4160 if (IS_ERR(trans)) {
4161 ret = PTR_ERR(trans);
4162 break;
4163 }
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004164 continue;
4165 } else {
4166 goto skip;
4167 }
4168update_tree:
4169 if (!btrfs_is_empty_uuid(root_item.uuid)) {
Stefan Behrens803b2f52013-08-15 17:11:21 +02004170 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4171 root_item.uuid,
4172 BTRFS_UUID_KEY_SUBVOL,
4173 key.objectid);
4174 if (ret < 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05004175 btrfs_warn(fs_info, "uuid_tree_add failed %d",
Stefan Behrens803b2f52013-08-15 17:11:21 +02004176 ret);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004177 break;
4178 }
4179 }
4180
4181 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
Stefan Behrens803b2f52013-08-15 17:11:21 +02004182 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4183 root_item.received_uuid,
4184 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4185 key.objectid);
4186 if (ret < 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05004187 btrfs_warn(fs_info, "uuid_tree_add failed %d",
Stefan Behrens803b2f52013-08-15 17:11:21 +02004188 ret);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004189 break;
4190 }
4191 }
4192
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004193skip:
Stefan Behrens803b2f52013-08-15 17:11:21 +02004194 if (trans) {
4195 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004196 trans = NULL;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004197 if (ret)
4198 break;
4199 }
4200
Stefan Behrens803b2f52013-08-15 17:11:21 +02004201 btrfs_release_path(path);
4202 if (key.offset < (u64)-1) {
4203 key.offset++;
4204 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4205 key.offset = 0;
4206 key.type = BTRFS_ROOT_ITEM_KEY;
4207 } else if (key.objectid < (u64)-1) {
4208 key.offset = 0;
4209 key.type = BTRFS_ROOT_ITEM_KEY;
4210 key.objectid++;
4211 } else {
4212 break;
4213 }
4214 cond_resched();
4215 }
4216
4217out:
4218 btrfs_free_path(path);
Filipe David Borba Mananaf45388f2013-08-28 10:28:34 +01004219 if (trans && !IS_ERR(trans))
4220 btrfs_end_transaction(trans, fs_info->uuid_root);
Stefan Behrens803b2f52013-08-15 17:11:21 +02004221 if (ret)
Frank Holtonefe120a2013-12-20 11:37:06 -05004222 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
Stefan Behrens70f80172013-08-15 17:11:23 +02004223 else
4224 fs_info->update_uuid_tree_gen = 1;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004225 up(&fs_info->uuid_tree_rescan_sem);
4226 return 0;
4227}
4228
Stefan Behrens70f80172013-08-15 17:11:23 +02004229/*
4230 * Callback for btrfs_uuid_tree_iterate().
4231 * returns:
4232 * 0 check succeeded, the entry is not outdated.
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08004233 * < 0 if an error occurred.
Stefan Behrens70f80172013-08-15 17:11:23 +02004234 * > 0 if the check failed, which means the caller shall remove the entry.
4235 */
4236static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4237 u8 *uuid, u8 type, u64 subid)
4238{
4239 struct btrfs_key key;
4240 int ret = 0;
4241 struct btrfs_root *subvol_root;
4242
4243 if (type != BTRFS_UUID_KEY_SUBVOL &&
4244 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4245 goto out;
4246
4247 key.objectid = subid;
4248 key.type = BTRFS_ROOT_ITEM_KEY;
4249 key.offset = (u64)-1;
4250 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4251 if (IS_ERR(subvol_root)) {
4252 ret = PTR_ERR(subvol_root);
4253 if (ret == -ENOENT)
4254 ret = 1;
4255 goto out;
4256 }
4257
4258 switch (type) {
4259 case BTRFS_UUID_KEY_SUBVOL:
4260 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4261 ret = 1;
4262 break;
4263 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4264 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4265 BTRFS_UUID_SIZE))
4266 ret = 1;
4267 break;
4268 }
4269
4270out:
4271 return ret;
4272}
4273
4274static int btrfs_uuid_rescan_kthread(void *data)
4275{
4276 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4277 int ret;
4278
4279 /*
4280 * 1st step is to iterate through the existing UUID tree and
4281 * to delete all entries that contain outdated data.
4282 * 2nd step is to add all missing entries to the UUID tree.
4283 */
4284 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4285 if (ret < 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05004286 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
Stefan Behrens70f80172013-08-15 17:11:23 +02004287 up(&fs_info->uuid_tree_rescan_sem);
4288 return ret;
4289 }
4290 return btrfs_uuid_scan_kthread(data);
4291}
4292
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004293int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4294{
4295 struct btrfs_trans_handle *trans;
4296 struct btrfs_root *tree_root = fs_info->tree_root;
4297 struct btrfs_root *uuid_root;
Stefan Behrens803b2f52013-08-15 17:11:21 +02004298 struct task_struct *task;
4299 int ret;
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004300
4301 /*
4302 * 1 - root node
4303 * 1 - root item
4304 */
4305 trans = btrfs_start_transaction(tree_root, 2);
4306 if (IS_ERR(trans))
4307 return PTR_ERR(trans);
4308
4309 uuid_root = btrfs_create_tree(trans, fs_info,
4310 BTRFS_UUID_TREE_OBJECTID);
4311 if (IS_ERR(uuid_root)) {
David Sterba6d13f542015-04-24 19:12:01 +02004312 ret = PTR_ERR(uuid_root);
Jeff Mahoney66642832016-06-10 18:19:25 -04004313 btrfs_abort_transaction(trans, ret);
Josef Bacik65d4f4c2015-09-23 15:00:37 -04004314 btrfs_end_transaction(trans, tree_root);
David Sterba6d13f542015-04-24 19:12:01 +02004315 return ret;
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004316 }
4317
4318 fs_info->uuid_root = uuid_root;
4319
Stefan Behrens803b2f52013-08-15 17:11:21 +02004320 ret = btrfs_commit_transaction(trans, tree_root);
4321 if (ret)
4322 return ret;
4323
4324 down(&fs_info->uuid_tree_rescan_sem);
4325 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4326 if (IS_ERR(task)) {
Stefan Behrens70f80172013-08-15 17:11:23 +02004327 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
Frank Holtonefe120a2013-12-20 11:37:06 -05004328 btrfs_warn(fs_info, "failed to start uuid_scan task");
Stefan Behrens803b2f52013-08-15 17:11:21 +02004329 up(&fs_info->uuid_tree_rescan_sem);
4330 return PTR_ERR(task);
4331 }
4332
4333 return 0;
Stefan Behrensf7a81ea2013-08-15 17:11:19 +02004334}
Stefan Behrens803b2f52013-08-15 17:11:21 +02004335
Stefan Behrens70f80172013-08-15 17:11:23 +02004336int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4337{
4338 struct task_struct *task;
4339
4340 down(&fs_info->uuid_tree_rescan_sem);
4341 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4342 if (IS_ERR(task)) {
4343 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
Frank Holtonefe120a2013-12-20 11:37:06 -05004344 btrfs_warn(fs_info, "failed to start uuid_rescan task");
Stefan Behrens70f80172013-08-15 17:11:23 +02004345 up(&fs_info->uuid_tree_rescan_sem);
4346 return PTR_ERR(task);
4347 }
4348
4349 return 0;
4350}
4351
Chris Mason8f18cf12008-04-25 16:53:30 -04004352/*
4353 * shrinking a device means finding all of the device extents past
4354 * the new size, and then following the back refs to the chunks.
4355 * The chunk relocation code actually frees the device extent
4356 */
4357int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4358{
4359 struct btrfs_trans_handle *trans;
4360 struct btrfs_root *root = device->dev_root;
4361 struct btrfs_dev_extent *dev_extent = NULL;
4362 struct btrfs_path *path;
4363 u64 length;
Chris Mason8f18cf12008-04-25 16:53:30 -04004364 u64 chunk_offset;
4365 int ret;
4366 int slot;
Josef Bacikba1bf482009-09-11 16:11:19 -04004367 int failed = 0;
4368 bool retried = false;
Filipe Manana53e489b2015-06-02 14:43:21 +01004369 bool checked_pending_chunks = false;
Chris Mason8f18cf12008-04-25 16:53:30 -04004370 struct extent_buffer *l;
4371 struct btrfs_key key;
David Sterba6c417612011-04-13 15:41:04 +02004372 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Mason8f18cf12008-04-25 16:53:30 -04004373 u64 old_total = btrfs_super_total_bytes(super_copy);
Miao Xie7cc8e582014-09-03 21:35:38 +08004374 u64 old_size = btrfs_device_get_total_bytes(device);
4375 u64 diff = old_size - new_size;
Chris Mason8f18cf12008-04-25 16:53:30 -04004376
Stefan Behrens63a212a2012-11-05 18:29:28 +01004377 if (device->is_tgtdev_for_dev_replace)
4378 return -EINVAL;
4379
Chris Mason8f18cf12008-04-25 16:53:30 -04004380 path = btrfs_alloc_path();
4381 if (!path)
4382 return -ENOMEM;
4383
David Sterbae4058b52015-11-27 16:31:35 +01004384 path->reada = READA_FORWARD;
Chris Mason8f18cf12008-04-25 16:53:30 -04004385
Chris Mason7d9eb122008-07-08 14:19:17 -04004386 lock_chunks(root);
4387
Miao Xie7cc8e582014-09-03 21:35:38 +08004388 btrfs_device_set_total_bytes(device, new_size);
Josef Bacik2bf64752011-09-26 17:12:22 -04004389 if (device->writeable) {
Yan Zheng2b820322008-11-17 21:11:30 -05004390 device->fs_devices->total_rw_bytes -= diff;
Josef Bacik2bf64752011-09-26 17:12:22 -04004391 spin_lock(&root->fs_info->free_chunk_lock);
4392 root->fs_info->free_chunk_space -= diff;
4393 spin_unlock(&root->fs_info->free_chunk_lock);
4394 }
Chris Mason7d9eb122008-07-08 14:19:17 -04004395 unlock_chunks(root);
Chris Mason8f18cf12008-04-25 16:53:30 -04004396
Josef Bacikba1bf482009-09-11 16:11:19 -04004397again:
Chris Mason8f18cf12008-04-25 16:53:30 -04004398 key.objectid = device->devid;
4399 key.offset = (u64)-1;
4400 key.type = BTRFS_DEV_EXTENT_KEY;
4401
Ilya Dryomov213e64d2012-03-27 17:09:18 +03004402 do {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004403 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
Chris Mason8f18cf12008-04-25 16:53:30 -04004404 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004405 if (ret < 0) {
4406 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Chris Mason8f18cf12008-04-25 16:53:30 -04004407 goto done;
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004408 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004409
4410 ret = btrfs_previous_item(root, path, 0, key.type);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004411 if (ret)
4412 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Chris Mason8f18cf12008-04-25 16:53:30 -04004413 if (ret < 0)
4414 goto done;
4415 if (ret) {
4416 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02004417 btrfs_release_path(path);
Yan Zhengbf1fb512009-07-22 09:59:00 -04004418 break;
Chris Mason8f18cf12008-04-25 16:53:30 -04004419 }
4420
4421 l = path->nodes[0];
4422 slot = path->slots[0];
4423 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4424
Josef Bacikba1bf482009-09-11 16:11:19 -04004425 if (key.objectid != device->devid) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004426 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
David Sterbab3b4aa72011-04-21 01:20:15 +02004427 btrfs_release_path(path);
Yan Zhengbf1fb512009-07-22 09:59:00 -04004428 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04004429 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004430
4431 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4432 length = btrfs_dev_extent_length(l, dev_extent);
4433
Josef Bacikba1bf482009-09-11 16:11:19 -04004434 if (key.offset + length <= new_size) {
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004435 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
David Sterbab3b4aa72011-04-21 01:20:15 +02004436 btrfs_release_path(path);
Chris Balld6397ba2009-04-27 07:29:03 -04004437 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04004438 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004439
Chris Mason8f18cf12008-04-25 16:53:30 -04004440 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
David Sterbab3b4aa72011-04-21 01:20:15 +02004441 btrfs_release_path(path);
Chris Mason8f18cf12008-04-25 16:53:30 -04004442
Zhaoleidc2ee4e2015-08-05 18:00:04 +08004443 ret = btrfs_relocate_chunk(root, chunk_offset);
Filipe Manana67c5e7d2015-06-11 00:58:53 +01004444 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
Josef Bacikba1bf482009-09-11 16:11:19 -04004445 if (ret && ret != -ENOSPC)
Chris Mason8f18cf12008-04-25 16:53:30 -04004446 goto done;
Josef Bacikba1bf482009-09-11 16:11:19 -04004447 if (ret == -ENOSPC)
4448 failed++;
Ilya Dryomov213e64d2012-03-27 17:09:18 +03004449 } while (key.offset-- > 0);
Josef Bacikba1bf482009-09-11 16:11:19 -04004450
4451 if (failed && !retried) {
4452 failed = 0;
4453 retried = true;
4454 goto again;
4455 } else if (failed && retried) {
4456 ret = -ENOSPC;
Josef Bacikba1bf482009-09-11 16:11:19 -04004457 goto done;
Chris Mason8f18cf12008-04-25 16:53:30 -04004458 }
4459
Chris Balld6397ba2009-04-27 07:29:03 -04004460 /* Shrinking succeeded, else we would be at "done". */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004461 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00004462 if (IS_ERR(trans)) {
4463 ret = PTR_ERR(trans);
4464 goto done;
4465 }
4466
Chris Balld6397ba2009-04-27 07:29:03 -04004467 lock_chunks(root);
Filipe Manana53e489b2015-06-02 14:43:21 +01004468
4469 /*
4470 * We checked in the above loop all device extents that were already in
4471 * the device tree. However before we have updated the device's
4472 * total_bytes to the new size, we might have had chunk allocations that
4473 * have not complete yet (new block groups attached to transaction
4474 * handles), and therefore their device extents were not yet in the
4475 * device tree and we missed them in the loop above. So if we have any
4476 * pending chunk using a device extent that overlaps the device range
4477 * that we can not use anymore, commit the current transaction and
4478 * repeat the search on the device tree - this way we guarantee we will
4479 * not have chunks using device extents that end beyond 'new_size'.
