5f01dad4b696352f211106b4c9bd0024268913b6
[linux-2.6.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/iocontext.h>
24 #include <asm/div64.h>
25 #include "compat.h"
26 #include "ctree.h"
27 #include "extent_map.h"
28 #include "disk-io.h"
29 #include "transaction.h"
30 #include "print-tree.h"
31 #include "volumes.h"
32 #include "async-thread.h"
33
34 struct map_lookup {
35         u64 type;
36         int io_align;
37         int io_width;
38         int stripe_len;
39         int sector_size;
40         int num_stripes;
41         int sub_stripes;
42         struct btrfs_bio_stripe stripes[];
43 };
44
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46                                 struct btrfs_root *root,
47                                 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51                             (sizeof(struct btrfs_bio_stripe) * (n)))
52
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
55
56 void btrfs_lock_volumes(void)
57 {
58         mutex_lock(&uuid_mutex);
59 }
60
61 void btrfs_unlock_volumes(void)
62 {
63         mutex_unlock(&uuid_mutex);
64 }
65
66 static void lock_chunks(struct btrfs_root *root)
67 {
68         mutex_lock(&root->fs_info->chunk_mutex);
69 }
70
71 static void unlock_chunks(struct btrfs_root *root)
72 {
73         mutex_unlock(&root->fs_info->chunk_mutex);
74 }
75
76 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
77 {
78         struct btrfs_device *device;
79         WARN_ON(fs_devices->opened);
80         while (!list_empty(&fs_devices->devices)) {
81                 device = list_entry(fs_devices->devices.next,
82                                     struct btrfs_device, dev_list);
83                 list_del(&device->dev_list);
84                 kfree(device->name);
85                 kfree(device);
86         }
87         kfree(fs_devices);
88 }
89
90 int btrfs_cleanup_fs_uuids(void)
91 {
92         struct btrfs_fs_devices *fs_devices;
93
94         while (!list_empty(&fs_uuids)) {
95                 fs_devices = list_entry(fs_uuids.next,
96                                         struct btrfs_fs_devices, list);
97                 list_del(&fs_devices->list);
98                 free_fs_devices(fs_devices);
99         }
100         return 0;
101 }
102
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
104                                                    u64 devid, u8 *uuid)
105 {
106         struct btrfs_device *dev;
107
108         list_for_each_entry(dev, head, dev_list) {
109                 if (dev->devid == devid &&
110                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
111                         return dev;
112                 }
113         }
114         return NULL;
115 }
116
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
118 {
119         struct btrfs_fs_devices *fs_devices;
120
121         list_for_each_entry(fs_devices, &fs_uuids, list) {
122                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
123                         return fs_devices;
124         }
125         return NULL;
126 }
127
128 static void requeue_list(struct btrfs_pending_bios *pending_bios,
129                         struct bio *head, struct bio *tail)
130 {
131
132         struct bio *old_head;
133
134         old_head = pending_bios->head;
135         pending_bios->head = head;
136         if (pending_bios->tail)
137                 tail->bi_next = old_head;
138         else
139                 pending_bios->tail = tail;
140 }
141
142 /*
143  * we try to collect pending bios for a device so we don't get a large
144  * number of procs sending bios down to the same device.  This greatly
145  * improves the schedulers ability to collect and merge the bios.
146  *
147  * But, it also turns into a long list of bios to process and that is sure
148  * to eventually make the worker thread block.  The solution here is to
149  * make some progress and then put this work struct back at the end of
150  * the list if the block device is congested.  This way, multiple devices
151  * can make progress from a single worker thread.
152  */
153 static noinline int run_scheduled_bios(struct btrfs_device *device)
154 {
155         struct bio *pending;
156         struct backing_dev_info *bdi;
157         struct btrfs_fs_info *fs_info;
158         struct btrfs_pending_bios *pending_bios;
159         struct bio *tail;
160         struct bio *cur;
161         int again = 0;
162         unsigned long num_run;
163         unsigned long num_sync_run;
164         unsigned long limit;
165         unsigned long last_waited = 0;
166
167         bdi = blk_get_backing_dev_info(device->bdev);
168         fs_info = device->dev_root->fs_info;
169         limit = btrfs_async_submit_limit(fs_info);
170         limit = limit * 2 / 3;
171
172         /* we want to make sure that every time we switch from the sync
173          * list to the normal list, we unplug
174          */
175         num_sync_run = 0;
176
177 loop:
178         spin_lock(&device->io_lock);
179         num_run = 0;
180
181 loop_lock:
182
183         /* take all the bios off the list at once and process them
184          * later on (without the lock held).  But, remember the
185          * tail and other pointers so the bios can be properly reinserted
186          * into the list if we hit congestion
187          */
188         if (device->pending_sync_bios.head)
189                 pending_bios = &device->pending_sync_bios;
190         else
191                 pending_bios = &device->pending_bios;
192
193         pending = pending_bios->head;
194         tail = pending_bios->tail;
195         WARN_ON(pending && !tail);
196
197         /*
198          * if pending was null this time around, no bios need processing
199          * at all and we can stop.  Otherwise it'll loop back up again
200          * and do an additional check so no bios are missed.
201          *
202          * device->running_pending is used to synchronize with the
203          * schedule_bio code.
204          */
205         if (device->pending_sync_bios.head == NULL &&
206             device->pending_bios.head == NULL) {
207                 again = 0;
208                 device->running_pending = 0;
209         } else {
210                 again = 1;
211                 device->running_pending = 1;
212         }
213
214         pending_bios->head = NULL;
215         pending_bios->tail = NULL;
216
217         spin_unlock(&device->io_lock);
218
219         /*
220          * if we're doing the regular priority list, make sure we unplug
221          * for any high prio bios we've sent down
222          */
223         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
224                 num_sync_run = 0;
225                 blk_run_backing_dev(bdi, NULL);
226         }
227
228         while (pending) {
229
230                 rmb();
231                 if (pending_bios != &device->pending_sync_bios &&
232                     device->pending_sync_bios.head &&
233                     num_run > 16) {
234                         cond_resched();
235                         spin_lock(&device->io_lock);
236                         requeue_list(pending_bios, pending, tail);
237                         goto loop_lock;
238                 }
239
240                 cur = pending;
241                 pending = pending->bi_next;
242                 cur->bi_next = NULL;
243                 atomic_dec(&fs_info->nr_async_bios);
244
245                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
246                     waitqueue_active(&fs_info->async_submit_wait))
247                         wake_up(&fs_info->async_submit_wait);
248
249                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
250                 submit_bio(cur->bi_rw, cur);
251                 num_run++;
252                 if (bio_sync(cur))
253                         num_sync_run++;
254
255                 if (need_resched()) {
256                         if (num_sync_run) {
257                                 blk_run_backing_dev(bdi, NULL);
258                                 num_sync_run = 0;
259                         }
260                         cond_resched();
261                 }
262
263                 /*
264                  * we made progress, there is more work to do and the bdi
265                  * is now congested.  Back off and let other work structs
266                  * run instead
267                  */
268                 if (pending && bdi_write_congested(bdi) && num_run > 16 &&
269                     fs_info->fs_devices->open_devices > 1) {
270                         struct io_context *ioc;
271
272                         ioc = current->io_context;
273
274                         /*
275                          * the main goal here is that we don't want to
276                          * block if we're going to be able to submit
277                          * more requests without blocking.
278                          *
279                          * This code does two great things, it pokes into
280                          * the elevator code from a filesystem _and_
281                          * it makes assumptions about how batching works.
282                          */
283                         if (ioc && ioc->nr_batch_requests > 0 &&
284                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
285                             (last_waited == 0 ||
286                              ioc->last_waited == last_waited)) {
287                                 /*
288                                  * we want to go through our batch of
289                                  * requests and stop.  So, we copy out
290                                  * the ioc->last_waited time and test
291                                  * against it before looping
292                                  */
293                                 last_waited = ioc->last_waited;
294                                 if (need_resched()) {
295                                         if (num_sync_run) {
296                                                 blk_run_backing_dev(bdi, NULL);
297                                                 num_sync_run = 0;
298                                         }
299                                         cond_resched();
300                                 }
301                                 continue;
302                         }
303                         spin_lock(&device->io_lock);
304                         requeue_list(pending_bios, pending, tail);
305                         device->running_pending = 1;
306
307                         spin_unlock(&device->io_lock);
308                         btrfs_requeue_work(&device->work);
309                         goto done;
310                 }
311         }
312
313         if (num_sync_run) {
314                 num_sync_run = 0;
315                 blk_run_backing_dev(bdi, NULL);
316         }
317
318         cond_resched();
319         if (again)
320                 goto loop;
321
322         spin_lock(&device->io_lock);
323         if (device->pending_bios.head || device->pending_sync_bios.head)
324                 goto loop_lock;
325         spin_unlock(&device->io_lock);
326
327         /*
328          * IO has already been through a long path to get here.  Checksumming,
329          * async helper threads, perhaps compression.  We've done a pretty
330          * good job of collecting a batch of IO and should just unplug
331          * the device right away.
332          *
333          * This will help anyone who is waiting on the IO, they might have
334          * already unplugged, but managed to do so before the bio they
335          * cared about found its way down here.