4480 */
4481 if (!checked_pending_chunks) {
4482 u64 start = new_size;
4483 u64 len = old_size - new_size;
4484
Jeff Mahoney499f3772015-06-15 09:41:17 -04004485 if (contains_pending_extent(trans->transaction, device,
4486 &start, len)) {
Filipe Manana53e489b2015-06-02 14:43:21 +01004487 unlock_chunks(root);
4488 checked_pending_chunks = true;
4489 failed = 0;
4490 retried = false;
4491 ret = btrfs_commit_transaction(trans, root);
4492 if (ret)
4493 goto done;
4494 goto again;
4495 }
4496 }
4497
Miao Xie7cc8e582014-09-03 21:35:38 +08004498 btrfs_device_set_disk_total_bytes(device, new_size);
Miao Xie935e5cc2014-09-03 21:35:33 +08004499 if (list_empty(&device->resized_list))
4500 list_add_tail(&device->resized_list,
4501 &root->fs_info->fs_devices->resized_devices);
Chris Balld6397ba2009-04-27 07:29:03 -04004502
Chris Balld6397ba2009-04-27 07:29:03 -04004503 WARN_ON(diff > old_total);
4504 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4505 unlock_chunks(root);
Miao Xie2196d6e2014-09-03 21:35:41 +08004506
4507 /* Now btrfs_update_device() will change the on-disk size. */
4508 ret = btrfs_update_device(trans, device);
Chris Balld6397ba2009-04-27 07:29:03 -04004509 btrfs_end_transaction(trans, root);
Chris Mason8f18cf12008-04-25 16:53:30 -04004510done:
4511 btrfs_free_path(path);
Filipe Manana53e489b2015-06-02 14:43:21 +01004512 if (ret) {
4513 lock_chunks(root);
4514 btrfs_device_set_total_bytes(device, old_size);
4515 if (device->writeable)
4516 device->fs_devices->total_rw_bytes += diff;
4517 spin_lock(&root->fs_info->free_chunk_lock);
4518 root->fs_info->free_chunk_space += diff;
4519 spin_unlock(&root->fs_info->free_chunk_lock);
4520 unlock_chunks(root);
4521 }
Chris Mason8f18cf12008-04-25 16:53:30 -04004522 return ret;
4523}
4524
Li Zefan125ccb02011-12-08 15:07:24 +08004525static int btrfs_add_system_chunk(struct btrfs_root *root,
Chris Mason0b86a832008-03-24 15:01:56 -04004526 struct btrfs_key *key,
4527 struct btrfs_chunk *chunk, int item_size)
4528{
David Sterba6c417612011-04-13 15:41:04 +02004529 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Mason0b86a832008-03-24 15:01:56 -04004530 struct btrfs_disk_key disk_key;
4531 u32 array_size;
4532 u8 *ptr;
4533
Miao Xiefe48a5c2014-09-03 21:35:39 +08004534 lock_chunks(root);
Chris Mason0b86a832008-03-24 15:01:56 -04004535 array_size = btrfs_super_sys_array_size(super_copy);
Gui Hecheng5f43f862014-04-21 20:13:11 +08004536 if (array_size + item_size + sizeof(disk_key)
Miao Xiefe48a5c2014-09-03 21:35:39 +08004537 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4538 unlock_chunks(root);
Chris Mason0b86a832008-03-24 15:01:56 -04004539 return -EFBIG;
Miao Xiefe48a5c2014-09-03 21:35:39 +08004540 }
Chris Mason0b86a832008-03-24 15:01:56 -04004541
4542 ptr = super_copy->sys_chunk_array + array_size;
4543 btrfs_cpu_key_to_disk(&disk_key, key);
4544 memcpy(ptr, &disk_key, sizeof(disk_key));
4545 ptr += sizeof(disk_key);
4546 memcpy(ptr, chunk, item_size);
4547 item_size += sizeof(disk_key);
4548 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
Miao Xiefe48a5c2014-09-03 21:35:39 +08004549 unlock_chunks(root);
4550
Chris Mason0b86a832008-03-24 15:01:56 -04004551 return 0;
4552}
4553
Miao Xieb2117a32011-01-05 10:07:28 +00004554/*
Arne Jansen73c5de02011-04-12 12:07:57 +02004555 * sort the devices in descending order by max_avail, total_avail
Miao Xieb2117a32011-01-05 10:07:28 +00004556 */
Arne Jansen73c5de02011-04-12 12:07:57 +02004557static int btrfs_cmp_device_info(const void *a, const void *b)
Miao Xieb2117a32011-01-05 10:07:28 +00004558{
Arne Jansen73c5de02011-04-12 12:07:57 +02004559 const struct btrfs_device_info *di_a = a;
4560 const struct btrfs_device_info *di_b = b;
Miao Xieb2117a32011-01-05 10:07:28 +00004561
Arne Jansen73c5de02011-04-12 12:07:57 +02004562 if (di_a->max_avail > di_b->max_avail)
4563 return -1;
4564 if (di_a->max_avail < di_b->max_avail)
4565 return 1;
4566 if (di_a->total_avail > di_b->total_avail)
4567 return -1;
4568 if (di_a->total_avail < di_b->total_avail)
4569 return 1;
Miao Xieb2117a32011-01-05 10:07:28 +00004570 return 0;
4571}
4572
David Woodhouse53b381b2013-01-29 18:40:14 -05004573static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4574{
4575 /* TODO allow them to set a preferred stripe size */
Byongho Leeee221842015-12-15 01:42:10 +09004576 return SZ_64K;
David Woodhouse53b381b2013-01-29 18:40:14 -05004577}
4578
4579static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4580{
Zhao Leiffe2d202015-01-20 15:11:44 +08004581 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
David Woodhouse53b381b2013-01-29 18:40:14 -05004582 return;
4583
Miao Xieceda0862013-04-11 10:30:16 +00004584 btrfs_set_fs_incompat(info, RAID56);
David Woodhouse53b381b2013-01-29 18:40:14 -05004585}
4586
Jeff Mahoney14a1e062016-06-15 10:25:38 -04004587#define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004588 - sizeof(struct btrfs_chunk)) \
4589 / sizeof(struct btrfs_stripe) + 1)
4590
4591#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4592 - 2 * sizeof(struct btrfs_disk_key) \
4593 - 2 * sizeof(struct btrfs_chunk)) \
4594 / sizeof(struct btrfs_stripe) + 1)
4595
Miao Xieb2117a32011-01-05 10:07:28 +00004596static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
Josef Bacik6df9a952013-06-27 13:22:46 -04004597 struct btrfs_root *extent_root, u64 start,
4598 u64 type)
Miao Xieb2117a32011-01-05 10:07:28 +00004599{
4600 struct btrfs_fs_info *info = extent_root->fs_info;
Miao Xieb2117a32011-01-05 10:07:28 +00004601 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4602 struct list_head *cur;
Arne Jansen73c5de02011-04-12 12:07:57 +02004603 struct map_lookup *map = NULL;
Miao Xieb2117a32011-01-05 10:07:28 +00004604 struct extent_map_tree *em_tree;
4605 struct extent_map *em;
Arne Jansen73c5de02011-04-12 12:07:57 +02004606 struct btrfs_device_info *devices_info = NULL;
4607 u64 total_avail;
4608 int num_stripes; /* total number of stripes to allocate */
David Woodhouse53b381b2013-01-29 18:40:14 -05004609 int data_stripes; /* number of stripes that count for
4610 block group size */
Arne Jansen73c5de02011-04-12 12:07:57 +02004611 int sub_stripes; /* sub_stripes info for map */
4612 int dev_stripes; /* stripes per dev */
4613 int devs_max; /* max devs to use */
4614 int devs_min; /* min devs needed */
4615 int devs_increment; /* ndevs has to be a multiple of this */
4616 int ncopies; /* how many copies to data has */
Miao Xieb2117a32011-01-05 10:07:28 +00004617 int ret;
Arne Jansen73c5de02011-04-12 12:07:57 +02004618 u64 max_stripe_size;
4619 u64 max_chunk_size;
4620 u64 stripe_size;
4621 u64 num_bytes;
David Woodhouse53b381b2013-01-29 18:40:14 -05004622 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
Arne Jansen73c5de02011-04-12 12:07:57 +02004623 int ndevs;
4624 int i;
4625 int j;
Liu Bo31e50222012-11-21 14:18:10 +00004626 int index;
Miao Xieb2117a32011-01-05 10:07:28 +00004627
Ilya Dryomov0c460c02012-03-27 17:09:17 +03004628 BUG_ON(!alloc_profile_is_valid(type, 0));
Arne Jansen73c5de02011-04-12 12:07:57 +02004629
Miao Xieb2117a32011-01-05 10:07:28 +00004630 if (list_empty(&fs_devices->alloc_list))
4631 return -ENOSPC;
4632
Liu Bo31e50222012-11-21 14:18:10 +00004633 index = __get_raid_index(type);
Arne Jansen73c5de02011-04-12 12:07:57 +02004634
Liu Bo31e50222012-11-21 14:18:10 +00004635 sub_stripes = btrfs_raid_array[index].sub_stripes;
4636 dev_stripes = btrfs_raid_array[index].dev_stripes;
4637 devs_max = btrfs_raid_array[index].devs_max;
4638 devs_min = btrfs_raid_array[index].devs_min;
4639 devs_increment = btrfs_raid_array[index].devs_increment;
4640 ncopies = btrfs_raid_array[index].ncopies;
Arne Jansen73c5de02011-04-12 12:07:57 +02004641
4642 if (type & BTRFS_BLOCK_GROUP_DATA) {
Byongho Leeee221842015-12-15 01:42:10 +09004643 max_stripe_size = SZ_1G;
Arne Jansen73c5de02011-04-12 12:07:57 +02004644 max_chunk_size = 10 * max_stripe_size;
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004645 if (!devs_max)
4646 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
Arne Jansen73c5de02011-04-12 12:07:57 +02004647 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
Chris Mason11003732012-01-06 15:47:38 -05004648 /* for larger filesystems, use larger metadata chunks */
Byongho Leeee221842015-12-15 01:42:10 +09004649 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4650 max_stripe_size = SZ_1G;
Chris Mason11003732012-01-06 15:47:38 -05004651 else
Byongho Leeee221842015-12-15 01:42:10 +09004652 max_stripe_size = SZ_256M;
Arne Jansen73c5de02011-04-12 12:07:57 +02004653 max_chunk_size = max_stripe_size;
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004654 if (!devs_max)
4655 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
Arne Jansen73c5de02011-04-12 12:07:57 +02004656 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
Byongho Leeee221842015-12-15 01:42:10 +09004657 max_stripe_size = SZ_32M;
Arne Jansen73c5de02011-04-12 12:07:57 +02004658 max_chunk_size = 2 * max_stripe_size;
Gui Hecheng23f8f9b2014-04-21 20:13:12 +08004659 if (!devs_max)
4660 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
Arne Jansen73c5de02011-04-12 12:07:57 +02004661 } else {
David Sterba351fd352014-05-15 16:48:20 +02004662 btrfs_err(info, "invalid chunk type 0x%llx requested",
Arne Jansen73c5de02011-04-12 12:07:57 +02004663 type);
4664 BUG_ON(1);
4665 }
4666
4667 /* we don't want a chunk larger than 10% of writeable space */
4668 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4669 max_chunk_size);
Miao Xieb2117a32011-01-05 10:07:28 +00004670
David Sterba31e818f2015-02-20 18:00:26 +01004671 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
Miao Xieb2117a32011-01-05 10:07:28 +00004672 GFP_NOFS);
4673 if (!devices_info)
4674 return -ENOMEM;
4675
Arne Jansen73c5de02011-04-12 12:07:57 +02004676 cur = fs_devices->alloc_list.next;
4677
4678 /*
4679 * in the first pass through the devices list, we gather information
4680 * about the available holes on each device.
4681 */
4682 ndevs = 0;
4683 while (cur != &fs_devices->alloc_list) {
4684 struct btrfs_device *device;
4685 u64 max_avail;
4686 u64 dev_offset;
4687
4688 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4689
4690 cur = cur->next;
4691
4692 if (!device->writeable) {
Julia Lawall31b1a2b2012-11-03 10:58:34 +00004693 WARN(1, KERN_ERR
Frank Holtonefe120a2013-12-20 11:37:06 -05004694 "BTRFS: read-only device in alloc_list\n");
Arne Jansen73c5de02011-04-12 12:07:57 +02004695 continue;
4696 }
4697
Stefan Behrens63a212a2012-11-05 18:29:28 +01004698 if (!device->in_fs_metadata ||
4699 device->is_tgtdev_for_dev_replace)
Arne Jansen73c5de02011-04-12 12:07:57 +02004700 continue;
4701
4702 if (device->total_bytes > device->bytes_used)
4703 total_avail = device->total_bytes - device->bytes_used;
4704 else
4705 total_avail = 0;
liubo38c01b92011-08-02 02:39:03 +00004706
4707 /* If there is no space on this device, skip it. */
4708 if (total_avail == 0)
4709 continue;
Arne Jansen73c5de02011-04-12 12:07:57 +02004710
Josef Bacik6df9a952013-06-27 13:22:46 -04004711 ret = find_free_dev_extent(trans, device,
Arne Jansen73c5de02011-04-12 12:07:57 +02004712 max_stripe_size * dev_stripes,
4713 &dev_offset, &max_avail);
4714 if (ret && ret != -ENOSPC)
4715 goto error;
4716
4717 if (ret == 0)
4718 max_avail = max_stripe_size * dev_stripes;
4719
4720 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4721 continue;
4722
Eric Sandeen063d0062013-01-31 00:55:01 +00004723 if (ndevs == fs_devices->rw_devices) {
4724 WARN(1, "%s: found more than %llu devices\n",
4725 __func__, fs_devices->rw_devices);
4726 break;
4727 }
Arne Jansen73c5de02011-04-12 12:07:57 +02004728 devices_info[ndevs].dev_offset = dev_offset;
4729 devices_info[ndevs].max_avail = max_avail;
4730 devices_info[ndevs].total_avail = total_avail;
4731 devices_info[ndevs].dev = device;
4732 ++ndevs;
4733 }
4734
4735 /*
4736 * now sort the devices by hole size / available space
4737 */
4738 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4739 btrfs_cmp_device_info, NULL);
4740
4741 /* round down to number of usable stripes */
4742 ndevs -= ndevs % devs_increment;
4743
4744 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4745 ret = -ENOSPC;
4746 goto error;
4747 }
4748
4749 if (devs_max && ndevs > devs_max)
4750 ndevs = devs_max;
4751 /*
4752 * the primary goal is to maximize the number of stripes, so use as many
4753 * devices as possible, even if the stripes are not maximum sized.
4754 */
4755 stripe_size = devices_info[ndevs-1].max_avail;
4756 num_stripes = ndevs * dev_stripes;
4757
David Woodhouse53b381b2013-01-29 18:40:14 -05004758 /*
4759 * this will have to be fixed for RAID1 and RAID10 over
4760 * more drives
4761 */
4762 data_stripes = num_stripes / ncopies;
4763
David Woodhouse53b381b2013-01-29 18:40:14 -05004764 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4765 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
Chandan Rajendrab7f67052016-06-23 15:16:44 +05304766 extent_root->stripesize);
David Woodhouse53b381b2013-01-29 18:40:14 -05004767 data_stripes = num_stripes - 1;
4768 }
4769 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4770 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
Chandan Rajendrab7f67052016-06-23 15:16:44 +05304771 extent_root->stripesize);
David Woodhouse53b381b2013-01-29 18:40:14 -05004772 data_stripes = num_stripes - 2;
4773 }
Chris Mason86db2572013-02-20 16:23:40 -05004774
4775 /*
4776 * Use the number of data stripes to figure out how big this chunk
4777 * is really going to be in terms of logical address space,
4778 * and compare that answer with the max chunk size
4779 */
4780 if (stripe_size * data_stripes > max_chunk_size) {
4781 u64 mask = (1ULL << 24) - 1;
David Sterbab8b93ad2015-01-16 17:26:13 +01004782
4783 stripe_size = div_u64(max_chunk_size, data_stripes);
Chris Mason86db2572013-02-20 16:23:40 -05004784
4785 /* bump the answer up to a 16MB boundary */
4786 stripe_size = (stripe_size + mask) & ~mask;
4787
4788 /* but don't go higher than the limits we found
4789 * while searching for free extents
4790 */
4791 if (stripe_size > devices_info[ndevs-1].max_avail)
4792 stripe_size = devices_info[ndevs-1].max_avail;
4793 }
4794
David Sterbab8b93ad2015-01-16 17:26:13 +01004795 stripe_size = div_u64(stripe_size, dev_stripes);
Ilya Dryomov37db63a2012-04-13 17:05:08 +03004796
4797 /* align to BTRFS_STRIPE_LEN */
David Sterbab8b93ad2015-01-16 17:26:13 +01004798 stripe_size = div_u64(stripe_size, raid_stripe_len);
David Woodhouse53b381b2013-01-29 18:40:14 -05004799 stripe_size *= raid_stripe_len;
Arne Jansen73c5de02011-04-12 12:07:57 +02004800
Miao Xieb2117a32011-01-05 10:07:28 +00004801 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4802 if (!map) {
4803 ret = -ENOMEM;
4804 goto error;
4805 }
4806 map->num_stripes = num_stripes;
Chris Mason9b3f68b2008-04-18 10:29:51 -04004807
Arne Jansen73c5de02011-04-12 12:07:57 +02004808 for (i = 0; i < ndevs; ++i) {
4809 for (j = 0; j < dev_stripes; ++j) {
4810 int s = i * dev_stripes + j;
4811 map->stripes[s].dev = devices_info[i].dev;
4812 map->stripes[s].physical = devices_info[i].dev_offset +
4813 j * stripe_size;
Chris Masona40a90a2008-04-18 11:55:51 -04004814 }
Chris Mason6324fbf2008-03-24 15:01:59 -04004815 }
Chris Mason593060d2008-03-25 16:50:33 -04004816 map->sector_size = extent_root->sectorsize;
David Woodhouse53b381b2013-01-29 18:40:14 -05004817 map->stripe_len = raid_stripe_len;
4818 map->io_align = raid_stripe_len;
4819 map->io_width = raid_stripe_len;
Chris Mason593060d2008-03-25 16:50:33 -04004820 map->type = type;
Chris Mason321aecc2008-04-16 10:49:51 -04004821 map->sub_stripes = sub_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04004822
David Woodhouse53b381b2013-01-29 18:40:14 -05004823 num_bytes = stripe_size * data_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04004824
Arne Jansen73c5de02011-04-12 12:07:57 +02004825 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
liubo1abe9b82011-03-24 11:18:59 +00004826
David Sterba172ddd62011-04-21 00:48:27 +02004827 em = alloc_extent_map();
Yan Zheng2b820322008-11-17 21:11:30 -05004828 if (!em) {
Wang Shilong298a8f92014-06-19 10:42:52 +08004829 kfree(map);
Miao Xieb2117a32011-01-05 10:07:28 +00004830 ret = -ENOMEM;
4831 goto error;
Yan Zheng2b820322008-11-17 21:11:30 -05004832 }
Wang Shilong298a8f92014-06-19 10:42:52 +08004833 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
Jeff Mahoney95617d62015-06-03 10:55:48 -04004834 em->map_lookup = map;
Yan Zheng2b820322008-11-17 21:11:30 -05004835 em->start = start;
Arne Jansen73c5de02011-04-12 12:07:57 +02004836 em->len = num_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04004837 em->block_start = 0;
Chris Masonc8b97812008-10-29 14:49:59 -04004838 em->block_len = em->len;
Josef Bacik6df9a952013-06-27 13:22:46 -04004839 em->orig_block_len = stripe_size;
Chris Mason0b86a832008-03-24 15:01:56 -04004840
Chris Mason0b86a832008-03-24 15:01:56 -04004841 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
Chris Mason890871b2009-09-02 16:24:52 -04004842 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04004843 ret = add_extent_mapping(em_tree, em, 0);
Josef Bacik6df9a952013-06-27 13:22:46 -04004844 if (!ret) {
4845 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4846 atomic_inc(&em->refs);
4847 }
Chris Mason890871b2009-09-02 16:24:52 -04004848 write_unlock(&em_tree->lock);
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004849 if (ret) {
4850 free_extent_map(em);
Mark Fasheh1dd46022011-09-08 17:29:00 -07004851 goto error;
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004852 }
Yan Zheng2b820322008-11-17 21:11:30 -05004853
Josef Bacik04487482013-01-29 15:03:37 -05004854 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4855 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4856 start, num_bytes);
Josef Bacik6df9a952013-06-27 13:22:46 -04004857 if (ret)
4858 goto error_del_extent;
Yan Zheng2b820322008-11-17 21:11:30 -05004859
Miao Xie7cc8e582014-09-03 21:35:38 +08004860 for (i = 0; i < map->num_stripes; i++) {
4861 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4862 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4863 }
Miao Xie43530c42014-09-03 21:35:36 +08004864
Miao Xie1c116182014-09-03 21:35:37 +08004865 spin_lock(&extent_root->fs_info->free_chunk_lock);
4866 extent_root->fs_info->free_chunk_space -= (stripe_size *
4867 map->num_stripes);
4868 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4869
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004870 free_extent_map(em);
David Woodhouse53b381b2013-01-29 18:40:14 -05004871 check_raid56_incompat_flag(extent_root->fs_info, type);
4872
Miao Xieb2117a32011-01-05 10:07:28 +00004873 kfree(devices_info);
Yan Zheng2b820322008-11-17 21:11:30 -05004874 return 0;
Miao Xieb2117a32011-01-05 10:07:28 +00004875
Josef Bacik6df9a952013-06-27 13:22:46 -04004876error_del_extent:
Josef Bacik0f5d42b2013-01-31 10:23:04 -05004877 write_lock(&em_tree->lock);
4878 remove_extent_mapping(em_tree, em);
4879 write_unlock(&em_tree->lock);
4880
4881 /* One for our allocation */
4882 free_extent_map(em);
4883 /* One for the tree reference */
4884 free_extent_map(em);
Filipe Manana495e64f2014-12-02 18:07:30 +00004885 /* One for the pending_chunks list reference */
4886 free_extent_map(em);
Miao Xieb2117a32011-01-05 10:07:28 +00004887error:
Miao Xieb2117a32011-01-05 10:07:28 +00004888 kfree(devices_info);
4889 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05004890}
4891
Josef Bacik6df9a952013-06-27 13:22:46 -04004892int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
Yan Zheng2b820322008-11-17 21:11:30 -05004893 struct btrfs_root *extent_root,
Josef Bacik6df9a952013-06-27 13:22:46 -04004894 u64 chunk_offset, u64 chunk_size)
Yan Zheng2b820322008-11-17 21:11:30 -05004895{
Yan Zheng2b820322008-11-17 21:11:30 -05004896 struct btrfs_key key;
4897 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4898 struct btrfs_device *device;
4899 struct btrfs_chunk *chunk;
4900 struct btrfs_stripe *stripe;
Josef Bacik6df9a952013-06-27 13:22:46 -04004901 struct extent_map_tree *em_tree;
4902 struct extent_map *em;
4903 struct map_lookup *map;
4904 size_t item_size;
4905 u64 dev_offset;
4906 u64 stripe_size;
4907 int i = 0;
Chris Mason140e6392015-12-23 13:30:51 -08004908 int ret = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05004909
Josef Bacik6df9a952013-06-27 13:22:46 -04004910 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4911 read_lock(&em_tree->lock);
4912 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4913 read_unlock(&em_tree->lock);
Yan Zheng2b820322008-11-17 21:11:30 -05004914
Josef Bacik6df9a952013-06-27 13:22:46 -04004915 if (!em) {
4916 btrfs_crit(extent_root->fs_info, "unable to find logical "
4917 "%Lu len %Lu", chunk_offset, chunk_size);
4918 return -EINVAL;
4919 }
4920
4921 if (em->start != chunk_offset || em->len != chunk_size) {
4922 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
David Sterba351fd352014-05-15 16:48:20 +02004923 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
Josef Bacik6df9a952013-06-27 13:22:46 -04004924 chunk_size, em->start, em->len);
4925 free_extent_map(em);
4926 return -EINVAL;
4927 }
4928
Jeff Mahoney95617d62015-06-03 10:55:48 -04004929 map = em->map_lookup;
Josef Bacik6df9a952013-06-27 13:22:46 -04004930 item_size = btrfs_chunk_item_size(map->num_stripes);
4931 stripe_size = em->orig_block_len;
4932
4933 chunk = kzalloc(item_size, GFP_NOFS);
4934 if (!chunk) {
4935 ret = -ENOMEM;
4936 goto out;
4937 }
4938
Filipe Manana50460e32015-11-20 10:42:47 +00004939 /*
4940 * Take the device list mutex to prevent races with the final phase of
4941 * a device replace operation that replaces the device object associated
4942 * with the map's stripes, because the device object's id can change
4943 * at any time during that final phase of the device replace operation
4944 * (dev-replace.c:btrfs_dev_replace_finishing()).