336          */
337         blk_run_backing_dev(bdi, NULL);
338 done:
339         return 0;
340 }
341
342 static void pending_bios_fn(struct btrfs_work *work)
343 {
344         struct btrfs_device *device;
345
346         device = container_of(work, struct btrfs_device, work);
347         run_scheduled_bios(device);
348 }
349
350 static noinline int device_list_add(const char *path,
351                            struct btrfs_super_block *disk_super,
352                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
353 {
354         struct btrfs_device *device;
355         struct btrfs_fs_devices *fs_devices;
356         u64 found_transid = btrfs_super_generation(disk_super);
357
358         fs_devices = find_fsid(disk_super->fsid);
359         if (!fs_devices) {
360                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
361                 if (!fs_devices)
362                         return -ENOMEM;
363                 INIT_LIST_HEAD(&fs_devices->devices);
364                 INIT_LIST_HEAD(&fs_devices->alloc_list);
365                 list_add(&fs_devices->list, &fs_uuids);
366                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
367                 fs_devices->latest_devid = devid;
368                 fs_devices->latest_trans = found_transid;
369                 device = NULL;
370         } else {
371                 device = __find_device(&fs_devices->devices, devid,
372                                        disk_super->dev_item.uuid);
373         }
374         if (!device) {
375                 if (fs_devices->opened)
376                         return -EBUSY;
377
378                 device = kzalloc(sizeof(*device), GFP_NOFS);
379                 if (!device) {
380                         /* we can safely leave the fs_devices entry around */
381                         return -ENOMEM;
382                 }
383                 device->devid = devid;
384                 device->work.func = pending_bios_fn;
385                 memcpy(device->uuid, disk_super->dev_item.uuid,
386                        BTRFS_UUID_SIZE);
387                 device->barriers = 1;
388                 spin_lock_init(&device->io_lock);
389                 device->name = kstrdup(path, GFP_NOFS);
390                 if (!device->name) {
391                         kfree(device);
392                         return -ENOMEM;
393                 }
394                 INIT_LIST_HEAD(&device->dev_alloc_list);
395                 list_add(&device->dev_list, &fs_devices->devices);
396                 device->fs_devices = fs_devices;
397                 fs_devices->num_devices++;
398         }
399
400         if (found_transid > fs_devices->latest_trans) {
401                 fs_devices->latest_devid = devid;
402                 fs_devices->latest_trans = found_transid;
403         }
404         *fs_devices_ret = fs_devices;
405         return 0;
406 }
407
408 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
409 {
410         struct btrfs_fs_devices *fs_devices;
411         struct btrfs_device *device;
412         struct btrfs_device *orig_dev;
413
414         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
415         if (!fs_devices)
416                 return ERR_PTR(-ENOMEM);
417
418         INIT_LIST_HEAD(&fs_devices->devices);
419         INIT_LIST_HEAD(&fs_devices->alloc_list);
420         INIT_LIST_HEAD(&fs_devices->list);
421         fs_devices->latest_devid = orig->latest_devid;
422         fs_devices->latest_trans = orig->latest_trans;
423         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
424
425         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
426                 device = kzalloc(sizeof(*device), GFP_NOFS);
427                 if (!device)
428                         goto error;
429
430                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
431                 if (!device->name)
432                         goto error;
433
434                 device->devid = orig_dev->devid;
435                 device->work.func = pending_bios_fn;
436                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
437                 device->barriers = 1;
438                 spin_lock_init(&device->io_lock);
439                 INIT_LIST_HEAD(&device->dev_list);
440                 INIT_LIST_HEAD(&device->dev_alloc_list);
441
442                 list_add(&device->dev_list, &fs_devices->devices);
443                 device->fs_devices = fs_devices;
444                 fs_devices->num_devices++;
445         }
446         return fs_devices;
447 error:
448         free_fs_devices(fs_devices);
449         return ERR_PTR(-ENOMEM);
450 }
451
452 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
453 {
454         struct btrfs_device *device, *next;
455
456         mutex_lock(&uuid_mutex);
457 again:
458         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
459                 if (device->in_fs_metadata)
460                         continue;
461
462                 if (device->bdev) {
463                         close_bdev_exclusive(device->bdev, device->mode);
464                         device->bdev = NULL;
465                         fs_devices->open_devices--;
466                 }
467                 if (device->writeable) {
468                         list_del_init(&device->dev_alloc_list);
469                         device->writeable = 0;
470                         fs_devices->rw_devices--;
471                 }
472                 list_del_init(&device->dev_list);
473                 fs_devices->num_devices--;
474                 kfree(device->name);
475                 kfree(device);
476         }
477
478         if (fs_devices->seed) {
479                 fs_devices = fs_devices->seed;
480                 goto again;
481         }
482
483         mutex_unlock(&uuid_mutex);
484         return 0;
485 }
486
487 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
488 {
489         struct btrfs_device *device;
490
491         if (--fs_devices->opened > 0)
492                 return 0;
493
494         list_for_each_entry(device, &fs_devices->devices, dev_list) {
495                 if (device->bdev) {
496                         close_bdev_exclusive(device->bdev, device->mode);
497                         fs_devices->open_devices--;
498                 }
499                 if (device->writeable) {
500                         list_del_init(&device->dev_alloc_list);
501                         fs_devices->rw_devices--;
502                 }
503
504                 device->bdev = NULL;
505                 device->writeable = 0;
506                 device->in_fs_metadata = 0;
507         }
508         WARN_ON(fs_devices->open_devices);
509         WARN_ON(fs_devices->rw_devices);
510         fs_devices->opened = 0;
511         fs_devices->seeding = 0;
512
513         return 0;
514 }
515
516 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
517 {
518         struct btrfs_fs_devices *seed_devices = NULL;
519         int ret;
520
521         mutex_lock(&uuid_mutex);
522         ret = __btrfs_close_devices(fs_devices);
523         if (!fs_devices->opened) {
524                 seed_devices = fs_devices->seed;
525                 fs_devices->seed = NULL;
526         }
527         mutex_unlock(&uuid_mutex);
528
529         while (seed_devices) {
530                 fs_devices = seed_devices;
531                 seed_devices = fs_devices->seed;
532                 __btrfs_close_devices(fs_devices);
533                 free_fs_devices(fs_devices);
534         }
535         return ret;
536 }
537
538 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
539                                 fmode_t flags, void *holder)
540 {
541         struct block_device *bdev;
542         struct list_head *head = &fs_devices->devices;
543         struct btrfs_device *device;
544         struct block_device *latest_bdev = NULL;
545         struct buffer_head *bh;
546         struct btrfs_super_block *disk_super;
547         u64 latest_devid = 0;
548         u64 latest_transid = 0;
549         u64 devid;
550         int seeding = 1;
551         int ret = 0;
552
553         list_for_each_entry(device, head, dev_list) {
554                 if (device->bdev)
555                         continue;
556                 if (!device->name)
557                         continue;
558
559                 bdev = open_bdev_exclusive(device->name, flags, holder);
560                 if (IS_ERR(bdev)) {
561                         printk(KERN_INFO "open %s failed\n", device->name);
562                         goto error;
563                 }
564                 set_blocksize(bdev, 4096);
565
566                 bh = btrfs_read_dev_super(bdev);
567                 if (!bh)
568                         goto error_close;
569
570                 disk_super = (struct btrfs_super_block *)bh->b_data;
571                 devid = le64_to_cpu(disk_super->dev_item.devid);
572                 if (devid != device->devid)
573                         goto error_brelse;
574
575                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
576                            BTRFS_UUID_SIZE))
577                         goto error_brelse;
578
579                 device->generation = btrfs_super_generation(disk_super);
580                 if (!latest_transid || device->generation > latest_transid) {
581                         latest_devid = devid;
582                         latest_transid = device->generation;
583                         latest_bdev = bdev;
584                 }
585
586                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
587                         device->writeable = 0;
588                 } else {
589                         device->writeable = !bdev_read_only(bdev);
590                         seeding = 0;
591                 }
592
593                 device->bdev = bdev;
594                 device->in_fs_metadata = 0;
595                 device->mode = flags;
596
597                 fs_devices->open_devices++;
598                 if (device->writeable) {
599                         fs_devices->rw_devices++;
600                         list_add(&device->dev_alloc_list,
601                                  &fs_devices->alloc_list);
602                 }
603                 continue;
604
605 error_brelse:
606                 brelse(bh);
607 error_close:
608                 close_bdev_exclusive(bdev, FMODE_READ);
609 error:
610                 continue;
611         }
612         if (fs_devices->open_devices == 0) {
613                 ret = -EIO;
614                 goto out;
615         }
616         fs_devices->seeding = seeding;
617         fs_devices->opened = 1;
618         fs_devices->latest_bdev = latest_bdev;
619         fs_devices->latest_devid = latest_devid;
620         fs_devices->latest_trans = latest_transid;
621         fs_devices->total_rw_bytes = 0;
622 out:
623         return ret;
624 }
625
626 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
627                        fmode_t flags, void *holder)
628 {
629         int ret;
630
631         mutex_lock(&uuid_mutex);
632         if (fs_devices->opened) {
633                 fs_devices->opened++;
634                 ret = 0;
635         } else {
636                 ret = __btrfs_open_devices(fs_devices, flags, holder);
637         }
638         mutex_unlock(&uuid_mutex);
639         return ret;
640 }
641
642 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
643                           struct btrfs_fs_devices **fs_devices_ret)
644 {
645         struct btrfs_super_block *disk_super;
646         struct block_device *bdev;
647         struct buffer_head *bh;
648         int ret;
649         u64 devid;
650         u64 transid;
651
652         mutex_lock(&uuid_mutex);
653
654         bdev = open_bdev_exclusive(path, flags, holder);
655
656         if (IS_ERR(bdev)) {
657                 ret = PTR_ERR(bdev);
658                 goto error;
659         }
660
661         ret = set_blocksize(bdev, 4096);
662         if (ret)
663                 goto error_close;
664         bh = btrfs_read_dev_super(bdev);
665         if (!bh) {
666                 ret = -EIO;
667                 goto error_close;
668         }
669         disk_super = (struct btrfs_super_block *)bh->b_data;
670         devid = le64_to_cpu(disk_super->dev_item.devid);
671         transid = btrfs_super_generation(disk_super);
672         if (disk_super->label[0])
673                 printk(KERN_INFO "device label %s ", disk_super->label);
674         else {
675                 /* FIXME, make a readl uuid parser */
676                 printk(KERN_INFO "device fsid %llx-%llx ",
677                        *(unsigned long long *)disk_super->fsid,
678                        *(unsigned long long *)(disk_super->fsid + 8));
679         }
680         printk(KERN_CONT "devid %llu transid %llu %s\n",
681                (unsigned long long)devid, (unsigned long long)transid, path);
682         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
683
684         brelse(bh);
685 error_close:
686         close_bdev_exclusive(bdev, flags);
687 error:
688         mutex_unlock(&uuid_mutex);
689         return ret;
690 }
691
692 /*
693  * this uses a pretty simple search, the expectation is that it is
694  * called very infrequently and that a given device has a small number
695  * of extents
696  */
697 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
698                                          struct btrfs_device *device,
699                                          u64 num_bytes, u64 *start)
700 {
701         struct btrfs_key key;
702         struct btrfs_root *root = device->dev_root;
703         struct btrfs_dev_extent *dev_extent = NULL;
704         struct btrfs_path *path;
705         u64 hole_size = 0;
706         u64 last_byte = 0;
707         u64 search_start = 0;
708         u64 search_end = device->total_bytes;
709         int ret;
710         int slot = 0;
711         int start_found;
712         struct extent_buffer *l;
713
714         path = btrfs_alloc_path();
715         if (!path)
716                 return -ENOMEM;
717         path->reada = 2;
718         start_found = 0;
719
720         /* FIXME use last free of some kind */
721
722         /* we don't want to overwrite the superblock on the drive,
723          * so we make sure to start at an offset of at least 1MB
724          */
725         search_start = max((u64)1024 * 1024, search_start);
726
727         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
728                 search_start = max(root->fs_info->alloc_start, search_start);
729
730         key.objectid = device->devid;
731         key.offset = search_start;
732         key.type = BTRFS_DEV_EXTENT_KEY;
733         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
734         if (ret < 0)
735                 goto error;
736         ret = btrfs_previous_item(root, path, 0, key.type);
737         if (ret < 0)
738                 goto error;
739         l = path->nodes[0];
740         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
741         while (1) {
742                 l = path->nodes[0];
743                 slot = path->slots[0];
744                 if (slot >= btrfs_header_nritems(l)) {
745                         ret = btrfs_next_leaf(root, path);
746                         if (ret == 0)
747                                 continue;
748                         if (ret < 0)
749                                 goto error;
750 no_more_items:
751                         if (!start_found) {
752                                 if (search_start >= search_end) {
753                                         ret = -ENOSPC;
754                                         goto error;
755                                 }
756                                 *start = search_start;
757                                 start_found = 1;
758                                 goto check_pending;
759                         }
760                         *start = last_byte > search_start ?
761                                 last_byte : search_start;
762                         if (search_end <= *start) {
763                                 ret = -ENOSPC;
764                                 goto error;
765                         }
766                         goto check_pending;
767                 }
768                 btrfs_item_key_to_cpu(l, &key, slot);
769
770                 if (key.objectid < device->devid)
771                         goto next;
772
773                 if (key.objectid > device->devid)
774                         goto no_more_items;
775
776                 if (key.offset >= search_start && key.offset > last_byte &&
777                     start_found) {
778                         if (last_byte < search_start)
779                                 last_byte = search_start;
780                         hole_size = key.offset - last_byte;
781                         if (key.offset > last_byte &&
782                             hole_size >= num_bytes) {
783                                 *start = last_byte;
784                                 goto check_pending;
785                         }
786                 }
787                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
788                         goto next;
789
790                 start_found = 1;
791                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
792                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
793 next:
794                 path->slots[0]++;
795                 cond_resched();
796         }
797 check_pending:
798         /* we have to make sure we didn't find an extent that has already
799          * been allocated by the map tree or the original allocation
800          */
801         BUG_ON(*start < search_start);
802
803         if (*start + num_bytes > search_end) {
804                 ret = -ENOSPC;
805                 goto error;
806         }
807         /* check for pending inserts here */
808         ret = 0;
809
810 error:
811         btrfs_free_path(path);
812         return ret;
813 }
814
815 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
816                           struct btrfs_device *device,
817                           u64 start)
818 {
819         int ret;
820         struct btrfs_path *path;
821         struct btrfs_root *root = device->dev_root;
822         struct btrfs_key key;
823         struct btrfs_key found_key;
824         struct extent_buffer *leaf = NULL;
825         struct btrfs_dev_extent *extent = NULL;
826
827         path = btrfs_alloc_path();
828         if (!path)
829                 return -ENOMEM;
830
831         key.objectid = device->devid;
832         key.offset = start;
833         key.type = BTRFS_DEV_EXTENT_KEY;
834
835         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
836         if (ret > 0) {
837                 ret = btrfs_previous_item(root, path, key.objectid,
838                                           BTRFS_DEV_EXTENT_KEY);
839                 BUG_ON(ret);
840                 leaf = path->nodes[0];
841                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
842                 extent = btrfs_item_ptr(leaf, path->slots[0],
843                                         struct btrfs_dev_extent);
844                 BUG_ON(found_key.offset > start || found_key.offset +
845                        btrfs_dev_extent_length(leaf, extent) < start);
846                 ret = 0;
847         } else if (ret == 0) {
848                 leaf = path->nodes[0];
849                 extent = btrfs_item_ptr(leaf, path->slots[0],
850                                         struct btrfs_dev_extent);
851         }
852         BUG_ON(ret);
853
854         if (device->bytes_used > 0)
855                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
856         ret = btrfs_del_item(trans, root, path);
857         BUG_ON(ret);
858
859         btrfs_free_path(path);
860         return ret;
861 }
862
863 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
864                            struct btrfs_device *device,
865                            u64 chunk_tree, u64 chunk_objectid,
866                            u64 chunk_offset, u64 start, u64 num_bytes)
867 {
868         int ret;
869         struct btrfs_path *path;
870         struct btrfs_root *root = device->dev_root;
871         struct btrfs_dev_extent *extent;
872         struct extent_buffer *leaf;
873         struct btrfs_key key;
874
875         WARN_ON(!device->in_fs_metadata);
876         path = btrfs_alloc_path();
877         if (!path)
878                 return -ENOMEM;
879
880         key.objectid = device->devid;
881         key.offset = start;
882         key.type = BTRFS_DEV_EXTENT_KEY;
883         ret = btrfs_insert_empty_item(trans, root, path, &key,
884                                       sizeof(*extent));
885         BUG_ON(ret);
886
887         leaf = path->nodes[0];
888         extent = btrfs_item_ptr(leaf, path->slots[0],
889                                 struct btrfs_dev_extent);
890         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
891         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
892         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
893
894         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
895                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
896                     BTRFS_UUID_SIZE);
897
898         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
899         btrfs_mark_buffer_dirty(leaf);
900         btrfs_free_path(path);
901         return ret;
902 }
903
904 static noinline int find_next_chunk(struct btrfs_root *root,
905                                     u64 objectid, u64 *offset)
906 {
907         struct btrfs_path *path;
908         int ret;
909         struct btrfs_key key;
910         struct btrfs_chunk *chunk;
911         struct btrfs_key found_key;
912
913         path = btrfs_alloc_path();
914         BUG_ON(!path);
915
916         key.objectid = objectid;
917         key.offset = (u64)-1;
918         key.type = BTRFS_CHUNK_ITEM_KEY;
919
920         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
921         if (ret < 0)
922                 goto error;
923
924         BUG_ON(ret == 0);
925
926         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
927         if (ret) {
928                 *offset = 0;
929         } else {
930                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
931                                       path->slots[0]);
932                 if (found_key.objectid != objectid)
933                         *offset = 0;
934                 else {
935                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
936                                                struct btrfs_chunk);
937                         *offset = found_key.offset +
938                                 btrfs_chunk_length(path->nodes[0], chunk);
939                 }
940         }
941         ret = 0;
942 error:
943         btrfs_free_path(path);
944         return ret;
945 }
946
947 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
948 {
949         int ret;
950         struct btrfs_key key;
951         struct btrfs_key found_key;
952         struct btrfs_path *path;
953
954         root = root->fs_info->chunk_root;
955
956         path = btrfs_alloc_path();
957         if (!path)
958                 return -ENOMEM;
959
960         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
961         key.type = BTRFS_DEV_ITEM_KEY;
962         key.offset = (u64)-1;
963
964         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
965         if (ret < 0)
966                 goto error;
967
968         BUG_ON(ret == 0);
969
970         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
971                                   BTRFS_DEV_ITEM_KEY);
972         if (ret) {
973                 *objectid = 1;
974         } else {
975                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
976                                       path->slots[0]);
977                 *objectid = found_key.offset + 1;
978         }
979         ret = 0;
980 error:
981         btrfs_free_path(path);
982         return ret;
983 }
984
985 /*
986  * the device information is stored in the chunk root
987  * the btrfs_device struct should be fully filled in
988  */
989 int btrfs_add_device(struct btrfs_trans_handle *trans,
990                      struct btrfs_root *root,
991                      struct btrfs_device *device)
992 {
993         int ret;
994         struct btrfs_path *path;
995         struct btrfs_dev_item *dev_item;
996         struct extent_buffer *leaf;
997         struct btrfs_key key;
998         unsigned long ptr;
999
1000         root = root->fs_info->chunk_root;
1001
1002         path = btrfs_alloc_path();
1003         if (!path)
1004                 return -ENOMEM;
1005
1006         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1007         key.type = BTRFS_DEV_ITEM_KEY;
1008         key.offset = device->devid;
1009
1010         ret = btrfs_insert_empty_item(trans, root, path, &key,
1011                                       sizeof(*dev_item));
1012         if (ret)
1013                 goto out;
1014
1015         leaf = path->nodes[0];
1016         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1017
1018         btrfs_set_device_id(leaf, dev_item, device->devid);
1019         btrfs_set_device_generation(leaf, dev_item, 0);
1020         btrfs_set_device_type(leaf, dev_item, device->type);
1021         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1022         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1023         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1024         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1025         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1026         btrfs_set_device_group(leaf, dev_item, 0);
1027         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1028         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1029         btrfs_set_device_start_offset(leaf, dev_item, 0);
1030
1031         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1032         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1033         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1034         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1035         btrfs_mark_buffer_dirty(leaf);
1036
1037         ret = 0;
1038 out:
1039         btrfs_free_path(path);
1040         return ret;
1041 }
1042
1043 static int btrfs_rm_dev_item(struct btrfs_root *root,
1044                              struct btrfs_device *device)
1045 {
1046         int ret;
1047         struct btrfs_path *path;
1048         struct btrfs_key key;
1049         struct btrfs_trans_handle *trans;
1050
1051         root = root->fs_info->chunk_root;
1052
1053         path = btrfs_alloc_path();
1054         if (!path)
1055                 return -ENOMEM;
1056
1057         trans = btrfs_start_transaction(root, 1);
1058         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1059         key.type = BTRFS_DEV_ITEM_KEY;
1060         key.offset = device->devid;
1061         lock_chunks(root);
1062
1063         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064         if (ret < 0)
1065                 goto out;
1066
1067         if (ret > 0) {
1068                 ret = -ENOENT;
1069                 goto out;
1070         }
1071
1072         ret = btrfs_del_item(trans, root, path);
1073         if (ret)
1074                 goto out;
1075 out:
1076         btrfs_free_path(path);
1077         unlock_chunks(root);
1078         btrfs_commit_transaction(trans, root);
1079         return ret;
1080 }
1081
1082 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1083 {
1084         struct btrfs_device *device;
1085         struct btrfs_device *next_device;
1086         struct block_device *bdev;
1087         struct buffer_head *bh = NULL;
1088         struct btrfs_super_block *disk_super;
1089         u64 all_avail;
1090         u64 devid;
1091         u64 num_devices;
1092         u8 *dev_uuid;
1093         int ret = 0;
1094
1095         mutex_lock(&uuid_mutex);
1096         mutex_lock(&root->fs_info->volume_mutex);
1097
1098         all_avail = root->fs_info->avail_data_alloc_bits |
1099                 root->fs_info->avail_system_alloc_bits |
1100                 root->fs_info->avail_metadata_alloc_bits;
1101
1102         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1103             root->fs_info->fs_devices->rw_devices <= 4) {
1104                 printk(KERN_ERR "btrfs: unable to go below four devices "
1105                        "on raid10\n");
1106                 ret = -EINVAL;
1107                 goto out;
1108         }
1109
1110         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1111             root->fs_info->fs_devices->rw_devices <= 2) {
1112                 printk(KERN_ERR "btrfs: unable to go below two "
1113                        "devices on raid1\n");
1114                 ret = -EINVAL;
1115                 goto out;
1116         }
1117
1118         if (strcmp(device_path, "missing") == 0) {
1119                 struct list_head *devices;
1120                 struct btrfs_device *tmp;
1121
1122                 device = NULL;
1123                 devices = &root->fs_info->fs_devices->devices;
1124                 list_for_each_entry(tmp, devices, dev_list) {
1125                         if (tmp->in_fs_metadata && !tmp->bdev) {
1126                                 device = tmp;
1127                                 break;
1128                         }
1129                 }
1130                 bdev = NULL;
1131                 bh = NULL;
1132                 disk_super = NULL;
1133                 if (!device) {
1134                         printk(KERN_ERR "btrfs: no missing devices found to "
1135                                "remove\n");
1136                         goto out;
1137                 }
1138         } else {
1139                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1140                                       root->fs_info->bdev_holder);
1141                 if (IS_ERR(bdev)) {
1142                         ret = PTR_ERR(bdev);
1143                         goto out;
1144                 }
1145
1146                 set_blocksize(bdev, 4096);
1147                 bh = btrfs_read_dev_super(bdev);
1148                 if (!bh) {
1149                         ret = -EIO;
1150                         goto error_close;
1151                 }
1152                 disk_super = (struct btrfs_super_block *)bh->b_data;
1153                 devid = le64_to_cpu(disk_super->dev_item.devid);
1154                 dev_uuid = disk_super->dev_item.uuid;
1155                 device = btrfs_find_device(root, devid, dev_uuid,
1156                                            disk_super->fsid);
1157                 if (!device) {
1158                         ret = -ENOENT;
1159                         goto error_brelse;
1160                 }
1161         }
1162
1163         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1164                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1165                        "device\n");
1166                 ret = -EINVAL;
1167                 goto error_brelse;
1168         }
1169
1170         if (device->writeable) {
1171                 list_del_init(&device->dev_alloc_list);
1172                 root->fs_info->fs_devices->rw_devices--;
1173         }
1174
1175         ret = btrfs_shrink_device(device, 0);
1176         if (ret)
1177                 goto error_brelse;
1178
1179         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1180         if (ret)
1181                 goto error_brelse;
1182
1183         device->in_fs_metadata = 0;
1184         list_del_init(&device->dev_list);
1185         device->fs_devices->num_devices--;
1186
1187         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1188                                  struct btrfs_device, dev_list);
1189         if (device->bdev == root->fs_info->sb->s_bdev)
1190                 root->fs_info->sb->s_bdev = next_device->bdev;
1191         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1192                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1193
1194         if (device->bdev) {
1195                 close_bdev_exclusive(device->bdev, device->mode);
1196                 device->bdev = NULL;
1197                 device->fs_devices->open_devices--;
1198         }
1199
1200         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1201         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1202
1203         if (device->fs_devices->open_devices == 0) {
1204                 struct btrfs_fs_devices *fs_devices;
1205                 fs_devices = root->fs_info->fs_devices;
1206                 while (fs_devices) {
1207                         if (fs_devices->seed == device->fs_devices)
1208                                 break;
1209                         fs_devices = fs_devices->seed;
1210                 }
1211                 fs_devices->seed = device->fs_devices->seed;
1212                 device->fs_devices->seed = NULL;
1213                 __btrfs_close_devices(device->fs_devices);
1214                 free_fs_devices(device->fs_devices);
1215         }
1216
1217         /*
1218          * at this point, the device is zero sized.  We want to
1219          * remove it from the devices list and zero out the old super
1220          */
1221         if (device->writeable) {
1222                 /* make sure this device isn't detected as part of
1223                  * the FS anymore
1224                  */
1225                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1226                 set_buffer_dirty(bh);
1227                 sync_dirty_buffer(bh);
1228         }
1229
1230         kfree(device->name);
1231         kfree(device);
1232         ret = 0;
1233
1234 error_brelse:
1235         brelse(bh);
1236 error_close:
1237         if (bdev)
1238                 close_bdev_exclusive(bdev, FMODE_READ);
1239 out:
1240         mutex_unlock(&root->fs_info->volume_mutex);
1241         mutex_unlock(&uuid_mutex);
1242         return ret;
1243 }
1244
1245 /*
1246  * does all the dirty work required for changing file system's UUID.
1247  */
1248 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1249                                 struct btrfs_root *root)
1250 {
1251         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1252         struct btrfs_fs_devices *old_devices;
1253         struct btrfs_fs_devices *seed_devices;
1254         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1255         struct btrfs_device *device;
1256         u64 super_flags;
1257
1258         BUG_ON(!mutex_is_locked(&uuid_mutex));
1259         if (!fs_devices->seeding)
1260                 return -EINVAL;
1261
1262         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1263         if (!seed_devices)
1264                 return -ENOMEM;
1265
1266         old_devices = clone_fs_devices(fs_devices);
1267         if (IS_ERR(old_devices)) {
1268                 kfree(seed_devices);
1269                 return PTR_ERR(old_devices);
1270         }
1271
1272         list_add(&old_devices->list, &fs_uuids);
1273
1274         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1275         seed_devices->opened = 1;
1276         INIT_LIST_HEAD(&seed_devices->devices);
1277         INIT_LIST_HEAD(&seed_devices->alloc_list);
1278         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1279         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1280         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1281                 device->fs_devices = seed_devices;
1282         }
1283
1284         fs_devices->seeding = 0;
1285         fs_devices->num_devices = 0;
1286         fs_devices->open_devices = 0;
1287         fs_devices->seed = seed_devices;
1288
1289         generate_random_uuid(fs_devices->fsid);
1290         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1291         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1292         super_flags = btrfs_super_flags(disk_super) &
1293                       ~BTRFS_SUPER_FLAG_SEEDING;
1294         btrfs_set_super_flags(disk_super, super_flags);
1295
1296         return 0;
1297 }
1298
1299 /*
1300  * strore the expected generation for seed devices in device items.