4945 */
4946 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
Josef Bacik6df9a952013-06-27 13:22:46 -04004947 for (i = 0; i < map->num_stripes; i++) {
4948 device = map->stripes[i].dev;
4949 dev_offset = map->stripes[i].physical;
4950
Yan Zheng2b820322008-11-17 21:11:30 -05004951 ret = btrfs_update_device(trans, device);
Mark Fasheh3acd3952011-09-08 17:40:01 -07004952 if (ret)
Filipe Manana50460e32015-11-20 10:42:47 +00004953 break;
Josef Bacik6df9a952013-06-27 13:22:46 -04004954 ret = btrfs_alloc_dev_extent(trans, device,
4955 chunk_root->root_key.objectid,
4956 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4957 chunk_offset, dev_offset,
4958 stripe_size);
4959 if (ret)
Filipe Manana50460e32015-11-20 10:42:47 +00004960 break;
4961 }
4962 if (ret) {
4963 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4964 goto out;
Yan Zheng2b820322008-11-17 21:11:30 -05004965 }
4966
Yan Zheng2b820322008-11-17 21:11:30 -05004967 stripe = &chunk->stripe;
Josef Bacik6df9a952013-06-27 13:22:46 -04004968 for (i = 0; i < map->num_stripes; i++) {
4969 device = map->stripes[i].dev;
4970 dev_offset = map->stripes[i].physical;
Yan Zheng2b820322008-11-17 21:11:30 -05004971
4972 btrfs_set_stack_stripe_devid(stripe, device->devid);
4973 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4974 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4975 stripe++;
Yan Zheng2b820322008-11-17 21:11:30 -05004976 }
Filipe Manana50460e32015-11-20 10:42:47 +00004977 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
Yan Zheng2b820322008-11-17 21:11:30 -05004978
4979 btrfs_set_stack_chunk_length(chunk, chunk_size);
4980 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4981 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4982 btrfs_set_stack_chunk_type(chunk, map->type);
4983 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4984 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4985 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4986 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4987 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4988
4989 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4990 key.type = BTRFS_CHUNK_ITEM_KEY;
4991 key.offset = chunk_offset;
4992
4993 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
Mark Fasheh4ed1d162011-08-10 12:32:10 -07004994 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4995 /*
4996 * TODO: Cleanup of inserted chunk root in case of
4997 * failure.
4998 */
Li Zefan125ccb02011-12-08 15:07:24 +08004999 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
Yan Zheng2b820322008-11-17 21:11:30 -05005000 item_size);
Yan Zheng2b820322008-11-17 21:11:30 -05005001 }
liubo1abe9b82011-03-24 11:18:59 +00005002
Josef Bacik6df9a952013-06-27 13:22:46 -04005003out:
Yan Zheng2b820322008-11-17 21:11:30 -05005004 kfree(chunk);
Josef Bacik6df9a952013-06-27 13:22:46 -04005005 free_extent_map(em);
Mark Fasheh4ed1d162011-08-10 12:32:10 -07005006 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05005007}
5008
5009/*
5010 * Chunk allocation falls into two parts. The first part does works
5011 * that make the new allocated chunk useable, but not do any operation
5012 * that modifies the chunk tree. The second part does the works that
5013 * require modifying the chunk tree. This division is important for the
5014 * bootstrap process of adding storage to a seed btrfs.
5015 */
5016int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5017 struct btrfs_root *extent_root, u64 type)
5018{
5019 u64 chunk_offset;
Yan Zheng2b820322008-11-17 21:11:30 -05005020
Filipe Mananaa9629592015-05-18 19:11:40 +01005021 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
Josef Bacik6df9a952013-06-27 13:22:46 -04005022 chunk_offset = find_next_chunk(extent_root->fs_info);
5023 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
Yan Zheng2b820322008-11-17 21:11:30 -05005024}
5025
Chris Masond3977122009-01-05 21:25:51 -05005026static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
Yan Zheng2b820322008-11-17 21:11:30 -05005027 struct btrfs_root *root,
5028 struct btrfs_device *device)
5029{
5030 u64 chunk_offset;
5031 u64 sys_chunk_offset;
Yan Zheng2b820322008-11-17 21:11:30 -05005032 u64 alloc_profile;
Yan Zheng2b820322008-11-17 21:11:30 -05005033 struct btrfs_fs_info *fs_info = root->fs_info;
5034 struct btrfs_root *extent_root = fs_info->extent_root;
5035 int ret;
5036
Josef Bacik6df9a952013-06-27 13:22:46 -04005037 chunk_offset = find_next_chunk(fs_info);
Miao Xiede98ced2013-01-29 10:13:12 +00005038 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
Josef Bacik6df9a952013-06-27 13:22:46 -04005039 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5040 alloc_profile);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005041 if (ret)
5042 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05005043
Josef Bacik6df9a952013-06-27 13:22:46 -04005044 sys_chunk_offset = find_next_chunk(root->fs_info);
Miao Xiede98ced2013-01-29 10:13:12 +00005045 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
Josef Bacik6df9a952013-06-27 13:22:46 -04005046 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5047 alloc_profile);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005048 return ret;
Yan Zheng2b820322008-11-17 21:11:30 -05005049}
5050
Miao Xied20983b2014-07-03 18:22:13 +08005051static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5052{
5053 int max_errors;
5054
5055 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5056 BTRFS_BLOCK_GROUP_RAID10 |
5057 BTRFS_BLOCK_GROUP_RAID5 |
5058 BTRFS_BLOCK_GROUP_DUP)) {
5059 max_errors = 1;
5060 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5061 max_errors = 2;
5062 } else {
5063 max_errors = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05005064 }
5065
Miao Xied20983b2014-07-03 18:22:13 +08005066 return max_errors;
Yan Zheng2b820322008-11-17 21:11:30 -05005067}
5068
5069int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5070{
5071 struct extent_map *em;
5072 struct map_lookup *map;
5073 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5074 int readonly = 0;
Miao Xied20983b2014-07-03 18:22:13 +08005075 int miss_ndevs = 0;
Yan Zheng2b820322008-11-17 21:11:30 -05005076 int i;
5077
Chris Mason890871b2009-09-02 16:24:52 -04005078 read_lock(&map_tree->map_tree.lock);
Yan Zheng2b820322008-11-17 21:11:30 -05005079 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
Chris Mason890871b2009-09-02 16:24:52 -04005080 read_unlock(&map_tree->map_tree.lock);
Yan Zheng2b820322008-11-17 21:11:30 -05005081 if (!em)
5082 return 1;
5083
Jeff Mahoney95617d62015-06-03 10:55:48 -04005084 map = em->map_lookup;
Yan Zheng2b820322008-11-17 21:11:30 -05005085 for (i = 0; i < map->num_stripes; i++) {
Miao Xied20983b2014-07-03 18:22:13 +08005086 if (map->stripes[i].dev->missing) {
5087 miss_ndevs++;
5088 continue;
5089 }
5090
Yan Zheng2b820322008-11-17 21:11:30 -05005091 if (!map->stripes[i].dev->writeable) {
5092 readonly = 1;
Miao Xied20983b2014-07-03 18:22:13 +08005093 goto end;
Yan Zheng2b820322008-11-17 21:11:30 -05005094 }
5095 }
Miao Xied20983b2014-07-03 18:22:13 +08005096
5097 /*
5098 * If the number of missing devices is larger than max errors,
5099 * we can not write the data into that chunk successfully, so
5100 * set it readonly.
5101 */
5102 if (miss_ndevs > btrfs_chunk_max_errors(map))
5103 readonly = 1;
5104end:
Yan Zheng2b820322008-11-17 21:11:30 -05005105 free_extent_map(em);
5106 return readonly;
Chris Mason0b86a832008-03-24 15:01:56 -04005107}
5108
5109void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5110{
David Sterbaa8067e02011-04-21 00:34:43 +02005111 extent_map_tree_init(&tree->map_tree);
Chris Mason0b86a832008-03-24 15:01:56 -04005112}
5113
5114void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5115{
5116 struct extent_map *em;
5117
Chris Masond3977122009-01-05 21:25:51 -05005118 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04005119 write_lock(&tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04005120 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5121 if (em)
5122 remove_extent_mapping(&tree->map_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04005123 write_unlock(&tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04005124 if (!em)
5125 break;
Chris Mason0b86a832008-03-24 15:01:56 -04005126 /* once for us */
5127 free_extent_map(em);
5128 /* once for the tree */
5129 free_extent_map(em);
5130 }
5131}
5132
Stefan Behrens5d964052012-11-05 14:59:07 +01005133int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
Chris Masonf1885912008-04-09 16:28:12 -04005134{
Stefan Behrens5d964052012-11-05 14:59:07 +01005135 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
Chris Masonf1885912008-04-09 16:28:12 -04005136 struct extent_map *em;
5137 struct map_lookup *map;
5138 struct extent_map_tree *em_tree = &map_tree->map_tree;
5139 int ret;
5140
Chris Mason890871b2009-09-02 16:24:52 -04005141 read_lock(&em_tree->lock);
Chris Masonf1885912008-04-09 16:28:12 -04005142 em = lookup_extent_mapping(em_tree, logical, len);
Chris Mason890871b2009-09-02 16:24:52 -04005143 read_unlock(&em_tree->lock);
Chris Masonf1885912008-04-09 16:28:12 -04005144
Josef Bacikfb7669b2013-04-23 10:53:18 -04005145 /*
5146 * We could return errors for these cases, but that could get ugly and
5147 * we'd probably do the same thing which is just not do anything else
5148 * and exit, so return 1 so the callers don't try to use other copies.
5149 */
5150 if (!em) {
David Sterba351fd352014-05-15 16:48:20 +02005151 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
Josef Bacikfb7669b2013-04-23 10:53:18 -04005152 logical+len);
5153 return 1;
5154 }
5155
5156 if (em->start > logical || em->start + em->len < logical) {
David Sterbaccf39f92013-07-11 15:41:11 +02005157 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
David Sterba351fd352014-05-15 16:48:20 +02005158 "%Lu-%Lu", logical, logical+len, em->start,
Josef Bacikfb7669b2013-04-23 10:53:18 -04005159 em->start + em->len);
Liu Bo7d3d1742013-09-29 10:33:16 +08005160 free_extent_map(em);
Josef Bacikfb7669b2013-04-23 10:53:18 -04005161 return 1;
5162 }
5163
Jeff Mahoney95617d62015-06-03 10:55:48 -04005164 map = em->map_lookup;
Chris Masonf1885912008-04-09 16:28:12 -04005165 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5166 ret = map->num_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04005167 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5168 ret = map->sub_stripes;
David Woodhouse53b381b2013-01-29 18:40:14 -05005169 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5170 ret = 2;
5171 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5172 ret = 3;
Chris Masonf1885912008-04-09 16:28:12 -04005173 else
5174 ret = 1;
5175 free_extent_map(em);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005176
Liu Bo73beece2015-07-17 16:49:19 +08005177 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005178 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5179 ret++;
Liu Bo73beece2015-07-17 16:49:19 +08005180 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005181
Chris Masonf1885912008-04-09 16:28:12 -04005182 return ret;
5183}
5184
David Woodhouse53b381b2013-01-29 18:40:14 -05005185unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5186 struct btrfs_mapping_tree *map_tree,
5187 u64 logical)
5188{
5189 struct extent_map *em;
5190 struct map_lookup *map;
5191 struct extent_map_tree *em_tree = &map_tree->map_tree;
5192 unsigned long len = root->sectorsize;
5193
5194 read_lock(&em_tree->lock);
5195 em = lookup_extent_mapping(em_tree, logical, len);
5196 read_unlock(&em_tree->lock);
5197 BUG_ON(!em);
5198
5199 BUG_ON(em->start > logical || em->start + em->len < logical);
Jeff Mahoney95617d62015-06-03 10:55:48 -04005200 map = em->map_lookup;
Zhao Leiffe2d202015-01-20 15:11:44 +08005201 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
David Woodhouse53b381b2013-01-29 18:40:14 -05005202 len = map->stripe_len * nr_data_stripes(map);
David Woodhouse53b381b2013-01-29 18:40:14 -05005203 free_extent_map(em);
5204 return len;
5205}
5206
5207int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5208 u64 logical, u64 len, int mirror_num)
5209{
5210 struct extent_map *em;
5211 struct map_lookup *map;
5212 struct extent_map_tree *em_tree = &map_tree->map_tree;
5213 int ret = 0;
5214
5215 read_lock(&em_tree->lock);
5216 em = lookup_extent_mapping(em_tree, logical, len);
5217 read_unlock(&em_tree->lock);
5218 BUG_ON(!em);
5219
5220 BUG_ON(em->start > logical || em->start + em->len < logical);
Jeff Mahoney95617d62015-06-03 10:55:48 -04005221 map = em->map_lookup;
Zhao Leiffe2d202015-01-20 15:11:44 +08005222 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
David Woodhouse53b381b2013-01-29 18:40:14 -05005223 ret = 1;
5224 free_extent_map(em);
5225 return ret;
5226}
5227
Stefan Behrens30d98612012-11-06 14:52:18 +01005228static int find_live_mirror(struct btrfs_fs_info *fs_info,
5229 struct map_lookup *map, int first, int num,
5230 int optimal, int dev_replace_is_ongoing)
Chris Masondfe25022008-05-13 13:46:40 -04005231{
5232 int i;
Stefan Behrens30d98612012-11-06 14:52:18 +01005233 int tolerance;
5234 struct btrfs_device *srcdev;
5235
5236 if (dev_replace_is_ongoing &&
5237 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5238 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5239 srcdev = fs_info->dev_replace.srcdev;
5240 else
5241 srcdev = NULL;
5242
5243 /*
5244 * try to avoid the drive that is the source drive for a
5245 * dev-replace procedure, only choose it if no other non-missing
5246 * mirror is available
5247 */
5248 for (tolerance = 0; tolerance < 2; tolerance++) {
5249 if (map->stripes[optimal].dev->bdev &&
5250 (tolerance || map->stripes[optimal].dev != srcdev))
5251 return optimal;
5252 for (i = first; i < first + num; i++) {
5253 if (map->stripes[i].dev->bdev &&
5254 (tolerance || map->stripes[i].dev != srcdev))
5255 return i;
5256 }
Chris Masondfe25022008-05-13 13:46:40 -04005257 }
Stefan Behrens30d98612012-11-06 14:52:18 +01005258
Chris Masondfe25022008-05-13 13:46:40 -04005259 /* we couldn't find one that doesn't fail. Just return something
5260 * and the io error handling code will clean up eventually
5261 */
5262 return optimal;
5263}
5264
David Woodhouse53b381b2013-01-29 18:40:14 -05005265static inline int parity_smaller(u64 a, u64 b)
5266{
5267 return a > b;
5268}
5269
5270/* Bubble-sort the stripe set to put the parity/syndrome stripes last */
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005271static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
David Woodhouse53b381b2013-01-29 18:40:14 -05005272{
5273 struct btrfs_bio_stripe s;
5274 int i;
5275 u64 l;
5276 int again = 1;
5277
5278 while (again) {
5279 again = 0;
Zhao Leicc7539e2015-01-20 15:11:32 +08005280 for (i = 0; i < num_stripes - 1; i++) {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005281 if (parity_smaller(bbio->raid_map[i],
5282 bbio->raid_map[i+1])) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005283 s = bbio->stripes[i];
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005284 l = bbio->raid_map[i];
David Woodhouse53b381b2013-01-29 18:40:14 -05005285 bbio->stripes[i] = bbio->stripes[i+1];
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005286 bbio->raid_map[i] = bbio->raid_map[i+1];
David Woodhouse53b381b2013-01-29 18:40:14 -05005287 bbio->stripes[i+1] = s;
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005288 bbio->raid_map[i+1] = l;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005289
David Woodhouse53b381b2013-01-29 18:40:14 -05005290 again = 1;
5291 }
5292 }
5293 }
5294}
5295
Zhao Lei6e9606d2015-01-20 15:11:34 +08005296static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5297{
5298 struct btrfs_bio *bbio = kzalloc(
Chris Masone57cf212015-02-19 17:51:39 -08005299 /* the size of the btrfs_bio */
Zhao Lei6e9606d2015-01-20 15:11:34 +08005300 sizeof(struct btrfs_bio) +
Chris Masone57cf212015-02-19 17:51:39 -08005301 /* plus the variable array for the stripes */
Zhao Lei6e9606d2015-01-20 15:11:34 +08005302 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
Chris Masone57cf212015-02-19 17:51:39 -08005303 /* plus the variable array for the tgt dev */
Zhao Lei6e9606d2015-01-20 15:11:34 +08005304 sizeof(int) * (real_stripes) +
Chris Masone57cf212015-02-19 17:51:39 -08005305 /*
5306 * plus the raid_map, which includes both the tgt dev
5307 * and the stripes
5308 */
5309 sizeof(u64) * (total_stripes),
Michal Hocko277fb5f2015-08-19 14:17:41 +02005310 GFP_NOFS|__GFP_NOFAIL);
Zhao Lei6e9606d2015-01-20 15:11:34 +08005311
5312 atomic_set(&bbio->error, 0);
5313 atomic_set(&bbio->refs, 1);
5314
5315 return bbio;
5316}
5317
5318void btrfs_get_bbio(struct btrfs_bio *bbio)
5319{
5320 WARN_ON(!atomic_read(&bbio->refs));
5321 atomic_inc(&bbio->refs);
5322}
5323
5324void btrfs_put_bbio(struct btrfs_bio *bbio)
5325{
5326 if (!bbio)
5327 return;
5328 if (atomic_dec_and_test(&bbio->refs))
5329 kfree(bbio);
5330}
5331
Mike Christieb3d3fa52016-06-05 14:31:53 -05005332static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
Chris Masonf2d8d742008-04-21 10:03:05 -04005333 u64 logical, u64 *length,
Jan Schmidta1d3c472011-08-04 17:15:33 +02005334 struct btrfs_bio **bbio_ret,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005335 int mirror_num, int need_raid_map)
Chris Mason0b86a832008-03-24 15:01:56 -04005336{
5337 struct extent_map *em;
5338 struct map_lookup *map;
Stefan Behrens3ec706c2012-11-05 15:46:42 +01005339 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
Chris Mason0b86a832008-03-24 15:01:56 -04005340 struct extent_map_tree *em_tree = &map_tree->map_tree;
5341 u64 offset;
Chris Mason593060d2008-03-25 16:50:33 -04005342 u64 stripe_offset;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005343 u64 stripe_end_offset;
Chris Mason593060d2008-03-25 16:50:33 -04005344 u64 stripe_nr;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005345 u64 stripe_nr_orig;
5346 u64 stripe_nr_end;
David Woodhouse53b381b2013-01-29 18:40:14 -05005347 u64 stripe_len;
David Sterba9d644a62015-02-20 18:42:11 +01005348 u32 stripe_index;
Chris Masoncea9e442008-04-09 16:28:12 -04005349 int i;
Li Zefande11cc12011-12-01 12:55:47 +08005350 int ret = 0;
Chris Masonf2d8d742008-04-21 10:03:05 -04005351 int num_stripes;
Chris Masona236aed2008-04-29 09:38:00 -04005352 int max_errors = 0;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005353 int tgtdev_indexes = 0;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005354 struct btrfs_bio *bbio = NULL;
Stefan Behrens472262f2012-11-06 14:43:46 +01005355 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5356 int dev_replace_is_ongoing = 0;
5357 int num_alloc_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005358 int patch_the_first_stripe_for_dev_replace = 0;
5359 u64 physical_to_patch_in_first_stripe = 0;
David Woodhouse53b381b2013-01-29 18:40:14 -05005360 u64 raid56_full_stripe_start = (u64)-1;
Chris Mason0b86a832008-03-24 15:01:56 -04005361
Chris Mason890871b2009-09-02 16:24:52 -04005362 read_lock(&em_tree->lock);
Chris Mason0b86a832008-03-24 15:01:56 -04005363 em = lookup_extent_mapping(em_tree, logical, *length);
Chris Mason890871b2009-09-02 16:24:52 -04005364 read_unlock(&em_tree->lock);
Chris Masonf2d8d742008-04-21 10:03:05 -04005365
Chris Mason3b951512008-04-17 11:29:12 -04005366 if (!em) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005367 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02005368 logical, *length);
Josef Bacik9bb91872013-04-19 23:45:33 -04005369 return -EINVAL;
Chris Mason3b951512008-04-17 11:29:12 -04005370 }
Chris Mason0b86a832008-03-24 15:01:56 -04005371
Josef Bacik9bb91872013-04-19 23:45:33 -04005372 if (em->start > logical || em->start + em->len < logical) {
5373 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
David Sterba351fd352014-05-15 16:48:20 +02005374 "found %Lu-%Lu", logical, em->start,
Josef Bacik9bb91872013-04-19 23:45:33 -04005375 em->start + em->len);
Liu Bo7d3d1742013-09-29 10:33:16 +08005376 free_extent_map(em);
Josef Bacik9bb91872013-04-19 23:45:33 -04005377 return -EINVAL;
5378 }
5379
Jeff Mahoney95617d62015-06-03 10:55:48 -04005380 map = em->map_lookup;
Chris Mason0b86a832008-03-24 15:01:56 -04005381 offset = logical - em->start;
Chris Mason593060d2008-03-25 16:50:33 -04005382
David Woodhouse53b381b2013-01-29 18:40:14 -05005383 stripe_len = map->stripe_len;
Chris Mason593060d2008-03-25 16:50:33 -04005384 stripe_nr = offset;
5385 /*
5386 * stripe_nr counts the total number of stripes we have to stride
5387 * to get to this block
5388 */
David Sterba47c57132015-02-20 18:43:47 +01005389 stripe_nr = div64_u64(stripe_nr, stripe_len);
Chris Mason593060d2008-03-25 16:50:33 -04005390
David Woodhouse53b381b2013-01-29 18:40:14 -05005391 stripe_offset = stripe_nr * stripe_len;
Josef Bacike042d1e2016-04-12 12:54:40 -04005392 if (offset < stripe_offset) {
5393 btrfs_crit(fs_info, "stripe math has gone wrong, "
5394 "stripe_offset=%llu, offset=%llu, start=%llu, "
5395 "logical=%llu, stripe_len=%llu",
5396 stripe_offset, offset, em->start, logical,
5397 stripe_len);
5398 free_extent_map(em);
5399 return -EINVAL;
5400 }
Chris Mason593060d2008-03-25 16:50:33 -04005401
5402 /* stripe_offset is the offset of this block in its stripe*/
5403 stripe_offset = offset - stripe_offset;
5404
David Woodhouse53b381b2013-01-29 18:40:14 -05005405 /* if we're here for raid56, we need to know the stripe aligned start */
Zhao Leiffe2d202015-01-20 15:11:44 +08005406 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005407 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5408 raid56_full_stripe_start = offset;
5409
5410 /* allow a write of a full stripe, but make sure we don't
5411 * allow straddling of stripes
5412 */
David Sterba47c57132015-02-20 18:43:47 +01005413 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5414 full_stripe_len);
David Woodhouse53b381b2013-01-29 18:40:14 -05005415 raid56_full_stripe_start *= full_stripe_len;
5416 }
5417
Mike Christieb3d3fa52016-06-05 14:31:53 -05005418 if (op == REQ_OP_DISCARD) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005419 /* we don't discard raid56 yet */
Zhao Leiffe2d202015-01-20 15:11:44 +08005420 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005421 ret = -EOPNOTSUPP;
5422 goto out;
5423 }
Li Dongyangfce3bb92011-03-24 10:24:26 +00005424 *length = min_t(u64, em->len - offset, *length);
David Woodhouse53b381b2013-01-29 18:40:14 -05005425 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5426 u64 max_len;
5427 /* For writes to RAID[56], allow a full stripeset across all disks.
5428 For other RAID types and for RAID[56] reads, just allow a single
5429 stripe (on a single disk). */
Zhao Leiffe2d202015-01-20 15:11:44 +08005430 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
Mike Christieb3d3fa52016-06-05 14:31:53 -05005431 (op == REQ_OP_WRITE)) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005432 max_len = stripe_len * nr_data_stripes(map) -
5433 (offset - raid56_full_stripe_start);
5434 } else {
5435 /* we limit the length of each bio to what fits in a stripe */
5436 max_len = stripe_len - stripe_offset;
5437 }
5438 *length = min_t(u64, em->len - offset, max_len);
Chris Masoncea9e442008-04-09 16:28:12 -04005439 } else {
5440 *length = em->len - offset;
5441 }
Chris Masonf2d8d742008-04-21 10:03:05 -04005442
David Woodhouse53b381b2013-01-29 18:40:14 -05005443 /* This is for when we're called from btrfs_merge_bio_hook() and all
5444 it cares about is the length */
Jan Schmidta1d3c472011-08-04 17:15:33 +02005445 if (!bbio_ret)
Chris Masoncea9e442008-04-09 16:28:12 -04005446 goto out;
5447
Liu Bo73beece2015-07-17 16:49:19 +08005448 btrfs_dev_replace_lock(dev_replace, 0);
Stefan Behrens472262f2012-11-06 14:43:46 +01005449 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5450 if (!dev_replace_is_ongoing)
Liu Bo73beece2015-07-17 16:49:19 +08005451 btrfs_dev_replace_unlock(dev_replace, 0);
5452 else
5453 btrfs_dev_replace_set_lock_blocking(dev_replace);
Stefan Behrens472262f2012-11-06 14:43:46 +01005454
Stefan Behrensad6d6202012-11-06 15:06:47 +01005455 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
Mike Christieb3d3fa52016-06-05 14:31:53 -05005456 op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5457 op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
Stefan Behrensad6d6202012-11-06 15:06:47 +01005458 /*
5459 * in dev-replace case, for repair case (that's the only
5460 * case where the mirror is selected explicitly when
5461 * calling btrfs_map_block), blocks left of the left cursor
5462 * can also be read from the target drive.
5463 * For REQ_GET_READ_MIRRORS, the target drive is added as
5464 * the last one to the array of stripes. For READ, it also
5465 * needs to be supported using the same mirror number.
5466 * If the requested block is not left of the left cursor,
5467 * EIO is returned. This can happen because btrfs_num_copies()
5468 * returns one more in the dev-replace case.
5469 */
5470 u64 tmp_length = *length;
5471 struct btrfs_bio *tmp_bbio = NULL;
5472 int tmp_num_stripes;
5473 u64 srcdev_devid = dev_replace->srcdev->devid;
5474 int index_srcdev = 0;
5475 int found = 0;
5476 u64 physical_of_found = 0;
5477
5478 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005479 logical, &tmp_length, &tmp_bbio, 0, 0);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005480 if (ret) {
5481 WARN_ON(tmp_bbio != NULL);
5482 goto out;
5483 }
5484
5485 tmp_num_stripes = tmp_bbio->num_stripes;
5486 if (mirror_num > tmp_num_stripes) {
5487 /*
5488 * REQ_GET_READ_MIRRORS does not contain this
5489 * mirror, that means that the requested area
5490 * is not left of the left cursor
5491 */
5492 ret = -EIO;
Zhao Lei6e9606d2015-01-20 15:11:34 +08005493 btrfs_put_bbio(tmp_bbio);
Stefan Behrensad6d6202012-11-06 15:06:47 +01005494 goto out;
5495 }
5496
5497 /*
5498 * process the rest of the function using the mirror_num
5499 * of the source drive. Therefore look it up first.
5500 * At the end, patch the device pointer to the one of the
5501 * target drive.
5502 */
5503 for (i = 0; i < tmp_num_stripes; i++) {
Zhao Lei94a97df2015-12-09 21:03:49 +08005504 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5505 continue;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005506
Zhao Lei94a97df2015-12-09 21:03:49 +08005507 /*
5508 * In case of DUP, in order to keep it simple, only add
5509 * the mirror with the lowest physical address
5510 */
5511 if (found &&
5512 physical_of_found <= tmp_bbio->stripes[i].physical)
5513 continue;
5514
5515 index_srcdev = i;
5516 found = 1;
5517 physical_of_found = tmp_bbio->stripes[i].physical;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005518 }
5519
Zhao Lei6e9606d2015-01-20 15:11:34 +08005520 btrfs_put_bbio(tmp_bbio);
Zhao Lei94a97df2015-12-09 21:03:49 +08005521
5522 if (!found) {
5523 WARN_ON(1);
5524 ret = -EIO;
5525 goto out;
5526 }
5527
5528 mirror_num = index_srcdev + 1;
5529 patch_the_first_stripe_for_dev_replace = 1;
5530 physical_to_patch_in_first_stripe = physical_of_found;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005531 } else if (mirror_num > map->num_stripes) {
5532 mirror_num = 0;
5533 }
5534
Chris Masonf2d8d742008-04-21 10:03:05 -04005535 num_stripes = 1;
Chris Masoncea9e442008-04-09 16:28:12 -04005536 stripe_index = 0;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005537 stripe_nr_orig = stripe_nr;
Qu Wenruofda28322013-02-26 08:10:22 +00005538 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
David Sterbab8b93ad2015-01-16 17:26:13 +01005539 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
Li Dongyangfce3bb92011-03-24 10:24:26 +00005540 stripe_end_offset = stripe_nr_end * map->stripe_len -
5541 (offset + *length);
David Woodhouse53b381b2013-01-29 18:40:14 -05005542
Li Dongyangfce3bb92011-03-24 10:24:26 +00005543 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
Mike Christieb3d3fa52016-06-05 14:31:53 -05005544 if (op == REQ_OP_DISCARD)
Li Dongyangfce3bb92011-03-24 10:24:26 +00005545 num_stripes = min_t(u64, map->num_stripes,
5546 stripe_nr_end - stripe_nr_orig);
David Sterba47c57132015-02-20 18:43:47 +01005547 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5548 &stripe_index);
Mike Christieb3d3fa52016-06-05 14:31:53 -05005549 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5550 op != REQ_GET_READ_MIRRORS)
Miao Xie28e1cc72014-09-12 18:44:02 +08005551 mirror_num = 1;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005552 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
Mike Christieb3d3fa52016-06-05 14:31:53 -05005553 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5554 op == REQ_GET_READ_MIRRORS)
Chris Masonf2d8d742008-04-21 10:03:05 -04005555 num_stripes = map->num_stripes;
Chris Mason2fff7342008-04-29 14:12:09 -04005556 else if (mirror_num)
Chris Masonf1885912008-04-09 16:28:12 -04005557 stripe_index = mirror_num - 1;
Chris Masondfe25022008-05-13 13:46:40 -04005558 else {
Stefan Behrens30d98612012-11-06 14:52:18 +01005559 stripe_index = find_live_mirror(fs_info, map, 0,
Chris Masondfe25022008-05-13 13:46:40 -04005560 map->num_stripes,
Stefan Behrens30d98612012-11-06 14:52:18 +01005561 current->pid % map->num_stripes,
5562 dev_replace_is_ongoing);
Jan Schmidta1d3c472011-08-04 17:15:33 +02005563 mirror_num = stripe_index + 1;
Chris Masondfe25022008-05-13 13:46:40 -04005564 }
Chris Mason2fff7342008-04-29 14:12:09 -04005565
Chris Mason611f0e02008-04-03 16:29:03 -04005566 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
Vincent Stehlédf5c82a2016-07-15 17:03:21 +02005567 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
Mike Christieb3d3fa52016-06-05 14:31:53 -05005568 op == REQ_GET_READ_MIRRORS) {
Chris Masonf2d8d742008-04-21 10:03:05 -04005569 num_stripes = map->num_stripes;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005570 } else if (mirror_num) {
Chris Masonf1885912008-04-09 16:28:12 -04005571 stripe_index = mirror_num - 1;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005572 } else {
5573 mirror_num = 1;
5574 }
Chris Mason2fff7342008-04-29 14:12:09 -04005575
Chris Mason321aecc2008-04-16 10:49:51 -04005576 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
David Sterba9d644a62015-02-20 18:42:11 +01005577 u32 factor = map->num_stripes / map->sub_stripes;
Chris Mason321aecc2008-04-16 10:49:51 -04005578
David Sterba47c57132015-02-20 18:43:47 +01005579 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
Chris Mason321aecc2008-04-16 10:49:51 -04005580 stripe_index *= map->sub_stripes;
5581
Mike Christieb3d3fa52016-06-05 14:31:53 -05005582 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
Chris Masonf2d8d742008-04-21 10:03:05 -04005583 num_stripes = map->sub_stripes;
Mike Christieb3d3fa52016-06-05 14:31:53 -05005584 else if (op == REQ_OP_DISCARD)
Li Dongyangfce3bb92011-03-24 10:24:26 +00005585 num_stripes = min_t(u64, map->sub_stripes *
5586 (stripe_nr_end - stripe_nr_orig),
5587 map->num_stripes);
Chris Mason321aecc2008-04-16 10:49:51 -04005588 else if (mirror_num)
5589 stripe_index += mirror_num - 1;
Chris Masondfe25022008-05-13 13:46:40 -04005590 else {
Jan Schmidt3e743172012-04-27 12:41:45 -04005591 int old_stripe_index = stripe_index;
Stefan Behrens30d98612012-11-06 14:52:18 +01005592 stripe_index = find_live_mirror(fs_info, map,
5593 stripe_index,
Chris Masondfe25022008-05-13 13:46:40 -04005594 map->sub_stripes, stripe_index +
Stefan Behrens30d98612012-11-06 14:52:18 +01005595 current->pid % map->sub_stripes,
5596 dev_replace_is_ongoing);
Jan Schmidt3e743172012-04-27 12:41:45 -04005597 mirror_num = stripe_index - old_stripe_index + 1;
Chris Masondfe25022008-05-13 13:46:40 -04005598 }
David Woodhouse53b381b2013-01-29 18:40:14 -05005599
Zhao Leiffe2d202015-01-20 15:11:44 +08005600 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005601 if (need_raid_map &&
Mike Christieb3d3fa52016-06-05 14:31:53 -05005602 (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
Miao Xieaf8e2d12014-10-23 14:42:50 +08005603 mirror_num > 1)) {
David Woodhouse53b381b2013-01-29 18:40:14 -05005604 /* push stripe_nr back to the start of the full stripe */
David Sterbab8b93ad2015-01-16 17:26:13 +01005605 stripe_nr = div_u64(raid56_full_stripe_start,
5606 stripe_len * nr_data_stripes(map));
David Woodhouse53b381b2013-01-29 18:40:14 -05005607
5608 /* RAID[56] write or recovery. Return all stripes */
5609 num_stripes = map->num_stripes;
5610 max_errors = nr_parity_stripes(map);
5611
David Woodhouse53b381b2013-01-29 18:40:14 -05005612 *length = map->stripe_len;
5613 stripe_index = 0;
5614 stripe_offset = 0;
5615 } else {
5616 /*
5617 * Mirror #0 or #1 means the original data block.
5618 * Mirror #2 is RAID5 parity block.
5619 * Mirror #3 is RAID6 Q block.