1301  */
1302 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1303                                struct btrfs_root *root)
1304 {
1305         struct btrfs_path *path;
1306         struct extent_buffer *leaf;
1307         struct btrfs_dev_item *dev_item;
1308         struct btrfs_device *device;
1309         struct btrfs_key key;
1310         u8 fs_uuid[BTRFS_UUID_SIZE];
1311         u8 dev_uuid[BTRFS_UUID_SIZE];
1312         u64 devid;
1313         int ret;
1314
1315         path = btrfs_alloc_path();
1316         if (!path)
1317                 return -ENOMEM;
1318
1319         root = root->fs_info->chunk_root;
1320         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1321         key.offset = 0;
1322         key.type = BTRFS_DEV_ITEM_KEY;
1323
1324         while (1) {
1325                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1326                 if (ret < 0)
1327                         goto error;
1328
1329                 leaf = path->nodes[0];
1330 next_slot:
1331                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332                         ret = btrfs_next_leaf(root, path);
1333                         if (ret > 0)
1334                                 break;
1335                         if (ret < 0)
1336                                 goto error;
1337                         leaf = path->nodes[0];
1338                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339                         btrfs_release_path(root, path);
1340                         continue;
1341                 }
1342
1343                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1344                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1345                     key.type != BTRFS_DEV_ITEM_KEY)
1346                         break;
1347
1348                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1349                                           struct btrfs_dev_item);
1350                 devid = btrfs_device_id(leaf, dev_item);
1351                 read_extent_buffer(leaf, dev_uuid,
1352                                    (unsigned long)btrfs_device_uuid(dev_item),
1353                                    BTRFS_UUID_SIZE);
1354                 read_extent_buffer(leaf, fs_uuid,
1355                                    (unsigned long)btrfs_device_fsid(dev_item),
1356                                    BTRFS_UUID_SIZE);
1357                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1358                 BUG_ON(!device);
1359
1360                 if (device->fs_devices->seeding) {
1361                         btrfs_set_device_generation(leaf, dev_item,
1362                                                     device->generation);
1363                         btrfs_mark_buffer_dirty(leaf);
1364                 }
1365
1366                 path->slots[0]++;
1367                 goto next_slot;
1368         }
1369         ret = 0;
1370 error:
1371         btrfs_free_path(path);
1372         return ret;
1373 }
1374
1375 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1376 {
1377         struct btrfs_trans_handle *trans;
1378         struct btrfs_device *device;
1379         struct block_device *bdev;
1380         struct list_head *devices;
1381         struct super_block *sb = root->fs_info->sb;
1382         u64 total_bytes;
1383         int seeding_dev = 0;
1384         int ret = 0;
1385
1386         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1387                 return -EINVAL;
1388
1389         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1390         if (!bdev)
1391                 return -EIO;
1392
1393         if (root->fs_info->fs_devices->seeding) {
1394                 seeding_dev = 1;
1395                 down_write(&sb->s_umount);
1396                 mutex_lock(&uuid_mutex);
1397         }
1398
1399         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1400         mutex_lock(&root->fs_info->volume_mutex);
1401
1402         devices = &root->fs_info->fs_devices->devices;
1403         list_for_each_entry(device, devices, dev_list) {
1404                 if (device->bdev == bdev) {
1405                         ret = -EEXIST;
1406                         goto error;
1407                 }
1408         }
1409
1410         device = kzalloc(sizeof(*device), GFP_NOFS);
1411         if (!device) {
1412                 /* we can safely leave the fs_devices entry around */
1413                 ret = -ENOMEM;
1414                 goto error;
1415         }
1416
1417         device->name = kstrdup(device_path, GFP_NOFS);
1418         if (!device->name) {
1419                 kfree(device);
1420                 ret = -ENOMEM;
1421                 goto error;
1422         }
1423
1424         ret = find_next_devid(root, &device->devid);
1425         if (ret) {
1426                 kfree(device);
1427                 goto error;
1428         }
1429
1430         trans = btrfs_start_transaction(root, 1);
1431         lock_chunks(root);
1432
1433         device->barriers = 1;
1434         device->writeable = 1;
1435         device->work.func = pending_bios_fn;
1436         generate_random_uuid(device->uuid);
1437         spin_lock_init(&device->io_lock);
1438         device->generation = trans->transid;
1439         device->io_width = root->sectorsize;
1440         device->io_align = root->sectorsize;
1441         device->sector_size = root->sectorsize;
1442         device->total_bytes = i_size_read(bdev->bd_inode);
1443         device->dev_root = root->fs_info->dev_root;
1444         device->bdev = bdev;
1445         device->in_fs_metadata = 1;
1446         device->mode = 0;
1447         set_blocksize(device->bdev, 4096);
1448
1449         if (seeding_dev) {
1450                 sb->s_flags &= ~MS_RDONLY;
1451                 ret = btrfs_prepare_sprout(trans, root);
1452                 BUG_ON(ret);
1453         }
1454
1455         device->fs_devices = root->fs_info->fs_devices;
1456         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1457         list_add(&device->dev_alloc_list,
1458                  &root->fs_info->fs_devices->alloc_list);
1459         root->fs_info->fs_devices->num_devices++;
1460         root->fs_info->fs_devices->open_devices++;
1461         root->fs_info->fs_devices->rw_devices++;
1462         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1463
1464         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1465         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1466                                     total_bytes + device->total_bytes);
1467
1468         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1469         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1470                                     total_bytes + 1);
1471
1472         if (seeding_dev) {
1473                 ret = init_first_rw_device(trans, root, device);
1474                 BUG_ON(ret);
1475                 ret = btrfs_finish_sprout(trans, root);
1476                 BUG_ON(ret);
1477         } else {
1478                 ret = btrfs_add_device(trans, root, device);
1479         }
1480
1481         /*
1482          * we've got more storage, clear any full flags on the space
1483          * infos
1484          */
1485         btrfs_clear_space_info_full(root->fs_info);
1486
1487         unlock_chunks(root);
1488         btrfs_commit_transaction(trans, root);
1489
1490         if (seeding_dev) {
1491                 mutex_unlock(&uuid_mutex);
1492                 up_write(&sb->s_umount);
1493
1494                 ret = btrfs_relocate_sys_chunks(root);
1495                 BUG_ON(ret);
1496         }
1497 out:
1498         mutex_unlock(&root->fs_info->volume_mutex);
1499         return ret;
1500 error:
1501         close_bdev_exclusive(bdev, 0);
1502         if (seeding_dev) {
1503                 mutex_unlock(&uuid_mutex);
1504                 up_write(&sb->s_umount);
1505         }
1506         goto out;
1507 }
1508
1509 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1510                                         struct btrfs_device *device)
1511 {
1512         int ret;
1513         struct btrfs_path *path;
1514         struct btrfs_root *root;
1515         struct btrfs_dev_item *dev_item;
1516         struct extent_buffer *leaf;
1517         struct btrfs_key key;
1518
1519         root = device->dev_root->fs_info->chunk_root;
1520
1521         path = btrfs_alloc_path();
1522         if (!path)
1523                 return -ENOMEM;
1524
1525         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1526         key.type = BTRFS_DEV_ITEM_KEY;
1527         key.offset = device->devid;
1528
1529         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1530         if (ret < 0)
1531                 goto out;
1532
1533         if (ret > 0) {
1534                 ret = -ENOENT;
1535                 goto out;
1536         }
1537
1538         leaf = path->nodes[0];
1539         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1540
1541         btrfs_set_device_id(leaf, dev_item, device->devid);
1542         btrfs_set_device_type(leaf, dev_item, device->type);
1543         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1544         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1545         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1546         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1547         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1548         btrfs_mark_buffer_dirty(leaf);
1549
1550 out:
1551         btrfs_free_path(path);
1552         return ret;
1553 }
1554
1555 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1556                       struct btrfs_device *device, u64 new_size)
1557 {
1558         struct btrfs_super_block *super_copy =
1559                 &device->dev_root->fs_info->super_copy;
1560         u64 old_total = btrfs_super_total_bytes(super_copy);
1561         u64 diff = new_size - device->total_bytes;
1562
1563         if (!device->writeable)
1564                 return -EACCES;
1565         if (new_size <= device->total_bytes)
1566                 return -EINVAL;
1567
1568         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1569         device->fs_devices->total_rw_bytes += diff;
1570
1571         device->total_bytes = new_size;
1572         btrfs_clear_space_info_full(device->dev_root->fs_info);
1573
1574         return btrfs_update_device(trans, device);
1575 }
1576
1577 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1578                       struct btrfs_device *device, u64 new_size)
1579 {
1580         int ret;
1581         lock_chunks(device->dev_root);
1582         ret = __btrfs_grow_device(trans, device, new_size);
1583         unlock_chunks(device->dev_root);
1584         return ret;
1585 }
1586
1587 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1588                             struct btrfs_root *root,
1589                             u64 chunk_tree, u64 chunk_objectid,
1590                             u64 chunk_offset)
1591 {
1592         int ret;
1593         struct btrfs_path *path;
1594         struct btrfs_key key;
1595
1596         root = root->fs_info->chunk_root;
1597         path = btrfs_alloc_path();
1598         if (!path)
1599                 return -ENOMEM;
1600
1601         key.objectid = chunk_objectid;
1602         key.offset = chunk_offset;
1603         key.type = BTRFS_CHUNK_ITEM_KEY;
1604
1605         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1606         BUG_ON(ret);
1607
1608         ret = btrfs_del_item(trans, root, path);
1609         BUG_ON(ret);
1610
1611         btrfs_free_path(path);
1612         return 0;
1613 }
1614
1615 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1616                         chunk_offset)
1617 {
1618         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1619         struct btrfs_disk_key *disk_key;
1620         struct btrfs_chunk *chunk;
1621         u8 *ptr;
1622         int ret = 0;
1623         u32 num_stripes;
1624         u32 array_size;
1625         u32 len = 0;
1626         u32 cur;
1627         struct btrfs_key key;
1628
1629         array_size = btrfs_super_sys_array_size(super_copy);
1630
1631         ptr = super_copy->sys_chunk_array;
1632         cur = 0;
1633
1634         while (cur < array_size) {
1635                 disk_key = (struct btrfs_disk_key *)ptr;
1636                 btrfs_disk_key_to_cpu(&key, disk_key);
1637
1638                 len = sizeof(*disk_key);
1639
1640                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1641                         chunk = (struct btrfs_chunk *)(ptr + len);
1642                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1643                         len += btrfs_chunk_item_size(num_stripes);
1644                 } else {
1645                         ret = -EIO;
1646                         break;
1647                 }
1648                 if (key.objectid == chunk_objectid &&
1649                     key.offset == chunk_offset) {
1650                         memmove(ptr, ptr + len, array_size - (cur + len));
1651                         array_size -= len;
1652                         btrfs_set_super_sys_array_size(super_copy, array_size);
1653                 } else {
1654                         ptr += len;
1655                         cur += len;
1656                 }
1657         }
1658         return ret;
1659 }
1660
1661 static int btrfs_relocate_chunk(struct btrfs_root *root,
1662                          u64 chunk_tree, u64 chunk_objectid,
1663                          u64 chunk_offset)
1664 {
1665         struct extent_map_tree *em_tree;
1666         struct btrfs_root *extent_root;
1667         struct btrfs_trans_handle *trans;
1668         struct extent_map *em;
1669         struct map_lookup *map;
1670         int ret;
1671         int i;
1672
1673         printk(KERN_INFO "btrfs relocating chunk %llu\n",
1674                (unsigned long long)chunk_offset);
1675         root = root->fs_info->chunk_root;
1676         extent_root = root->fs_info->extent_root;
1677         em_tree = &root->fs_info->mapping_tree.map_tree;
1678
1679         /* step one, relocate all the extents inside this chunk */
1680         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1681         BUG_ON(ret);
1682
1683         trans = btrfs_start_transaction(root, 1);
1684         BUG_ON(!trans);
1685
1686         lock_chunks(root);