5620 */
David Sterba47c57132015-02-20 18:43:47 +01005621 stripe_nr = div_u64_rem(stripe_nr,
5622 nr_data_stripes(map), &stripe_index);
David Woodhouse53b381b2013-01-29 18:40:14 -05005623 if (mirror_num > 1)
5624 stripe_index = nr_data_stripes(map) +
5625 mirror_num - 2;
5626
5627 /* We distribute the parity blocks across stripes */
David Sterba47c57132015-02-20 18:43:47 +01005628 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5629 &stripe_index);
Mike Christieb3d3fa52016-06-05 14:31:53 -05005630 if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5631 op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
Miao Xie28e1cc72014-09-12 18:44:02 +08005632 mirror_num = 1;
David Woodhouse53b381b2013-01-29 18:40:14 -05005633 }
Chris Mason8790d502008-04-03 16:29:03 -04005634 } else {
5635 /*
David Sterba47c57132015-02-20 18:43:47 +01005636 * after this, stripe_nr is the number of stripes on this
5637 * device we have to walk to find the data, and stripe_index is
5638 * the number of our device in the stripe array
Chris Mason8790d502008-04-03 16:29:03 -04005639 */
David Sterba47c57132015-02-20 18:43:47 +01005640 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5641 &stripe_index);
Jan Schmidta1d3c472011-08-04 17:15:33 +02005642 mirror_num = stripe_index + 1;
Chris Mason8790d502008-04-03 16:29:03 -04005643 }
Josef Bacike042d1e2016-04-12 12:54:40 -04005644 if (stripe_index >= map->num_stripes) {
5645 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5646 "got stripe_index=%u, num_stripes=%u",
5647 stripe_index, map->num_stripes);
5648 ret = -EINVAL;
5649 goto out;
5650 }
Chris Mason593060d2008-03-25 16:50:33 -04005651
Stefan Behrens472262f2012-11-06 14:43:46 +01005652 num_alloc_stripes = num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005653 if (dev_replace_is_ongoing) {
Mike Christieb3d3fa52016-06-05 14:31:53 -05005654 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
Stefan Behrensad6d6202012-11-06 15:06:47 +01005655 num_alloc_stripes <<= 1;
Mike Christieb3d3fa52016-06-05 14:31:53 -05005656 if (op == REQ_GET_READ_MIRRORS)
Stefan Behrensad6d6202012-11-06 15:06:47 +01005657 num_alloc_stripes++;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005658 tgtdev_indexes = num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005659 }
Miao Xie2c8cdd62014-11-14 16:06:25 +08005660
Zhao Lei6e9606d2015-01-20 15:11:34 +08005661 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
Li Zefande11cc12011-12-01 12:55:47 +08005662 if (!bbio) {
5663 ret = -ENOMEM;
5664 goto out;
5665 }
Miao Xie2c8cdd62014-11-14 16:06:25 +08005666 if (dev_replace_is_ongoing)
5667 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
Li Zefande11cc12011-12-01 12:55:47 +08005668
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005669 /* build raid_map */
Zhao Leiffe2d202015-01-20 15:11:44 +08005670 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
Mike Christieb3d3fa52016-06-05 14:31:53 -05005671 need_raid_map &&
5672 ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005673 mirror_num > 1)) {
5674 u64 tmp;
David Sterba9d644a62015-02-20 18:42:11 +01005675 unsigned rot;
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005676
5677 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5678 sizeof(struct btrfs_bio_stripe) *
5679 num_alloc_stripes +
5680 sizeof(int) * tgtdev_indexes);
5681
5682 /* Work out the disk rotation on this stripe-set */
David Sterba47c57132015-02-20 18:43:47 +01005683 div_u64_rem(stripe_nr, num_stripes, &rot);
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005684
5685 /* Fill in the logical address of each stripe */
5686 tmp = stripe_nr * nr_data_stripes(map);
5687 for (i = 0; i < nr_data_stripes(map); i++)
5688 bbio->raid_map[(i+rot) % num_stripes] =
5689 em->start + (tmp + i) * map->stripe_len;
5690
5691 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5692 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5693 bbio->raid_map[(i+rot+1) % num_stripes] =
5694 RAID6_Q_STRIPE;
5695 }
5696
Mike Christieb3d3fa52016-06-05 14:31:53 -05005697 if (op == REQ_OP_DISCARD) {
David Sterba9d644a62015-02-20 18:42:11 +01005698 u32 factor = 0;
5699 u32 sub_stripes = 0;
Li Zefanec9ef7a2011-12-01 14:06:42 +08005700 u64 stripes_per_dev = 0;
5701 u32 remaining_stripes = 0;
Liu Bob89203f2012-04-12 16:03:56 -04005702 u32 last_stripe = 0;
Li Zefanec9ef7a2011-12-01 14:06:42 +08005703
5704 if (map->type &
5705 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5706 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5707 sub_stripes = 1;
5708 else
5709 sub_stripes = map->sub_stripes;
5710
5711 factor = map->num_stripes / sub_stripes;
5712 stripes_per_dev = div_u64_rem(stripe_nr_end -
5713 stripe_nr_orig,
5714 factor,
5715 &remaining_stripes);
Liu Bob89203f2012-04-12 16:03:56 -04005716 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5717 last_stripe *= sub_stripes;
Li Zefanec9ef7a2011-12-01 14:06:42 +08005718 }
5719
Li Dongyangfce3bb92011-03-24 10:24:26 +00005720 for (i = 0; i < num_stripes; i++) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02005721 bbio->stripes[i].physical =
Chris Masonf2d8d742008-04-21 10:03:05 -04005722 map->stripes[stripe_index].physical +
5723 stripe_offset + stripe_nr * map->stripe_len;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005724 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005725
Li Zefanec9ef7a2011-12-01 14:06:42 +08005726 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5727 BTRFS_BLOCK_GROUP_RAID10)) {
5728 bbio->stripes[i].length = stripes_per_dev *
5729 map->stripe_len;
Liu Bob89203f2012-04-12 16:03:56 -04005730
Li Zefanec9ef7a2011-12-01 14:06:42 +08005731 if (i / sub_stripes < remaining_stripes)
5732 bbio->stripes[i].length +=
5733 map->stripe_len;
Liu Bob89203f2012-04-12 16:03:56 -04005734
5735 /*
5736 * Special for the first stripe and
5737 * the last stripe:
5738 *
5739 * |-------|...|-------|
5740 * |----------|
5741 * off end_off
5742 */
Li Zefanec9ef7a2011-12-01 14:06:42 +08005743 if (i < sub_stripes)
Jan Schmidta1d3c472011-08-04 17:15:33 +02005744 bbio->stripes[i].length -=
Li Dongyangfce3bb92011-03-24 10:24:26 +00005745 stripe_offset;
Liu Bob89203f2012-04-12 16:03:56 -04005746
5747 if (stripe_index >= last_stripe &&
5748 stripe_index <= (last_stripe +
5749 sub_stripes - 1))
Li Zefanec9ef7a2011-12-01 14:06:42 +08005750 bbio->stripes[i].length -=
5751 stripe_end_offset;
Liu Bob89203f2012-04-12 16:03:56 -04005752
Li Zefanec9ef7a2011-12-01 14:06:42 +08005753 if (i == sub_stripes - 1)
Li Dongyangfce3bb92011-03-24 10:24:26 +00005754 stripe_offset = 0;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005755 } else
Jan Schmidta1d3c472011-08-04 17:15:33 +02005756 bbio->stripes[i].length = *length;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005757
5758 stripe_index++;
5759 if (stripe_index == map->num_stripes) {
5760 /* This could only happen for RAID0/10 */
5761 stripe_index = 0;
5762 stripe_nr++;
5763 }
Chris Masonf2d8d742008-04-21 10:03:05 -04005764 }
Li Dongyangfce3bb92011-03-24 10:24:26 +00005765 } else {
5766 for (i = 0; i < num_stripes; i++) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02005767 bbio->stripes[i].physical =
Linus Torvalds212a17a2011-03-28 15:31:05 -07005768 map->stripes[stripe_index].physical +
5769 stripe_offset +
5770 stripe_nr * map->stripe_len;
Jan Schmidta1d3c472011-08-04 17:15:33 +02005771 bbio->stripes[i].dev =
Linus Torvalds212a17a2011-03-28 15:31:05 -07005772 map->stripes[stripe_index].dev;
Li Dongyangfce3bb92011-03-24 10:24:26 +00005773 stripe_index++;
5774 }
Chris Mason593060d2008-03-25 16:50:33 -04005775 }
Li Zefande11cc12011-12-01 12:55:47 +08005776
Mike Christieb3d3fa52016-06-05 14:31:53 -05005777 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
Miao Xied20983b2014-07-03 18:22:13 +08005778 max_errors = btrfs_chunk_max_errors(map);
Li Zefande11cc12011-12-01 12:55:47 +08005779
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005780 if (bbio->raid_map)
5781 sort_parity_stripes(bbio, num_stripes);
Zhao Leicc7539e2015-01-20 15:11:32 +08005782
Miao Xie2c8cdd62014-11-14 16:06:25 +08005783 tgtdev_indexes = 0;
Mike Christieb3d3fa52016-06-05 14:31:53 -05005784 if (dev_replace_is_ongoing &&
5785 (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
Stefan Behrens472262f2012-11-06 14:43:46 +01005786 dev_replace->tgtdev != NULL) {
5787 int index_where_to_add;
5788 u64 srcdev_devid = dev_replace->srcdev->devid;
5789
5790 /*
5791 * duplicate the write operations while the dev replace
5792 * procedure is running. Since the copying of the old disk
5793 * to the new disk takes place at run time while the
5794 * filesystem is mounted writable, the regular write
5795 * operations to the old disk have to be duplicated to go
5796 * to the new disk as well.
5797 * Note that device->missing is handled by the caller, and
5798 * that the write to the old disk is already set up in the
5799 * stripes array.
5800 */
5801 index_where_to_add = num_stripes;
5802 for (i = 0; i < num_stripes; i++) {
5803 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5804 /* write to new disk, too */
5805 struct btrfs_bio_stripe *new =
5806 bbio->stripes + index_where_to_add;
5807 struct btrfs_bio_stripe *old =
5808 bbio->stripes + i;
5809
5810 new->physical = old->physical;
5811 new->length = old->length;
5812 new->dev = dev_replace->tgtdev;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005813 bbio->tgtdev_map[i] = index_where_to_add;
Stefan Behrens472262f2012-11-06 14:43:46 +01005814 index_where_to_add++;
5815 max_errors++;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005816 tgtdev_indexes++;
Stefan Behrens472262f2012-11-06 14:43:46 +01005817 }
5818 }
5819 num_stripes = index_where_to_add;
Mike Christieb3d3fa52016-06-05 14:31:53 -05005820 } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
Stefan Behrensad6d6202012-11-06 15:06:47 +01005821 dev_replace->tgtdev != NULL) {
5822 u64 srcdev_devid = dev_replace->srcdev->devid;
5823 int index_srcdev = 0;
5824 int found = 0;
5825 u64 physical_of_found = 0;
5826
5827 /*
5828 * During the dev-replace procedure, the target drive can
5829 * also be used to read data in case it is needed to repair
5830 * a corrupt block elsewhere. This is possible if the
5831 * requested area is left of the left cursor. In this area,
5832 * the target drive is a full copy of the source drive.
5833 */
5834 for (i = 0; i < num_stripes; i++) {
5835 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5836 /*
5837 * In case of DUP, in order to keep it
5838 * simple, only add the mirror with the
5839 * lowest physical address
5840 */
5841 if (found &&
5842 physical_of_found <=
5843 bbio->stripes[i].physical)
5844 continue;
5845 index_srcdev = i;
5846 found = 1;
5847 physical_of_found = bbio->stripes[i].physical;
5848 }
5849 }
5850 if (found) {
Filipe Manana22ab04e2016-05-18 20:29:44 +01005851 struct btrfs_bio_stripe *tgtdev_stripe =
5852 bbio->stripes + num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005853
Filipe Manana22ab04e2016-05-18 20:29:44 +01005854 tgtdev_stripe->physical = physical_of_found;
5855 tgtdev_stripe->length =
5856 bbio->stripes[index_srcdev].length;
5857 tgtdev_stripe->dev = dev_replace->tgtdev;
5858 bbio->tgtdev_map[index_srcdev] = num_stripes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005859
Filipe Manana22ab04e2016-05-18 20:29:44 +01005860 tgtdev_indexes++;
5861 num_stripes++;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005862 }
Stefan Behrens472262f2012-11-06 14:43:46 +01005863 }
5864
Li Zefande11cc12011-12-01 12:55:47 +08005865 *bbio_ret = bbio;
Zhao Lei10f11902015-01-20 15:11:43 +08005866 bbio->map_type = map->type;
Li Zefande11cc12011-12-01 12:55:47 +08005867 bbio->num_stripes = num_stripes;
5868 bbio->max_errors = max_errors;
5869 bbio->mirror_num = mirror_num;
Miao Xie2c8cdd62014-11-14 16:06:25 +08005870 bbio->num_tgtdevs = tgtdev_indexes;
Stefan Behrensad6d6202012-11-06 15:06:47 +01005871
5872 /*
5873 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5874 * mirror_num == num_stripes + 1 && dev_replace target drive is
5875 * available as a mirror
5876 */
5877 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5878 WARN_ON(num_stripes > 1);
5879 bbio->stripes[0].dev = dev_replace->tgtdev;
5880 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5881 bbio->mirror_num = map->num_stripes + 1;
5882 }
Chris Masoncea9e442008-04-09 16:28:12 -04005883out:
Liu Bo73beece2015-07-17 16:49:19 +08005884 if (dev_replace_is_ongoing) {
5885 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5886 btrfs_dev_replace_unlock(dev_replace, 0);
5887 }
Chris Mason0b86a832008-03-24 15:01:56 -04005888 free_extent_map(em);
Li Zefande11cc12011-12-01 12:55:47 +08005889 return ret;
Chris Mason0b86a832008-03-24 15:01:56 -04005890}
5891
Mike Christieb3d3fa52016-06-05 14:31:53 -05005892int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
Chris Masonf2d8d742008-04-21 10:03:05 -04005893 u64 logical, u64 *length,
Jan Schmidta1d3c472011-08-04 17:15:33 +02005894 struct btrfs_bio **bbio_ret, int mirror_num)
Chris Masonf2d8d742008-04-21 10:03:05 -04005895{
Mike Christieb3d3fa52016-06-05 14:31:53 -05005896 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005897 mirror_num, 0);
Chris Masonf2d8d742008-04-21 10:03:05 -04005898}
5899
Miao Xieaf8e2d12014-10-23 14:42:50 +08005900/* For Scrub/replace */
Mike Christieb3d3fa52016-06-05 14:31:53 -05005901int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
Miao Xieaf8e2d12014-10-23 14:42:50 +08005902 u64 logical, u64 *length,
5903 struct btrfs_bio **bbio_ret, int mirror_num,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005904 int need_raid_map)
Miao Xieaf8e2d12014-10-23 14:42:50 +08005905{
Mike Christieb3d3fa52016-06-05 14:31:53 -05005906 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
Zhao Lei8e5cfb52015-01-20 15:11:33 +08005907 mirror_num, need_raid_map);
Miao Xieaf8e2d12014-10-23 14:42:50 +08005908}
5909
Yan Zhenga512bbf2008-12-08 16:46:26 -05005910int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5911 u64 chunk_start, u64 physical, u64 devid,
5912 u64 **logical, int *naddrs, int *stripe_len)
5913{
5914 struct extent_map_tree *em_tree = &map_tree->map_tree;
5915 struct extent_map *em;
5916 struct map_lookup *map;
5917 u64 *buf;
5918 u64 bytenr;
5919 u64 length;
5920 u64 stripe_nr;
David Woodhouse53b381b2013-01-29 18:40:14 -05005921 u64 rmap_len;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005922 int i, j, nr = 0;
5923
Chris Mason890871b2009-09-02 16:24:52 -04005924 read_lock(&em_tree->lock);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005925 em = lookup_extent_mapping(em_tree, chunk_start, 1);
Chris Mason890871b2009-09-02 16:24:52 -04005926 read_unlock(&em_tree->lock);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005927
Josef Bacik835d9742013-03-19 12:13:25 -04005928 if (!em) {
Frank Holtonefe120a2013-12-20 11:37:06 -05005929 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
Josef Bacik835d9742013-03-19 12:13:25 -04005930 chunk_start);
5931 return -EIO;
5932 }
5933
5934 if (em->start != chunk_start) {
Frank Holtonefe120a2013-12-20 11:37:06 -05005935 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
Josef Bacik835d9742013-03-19 12:13:25 -04005936 em->start, chunk_start);
5937 free_extent_map(em);
5938 return -EIO;
5939 }
Jeff Mahoney95617d62015-06-03 10:55:48 -04005940 map = em->map_lookup;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005941
5942 length = em->len;
David Woodhouse53b381b2013-01-29 18:40:14 -05005943 rmap_len = map->stripe_len;
5944
Yan Zhenga512bbf2008-12-08 16:46:26 -05005945 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
David Sterbab8b93ad2015-01-16 17:26:13 +01005946 length = div_u64(length, map->num_stripes / map->sub_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005947 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
David Sterbab8b93ad2015-01-16 17:26:13 +01005948 length = div_u64(length, map->num_stripes);
Zhao Leiffe2d202015-01-20 15:11:44 +08005949 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
David Sterbab8b93ad2015-01-16 17:26:13 +01005950 length = div_u64(length, nr_data_stripes(map));
David Woodhouse53b381b2013-01-29 18:40:14 -05005951 rmap_len = map->stripe_len * nr_data_stripes(map);
5952 }
Yan Zhenga512bbf2008-12-08 16:46:26 -05005953
David Sterba31e818f2015-02-20 18:00:26 +01005954 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005955 BUG_ON(!buf); /* -ENOMEM */
Yan Zhenga512bbf2008-12-08 16:46:26 -05005956
5957 for (i = 0; i < map->num_stripes; i++) {
5958 if (devid && map->stripes[i].dev->devid != devid)
5959 continue;
5960 if (map->stripes[i].physical > physical ||
5961 map->stripes[i].physical + length <= physical)
5962 continue;
5963
5964 stripe_nr = physical - map->stripes[i].physical;
David Sterbab8b93ad2015-01-16 17:26:13 +01005965 stripe_nr = div_u64(stripe_nr, map->stripe_len);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005966
5967 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5968 stripe_nr = stripe_nr * map->num_stripes + i;
David Sterbab8b93ad2015-01-16 17:26:13 +01005969 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005970 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5971 stripe_nr = stripe_nr * map->num_stripes + i;
David Woodhouse53b381b2013-01-29 18:40:14 -05005972 } /* else if RAID[56], multiply by nr_data_stripes().