1687
1688         /*
1689          * step two, delete the device extents and the
1690          * chunk tree entries
1691          */
1692         spin_lock(&em_tree->lock);
1693         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1694         spin_unlock(&em_tree->lock);
1695
1696         BUG_ON(em->start > chunk_offset ||
1697                em->start + em->len < chunk_offset);
1698         map = (struct map_lookup *)em->bdev;
1699
1700         for (i = 0; i < map->num_stripes; i++) {
1701                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1702                                             map->stripes[i].physical);
1703                 BUG_ON(ret);
1704
1705                 if (map->stripes[i].dev) {
1706                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1707                         BUG_ON(ret);
1708                 }
1709         }
1710         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1711                                chunk_offset);
1712
1713         BUG_ON(ret);
1714
1715         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1716                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1717                 BUG_ON(ret);
1718         }
1719
1720         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1721         BUG_ON(ret);
1722
1723         spin_lock(&em_tree->lock);
1724         remove_extent_mapping(em_tree, em);
1725         spin_unlock(&em_tree->lock);
1726
1727         kfree(map);
1728         em->bdev = NULL;
1729
1730         /* once for the tree */
1731         free_extent_map(em);
1732         /* once for us */
1733         free_extent_map(em);
1734
1735         unlock_chunks(root);
1736         btrfs_end_transaction(trans, root);
1737         return 0;
1738 }
1739
1740 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1741 {
1742         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1743         struct btrfs_path *path;
1744         struct extent_buffer *leaf;
1745         struct btrfs_chunk *chunk;
1746         struct btrfs_key key;
1747         struct btrfs_key found_key;
1748         u64 chunk_tree = chunk_root->root_key.objectid;
1749         u64 chunk_type;
1750         int ret;
1751
1752         path = btrfs_alloc_path();
1753         if (!path)
1754                 return -ENOMEM;
1755
1756         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1757         key.offset = (u64)-1;
1758         key.type = BTRFS_CHUNK_ITEM_KEY;
1759
1760         while (1) {
1761                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1762                 if (ret < 0)
1763                         goto error;
1764                 BUG_ON(ret == 0);
1765
1766                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1767                                           key.type);
1768                 if (ret < 0)
1769                         goto error;
1770                 if (ret > 0)
1771                         break;
1772
1773                 leaf = path->nodes[0];
1774                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1775
1776                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1777                                        struct btrfs_chunk);
1778                 chunk_type = btrfs_chunk_type(leaf, chunk);
1779                 btrfs_release_path(chunk_root, path);
1780
1781                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1782                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1783                                                    found_key.objectid,
1784                                                    found_key.offset);
1785                         BUG_ON(ret);
1786                 }
1787
1788                 if (found_key.offset == 0)
1789                         break;
1790                 key.offset = found_key.offset - 1;
1791         }
1792         ret = 0;
1793 error:
1794         btrfs_free_path(path);
1795         return ret;
1796 }
1797
1798 static u64 div_factor(u64 num, int factor)
1799 {
1800         if (factor == 10)
1801                 return num;
1802         num *= factor;
1803         do_div(num, 10);
1804         return num;
1805 }
1806
1807 int btrfs_balance(struct btrfs_root *dev_root)
1808 {
1809         int ret;
1810         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1811         struct btrfs_device *device;
1812         u64 old_size;
1813         u64 size_to_free;
1814         struct btrfs_path *path;
1815         struct btrfs_key key;
1816         struct btrfs_chunk *chunk;
1817         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1818         struct btrfs_trans_handle *trans;
1819         struct btrfs_key found_key;
1820
1821         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1822                 return -EROFS;
1823
1824         mutex_lock(&dev_root->fs_info->volume_mutex);
1825         dev_root = dev_root->fs_info->dev_root;
1826
1827         /* step one make some room on all the devices */
1828         list_for_each_entry(device, devices, dev_list) {
1829                 old_size = device->total_bytes;
1830                 size_to_free = div_factor(old_size, 1);
1831                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1832                 if (!device->writeable ||
1833                     device->total_bytes - device->bytes_used > size_to_free)
1834                         continue;
1835
1836                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1837                 BUG_ON(ret);
1838
1839                 trans = btrfs_start_transaction(dev_root, 1);
1840                 BUG_ON(!trans);
1841
1842                 ret = btrfs_grow_device(trans, device, old_size);
1843                 BUG_ON(ret);
1844
1845                 btrfs_end_transaction(trans, dev_root);
1846         }
1847
1848         /* step two, relocate all the chunks */
1849         path = btrfs_alloc_path();
1850         BUG_ON(!path);
1851
1852         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1853         key.offset = (u64)-1;
1854         key.type = BTRFS_CHUNK_ITEM_KEY;
1855
1856         while (1) {
1857                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1858                 if (ret < 0)
1859                         goto error;
1860
1861                 /*
1862                  * this shouldn't happen, it means the last relocate
1863                  * failed
1864                  */
1865                 if (ret == 0)
1866                         break;
1867
1868                 ret = btrfs_previous_item(chunk_root, path, 0,
1869                                           BTRFS_CHUNK_ITEM_KEY);
1870                 if (ret)
1871                         break;
1872
1873                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1874                                       path->slots[0]);
1875                 if (found_key.objectid != key.objectid)
1876                         break;
1877
1878                 chunk = btrfs_item_ptr(path->nodes[0],
1879                                        path->slots[0],
1880                                        struct btrfs_chunk);
1881                 key.offset = found_key.offset;
1882                 /* chunk zero is special */
1883                 if (key.offset == 0)
1884                         break;
1885
1886                 btrfs_release_path(chunk_root, path);
1887                 ret = btrfs_relocate_chunk(chunk_root,
1888                                            chunk_root->root_key.objectid,
1889                                            found_key.objectid,
1890                                            found_key.offset);
1891                 BUG_ON(ret);
1892         }
1893         ret = 0;
1894 error:
1895         btrfs_free_path(path);
1896         mutex_unlock(&dev_root->fs_info->volume_mutex);
1897         return ret;
1898 }
1899
1900 /*
1901  * shrinking a device means finding all of the device extents past
1902  * the new size, and then following the back refs to the chunks.
1903  * The chunk relocation code actually frees the device extent
1904  */
1905 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1906 {
1907         struct btrfs_trans_handle *trans;
1908         struct btrfs_root *root = device->dev_root;
1909         struct btrfs_dev_extent *dev_extent = NULL;
1910         struct btrfs_path *path;
1911         u64 length;
1912         u64 chunk_tree;
1913         u64 chunk_objectid;
1914         u64 chunk_offset;
1915         int ret;
1916         int slot;
1917         struct extent_buffer *l;
1918         struct btrfs_key key;
1919         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1920         u64 old_total = btrfs_super_total_bytes(super_copy);
1921         u64 diff = device->total_bytes - new_size;
1922
1923         if (new_size >= device->total_bytes)
1924                 return -EINVAL;
1925
1926         path = btrfs_alloc_path();
1927         if (!path)
1928                 return -ENOMEM;
1929
1930         trans = btrfs_start_transaction(root, 1);
1931         if (!trans) {
1932                 ret = -ENOMEM;
1933                 goto done;
1934         }
1935
1936         path->reada = 2;
1937
1938         lock_chunks(root);
1939
1940         device->total_bytes = new_size;
1941         if (device->writeable)
1942                 device->fs_devices->total_rw_bytes -= diff;
1943         unlock_chunks(root);
1944         btrfs_end_transaction(trans, root);
1945
1946         key.objectid = device->devid;
1947         key.offset = (u64)-1;
1948         key.type = BTRFS_DEV_EXTENT_KEY;
1949
1950         while (1) {
1951                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1952                 if (ret < 0)
1953                         goto done;
1954
1955                 ret = btrfs_previous_item(root, path, 0, key.type);
1956                 if (ret < 0)
1957                         goto done;
1958                 if (ret) {
1959                         ret = 0;
1960                         goto done;
1961                 }
1962
1963                 l = path->nodes[0];
1964                 slot = path->slots[0];
1965                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1966
1967                 if (key.objectid != device->devid)
1968                         goto done;
1969
1970                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1971                 length = btrfs_dev_extent_length(l, dev_extent);
1972
1973                 if (key.offset + length <= new_size)
1974                         break;
1975
1976                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1977                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1978                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1979                 btrfs_release_path(root, path);
1980
1981                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1982                                            chunk_offset);
1983                 if (ret)
1984                         goto done;
1985         }
1986
1987         /* Shrinking succeeded, else we would be at "done". */
1988         trans = btrfs_start_transaction(root, 1);
1989         if (!trans) {
1990                 ret = -ENOMEM;
1991                 goto done;
1992         }
1993         lock_chunks(root);
1994
1995         device->disk_total_bytes = new_size;
1996         /* Now btrfs_update_device() will change the on-disk size. */
1997         ret = btrfs_update_device(trans, device);
1998         if (ret) {
1999                 unlock_chunks(root);
2000                 btrfs_end_transaction(trans, root);
2001                 goto done;
2002         }
2003         WARN_ON(diff > old_total);
2004         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2005         unlock_chunks(root);
2006         btrfs_end_transaction(trans, root);
2007 done:
2008         btrfs_free_path(path);
2009         return ret;
2010 }
2011
2012 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2013                            struct btrfs_root *root,
2014                            struct btrfs_key *key,
2015                            struct btrfs_chunk *chunk, int item_size)
2016 {
2017         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2018         struct btrfs_disk_key disk_key;
2019         u32 array_size;
2020         u8 *ptr;
2021
2022         array_size = btrfs_super_sys_array_size(super_copy);
2023         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2024                 return -EFBIG;
2025
2026         ptr = super_copy->sys_chunk_array + array_size;
2027         btrfs_cpu_key_to_disk(&disk_key, key);
2028         memcpy(ptr, &disk_key, sizeof(disk_key));
2029         ptr += sizeof(disk_key);
2030         memcpy(ptr, chunk, item_size);
2031         item_size += sizeof(disk_key);
2032         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2033         return 0;
2034 }
2035
2036 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2037                                         int num_stripes, int sub_stripes)
2038 {
2039         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2040                 return calc_size;
2041         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2042                 return calc_size * (num_stripes / sub_stripes);
2043         else
2044                 return calc_size * num_stripes;
2045 }
2046
2047 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2048                                struct btrfs_root *extent_root,
2049                                struct map_lookup **map_ret,
2050                                u64 *num_bytes, u64 *stripe_size,
2051                                u64 start, u64 type)
2052 {
2053         struct btrfs_fs_info *info = extent_root->fs_info;
2054         struct btrfs_device *device = NULL;
2055         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2056         struct list_head *cur;
2057         struct map_lookup *map = NULL;
2058         struct extent_map_tree *em_tree;
2059         struct extent_map *em;
2060         struct list_head private_devs;
2061         int min_stripe_size = 1 * 1024 * 1024;
2062         u64 calc_size = 1024 * 1024 * 1024;
2063         u64 max_chunk_size = calc_size;
2064         u64 min_free;
2065         u64 avail;
2066         u64 max_avail = 0;
2067         u64 dev_offset;
2068         int num_stripes = 1;
2069         int min_stripes = 1;
2070         int sub_stripes = 0;
2071         int looped = 0;
2072         int ret;
2073         int index;
2074         int stripe_len = 64 * 1024;
2075
2076         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2077             (type & BTRFS_BLOCK_GROUP_DUP)) {
2078                 WARN_ON(1);
2079                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2080         }
2081         if (list_empty(&fs_devices->alloc_list))
2082                 return -ENOSPC;
2083
2084         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2085                 num_stripes = fs_devices->rw_devices;
2086                 min_stripes = 2;
2087         }
2088         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2089                 num_stripes = 2;
2090                 min_stripes = 2;
2091         }
2092         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2093                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2094                 if (num_stripes < 2)
2095                         return -ENOSPC;
2096                 min_stripes = 2;
2097         }
2098         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2099                 num_stripes = fs_devices->rw_devices;
2100                 if (num_stripes < 4)
2101                         return -ENOSPC;
2102                 num_stripes &= ~(u32)1;
2103                 sub_stripes = 2;
2104                 min_stripes = 4;
2105         }
2106
2107         if (type & BTRFS_BLOCK_GROUP_DATA) {
2108                 max_chunk_size = 10 * calc_size;
2109                 min_stripe_size = 64 * 1024 * 1024;
2110         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2111                 max_chunk_size = 4 * calc_size;
2112                 min_stripe_size = 32 * 1024 * 1024;
2113         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2114                 calc_size = 8 * 1024 * 1024;
2115                 max_chunk_size = calc_size * 2;
2116                 min_stripe_size = 1 * 1024 * 1024;
2117         }
2118
2119         /* we don't want a chunk larger than 10% of writeable space */
2120         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2121                              max_chunk_size);
2122
2123 again:
2124         if (!map || map->num_stripes != num_stripes) {
2125                 kfree(map);
2126                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2127                 if (!map)
2128                         return -ENOMEM;
2129                 map->num_stripes = num_stripes;
2130         }
2131
2132         if (calc_size * num_stripes > max_chunk_size) {
2133                 calc_size = max_chunk_size;
2134                 do_div(calc_size, num_stripes);
2135                 do_div(calc_size, stripe_len);
2136                 calc_size *= stripe_len;
2137         }
2138         /* we don't want tiny stripes */
2139         calc_size = max_t(u64, min_stripe_size, calc_size);
2140
2141         do_div(calc_size, stripe_len);
2142         calc_size *= stripe_len;
2143
2144         cur = fs_devices->alloc_list.next;
2145         index = 0;
2146
2147         if (type & BTRFS_BLOCK_GROUP_DUP)
2148                 min_free = calc_size * 2;
2149         else
2150                 min_free = calc_size;
2151
2152         /*
2153          * we add 1MB because we never use the first 1MB of the device, unless
2154          * we've looped, then we are likely allocating the maximum amount of
2155          * space left already
2156          */
2157         if (!looped)
2158                 min_free += 1024 * 1024;
2159
2160         INIT_LIST_HEAD(&private_devs);
2161         while (index < num_stripes) {
2162                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2163                 BUG_ON(!device->writeable);
2164                 if (device->total_bytes > device->bytes_used)
2165                         avail = device->total_bytes - device->bytes_used;
2166                 else
2167                         avail = 0;
2168                 cur = cur->next;
2169
2170                 if (device->in_fs_metadata && avail >= min_free) {
2171                         ret = find_free_dev_extent(trans, device,
2172                                                    min_free, &dev_offset);
2173                         if (ret == 0) {
2174                                 list_move_tail(&device->dev_alloc_list,
2175                                                &private_devs);
2176                                 map->stripes[index].dev = device;
2177                                 map->stripes[index].physical = dev_offset;
2178                                 index++;
2179                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2180                                         map->stripes[index].dev = device;
2181                                         map->stripes[index].physical =
2182                                                 dev_offset + calc_size;
2183                                         index++;
2184                                 }
2185                         }
2186                 } else if (device->in_fs_metadata && avail > max_avail)
2187                         max_avail = avail;
2188                 if (cur == &fs_devices->alloc_list)
2189                         break;
2190         }
2191         list_splice(&private_devs, &fs_devices->alloc_list);
2192         if (index < num_stripes) {
2193                 if (index >= min_stripes) {
2194                         num_stripes = index;
2195                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2196                                 num_stripes /= sub_stripes;
2197                                 num_stripes *= sub_stripes;
2198                         }
2199                         looped = 1;
2200                         goto again;
2201                 }
2202                 if (!looped && max_avail > 0) {
2203                         looped = 1;
2204                         calc_size = max_avail;
2205                         goto again;
2206                 }
2207                 kfree(map);
2208                 return -ENOSPC;
2209         }
2210         map->sector_size = extent_root->sectorsize;
2211         map->stripe_len = stripe_len;
2212         map->io_align = stripe_len;
2213         map->io_width = stripe_len;
2214         map->type = type;
2215         map->num_stripes = num_stripes;
2216         map->sub_stripes = sub_stripes;
2217
2218         *map_ret = map;
2219         *stripe_size = calc_size;
2220         *num_bytes = chunk_bytes_by_type(type, calc_size,
2221                                          num_stripes, sub_stripes);
2222
2223         em = alloc_extent_map(GFP_NOFS);
2224         if (!em) {
2225                 kfree(map);
2226                 return -ENOMEM;
2227         }
2228         em->bdev = (struct block_device *)map;
2229         em->start = start;
2230         em->len = *num_bytes;
2231         em->block_start = 0;
2232         em->block_len = em->len;
2233
2234         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2235         spin_lock(&em_tree->lock);
2236         ret = add_extent_mapping(em_tree, em);
2237         spin_unlock(&em_tree->lock);
2238         BUG_ON(ret);
2239         free_extent_map(em);
2240
2241         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2242                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2243                                      start, *num_bytes);
2244         BUG_ON(ret);
2245
2246         index = 0;
2247         while (index < map->num_stripes) {
2248                 device = map->stripes[index].dev;
2249                 dev_offset = map->stripes[index].physical;
2250
2251                 ret = btrfs_alloc_dev_extent(trans, device,
2252                                 info->chunk_root->root_key.objectid,
2253                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2254                                 start, dev_offset, calc_size);
2255                 BUG_ON(ret);
2256                 index++;
2257         }
2258
2259         return 0;
2260 }
2261
2262 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2263                                 struct btrfs_root *extent_root,
2264                                 struct map_lookup *map, u64 chunk_offset,
2265                                 u64 chunk_size, u64 stripe_size)
2266 {
2267         u64 dev_offset;
2268         struct btrfs_key key;
2269         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2270         struct btrfs_device *device;
2271         struct btrfs_chunk *chunk;
2272         struct btrfs_stripe *stripe;
2273         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2274         int index = 0;
2275         int ret;
2276
2277         chunk = kzalloc(item_size, GFP_NOFS);
2278         if (!chunk)
2279                 return -ENOMEM;
2280
2281         index = 0;
2282         while (index < map->num_stripes) {
2283                 device = map->stripes[index].dev;
2284                 device->bytes_used += stripe_size;
2285                 ret = btrfs_update_device(trans, device);
2286                 BUG_ON(ret);
2287                 index++;
2288         }
2289
2290         index = 0;
2291         stripe = &chunk->stripe;
2292         while (index < map->num_stripes) {
2293                 device = map->stripes[index].dev;
2294                 dev_offset = map->stripes[index].physical;
2295
2296                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2297                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2298                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2299                 stripe++;
2300                 index++;
2301         }
2302
2303         btrfs_set_stack_chunk_length(chunk, chunk_size);
2304         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2305         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2306         btrfs_set_stack_chunk_type(chunk, map->type);
2307         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2308         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2309         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2310         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2311         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2312
2313         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2314         key.type = BTRFS_CHUNK_ITEM_KEY;
2315         key.offset = chunk_offset;
2316
2317         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2318         BUG_ON(ret);
2319
2320         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2321                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2322                                              item_size);
2323                 BUG_ON(ret);
2324         }
2325         kfree(chunk);
2326         return 0;
2327 }
2328
2329 /*
2330  * Chunk allocation falls into two parts. The first part does works
2331  * that make the new allocated chunk useable, but not do any operation
2332  * that modifies the chunk tree. The second part does the works that
2333  * require modifying the chunk tree. This division is important for the
2334  * bootstrap process of adding storage to a seed btrfs.
2335  */
2336 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2337                       struct btrfs_root *extent_root, u64 type)
2338 {
2339         u64 chunk_offset;
2340         u64 chunk_size;
2341         u64 stripe_size;
2342         struct map_lookup *map;
2343         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2344         int ret;
2345
2346         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2347                               &chunk_offset);
2348         if (ret)
2349                 return ret;
2350
2351         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2352                                   &stripe_size, chunk_offset, type);
2353         if (ret)
2354                 return ret;
2355
2356         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2357                                    chunk_size, stripe_size);
2358         BUG_ON(ret);
2359         return 0;
2360 }
2361
2362 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2363                                          struct btrfs_root *root,
2364                                          struct btrfs_device *device)
2365 {
2366         u64 chunk_offset;
2367         u64 sys_chunk_offset;
2368         u64 chunk_size;
2369         u64 sys_chunk_size;
2370         u64 stripe_size;
2371         u64 sys_stripe_size;
2372         u64 alloc_profile;
2373         struct map_lookup *map;
2374         struct map_lookup *sys_map;
2375         struct btrfs_fs_info *fs_info = root->fs_info;
2376         struct btrfs_root *extent_root = fs_info->extent_root;
2377         int ret;
2378
2379         ret = find_next_chunk(fs_info->chunk_root,
2380                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2381         BUG_ON(ret);
2382
2383         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2384                         (fs_info->metadata_alloc_profile &
2385                          fs_info->avail_metadata_alloc_bits);
2386         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2387
2388         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2389                                   &stripe_size, chunk_offset, alloc_profile);
2390         BUG_ON(ret);
2391
2392         sys_chunk_offset = chunk_offset + chunk_size;
2393
2394         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2395                         (fs_info->system_alloc_profile &
2396                          fs_info->avail_system_alloc_bits);
2397         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2398
2399         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2400                                   &sys_chunk_size, &sys_stripe_size,
2401                                   sys_chunk_offset, alloc_profile);
2402         BUG_ON(ret);
2403
2404         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2405         BUG_ON(ret);
2406
2407         /*
2408          * Modifying chunk tree needs allocating new blocks from both
2409          * system block group and metadata block group. So we only can
2410          * do operations require modifying the chunk tree after both
2411          * block groups were created.