5973 * Alternatively, just use rmap_len below instead of
5974 * map->stripe_len */
5975
5976 bytenr = chunk_start + stripe_nr * rmap_len;
Chris Mason934d3752008-12-08 16:43:10 -05005977 WARN_ON(nr >= map->num_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005978 for (j = 0; j < nr; j++) {
5979 if (buf[j] == bytenr)
5980 break;
5981 }
Chris Mason934d3752008-12-08 16:43:10 -05005982 if (j == nr) {
5983 WARN_ON(nr >= map->num_stripes);
Yan Zhenga512bbf2008-12-08 16:46:26 -05005984 buf[nr++] = bytenr;
Chris Mason934d3752008-12-08 16:43:10 -05005985 }
Yan Zhenga512bbf2008-12-08 16:46:26 -05005986 }
5987
Yan Zhenga512bbf2008-12-08 16:46:26 -05005988 *logical = buf;
5989 *naddrs = nr;
David Woodhouse53b381b2013-01-29 18:40:14 -05005990 *stripe_len = rmap_len;
Yan Zhenga512bbf2008-12-08 16:46:26 -05005991
5992 free_extent_map(em);
5993 return 0;
5994}
5995
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02005996static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
Miao Xie8408c712014-06-19 10:42:55 +08005997{
Mike Snitzer326e1db2015-05-22 09:14:03 -04005998 bio->bi_private = bbio->private;
5999 bio->bi_end_io = bbio->end_io;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006000 bio_endio(bio);
Mike Snitzer326e1db2015-05-22 09:14:03 -04006001
Zhao Lei6e9606d2015-01-20 15:11:34 +08006002 btrfs_put_bbio(bbio);
Miao Xie8408c712014-06-19 10:42:55 +08006003}
6004
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006005static void btrfs_end_bio(struct bio *bio)
Chris Mason8790d502008-04-03 16:29:03 -04006006{
Chris Mason9be33952013-05-17 18:30:14 -04006007 struct btrfs_bio *bbio = bio->bi_private;
Chris Mason7d2b4da2008-08-05 10:13:57 -04006008 int is_orig_bio = 0;
Chris Mason8790d502008-04-03 16:29:03 -04006009
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006010 if (bio->bi_error) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02006011 atomic_inc(&bbio->error);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006012 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
Stefan Behrens442a4f62012-05-25 16:06:08 +02006013 unsigned int stripe_index =
Chris Mason9be33952013-05-17 18:30:14 -04006014 btrfs_io_bio(bio)->stripe_index;
Zhao Lei65f53332015-06-08 20:05:50 +08006015 struct btrfs_device *dev;
Stefan Behrens442a4f62012-05-25 16:06:08 +02006016
6017 BUG_ON(stripe_index >= bbio->num_stripes);
6018 dev = bbio->stripes[stripe_index].dev;
Stefan Behrens597a60f2012-06-14 16:42:31 +02006019 if (dev->bdev) {
Mike Christie37226b22016-06-05 14:31:52 -05006020 if (bio_op(bio) == REQ_OP_WRITE)
Stefan Behrens597a60f2012-06-14 16:42:31 +02006021 btrfs_dev_stat_inc(dev,
6022 BTRFS_DEV_STAT_WRITE_ERRS);
6023 else
6024 btrfs_dev_stat_inc(dev,
6025 BTRFS_DEV_STAT_READ_ERRS);
Jens Axboe1eff9d32016-08-05 15:35:16 -06006026 if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH)
Stefan Behrens597a60f2012-06-14 16:42:31 +02006027 btrfs_dev_stat_inc(dev,
6028 BTRFS_DEV_STAT_FLUSH_ERRS);
6029 btrfs_dev_stat_print_on_error(dev);
6030 }
Stefan Behrens442a4f62012-05-25 16:06:08 +02006031 }
6032 }
Chris Mason8790d502008-04-03 16:29:03 -04006033
Jan Schmidta1d3c472011-08-04 17:15:33 +02006034 if (bio == bbio->orig_bio)
Chris Mason7d2b4da2008-08-05 10:13:57 -04006035 is_orig_bio = 1;
6036
Miao Xiec404e0d2014-01-30 16:46:55 +08006037 btrfs_bio_counter_dec(bbio->fs_info);
6038
Jan Schmidta1d3c472011-08-04 17:15:33 +02006039 if (atomic_dec_and_test(&bbio->stripes_pending)) {
Chris Mason7d2b4da2008-08-05 10:13:57 -04006040 if (!is_orig_bio) {
6041 bio_put(bio);
Jan Schmidta1d3c472011-08-04 17:15:33 +02006042 bio = bbio->orig_bio;
Chris Mason7d2b4da2008-08-05 10:13:57 -04006043 }
Muthu Kumarc7b22bb2014-01-08 14:19:52 -07006044
Chris Mason9be33952013-05-17 18:30:14 -04006045 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
Chris Masona236aed2008-04-29 09:38:00 -04006046 /* only send an error to the higher layers if it is
David Woodhouse53b381b2013-01-29 18:40:14 -05006047 * beyond the tolerance of the btrfs bio
Chris Masona236aed2008-04-29 09:38:00 -04006048 */
Jan Schmidta1d3c472011-08-04 17:15:33 +02006049 if (atomic_read(&bbio->error) > bbio->max_errors) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006050 bio->bi_error = -EIO;
Chris Mason5dbc8fc2011-12-09 11:07:37 -05006051 } else {
Chris Mason1259ab72008-05-12 13:39:03 -04006052 /*
6053 * this bio is actually up to date, we didn't
6054 * go over the max number of errors
6055 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006056 bio->bi_error = 0;
Chris Mason1259ab72008-05-12 13:39:03 -04006057 }
Miao Xiec55f1392014-06-19 10:42:54 +08006058
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006059 btrfs_end_bbio(bbio, bio);
Chris Mason7d2b4da2008-08-05 10:13:57 -04006060 } else if (!is_orig_bio) {
Chris Mason8790d502008-04-03 16:29:03 -04006061 bio_put(bio);
6062 }
Chris Mason8790d502008-04-03 16:29:03 -04006063}
6064
Chris Mason8b712842008-06-11 16:50:36 -04006065/*
6066 * see run_scheduled_bios for a description of why bios are collected for
6067 * async submit.
6068 *
6069 * This will add one bio to the pending list for a device and make sure
6070 * the work struct is scheduled.
6071 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00006072static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6073 struct btrfs_device *device,
Mike Christie4e49ea42016-06-05 14:31:41 -05006074 struct bio *bio)
Chris Mason8b712842008-06-11 16:50:36 -04006075{
6076 int should_queue = 1;
Chris Masonffbd5172009-04-20 15:50:09 -04006077 struct btrfs_pending_bios *pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -04006078
David Woodhouse53b381b2013-01-29 18:40:14 -05006079 if (device->missing || !device->bdev) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006080 bio_io_error(bio);
David Woodhouse53b381b2013-01-29 18:40:14 -05006081 return;
6082 }
6083
Chris Mason8b712842008-06-11 16:50:36 -04006084 /* don't bother with additional async steps for reads, right now */
Mike Christie37226b22016-06-05 14:31:52 -05006085 if (bio_op(bio) == REQ_OP_READ) {
Chris Mason492bb6de2008-07-31 16:29:02 -04006086 bio_get(bio);
Mike Christie4e49ea42016-06-05 14:31:41 -05006087 btrfsic_submit_bio(bio);
Chris Mason492bb6de2008-07-31 16:29:02 -04006088 bio_put(bio);
Jeff Mahoney143bede2012-03-01 14:56:26 +01006089 return;
Chris Mason8b712842008-06-11 16:50:36 -04006090 }
6091
6092 /*
Chris Mason0986fe9e2008-08-15 15:34:15 -04006093 * nr_async_bios allows us to reliably return congestion to the
Chris Mason8b712842008-06-11 16:50:36 -04006094 * higher layers. Otherwise, the async bio makes it appear we have
6095 * made progress against dirty pages when we've really just put it
6096 * on a queue for later
6097 */
Chris Mason0986fe9e2008-08-15 15:34:15 -04006098 atomic_inc(&root->fs_info->nr_async_bios);
Chris Mason492bb6de2008-07-31 16:29:02 -04006099 WARN_ON(bio->bi_next);
Chris Mason8b712842008-06-11 16:50:36 -04006100 bio->bi_next = NULL;
Chris Mason8b712842008-06-11 16:50:36 -04006101
6102 spin_lock(&device->io_lock);
Jens Axboe1eff9d32016-08-05 15:35:16 -06006103 if (bio->bi_opf & REQ_SYNC)
Chris Masonffbd5172009-04-20 15:50:09 -04006104 pending_bios = &device->pending_sync_bios;
6105 else
6106 pending_bios = &device->pending_bios;
Chris Mason8b712842008-06-11 16:50:36 -04006107
Chris Masonffbd5172009-04-20 15:50:09 -04006108 if (pending_bios->tail)
6109 pending_bios->tail->bi_next = bio;
Chris Mason8b712842008-06-11 16:50:36 -04006110
Chris Masonffbd5172009-04-20 15:50:09 -04006111 pending_bios->tail = bio;
6112 if (!pending_bios->head)
6113 pending_bios->head = bio;
Chris Mason8b712842008-06-11 16:50:36 -04006114 if (device->running_pending)
6115 should_queue = 0;
6116
6117 spin_unlock(&device->io_lock);
6118
6119 if (should_queue)
Qu Wenruoa8c93d42014-02-28 10:46:08 +08006120 btrfs_queue_work(root->fs_info->submit_workers,
6121 &device->work);
Chris Mason8b712842008-06-11 16:50:36 -04006122}
6123
Josef Bacikde1ee922012-10-19 16:50:56 -04006124static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6125 struct bio *bio, u64 physical, int dev_nr,
Mike Christie81a75f62016-06-05 14:31:54 -05006126 int async)
Josef Bacikde1ee922012-10-19 16:50:56 -04006127{
6128 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6129
6130 bio->bi_private = bbio;
Chris Mason9be33952013-05-17 18:30:14 -04006131 btrfs_io_bio(bio)->stripe_index = dev_nr;
Josef Bacikde1ee922012-10-19 16:50:56 -04006132 bio->bi_end_io = btrfs_end_bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07006133 bio->bi_iter.bi_sector = physical >> 9;
Josef Bacikde1ee922012-10-19 16:50:56 -04006134#ifdef DEBUG
6135 {
6136 struct rcu_string *name;
6137
6138 rcu_read_lock();
6139 name = rcu_dereference(dev->name);
Mike Christie81a75f62016-06-05 14:31:54 -05006140 pr_debug("btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu "
Jens Axboe1eff9d32016-08-05 15:35:16 -06006141 "(%s id %llu), size=%u\n", bio_op(bio), bio->bi_opf,
Fabian Frederick1b6e4462014-09-24 20:23:05 +02006142 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6143 name->str, dev->devid, bio->bi_iter.bi_size);
Josef Bacikde1ee922012-10-19 16:50:56 -04006144 rcu_read_unlock();
6145 }
6146#endif
6147 bio->bi_bdev = dev->bdev;
Miao Xiec404e0d2014-01-30 16:46:55 +08006148
6149 btrfs_bio_counter_inc_noblocked(root->fs_info);
6150
Josef Bacikde1ee922012-10-19 16:50:56 -04006151 if (async)
Mike Christie4e49ea42016-06-05 14:31:41 -05006152 btrfs_schedule_bio(root, dev, bio);
Josef Bacikde1ee922012-10-19 16:50:56 -04006153 else
Mike Christie4e49ea42016-06-05 14:31:41 -05006154 btrfsic_submit_bio(bio);
Josef Bacikde1ee922012-10-19 16:50:56 -04006155}
6156
Josef Bacikde1ee922012-10-19 16:50:56 -04006157static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6158{
6159 atomic_inc(&bbio->error);
6160 if (atomic_dec_and_test(&bbio->stripes_pending)) {
Nicholas D Steeves01327612016-05-19 21:18:45 -04006161 /* Should be the original bio. */
Miao Xie8408c712014-06-19 10:42:55 +08006162 WARN_ON(bio != bbio->orig_bio);
6163
Chris Mason9be33952013-05-17 18:30:14 -04006164 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
Kent Overstreet4f024f32013-10-11 15:44:27 -07006165 bio->bi_iter.bi_sector = logical >> 9;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02006166 bio->bi_error = -EIO;
6167 btrfs_end_bbio(bbio, bio);
Josef Bacikde1ee922012-10-19 16:50:56 -04006168 }
6169}
6170
Mike Christie81a75f62016-06-05 14:31:54 -05006171int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
Chris Mason8b712842008-06-11 16:50:36 -04006172 int mirror_num, int async_submit)
Chris Mason0b86a832008-03-24 15:01:56 -04006173{
Chris Mason0b86a832008-03-24 15:01:56 -04006174 struct btrfs_device *dev;
Chris Mason8790d502008-04-03 16:29:03 -04006175 struct bio *first_bio = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07006176 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
Chris Mason0b86a832008-03-24 15:01:56 -04006177 u64 length = 0;
6178 u64 map_length;
Chris Mason0b86a832008-03-24 15:01:56 -04006179 int ret;
Zhao Lei08da7572015-02-12 15:42:16 +08006180 int dev_nr;
6181 int total_devs;
Jan Schmidta1d3c472011-08-04 17:15:33 +02006182 struct btrfs_bio *bbio = NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04006183
Kent Overstreet4f024f32013-10-11 15:44:27 -07006184 length = bio->bi_iter.bi_size;
Chris Mason0b86a832008-03-24 15:01:56 -04006185 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04006186
Miao Xiec404e0d2014-01-30 16:46:55 +08006187 btrfs_bio_counter_inc_blocked(root->fs_info);
Mike Christie37226b22016-06-05 14:31:52 -05006188 ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6189 &map_length, &bbio, mirror_num, 1);
Miao Xiec404e0d2014-01-30 16:46:55 +08006190 if (ret) {
6191 btrfs_bio_counter_dec(root->fs_info);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006192 return ret;
Miao Xiec404e0d2014-01-30 16:46:55 +08006193 }
Chris Masoncea9e442008-04-09 16:28:12 -04006194
Jan Schmidta1d3c472011-08-04 17:15:33 +02006195 total_devs = bbio->num_stripes;
David Woodhouse53b381b2013-01-29 18:40:14 -05006196 bbio->orig_bio = first_bio;
6197 bbio->private = first_bio->bi_private;
6198 bbio->end_io = first_bio->bi_end_io;
Miao Xiec404e0d2014-01-30 16:46:55 +08006199 bbio->fs_info = root->fs_info;
David Woodhouse53b381b2013-01-29 18:40:14 -05006200 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6201
Zhao Leiad1ba2a2015-12-15 18:18:09 +08006202 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
Mike Christie37226b22016-06-05 14:31:52 -05006203 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
David Woodhouse53b381b2013-01-29 18:40:14 -05006204 /* In this case, map_length has been set to the length of
6205 a single stripe; not the whole write */
Mike Christie37226b22016-06-05 14:31:52 -05006206 if (bio_op(bio) == REQ_OP_WRITE) {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08006207 ret = raid56_parity_write(root, bio, bbio, map_length);
David Woodhouse53b381b2013-01-29 18:40:14 -05006208 } else {
Zhao Lei8e5cfb52015-01-20 15:11:33 +08006209 ret = raid56_parity_recover(root, bio, bbio, map_length,
Miao Xie42452152014-11-25 16:39:28 +08006210 mirror_num, 1);
David Woodhouse53b381b2013-01-29 18:40:14 -05006211 }
Miao Xie42452152014-11-25 16:39:28 +08006212
Miao Xiec404e0d2014-01-30 16:46:55 +08006213 btrfs_bio_counter_dec(root->fs_info);
6214 return ret;
David Woodhouse53b381b2013-01-29 18:40:14 -05006215 }
6216
Chris Masoncea9e442008-04-09 16:28:12 -04006217 if (map_length < length) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006218 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02006219 logical, length, map_length);
Chris Masoncea9e442008-04-09 16:28:12 -04006220 BUG();
6221 }
Jan Schmidta1d3c472011-08-04 17:15:33 +02006222
Zhao Lei08da7572015-02-12 15:42:16 +08006223 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
Josef Bacikde1ee922012-10-19 16:50:56 -04006224 dev = bbio->stripes[dev_nr].dev;
Mike Christie37226b22016-06-05 14:31:52 -05006225 if (!dev || !dev->bdev ||
6226 (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
Josef Bacikde1ee922012-10-19 16:50:56 -04006227 bbio_error(bbio, first_bio, logical);
Josef Bacikde1ee922012-10-19 16:50:56 -04006228 continue;
6229 }
6230
Jan Schmidta1d3c472011-08-04 17:15:33 +02006231 if (dev_nr < total_devs - 1) {
Chris Mason9be33952013-05-17 18:30:14 -04006232 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006233 BUG_ON(!bio); /* -ENOMEM */
Mike Snitzer326e1db2015-05-22 09:14:03 -04006234 } else
Jan Schmidta1d3c472011-08-04 17:15:33 +02006235 bio = first_bio;
Josef Bacik606686e2012-06-04 14:03:51 -04006236
Josef Bacikde1ee922012-10-19 16:50:56 -04006237 submit_stripe_bio(root, bbio, bio,
Mike Christie81a75f62016-06-05 14:31:54 -05006238 bbio->stripes[dev_nr].physical, dev_nr,
Josef Bacikde1ee922012-10-19 16:50:56 -04006239 async_submit);
Chris Mason239b14b2008-03-24 15:02:07 -04006240 }
Miao Xiec404e0d2014-01-30 16:46:55 +08006241 btrfs_bio_counter_dec(root->fs_info);
Chris Mason0b86a832008-03-24 15:01:56 -04006242 return 0;
6243}
6244
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006245struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
Yan Zheng2b820322008-11-17 21:11:30 -05006246 u8 *uuid, u8 *fsid)
Chris Mason0b86a832008-03-24 15:01:56 -04006247{
Yan Zheng2b820322008-11-17 21:11:30 -05006248 struct btrfs_device *device;
6249 struct btrfs_fs_devices *cur_devices;
Chris Mason0b86a832008-03-24 15:01:56 -04006250
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006251 cur_devices = fs_info->fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006252 while (cur_devices) {
6253 if (!fsid ||
6254 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6255 device = __find_device(&cur_devices->devices,
6256 devid, uuid);
6257 if (device)
6258 return device;
6259 }
6260 cur_devices = cur_devices->seed;
6261 }
6262 return NULL;
Chris Mason0b86a832008-03-24 15:01:56 -04006263}
6264
Chris Masondfe25022008-05-13 13:46:40 -04006265static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
Miao Xie5f375832014-09-03 21:35:46 +08006266 struct btrfs_fs_devices *fs_devices,
Chris Masondfe25022008-05-13 13:46:40 -04006267 u64 devid, u8 *dev_uuid)
6268{
6269 struct btrfs_device *device;
Chris Masondfe25022008-05-13 13:46:40 -04006270
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006271 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6272 if (IS_ERR(device))
yanhai zhu7cbd8a82008-11-12 14:38:54 -05006273 return NULL;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006274
6275 list_add(&device->dev_list, &fs_devices->devices);
Yan Zhenge4404d62008-12-12 10:03:26 -05006276 device->fs_devices = fs_devices;
Chris Masondfe25022008-05-13 13:46:40 -04006277 fs_devices->num_devices++;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006278
6279 device->missing = 1;
Chris Masoncd02dca2010-12-13 14:56:23 -05006280 fs_devices->missing_devices++;
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006281
Chris Masondfe25022008-05-13 13:46:40 -04006282 return device;
6283}
6284
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006285/**
6286 * btrfs_alloc_device - allocate struct btrfs_device
6287 * @fs_info: used only for generating a new devid, can be NULL if
6288 * devid is provided (i.e. @devid != NULL).
6289 * @devid: a pointer to devid for this device. If NULL a new devid
6290 * is generated.
6291 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6292 * is generated.
6293 *
6294 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6295 * on error. Returned struct is not linked onto any lists and can be
6296 * destroyed with kfree() right away.