2412          */
2413         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2414                                    chunk_size, stripe_size);
2415         BUG_ON(ret);
2416
2417         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2418                                    sys_chunk_offset, sys_chunk_size,
2419                                    sys_stripe_size);
2420         BUG_ON(ret);
2421         return 0;
2422 }
2423
2424 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2425 {
2426         struct extent_map *em;
2427         struct map_lookup *map;
2428         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2429         int readonly = 0;
2430         int i;
2431
2432         spin_lock(&map_tree->map_tree.lock);
2433         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2434         spin_unlock(&map_tree->map_tree.lock);
2435         if (!em)
2436                 return 1;
2437
2438         map = (struct map_lookup *)em->bdev;
2439         for (i = 0; i < map->num_stripes; i++) {
2440                 if (!map->stripes[i].dev->writeable) {
2441                         readonly = 1;
2442                         break;
2443                 }
2444         }
2445         free_extent_map(em);
2446         return readonly;
2447 }
2448
2449 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2450 {
2451         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2452 }
2453
2454 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2455 {
2456         struct extent_map *em;
2457
2458         while (1) {
2459                 spin_lock(&tree->map_tree.lock);
2460                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2461                 if (em)
2462                         remove_extent_mapping(&tree->map_tree, em);
2463                 spin_unlock(&tree->map_tree.lock);
2464                 if (!em)
2465                         break;
2466                 kfree(em->bdev);
2467                 /* once for us */
2468                 free_extent_map(em);
2469                 /* once for the tree */
2470                 free_extent_map(em);
2471         }
2472 }
2473
2474 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2475 {
2476         struct extent_map *em;
2477         struct map_lookup *map;
2478         struct extent_map_tree *em_tree = &map_tree->map_tree;
2479         int ret;
2480
2481         spin_lock(&em_tree->lock);
2482         em = lookup_extent_mapping(em_tree, logical, len);
2483         spin_unlock(&em_tree->lock);
2484         BUG_ON(!em);
2485
2486         BUG_ON(em->start > logical || em->start + em->len < logical);
2487         map = (struct map_lookup *)em->bdev;
2488         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2489                 ret = map->num_stripes;
2490         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2491                 ret = map->sub_stripes;
2492         else
2493                 ret = 1;
2494         free_extent_map(em);
2495         return ret;
2496 }
2497
2498 static int find_live_mirror(struct map_lookup *map, int first, int num,
2499                             int optimal)
2500 {
2501         int i;
2502         if (map->stripes[optimal].dev->bdev)
2503                 return optimal;
2504         for (i = first; i < first + num; i++) {
2505                 if (map->stripes[i].dev->bdev)
2506                         return i;
2507         }
2508         /* we couldn't find one that doesn't fail.  Just return something
2509          * and the io error handling code will clean up eventually
2510          */
2511         return optimal;
2512 }
2513
2514 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2515                              u64 logical, u64 *length,
2516                              struct btrfs_multi_bio **multi_ret,
2517                              int mirror_num, struct page *unplug_page)
2518 {
2519         struct extent_map *em;
2520         struct map_lookup *map;
2521         struct extent_map_tree *em_tree = &map_tree->map_tree;
2522         u64 offset;
2523         u64 stripe_offset;
2524         u64 stripe_nr;
2525         int stripes_allocated = 8;
2526         int stripes_required = 1;
2527         int stripe_index;
2528         int i;
2529         int num_stripes;
2530         int max_errors = 0;
2531         struct btrfs_multi_bio *multi = NULL;
2532
2533         if (multi_ret && !(rw & (1 << BIO_RW)))
2534                 stripes_allocated = 1;
2535 again:
2536         if (multi_ret) {
2537                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2538                                 GFP_NOFS);
2539                 if (!multi)
2540                         return -ENOMEM;
2541
2542                 atomic_set(&multi->error, 0);
2543         }
2544
2545         spin_lock(&em_tree->lock);
2546         em = lookup_extent_mapping(em_tree, logical, *length);
2547         spin_unlock(&em_tree->lock);
2548
2549         if (!em && unplug_page)
2550                 return 0;
2551
2552         if (!em) {
2553                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2554                        (unsigned long long)logical,
2555                        (unsigned long long)*length);
2556                 BUG();
2557         }
2558
2559         BUG_ON(em->start > logical || em->start + em->len < logical);
2560         map = (struct map_lookup *)em->bdev;
2561         offset = logical - em->start;
2562
2563         if (mirror_num > map->num_stripes)
2564                 mirror_num = 0;
2565
2566         /* if our multi bio struct is too small, back off and try again */
2567         if (rw & (1 << BIO_RW)) {
2568                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2569                                  BTRFS_BLOCK_GROUP_DUP)) {
2570                         stripes_required = map->num_stripes;
2571                         max_errors = 1;
2572                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2573                         stripes_required = map->sub_stripes;
2574                         max_errors = 1;
2575                 }
2576         }
2577         if (multi_ret && (rw & (1 << BIO_RW)) &&
2578             stripes_allocated < stripes_required) {
2579                 stripes_allocated = map->num_stripes;
2580                 free_extent_map(em);
2581                 kfree(multi);
2582                 goto again;
2583         }
2584         stripe_nr = offset;
2585         /*
2586          * stripe_nr counts the total number of stripes we have to stride
2587          * to get to this block
2588          */
2589         do_div(stripe_nr, map->stripe_len);
2590
2591         stripe_offset = stripe_nr * map->stripe_len;
2592         BUG_ON(offset < stripe_offset);
2593
2594         /* stripe_offset is the offset of this block in its stripe*/
2595         stripe_offset = offset - stripe_offset;
2596
2597         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2598                          BTRFS_BLOCK_GROUP_RAID10 |
2599                          BTRFS_BLOCK_GROUP_DUP)) {
2600                 /* we limit the length of each bio to what fits in a stripe */
2601                 *length = min_t(u64, em->len - offset,
2602                               map->stripe_len - stripe_offset);
2603         } else {
2604                 *length = em->len - offset;
2605         }
2606
2607         if (!multi_ret && !unplug_page)
2608                 goto out;
2609
2610         num_stripes = 1;
2611         stripe_index = 0;
2612         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2613                 if (unplug_page || (rw & (1 << BIO_RW)))
2614                         num_stripes = map->num_stripes;
2615                 else if (mirror_num)
2616                         stripe_index = mirror_num - 1;
2617                 else {
2618                         stripe_index = find_live_mirror(map, 0,
2619                                             map->num_stripes,
2620                                             current->pid % map->num_stripes);
2621                 }
2622
2623         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2624                 if (rw & (1 << BIO_RW))
2625                         num_stripes = map->num_stripes;
2626                 else if (mirror_num)
2627                         stripe_index = mirror_num - 1;
2628
2629         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2630                 int factor = map->num_stripes / map->sub_stripes;
2631
2632                 stripe_index = do_div(stripe_nr, factor);
2633                 stripe_index *= map->sub_stripes;
2634
2635                 if (unplug_page || (rw & (1 << BIO_RW)))
2636                         num_stripes = map->sub_stripes;
2637                 else if (mirror_num)
2638                         stripe_index += mirror_num - 1;
2639                 else {
2640                         stripe_index = find_live_mirror(map, stripe_index,
2641                                               map->sub_stripes, stripe_index +
2642                                               current->pid % map->sub_stripes);
2643                 }
2644         } else {
2645                 /*
2646                  * after this do_div call, stripe_nr is the number of stripes
2647                  * on this device we have to walk to find the data, and
2648                  * stripe_index is the number of our device in the stripe array
2649                  */
2650                 stripe_index = do_div(stripe_nr, map->num_stripes);
2651         }
2652         BUG_ON(stripe_index >= map->num_stripes);
2653
2654         for (i = 0; i < num_stripes; i++) {
2655                 if (unplug_page) {
2656                         struct btrfs_device *device;
2657                         struct backing_dev_info *bdi;
2658
2659                         device = map->stripes[stripe_index].dev;
2660                         if (device->bdev) {
2661                                 bdi = blk_get_backing_dev_info(device->bdev);
2662                                 if (bdi->unplug_io_fn)
2663                                         bdi->unplug_io_fn(bdi, unplug_page);
2664                         }
2665                 } else {
2666                         multi->stripes[i].physical =
2667                                 map->stripes[stripe_index].physical +
2668                                 stripe_offset + stripe_nr * map->stripe_len;
2669                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2670                 }
2671                 stripe_index++;
2672         }
2673         if (multi_ret) {
2674                 *multi_ret = multi;
2675                 multi->num_stripes = num_stripes;
2676                 multi->max_errors = max_errors;
2677         }
2678 out:
2679         free_extent_map(em);
2680         return 0;
2681 }
2682
2683 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2684                       u64 logical, u64 *length,
2685                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2686 {
2687         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2688                                  mirror_num, NULL);
2689 }
2690
2691 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2692                      u64 chunk_start, u64 physical, u64 devid,
2693                      u64 **logical, int *naddrs, int *stripe_len)
2694 {
2695         struct extent_map_tree *em_tree = &map_tree->map_tree;
2696         struct extent_map *em;
2697         struct map_lookup *map;
2698         u64 *buf;
2699         u64 bytenr;
2700         u64 length;
2701         u64 stripe_nr;
2702         int i, j, nr = 0;
2703
2704         spin_lock(&em_tree->lock);
2705         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2706         spin_unlock(&em_tree->lock);
2707
2708         BUG_ON(!em || em->start != chunk_start);
2709         map = (struct map_lookup *)em->bdev;
2710
2711         length = em->len;
2712         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2713                 do_div(length, map->num_stripes / map->sub_stripes);
2714         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2715                 do_div(length, map->num_stripes);
2716
2717         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2718         BUG_ON(!buf);
2719
2720         for (i = 0; i < map->num_stripes; i++) {
2721                 if (devid && map->stripes[i].dev->devid != devid)
2722                         continue;
2723                 if (map->stripes[i].physical > physical ||
2724                     map->stripes[i].physical + length <= physical)
2725                         continue;
2726
2727                 stripe_nr = physical - map->stripes[i].physical;
2728                 do_div(stripe_nr, map->stripe_len);
2729
2730                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2731                         stripe_nr = stripe_nr * map->num_stripes + i;
2732                         do_div(stripe_nr, map->sub_stripes);
2733                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2734                         stripe_nr = stripe_nr * map->num_stripes + i;
2735                 }
2736                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2737                 WARN_ON(nr >= map->num_stripes);
2738                 for (j = 0; j < nr; j++) {
2739                         if (buf[j] == bytenr)
2740                                 break;
2741                 }
2742                 if (j == nr) {
2743                         WARN_ON(nr >= map->num_stripes);
2744                         buf[nr++] = bytenr;
2745                 }
2746         }
2747
2748         for (i = 0; i > nr; i++) {
2749                 struct btrfs_multi_bio *multi;
2750                 struct btrfs_bio_stripe *stripe;
2751                 int ret;
2752
2753                 length = 1;
2754                 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2755                                       &length, &multi, 0);
2756                 BUG_ON(ret);
2757
2758                 stripe = multi->stripes;
2759                 for (j = 0; j < multi->num_stripes; j++) {
2760                         if (stripe->physical >= physical &&
2761                             physical < stripe->physical + length)
2762                                 break;
2763                 }
2764                 BUG_ON(j >= multi->num_stripes);
2765                 kfree(multi);
2766         }
2767
2768         *logical = buf;
2769         *naddrs = nr;
2770         *stripe_len = map->stripe_len;
2771
2772         free_extent_map(em);
2773         return 0;
2774 }
2775
2776 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2777                       u64 logical, struct page *page)
2778 {
2779         u64 length = PAGE_CACHE_SIZE;
2780         return __btrfs_map_block(map_tree, READ, logical, &length,
2781                                  NULL, 0, page);
2782 }
2783
2784 static void end_bio_multi_stripe(struct bio *bio, int err)
2785 {
2786         struct btrfs_multi_bio *multi = bio->bi_private;
2787         int is_orig_bio = 0;
2788
2789         if (err)
2790                 atomic_inc(&multi->error);
2791
2792         if (bio == multi->orig_bio)
2793                 is_orig_bio = 1;
2794
2795         if (atomic_dec_and_test(&multi->stripes_pending)) {
2796                 if (!is_orig_bio) {
2797                         bio_put(bio);
2798                         bio = multi->orig_bio;
2799                 }
2800                 bio->bi_private = multi->private;
2801                 bio->bi_end_io = multi->end_io;
2802                 /* only send an error to the higher layers if it is
2803                  * beyond the tolerance of the multi-bio
2804                  */
2805                 if (atomic_read(&multi->error) > multi->max_errors) {
2806                         err = -EIO;
2807                 } else if (err) {
2808                         /*
2809                          * this bio is actually up to date, we didn't
2810                          * go over the max number of errors
2811                          */
2812                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2813                         err = 0;
2814                 }
2815                 kfree(multi);
2816
2817                 bio_endio(bio, err);
2818         } else if (!is_orig_bio) {
2819                 bio_put(bio);
2820         }
2821 }
2822
2823 struct async_sched {
2824         struct bio *bio;
2825         int rw;
2826         struct btrfs_fs_info *info;
2827         struct btrfs_work work;
2828 };
2829
2830 /*
2831  * see run_scheduled_bios for a description of why bios are collected for
2832  * async submit.
2833  *
2834  * This will add one bio to the pending list for a device and make sure
2835  * the work struct is scheduled.