6297 */
6298struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6299 const u64 *devid,
6300 const u8 *uuid)
6301{
6302 struct btrfs_device *dev;
6303 u64 tmp;
6304
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05306305 if (WARN_ON(!devid && !fs_info))
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006306 return ERR_PTR(-EINVAL);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006307
6308 dev = __alloc_device();
6309 if (IS_ERR(dev))
6310 return dev;
6311
6312 if (devid)
6313 tmp = *devid;
6314 else {
6315 int ret;
6316
6317 ret = find_next_devid(fs_info, &tmp);
6318 if (ret) {
6319 kfree(dev);
6320 return ERR_PTR(ret);
6321 }
6322 }
6323 dev->devid = tmp;
6324
6325 if (uuid)
6326 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6327 else
6328 generate_random_uuid(dev->uuid);
6329
Liu Bo9e0af232014-08-15 23:36:53 +08006330 btrfs_init_work(&dev->work, btrfs_submit_helper,
6331 pending_bios_fn, NULL, NULL);
Ilya Dryomov12bd2fc2013-08-23 13:20:17 +03006332
6333 return dev;
6334}
6335
Liu Boe06cd3d2016-06-03 12:05:15 -07006336/* Return -EIO if any error, otherwise return 0. */
6337static int btrfs_check_chunk_valid(struct btrfs_root *root,
6338 struct extent_buffer *leaf,
6339 struct btrfs_chunk *chunk, u64 logical)
6340{
6341 u64 length;
6342 u64 stripe_len;
6343 u16 num_stripes;
6344 u16 sub_stripes;
6345 u64 type;
6346
6347 length = btrfs_chunk_length(leaf, chunk);
6348 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6349 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6350 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6351 type = btrfs_chunk_type(leaf, chunk);
6352
6353 if (!num_stripes) {
6354 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6355 num_stripes);
6356 return -EIO;
6357 }
6358 if (!IS_ALIGNED(logical, root->sectorsize)) {
6359 btrfs_err(root->fs_info,
6360 "invalid chunk logical %llu", logical);
6361 return -EIO;
6362 }
6363 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6364 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6365 btrfs_chunk_sector_size(leaf, chunk));
6366 return -EIO;
6367 }
6368 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6369 btrfs_err(root->fs_info,
6370 "invalid chunk length %llu", length);
6371 return -EIO;
6372 }
6373 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6374 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6375 stripe_len);
6376 return -EIO;
6377 }
6378 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6379 type) {
6380 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6381 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6382 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6383 btrfs_chunk_type(leaf, chunk));
6384 return -EIO;
6385 }
6386 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6387 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6388 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6389 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6390 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6391 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6392 num_stripes != 1)) {
6393 btrfs_err(root->fs_info,
6394 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6395 num_stripes, sub_stripes,
6396 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6397 return -EIO;
6398 }
6399
6400 return 0;
6401}
6402
Chris Mason0b86a832008-03-24 15:01:56 -04006403static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6404 struct extent_buffer *leaf,
6405 struct btrfs_chunk *chunk)
6406{
6407 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6408 struct map_lookup *map;
6409 struct extent_map *em;
6410 u64 logical;
6411 u64 length;
Qu Wenruof04b7722015-12-15 09:14:37 +08006412 u64 stripe_len;
Chris Mason0b86a832008-03-24 15:01:56 -04006413 u64 devid;
Chris Masona4437552008-04-18 10:29:38 -04006414 u8 uuid[BTRFS_UUID_SIZE];
Chris Mason593060d2008-03-25 16:50:33 -04006415 int num_stripes;
Chris Mason0b86a832008-03-24 15:01:56 -04006416 int ret;
Chris Mason593060d2008-03-25 16:50:33 -04006417 int i;
Chris Mason0b86a832008-03-24 15:01:56 -04006418
Chris Masone17cade2008-04-15 15:41:47 -04006419 logical = key->offset;
6420 length = btrfs_chunk_length(leaf, chunk);
Qu Wenruof04b7722015-12-15 09:14:37 +08006421 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6422 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
Liu Boe06cd3d2016-06-03 12:05:15 -07006423
6424 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6425 if (ret)
6426 return ret;
Chris Masona061fc82008-05-07 11:43:44 -04006427
Chris Mason890871b2009-09-02 16:24:52 -04006428 read_lock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04006429 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
Chris Mason890871b2009-09-02 16:24:52 -04006430 read_unlock(&map_tree->map_tree.lock);
Chris Mason0b86a832008-03-24 15:01:56 -04006431
6432 /* already mapped? */
6433 if (em && em->start <= logical && em->start + em->len > logical) {
6434 free_extent_map(em);
Chris Mason0b86a832008-03-24 15:01:56 -04006435 return 0;
6436 } else if (em) {
6437 free_extent_map(em);
6438 }
Chris Mason0b86a832008-03-24 15:01:56 -04006439
David Sterba172ddd62011-04-21 00:48:27 +02006440 em = alloc_extent_map();
Chris Mason0b86a832008-03-24 15:01:56 -04006441 if (!em)
6442 return -ENOMEM;
Chris Mason593060d2008-03-25 16:50:33 -04006443 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
Chris Mason0b86a832008-03-24 15:01:56 -04006444 if (!map) {
6445 free_extent_map(em);
6446 return -ENOMEM;
6447 }
6448
Wang Shilong298a8f92014-06-19 10:42:52 +08006449 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
Jeff Mahoney95617d62015-06-03 10:55:48 -04006450 em->map_lookup = map;
Chris Mason0b86a832008-03-24 15:01:56 -04006451 em->start = logical;
6452 em->len = length;
Josef Bacik70c8a912012-10-11 16:54:30 -04006453 em->orig_start = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006454 em->block_start = 0;
Chris Masonc8b97812008-10-29 14:49:59 -04006455 em->block_len = em->len;
Chris Mason0b86a832008-03-24 15:01:56 -04006456
Chris Mason593060d2008-03-25 16:50:33 -04006457 map->num_stripes = num_stripes;
6458 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6459 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6460 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6461 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6462 map->type = btrfs_chunk_type(leaf, chunk);
Chris Mason321aecc2008-04-16 10:49:51 -04006463 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
Chris Mason593060d2008-03-25 16:50:33 -04006464 for (i = 0; i < num_stripes; i++) {
6465 map->stripes[i].physical =
6466 btrfs_stripe_offset_nr(leaf, chunk, i);
6467 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
Chris Masona4437552008-04-18 10:29:38 -04006468 read_extent_buffer(leaf, uuid, (unsigned long)
6469 btrfs_stripe_dev_uuid_nr(chunk, i),
6470 BTRFS_UUID_SIZE);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006471 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6472 uuid, NULL);
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006473 if (!map->stripes[i].dev &&
6474 !btrfs_test_opt(root->fs_info, DEGRADED)) {
Chris Mason593060d2008-03-25 16:50:33 -04006475 free_extent_map(em);
6476 return -EIO;
6477 }
Chris Masondfe25022008-05-13 13:46:40 -04006478 if (!map->stripes[i].dev) {
6479 map->stripes[i].dev =
Miao Xie5f375832014-09-03 21:35:46 +08006480 add_missing_dev(root, root->fs_info->fs_devices,
6481 devid, uuid);
Chris Masondfe25022008-05-13 13:46:40 -04006482 if (!map->stripes[i].dev) {
Chris Masondfe25022008-05-13 13:46:40 -04006483 free_extent_map(em);
6484 return -EIO;
6485 }
Anand Jain816fceb2015-04-27 12:46:18 +08006486 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6487 devid, uuid);
Chris Masondfe25022008-05-13 13:46:40 -04006488 }
6489 map->stripes[i].dev->in_fs_metadata = 1;
Chris Mason0b86a832008-03-24 15:01:56 -04006490 }
6491
Chris Mason890871b2009-09-02 16:24:52 -04006492 write_lock(&map_tree->map_tree.lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006493 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
Chris Mason890871b2009-09-02 16:24:52 -04006494 write_unlock(&map_tree->map_tree.lock);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006495 BUG_ON(ret); /* Tree corruption */
Chris Mason0b86a832008-03-24 15:01:56 -04006496 free_extent_map(em);
6497
6498 return 0;
6499}
6500
Jeff Mahoney143bede2012-03-01 14:56:26 +01006501static void fill_device_from_item(struct extent_buffer *leaf,
Chris Mason0b86a832008-03-24 15:01:56 -04006502 struct btrfs_dev_item *dev_item,
6503 struct btrfs_device *device)
6504{
6505 unsigned long ptr;
Chris Mason0b86a832008-03-24 15:01:56 -04006506
6507 device->devid = btrfs_device_id(leaf, dev_item);
Chris Balld6397ba2009-04-27 07:29:03 -04006508 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6509 device->total_bytes = device->disk_total_bytes;
Miao Xie935e5cc2014-09-03 21:35:33 +08006510 device->commit_total_bytes = device->disk_total_bytes;
Chris Mason0b86a832008-03-24 15:01:56 -04006511 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
Miao Xiece7213c2014-09-03 21:35:34 +08006512 device->commit_bytes_used = device->bytes_used;
Chris Mason0b86a832008-03-24 15:01:56 -04006513 device->type = btrfs_device_type(leaf, dev_item);
6514 device->io_align = btrfs_device_io_align(leaf, dev_item);
6515 device->io_width = btrfs_device_io_width(leaf, dev_item);
6516 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
Stefan Behrens8dabb742012-11-06 13:15:27 +01006517 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
Stefan Behrens63a212a2012-11-05 18:29:28 +01006518 device->is_tgtdev_for_dev_replace = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006519
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02006520 ptr = btrfs_device_uuid(dev_item);
Chris Masone17cade2008-04-15 15:41:47 -04006521 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04006522}
6523
Miao Xie5f375832014-09-03 21:35:46 +08006524static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6525 u8 *fsid)
Yan Zheng2b820322008-11-17 21:11:30 -05006526{
6527 struct btrfs_fs_devices *fs_devices;
6528 int ret;
6529
Li Zefanb367e472011-12-07 11:38:24 +08006530 BUG_ON(!mutex_is_locked(&uuid_mutex));
Yan Zheng2b820322008-11-17 21:11:30 -05006531
6532 fs_devices = root->fs_info->fs_devices->seed;
6533 while (fs_devices) {
Miao Xie5f375832014-09-03 21:35:46 +08006534 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6535 return fs_devices;
6536
Yan Zheng2b820322008-11-17 21:11:30 -05006537 fs_devices = fs_devices->seed;
6538 }
6539
6540 fs_devices = find_fsid(fsid);
6541 if (!fs_devices) {
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006542 if (!btrfs_test_opt(root->fs_info, DEGRADED))
Miao Xie5f375832014-09-03 21:35:46 +08006543 return ERR_PTR(-ENOENT);
6544
6545 fs_devices = alloc_fs_devices(fsid);
6546 if (IS_ERR(fs_devices))
6547 return fs_devices;
6548
6549 fs_devices->seeding = 1;
6550 fs_devices->opened = 1;
6551 return fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006552 }
Yan Zhenge4404d62008-12-12 10:03:26 -05006553
6554 fs_devices = clone_fs_devices(fs_devices);
Miao Xie5f375832014-09-03 21:35:46 +08006555 if (IS_ERR(fs_devices))
6556 return fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006557
Christoph Hellwig97288f22008-12-02 06:36:09 -05006558 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
Chris Mason15916de2008-11-19 21:17:22 -05006559 root->fs_info->bdev_holder);
Julia Lawall48d28232012-04-14 11:24:33 +02006560 if (ret) {
6561 free_fs_devices(fs_devices);
Miao Xie5f375832014-09-03 21:35:46 +08006562 fs_devices = ERR_PTR(ret);
Yan Zheng2b820322008-11-17 21:11:30 -05006563 goto out;
Julia Lawall48d28232012-04-14 11:24:33 +02006564 }
Yan Zheng2b820322008-11-17 21:11:30 -05006565
6566 if (!fs_devices->seeding) {
6567 __btrfs_close_devices(fs_devices);
Yan Zhenge4404d62008-12-12 10:03:26 -05006568 free_fs_devices(fs_devices);
Miao Xie5f375832014-09-03 21:35:46 +08006569 fs_devices = ERR_PTR(-EINVAL);
Yan Zheng2b820322008-11-17 21:11:30 -05006570 goto out;
6571 }
6572
6573 fs_devices->seed = root->fs_info->fs_devices->seed;
6574 root->fs_info->fs_devices->seed = fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006575out:
Miao Xie5f375832014-09-03 21:35:46 +08006576 return fs_devices;
Yan Zheng2b820322008-11-17 21:11:30 -05006577}
6578
Chris Mason0d81ba52008-03-24 15:02:07 -04006579static int read_one_dev(struct btrfs_root *root,
Chris Mason0b86a832008-03-24 15:01:56 -04006580 struct extent_buffer *leaf,
6581 struct btrfs_dev_item *dev_item)
6582{
Miao Xie5f375832014-09-03 21:35:46 +08006583 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
Chris Mason0b86a832008-03-24 15:01:56 -04006584 struct btrfs_device *device;
6585 u64 devid;
6586 int ret;
Yan Zheng2b820322008-11-17 21:11:30 -05006587 u8 fs_uuid[BTRFS_UUID_SIZE];
Chris Masona4437552008-04-18 10:29:38 -04006588 u8 dev_uuid[BTRFS_UUID_SIZE];
6589
Chris Mason0b86a832008-03-24 15:01:56 -04006590 devid = btrfs_device_id(leaf, dev_item);
Geert Uytterhoeven410ba3a2013-08-20 13:20:11 +02006591 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
Chris Masona4437552008-04-18 10:29:38 -04006592 BTRFS_UUID_SIZE);
Geert Uytterhoeven1473b242013-08-20 13:20:12 +02006593 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
Yan Zheng2b820322008-11-17 21:11:30 -05006594 BTRFS_UUID_SIZE);
6595
6596 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
Miao Xie5f375832014-09-03 21:35:46 +08006597 fs_devices = open_seed_devices(root, fs_uuid);
6598 if (IS_ERR(fs_devices))
6599 return PTR_ERR(fs_devices);
Yan Zheng2b820322008-11-17 21:11:30 -05006600 }
6601
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01006602 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
Miao Xie5f375832014-09-03 21:35:46 +08006603 if (!device) {
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006604 if (!btrfs_test_opt(root->fs_info, DEGRADED))
Yan Zheng2b820322008-11-17 21:11:30 -05006605 return -EIO;
6606
Miao Xie5f375832014-09-03 21:35:46 +08006607 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6608 if (!device)
6609 return -ENOMEM;
Anand Jain33b97e42015-05-08 04:34:35 +08006610 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6611 devid, dev_uuid);
Miao Xie5f375832014-09-03 21:35:46 +08006612 } else {
Jeff Mahoney3cdde222016-06-09 21:38:35 -04006613 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
Miao Xie5f375832014-09-03 21:35:46 +08006614 return -EIO;
6615
6616 if(!device->bdev && !device->missing) {
Chris Masoncd02dca2010-12-13 14:56:23 -05006617 /*
6618 * this happens when a device that was properly setup
6619 * in the device info lists suddenly goes bad.
6620 * device->bdev is NULL, and so we have to set
6621 * device->missing to one here
6622 */
Miao Xie5f375832014-09-03 21:35:46 +08006623 device->fs_devices->missing_devices++;
Chris Masoncd02dca2010-12-13 14:56:23 -05006624 device->missing = 1;
Yan Zheng2b820322008-11-17 21:11:30 -05006625 }
Miao Xie5f375832014-09-03 21:35:46 +08006626
6627 /* Move the device to its own fs_devices */
6628 if (device->fs_devices != fs_devices) {
6629 ASSERT(device->missing);
6630
6631 list_move(&device->dev_list, &fs_devices->devices);
6632 device->fs_devices->num_devices--;
6633 fs_devices->num_devices++;
6634
6635 device->fs_devices->missing_devices--;
6636 fs_devices->missing_devices++;
6637
6638 device->fs_devices = fs_devices;
6639 }
Yan Zheng2b820322008-11-17 21:11:30 -05006640 }
6641
6642 if (device->fs_devices != root->fs_info->fs_devices) {
6643 BUG_ON(device->writeable);
6644 if (device->generation !=
6645 btrfs_device_generation(leaf, dev_item))
6646 return -EINVAL;
Chris Mason6324fbf2008-03-24 15:01:59 -04006647 }
Chris Mason0b86a832008-03-24 15:01:56 -04006648
6649 fill_device_from_item(leaf, dev_item, device);
Chris Masondfe25022008-05-13 13:46:40 -04006650 device->in_fs_metadata = 1;
Stefan Behrens63a212a2012-11-05 18:29:28 +01006651 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
Yan Zheng2b820322008-11-17 21:11:30 -05006652 device->fs_devices->total_rw_bytes += device->total_bytes;
Josef Bacik2bf64752011-09-26 17:12:22 -04006653 spin_lock(&root->fs_info->free_chunk_lock);
6654 root->fs_info->free_chunk_space += device->total_bytes -
6655 device->bytes_used;
6656 spin_unlock(&root->fs_info->free_chunk_lock);
6657 }
Chris Mason0b86a832008-03-24 15:01:56 -04006658 ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006659 return ret;
6660}
6661
Yan Zhenge4404d62008-12-12 10:03:26 -05006662int btrfs_read_sys_array(struct btrfs_root *root)
Chris Mason0b86a832008-03-24 15:01:56 -04006663{
David Sterba6c417612011-04-13 15:41:04 +02006664 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
Chris Masona061fc82008-05-07 11:43:44 -04006665 struct extent_buffer *sb;
Chris Mason0b86a832008-03-24 15:01:56 -04006666 struct btrfs_disk_key *disk_key;
Chris Mason0b86a832008-03-24 15:01:56 -04006667 struct btrfs_chunk *chunk;
David Sterba1ffb22c2014-10-31 19:02:42 +01006668 u8 *array_ptr;
6669 unsigned long sb_array_offset;
Chris Mason84eed902008-04-25 09:04:37 -04006670 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006671 u32 num_stripes;
6672 u32 array_size;
6673 u32 len = 0;
David Sterba1ffb22c2014-10-31 19:02:42 +01006674 u32 cur_offset;
Liu Boe06cd3d2016-06-03 12:05:15 -07006675 u64 type;
Chris Mason84eed902008-04-25 09:04:37 -04006676 struct btrfs_key key;
Chris Mason0b86a832008-03-24 15:01:56 -04006677
David Sterbaa83fffb2014-06-15 02:39:54 +02006678 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6679 /*
6680 * This will create extent buffer of nodesize, superblock size is
6681 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6682 * overallocate but we can keep it as-is, only the first page is used.
6683 */
6684 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
Liu Boc871b0f2016-06-06 12:01:23 -07006685 if (IS_ERR(sb))
6686 return PTR_ERR(sb);
David Sterba4db8c522015-12-03 13:06:46 +01006687 set_extent_buffer_uptodate(sb);
Chris Mason85d4e462011-07-26 16:11:19 -04006688 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
David Sterba8a334422011-10-07 18:06:13 +02006689 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04006690 * The sb extent buffer is artificial and just used to read the system array.
David Sterba4db8c522015-12-03 13:06:46 +01006691 * set_extent_buffer_uptodate() call does not properly mark all it's
David Sterba8a334422011-10-07 18:06:13 +02006692 * pages up-to-date when the page is larger: extent does not cover the
6693 * whole page and consequently check_page_uptodate does not find all
6694 * the page's extents up-to-date (the hole beyond sb),
6695 * write_extent_buffer then triggers a WARN_ON.
6696 *
6697 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6698 * but sb spans only this function. Add an explicit SetPageUptodate call
6699 * to silence the warning eg. on PowerPC 64.
6700 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006701 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
Chris Mason727011e2010-08-06 13:21:20 -04006702 SetPageUptodate(sb->pages[0]);
Chris Mason4008c042009-02-12 14:09:45 -05006703
Chris Masona061fc82008-05-07 11:43:44 -04006704 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
Chris Mason0b86a832008-03-24 15:01:56 -04006705 array_size = btrfs_super_sys_array_size(super_copy);
6706
David Sterba1ffb22c2014-10-31 19:02:42 +01006707 array_ptr = super_copy->sys_chunk_array;
6708 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6709 cur_offset = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006710
David Sterba1ffb22c2014-10-31 19:02:42 +01006711 while (cur_offset < array_size) {
6712 disk_key = (struct btrfs_disk_key *)array_ptr;
David Sterbae3540ea2014-11-05 15:24:51 +01006713 len = sizeof(*disk_key);
6714 if (cur_offset + len > array_size)
6715 goto out_short_read;
6716
Chris Mason0b86a832008-03-24 15:01:56 -04006717 btrfs_disk_key_to_cpu(&key, disk_key);
6718
David Sterba1ffb22c2014-10-31 19:02:42 +01006719 array_ptr += len;
6720 sb_array_offset += len;
6721 cur_offset += len;
Chris Mason0b86a832008-03-24 15:01:56 -04006722
Chris Mason0d81ba52008-03-24 15:02:07 -04006723 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
David Sterba1ffb22c2014-10-31 19:02:42 +01006724 chunk = (struct btrfs_chunk *)sb_array_offset;
David Sterbae3540ea2014-11-05 15:24:51 +01006725 /*
6726 * At least one btrfs_chunk with one stripe must be
6727 * present, exact stripe count check comes afterwards
6728 */
6729 len = btrfs_chunk_item_size(1);
6730 if (cur_offset + len > array_size)
6731 goto out_short_read;
6732
6733 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
David Sterbaf5cdedd2015-11-30 17:27:06 +01006734 if (!num_stripes) {
6735 printk(KERN_ERR
6736 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6737 num_stripes, cur_offset);
6738 ret = -EIO;
6739 break;
6740 }
6741
Liu Boe06cd3d2016-06-03 12:05:15 -07006742 type = btrfs_chunk_type(sb, chunk);
6743 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6744 btrfs_err(root->fs_info,
6745 "invalid chunk type %llu in sys_array at offset %u",
6746 type, cur_offset);
6747 ret = -EIO;
6748 break;
6749 }
6750
David Sterbae3540ea2014-11-05 15:24:51 +01006751 len = btrfs_chunk_item_size(num_stripes);
6752 if (cur_offset + len > array_size)
6753 goto out_short_read;
6754
Chris Mason0d81ba52008-03-24 15:02:07 -04006755 ret = read_one_chunk(root, &key, sb, chunk);
Chris Mason84eed902008-04-25 09:04:37 -04006756 if (ret)
6757 break;
Chris Mason0b86a832008-03-24 15:01:56 -04006758 } else {
David Sterba93a3d462015-11-30 17:27:09 +01006759 printk(KERN_ERR
6760 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6761 (u32)key.type, cur_offset);
Chris Mason84eed902008-04-25 09:04:37 -04006762 ret = -EIO;
6763 break;
Chris Mason0b86a832008-03-24 15:01:56 -04006764 }
David Sterba1ffb22c2014-10-31 19:02:42 +01006765 array_ptr += len;
6766 sb_array_offset += len;
6767 cur_offset += len;
Chris Mason0b86a832008-03-24 15:01:56 -04006768 }
Liu Bod8651772016-06-03 17:41:42 -07006769 clear_extent_buffer_uptodate(sb);
Liu Bo1c8b5b62016-05-13 17:06:59 -07006770 free_extent_buffer_stale(sb);
Chris Mason84eed902008-04-25 09:04:37 -04006771 return ret;
David Sterbae3540ea2014-11-05 15:24:51 +01006772
6773out_short_read:
6774 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6775 len, cur_offset);
Liu Bod8651772016-06-03 17:41:42 -07006776 clear_extent_buffer_uptodate(sb);
Liu Bo1c8b5b62016-05-13 17:06:59 -07006777 free_extent_buffer_stale(sb);
David Sterbae3540ea2014-11-05 15:24:51 +01006778 return -EIO;
Chris Mason0b86a832008-03-24 15:01:56 -04006779}
6780
6781int btrfs_read_chunk_tree(struct btrfs_root *root)
6782{
6783 struct btrfs_path *path;
6784 struct extent_buffer *leaf;
6785 struct btrfs_key key;
6786 struct btrfs_key found_key;
6787 int ret;
6788 int slot;
Liu Bo99e3ecf2016-06-03 12:05:14 -07006789 u64 total_dev = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006790
6791 root = root->fs_info->chunk_root;
6792
6793 path = btrfs_alloc_path();
6794 if (!path)
6795 return -ENOMEM;
6796
Li Zefanb367e472011-12-07 11:38:24 +08006797 mutex_lock(&uuid_mutex);
6798 lock_chunks(root);
6799
Filipe David Borba Manana395927a2013-07-30 12:03:04 +01006800 /*
6801 * Read all device items, and then all the chunk items. All
6802 * device items are found before any chunk item (their object id
6803 * is smaller than the lowest possible object id for a chunk
6804 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
Chris Mason0b86a832008-03-24 15:01:56 -04006805 */
6806 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6807 key.offset = 0;
6808 key.type = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006809 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Zhao Leiab593812010-03-25 12:34:49 +00006810 if (ret < 0)
6811 goto error;
Chris Masond3977122009-01-05 21:25:51 -05006812 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04006813 leaf = path->nodes[0];
6814 slot = path->slots[0];
6815 if (slot >= btrfs_header_nritems(leaf)) {
6816 ret = btrfs_next_leaf(root, path);
6817 if (ret == 0)
6818 continue;
6819 if (ret < 0)
6820 goto error;
6821 break;
6822 }
6823 btrfs_item_key_to_cpu(leaf, &found_key, slot);
Filipe David Borba Manana395927a2013-07-30 12:03:04 +01006824 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6825 struct btrfs_dev_item *dev_item;
6826 dev_item = btrfs_item_ptr(leaf, slot,
Chris Mason0b86a832008-03-24 15:01:56 -04006827 struct btrfs_dev_item);
Filipe David Borba Manana395927a2013-07-30 12:03:04 +01006828 ret = read_one_dev(root, leaf, dev_item);
6829 if (ret)
6830 goto error;
Liu Bo99e3ecf2016-06-03 12:05:14 -07006831 total_dev++;
Chris Mason0b86a832008-03-24 15:01:56 -04006832 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6833 struct btrfs_chunk *chunk;
6834 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6835 ret = read_one_chunk(root, &found_key, leaf, chunk);
Yan Zheng2b820322008-11-17 21:11:30 -05006836 if (ret)
6837 goto error;
Chris Mason0b86a832008-03-24 15:01:56 -04006838 }
6839 path->slots[0]++;
6840 }
Liu Bo99e3ecf2016-06-03 12:05:14 -07006841
6842 /*
6843 * After loading chunk tree, we've got all device information,
6844 * do another round of validation checks.
6845 */
6846 if (total_dev != root->fs_info->fs_devices->total_devices) {
6847 btrfs_err(root->fs_info,
6848 "super_num_devices %llu mismatch with num_devices %llu found here",
6849 btrfs_super_num_devices(root->fs_info->super_copy),
6850 total_dev);
6851 ret = -EINVAL;
6852 goto error;
6853 }
6854 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6855 root->fs_info->fs_devices->total_rw_bytes) {
6856 btrfs_err(root->fs_info,
6857 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6858 btrfs_super_total_bytes(root->fs_info->super_copy),
6859 root->fs_info->fs_devices->total_rw_bytes);
6860 ret = -EINVAL;
6861 goto error;
6862 }
Chris Mason0b86a832008-03-24 15:01:56 -04006863 ret = 0;
6864error:
Li Zefanb367e472011-12-07 11:38:24 +08006865 unlock_chunks(root);
6866 mutex_unlock(&uuid_mutex);
6867
Yan Zheng2b820322008-11-17 21:11:30 -05006868 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04006869 return ret;
6870}
Stefan Behrens442a4f62012-05-25 16:06:08 +02006871
Miao Xiecb517ea2013-05-15 07:48:19 +00006872void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6873{
6874 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6875 struct btrfs_device *device;
6876
Liu Bo29cc83f2014-05-11 23:14:59 +08006877 while (fs_devices) {
6878 mutex_lock(&fs_devices->device_list_mutex);
6879 list_for_each_entry(device, &fs_devices->devices, dev_list)
6880 device->dev_root = fs_info->dev_root;
6881 mutex_unlock(&fs_devices->device_list_mutex);
6882
6883 fs_devices = fs_devices->seed;
6884 }
Miao Xiecb517ea2013-05-15 07:48:19 +00006885}
6886
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006887static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6888{
6889 int i;
6890
6891 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6892 btrfs_dev_stat_reset(dev, i);
6893}
6894
6895int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6896{
6897 struct btrfs_key key;
6898 struct btrfs_key found_key;
6899 struct btrfs_root *dev_root = fs_info->dev_root;
6900 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6901 struct extent_buffer *eb;
6902 int slot;
6903 int ret = 0;
6904 struct btrfs_device *device;
6905 struct btrfs_path *path = NULL;
6906 int i;
6907
6908 path = btrfs_alloc_path();
6909 if (!path) {
6910 ret = -ENOMEM;
6911 goto out;
6912 }
6913
6914 mutex_lock(&fs_devices->device_list_mutex);
6915 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6916 int item_size;
6917 struct btrfs_dev_stats_item *ptr;
6918
David Sterba242e2952016-01-25 17:51:31 +01006919 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6920 key.type = BTRFS_PERSISTENT_ITEM_KEY;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006921 key.offset = device->devid;
6922 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6923 if (ret) {
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006924 __btrfs_reset_dev_stats(device);
6925 device->dev_stats_valid = 1;
6926 btrfs_release_path(path);
6927 continue;
6928 }
6929 slot = path->slots[0];
6930 eb = path->nodes[0];
6931 btrfs_item_key_to_cpu(eb, &found_key, slot);
6932 item_size = btrfs_item_size_nr(eb, slot);
6933
6934 ptr = btrfs_item_ptr(eb, slot,
6935 struct btrfs_dev_stats_item);
6936
6937 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6938 if (item_size >= (1 + i) * sizeof(__le64))
6939 btrfs_dev_stat_set(device, i,
6940 btrfs_dev_stats_value(eb, ptr, i));
6941 else
6942 btrfs_dev_stat_reset(device, i);
6943 }
6944
6945 device->dev_stats_valid = 1;
6946 btrfs_dev_stat_print_on_load(device);
6947 btrfs_release_path(path);
6948 }
6949 mutex_unlock(&fs_devices->device_list_mutex);
6950
6951out:
6952 btrfs_free_path(path);
6953 return ret < 0 ? ret : 0;
6954}
6955
6956static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6957 struct btrfs_root *dev_root,
6958 struct btrfs_device *device)
6959{
6960 struct btrfs_path *path;
6961 struct btrfs_key key;
6962 struct extent_buffer *eb;
6963 struct btrfs_dev_stats_item *ptr;
6964 int ret;
6965 int i;
6966
David Sterba242e2952016-01-25 17:51:31 +01006967 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6968 key.type = BTRFS_PERSISTENT_ITEM_KEY;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006969 key.offset = device->devid;
6970
6971 path = btrfs_alloc_path();
6972 BUG_ON(!path);
6973 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6974 if (ret < 0) {
David Sterbaecaeb142015-10-08 09:01:03 +02006975 btrfs_warn_in_rcu(dev_root->fs_info,
6976 "error %d while searching for dev_stats item for device %s",
Josef Bacik606686e2012-06-04 14:03:51 -04006977 ret, rcu_str_deref(device->name));
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006978 goto out;
6979 }
6980
6981 if (ret == 0 &&
6982 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6983 /* need to delete old one and insert a new one */
6984 ret = btrfs_del_item(trans, dev_root, path);
6985 if (ret != 0) {
David Sterbaecaeb142015-10-08 09:01:03 +02006986 btrfs_warn_in_rcu(dev_root->fs_info,
6987 "delete too small dev_stats item for device %s failed %d",
Josef Bacik606686e2012-06-04 14:03:51 -04006988 rcu_str_deref(device->name), ret);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02006989 goto out;
6990 }
6991 ret = 1;
6992 }
6993
6994 if (ret == 1) {
6995 /* need to insert a new item */
6996 btrfs_release_path(path);
6997 ret = btrfs_insert_empty_item(trans, dev_root, path,
6998 &key, sizeof(*ptr));
6999 if (ret < 0) {
David Sterbaecaeb142015-10-08 09:01:03 +02007000 btrfs_warn_in_rcu(dev_root->fs_info,
7001 "insert dev_stats item for device %s failed %d",
7002 rcu_str_deref(device->name), ret);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007003 goto out;
7004 }
7005 }
7006
7007 eb = path->nodes[0];
7008 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7009 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7010 btrfs_set_dev_stats_value(eb, ptr, i,
7011 btrfs_dev_stat_read(device, i));
7012 btrfs_mark_buffer_dirty(eb);
7013
7014out:
7015 btrfs_free_path(path);
7016 return ret;
7017}
7018
7019/*
7020 * called from commit_transaction. Writes all changed device stats to disk.
7021 */
7022int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7023 struct btrfs_fs_info *fs_info)
7024{
7025 struct btrfs_root *dev_root = fs_info->dev_root;
7026 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7027 struct btrfs_device *device;
Miao Xieaddc3fa2014-07-24 11:37:11 +08007028 int stats_cnt;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007029 int ret = 0;
7030
7031 mutex_lock(&fs_devices->device_list_mutex);
7032 list_for_each_entry(device, &fs_devices->devices, dev_list) {
Miao Xieaddc3fa2014-07-24 11:37:11 +08007033 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007034 continue;
7035
Miao Xieaddc3fa2014-07-24 11:37:11 +08007036 stats_cnt = atomic_read(&device->dev_stats_ccnt);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007037 ret = update_dev_stat_item(trans, dev_root, device);
7038 if (!ret)
Miao Xieaddc3fa2014-07-24 11:37:11 +08007039 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007040 }
7041 mutex_unlock(&fs_devices->device_list_mutex);
7042
7043 return ret;
7044}
7045
Stefan Behrens442a4f62012-05-25 16:06:08 +02007046void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7047{
7048 btrfs_dev_stat_inc(dev, index);
7049 btrfs_dev_stat_print_on_error(dev);
7050}
7051
Eric Sandeen48a3b632013-04-25 20:41:01 +00007052static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
Stefan Behrens442a4f62012-05-25 16:06:08 +02007053{
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007054 if (!dev->dev_stats_valid)
7055 return;
David Sterbab14af3b2015-10-08 10:43:10 +02007056 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7057 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
Josef Bacik606686e2012-06-04 14:03:51 -04007058 rcu_str_deref(dev->name),
Stefan Behrens442a4f62012-05-25 16:06:08 +02007059 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7060 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7061 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
Frank Holtonefe120a2013-12-20 11:37:06 -05007062 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7063 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
Stefan Behrens442a4f62012-05-25 16:06:08 +02007064}
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007065
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007066static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7067{
Stefan Behrensa98cdb82012-07-17 09:02:11 -06007068 int i;
7069
7070 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7071 if (btrfs_dev_stat_read(dev, i) != 0)
7072 break;
7073 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7074 return; /* all values == 0, suppress message */
7075
David Sterbaecaeb142015-10-08 09:01:03 +02007076 btrfs_info_in_rcu(dev->dev_root->fs_info,
7077 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
Josef Bacik606686e2012-06-04 14:03:51 -04007078 rcu_str_deref(dev->name),
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007079 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7080 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7081 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7082 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7083 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7084}
7085
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007086int btrfs_get_dev_stats(struct btrfs_root *root,
David Sterbab27f7c02012-06-22 06:30:39 -06007087 struct btrfs_ioctl_get_dev_stats *stats)
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007088{
7089 struct btrfs_device *dev;
7090 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7091 int i;
7092
7093 mutex_lock(&fs_devices->device_list_mutex);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01007094 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007095 mutex_unlock(&fs_devices->device_list_mutex);
7096
7097 if (!dev) {
Frank Holtonefe120a2013-12-20 11:37:06 -05007098 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007099 return -ENODEV;
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007100 } else if (!dev->dev_stats_valid) {
Frank Holtonefe120a2013-12-20 11:37:06 -05007101 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
Stefan Behrens733f4fb2012-05-25 16:06:10 +02007102 return -ENODEV;
David Sterbab27f7c02012-06-22 06:30:39 -06007103 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
Stefan Behrensc11d2c22012-05-25 16:06:09 +02007104 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7105 if (stats->nr_items > i)
7106 stats->values[i] =
7107 btrfs_dev_stat_read_and_reset(dev, i);
7108 else
7109 btrfs_dev_stat_reset(dev, i);
7110 }
7111 } else {
7112 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7113 if (stats->nr_items > i)
7114 stats->values[i] = btrfs_dev_stat_read(dev, i);
7115 }
7116 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7117 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7118 return 0;
7119}
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007120
Anand Jain12b1c262015-08-14 18:32:59 +08007121void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007122{
7123 struct buffer_head *bh;
7124 struct btrfs_super_block *disk_super;
Anand Jain12b1c262015-08-14 18:32:59 +08007125 int copy_num;
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007126
Anand Jain12b1c262015-08-14 18:32:59 +08007127 if (!bdev)
7128 return;
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007129
Anand Jain12b1c262015-08-14 18:32:59 +08007130 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7131 copy_num++) {
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007132
Anand Jain12b1c262015-08-14 18:32:59 +08007133 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7134 continue;
7135
7136 disk_super = (struct btrfs_super_block *)bh->b_data;
7137
7138 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7139 set_buffer_dirty(bh);
7140 sync_dirty_buffer(bh);
7141 brelse(bh);
7142 }
7143
7144 /* Notify udev that device has changed */
7145 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7146
7147 /* Update ctime/mtime for device path for libblkid */
7148 update_dev_time(device_path);
Stefan Behrensa8a6dab2012-11-05 15:50:14 +01007149}
Miao Xie935e5cc2014-09-03 21:35:33 +08007150
7151/*
7152 * Update the size of all devices, which is used for writing out the
7153 * super blocks.
7154 */
7155void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7156{
7157 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7158 struct btrfs_device *curr, *next;
7159
7160 if (list_empty(&fs_devices->resized_devices))
7161 return;
7162
7163 mutex_lock(&fs_devices->device_list_mutex);
7164 lock_chunks(fs_info->dev_root);
7165 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7166 resized_list) {
7167 list_del_init(&curr->resized_list);
7168 curr->commit_total_bytes = curr->disk_total_bytes;
7169 }
7170 unlock_chunks(fs_info->dev_root);
7171 mutex_unlock(&fs_devices->device_list_mutex);
7172}
Miao Xiece7213c2014-09-03 21:35:34 +08007173
7174/* Must be invoked during the transaction commit */
7175void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7176 struct btrfs_transaction *transaction)
7177{
7178 struct extent_map *em;
7179 struct map_lookup *map;
7180 struct btrfs_device *dev;
7181 int i;
7182
7183 if (list_empty(&transaction->pending_chunks))
7184 return;
7185
7186 /* In order to kick the device replace finish process */
7187 lock_chunks(root);
7188 list_for_each_entry(em, &transaction->pending_chunks, list) {
Jeff Mahoney95617d62015-06-03 10:55:48 -04007189 map = em->map_lookup;
Miao Xiece7213c2014-09-03 21:35:34 +08007190
7191 for (i = 0; i < map->num_stripes; i++) {
7192 dev = map->stripes[i].dev;
7193 dev->commit_bytes_used = dev->bytes_used;
7194 }
7195 }
7196 unlock_chunks(root);
7197}
Anand Jain5a13f432015-03-10 06:38:31 +08007198
7199void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7200{
7201 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7202 while (fs_devices) {
7203 fs_devices->fs_info = fs_info;
7204 fs_devices = fs_devices->seed;
7205 }
7206}
7207
7208void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7209{
7210 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7211 while (fs_devices) {
7212 fs_devices->fs_info = NULL;
7213 fs_devices = fs_devices->seed;
7214 }
7215}