2836  */
2837 static noinline int schedule_bio(struct btrfs_root *root,
2838                                  struct btrfs_device *device,
2839                                  int rw, struct bio *bio)
2840 {
2841         int should_queue = 1;
2842         struct btrfs_pending_bios *pending_bios;
2843
2844         /* don't bother with additional async steps for reads, right now */
2845         if (!(rw & (1 << BIO_RW))) {
2846                 bio_get(bio);
2847                 submit_bio(rw, bio);
2848                 bio_put(bio);
2849                 return 0;
2850         }
2851
2852         /*
2853          * nr_async_bios allows us to reliably return congestion to the
2854          * higher layers.  Otherwise, the async bio makes it appear we have
2855          * made progress against dirty pages when we've really just put it
2856          * on a queue for later
2857          */
2858         atomic_inc(&root->fs_info->nr_async_bios);
2859         WARN_ON(bio->bi_next);
2860         bio->bi_next = NULL;
2861         bio->bi_rw |= rw;
2862
2863         spin_lock(&device->io_lock);
2864         if (bio_sync(bio))
2865                 pending_bios = &device->pending_sync_bios;
2866         else
2867                 pending_bios = &device->pending_bios;
2868
2869         if (pending_bios->tail)
2870                 pending_bios->tail->bi_next = bio;
2871
2872         pending_bios->tail = bio;
2873         if (!pending_bios->head)
2874                 pending_bios->head = bio;
2875         if (device->running_pending)
2876                 should_queue = 0;
2877
2878         spin_unlock(&device->io_lock);
2879
2880         if (should_queue)
2881                 btrfs_queue_worker(&root->fs_info->submit_workers,
2882                                    &device->work);
2883         return 0;
2884 }
2885
2886 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2887                   int mirror_num, int async_submit)
2888 {
2889         struct btrfs_mapping_tree *map_tree;
2890         struct btrfs_device *dev;
2891         struct bio *first_bio = bio;
2892         u64 logical = (u64)bio->bi_sector << 9;
2893         u64 length = 0;
2894         u64 map_length;
2895         struct btrfs_multi_bio *multi = NULL;
2896         int ret;
2897         int dev_nr = 0;
2898         int total_devs = 1;
2899
2900         length = bio->bi_size;
2901         map_tree = &root->fs_info->mapping_tree;
2902         map_length = length;
2903
2904         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2905                               mirror_num);
2906         BUG_ON(ret);
2907
2908         total_devs = multi->num_stripes;
2909         if (map_length < length) {
2910                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2911                        "len %llu\n", (unsigned long long)logical,
2912                        (unsigned long long)length,
2913                        (unsigned long long)map_length);
2914                 BUG();
2915         }
2916         multi->end_io = first_bio->bi_end_io;
2917         multi->private = first_bio->bi_private;
2918         multi->orig_bio = first_bio;
2919         atomic_set(&multi->stripes_pending, multi->num_stripes);
2920
2921         while (dev_nr < total_devs) {
2922                 if (total_devs > 1) {
2923                         if (dev_nr < total_devs - 1) {
2924                                 bio = bio_clone(first_bio, GFP_NOFS);
2925                                 BUG_ON(!bio);
2926                         } else {
2927                                 bio = first_bio;
2928                         }
2929                         bio->bi_private = multi;
2930                         bio->bi_end_io = end_bio_multi_stripe;
2931                 }
2932                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2933                 dev = multi->stripes[dev_nr].dev;
2934                 BUG_ON(rw == WRITE && !dev->writeable);
2935                 if (dev && dev->bdev) {
2936                         bio->bi_bdev = dev->bdev;
2937                         if (async_submit)
2938                                 schedule_bio(root, dev, rw, bio);
2939                         else
2940                                 submit_bio(rw, bio);
2941                 } else {
2942                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2943                         bio->bi_sector = logical >> 9;
2944                         bio_endio(bio, -EIO);
2945                 }
2946                 dev_nr++;
2947         }
2948         if (total_devs == 1)
2949                 kfree(multi);
2950         return 0;
2951 }
2952
2953 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2954                                        u8 *uuid, u8 *fsid)
2955 {
2956         struct btrfs_device *device;
2957         struct btrfs_fs_devices *cur_devices;
2958
2959         cur_devices = root->fs_info->fs_devices;
2960         while (cur_devices) {
2961                 if (!fsid ||
2962                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2963                         device = __find_device(&cur_devices->devices,
2964                                                devid, uuid);
2965                         if (device)
2966                                 return device;
2967                 }
2968                 cur_devices = cur_devices->seed;
2969         }
2970         return NULL;
2971 }
2972
2973 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2974                                             u64 devid, u8 *dev_uuid)
2975 {
2976         struct btrfs_device *device;
2977         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2978
2979         device = kzalloc(sizeof(*device), GFP_NOFS);
2980         if (!device)
2981                 return NULL;
2982         list_add(&device->dev_list,
2983                  &fs_devices->devices);
2984         device->barriers = 1;
2985         device->dev_root = root->fs_info->dev_root;
2986         device->devid = devid;
2987         device->work.func = pending_bios_fn;
2988         device->fs_devices = fs_devices;
2989         fs_devices->num_devices++;
2990         spin_lock_init(&device->io_lock);
2991         INIT_LIST_HEAD(&device->dev_alloc_list);
2992         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2993         return device;
2994 }
2995
2996 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2997                           struct extent_buffer *leaf,
2998                           struct btrfs_chunk *chunk)
2999 {
3000         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3001         struct map_lookup *map;
3002         struct extent_map *em;
3003         u64 logical;
3004         u64 length;
3005         u64 devid;
3006         u8 uuid[BTRFS_UUID_SIZE];
3007         int num_stripes;
3008         int ret;
3009         int i;
3010
3011         logical = key->offset;
3012         length = btrfs_chunk_length(leaf, chunk);
3013
3014         spin_lock(&map_tree->map_tree.lock);
3015         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3016         spin_unlock(&map_tree->map_tree.lock);
3017
3018         /* already mapped? */
3019         if (em && em->start <= logical && em->start + em->len > logical) {
3020                 free_extent_map(em);
3021                 return 0;
3022         } else if (em) {
3023                 free_extent_map(em);
3024         }
3025
3026         em = alloc_extent_map(GFP_NOFS);
3027         if (!em)
3028                 return -ENOMEM;
3029         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3030         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3031         if (!map) {
3032                 free_extent_map(em);
3033                 return -ENOMEM;
3034         }
3035
3036         em->bdev = (struct block_device *)map;
3037         em->start = logical;
3038         em->len = length;
3039         em->block_start = 0;
3040         em->block_len = em->len;
3041
3042         map->num_stripes = num_stripes;
3043         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3044         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3045         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3046         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3047         map->type = btrfs_chunk_type(leaf, chunk);
3048         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3049         for (i = 0; i < num_stripes; i++) {
3050                 map->stripes[i].physical =
3051                         btrfs_stripe_offset_nr(leaf, chunk, i);
3052                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3053                 read_extent_buffer(leaf, uuid, (unsigned long)
3054                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3055                                    BTRFS_UUID_SIZE);
3056                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3057                                                         NULL);
3058                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3059                         kfree(map);
3060                         free_extent_map(em);
3061                         return -EIO;
3062                 }
3063                 if (!map->stripes[i].dev) {
3064                         map->stripes[i].dev =
3065                                 add_missing_dev(root, devid, uuid);
3066                         if (!map->stripes[i].dev) {
3067                                 kfree(map);
3068                                 free_extent_map(em);
3069                                 return -EIO;
3070                         }
3071                 }
3072                 map->stripes[i].dev->in_fs_metadata = 1;
3073         }
3074
3075         spin_lock(&map_tree->map_tree.lock);
3076         ret = add_extent_mapping(&map_tree->map_tree, em);
3077         spin_unlock(&map_tree->map_tree.lock);
3078         BUG_ON(ret);
3079         free_extent_map(em);
3080
3081         return 0;
3082 }
3083
3084 static int fill_device_from_item(struct extent_buffer *leaf,
3085                                  struct btrfs_dev_item *dev_item,
3086                                  struct btrfs_device *device)
3087 {
3088         unsigned long ptr;
3089
3090         device->devid = btrfs_device_id(leaf, dev_item);
3091         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3092         device->total_bytes = device->disk_total_bytes;
3093         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3094         device->type = btrfs_device_type(leaf, dev_item);
3095         device->io_align = btrfs_device_io_align(leaf, dev_item);
3096         device->io_width = btrfs_device_io_width(leaf, dev_item);
3097         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3098
3099         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3100         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3101
3102         return 0;
3103 }
3104
3105 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3106 {
3107         struct btrfs_fs_devices *fs_devices;
3108         int ret;
3109
3110         mutex_lock(&uuid_mutex);
3111
3112         fs_devices = root->fs_info->fs_devices->seed;
3113         while (fs_devices) {
3114                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3115                         ret = 0;
3116                         goto out;
3117                 }
3118                 fs_devices = fs_devices->seed;
3119         }
3120
3121         fs_devices = find_fsid(fsid);
3122         if (!fs_devices) {
3123                 ret = -ENOENT;
3124                 goto out;
3125         }
3126
3127         fs_devices = clone_fs_devices(fs_devices);
3128         if (IS_ERR(fs_devices)) {
3129                 ret = PTR_ERR(fs_devices);
3130                 goto out;
3131         }
3132
3133         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3134                                    root->fs_info->bdev_holder);
3135         if (ret)
3136                 goto out;
3137
3138         if (!fs_devices->seeding) {
3139                 __btrfs_close_devices(fs_devices);
3140                 free_fs_devices(fs_devices);
3141                 ret = -EINVAL;
3142                 goto out;
3143         }
3144
3145         fs_devices->seed = root->fs_info->fs_devices->seed;
3146         root->fs_info->fs_devices->seed = fs_devices;
3147 out:
3148         mutex_unlock(&uuid_mutex);
3149         return ret;
3150 }
3151
3152 static int read_one_dev(struct btrfs_root *root,
3153                         struct extent_buffer *leaf,
3154                         struct btrfs_dev_item *dev_item)
3155 {
3156         struct btrfs_device *device;
3157         u64 devid;
3158         int ret;
3159         u8 fs_uuid[BTRFS_UUID_SIZE];
3160         u8 dev_uuid[BTRFS_UUID_SIZE];
3161
3162         devid = btrfs_device_id(leaf, dev_item);
3163         read_extent_buffer(leaf, dev_uuid,
3164                            (unsigned long)btrfs_device_uuid(dev_item),
3165                            BTRFS_UUID_SIZE);
3166         read_extent_buffer(leaf, fs_uuid,
3167                            (unsigned long)btrfs_device_fsid(dev_item),
3168                            BTRFS_UUID_SIZE);
3169
3170         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3171                 ret = open_seed_devices(root, fs_uuid);
3172                 if (ret && !btrfs_test_opt(root, DEGRADED))
3173                         return ret;
3174         }
3175
3176         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3177         if (!device || !device->bdev) {
3178                 if (!btrfs_test_opt(root, DEGRADED))
3179                         return -EIO;
3180
3181                 if (!device) {
3182                         printk(KERN_WARNING "warning devid %llu missing\n",
3183                                (unsigned long long)devid);
3184                         device = add_missing_dev(root, devid, dev_uuid);
3185                         if (!device)
3186                                 return -ENOMEM;
3187                 }
3188         }
3189
3190         if (device->fs_devices != root->fs_info->fs_devices) {
3191                 BUG_ON(device->writeable);
3192                 if (device->generation !=
3193                     btrfs_device_generation(leaf, dev_item))
3194                         return -EINVAL;
3195         }
3196
3197         fill_device_from_item(leaf, dev_item, device);
3198         device->dev_root = root->fs_info->dev_root;
3199         device->in_fs_metadata = 1;
3200         if (device->writeable)
3201                 device->fs_devices->total_rw_bytes += device->total_bytes;
3202         ret = 0;
3203         return ret;
3204 }
3205
3206 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3207 {
3208         struct btrfs_dev_item *dev_item;
3209
3210         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3211                                                      dev_item);
3212         return read_one_dev(root, buf, dev_item);
3213 }
3214
3215 int btrfs_read_sys_array(struct btrfs_root *root)
3216 {
3217         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3218         struct extent_buffer *sb;
3219         struct btrfs_disk_key *disk_key;
3220         struct btrfs_chunk *chunk;
3221         u8 *ptr;
3222         unsigned long sb_ptr;
3223         int ret = 0;
3224         u32 num_stripes;
3225         u32 array_size;
3226         u32 len = 0;
3227         u32 cur;
3228         struct btrfs_key key;
3229
3230         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3231                                           BTRFS_SUPER_INFO_SIZE);
3232         if (!sb)
3233                 return -ENOMEM;
3234         btrfs_set_buffer_uptodate(sb);
3235         btrfs_set_buffer_lockdep_class(sb, 0);
3236
3237         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3238         array_size = btrfs_super_sys_array_size(super_copy);
3239
3240         ptr = super_copy->sys_chunk_array;
3241         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3242         cur = 0;
3243
3244         while (cur < array_size) {
3245                 disk_key = (struct btrfs_disk_key *)ptr;
3246                 btrfs_disk_key_to_cpu(&key, disk_key);
3247
3248                 len = sizeof(*disk_key); ptr += len;
3249                 sb_ptr += len;
3250                 cur += len;
3251
3252                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3253                         chunk = (struct btrfs_chunk *)sb_ptr;
3254                         ret = read_one_chunk(root, &key, sb, chunk);
3255                         if (ret)
3256                                 break;
3257                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3258                         len = btrfs_chunk_item_size(num_stripes);
3259                 } else {
3260                         ret = -EIO;
3261                         break;
3262                 }
3263                 ptr += len;
3264                 sb_ptr += len;
3265                 cur += len;
3266         }
3267         free_extent_buffer(sb);
3268         return ret;
3269 }
3270
3271 int btrfs_read_chunk_tree(struct btrfs_root *root)
3272 {
3273         struct btrfs_path *path;
3274         struct extent_buffer *leaf;
3275         struct btrfs_key key;
3276         struct btrfs_key found_key;
3277         int ret;
3278         int slot;
3279
3280         root = root->fs_info->chunk_root;
3281
3282         path = btrfs_alloc_path();
3283         if (!path)
3284                 return -ENOMEM;
3285
3286         /* first we search for all of the device items, and then we
3287          * read in all of the chunk items.  This way we can create chunk
3288          * mappings that reference all of the devices that are afound
3289          */
3290         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3291         key.offset = 0;
3292         key.type = 0;
3293 again:
3294         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3295         while (1) {
3296                 leaf = path->nodes[0];
3297                 slot = path->slots[0];
3298                 if (slot >= btrfs_header_nritems(leaf)) {
3299                         ret = btrfs_next_leaf(root, path);
3300                         if (ret == 0)
3301                                 continue;
3302                         if (ret < 0)
3303                                 goto error;
3304                         break;
3305                 }
3306                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3307                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3308                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3309                                 break;
3310                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3311                                 struct btrfs_dev_item *dev_item;
3312                                 dev_item = btrfs_item_ptr(leaf, slot,
3313                                                   struct btrfs_dev_item);
3314                                 ret = read_one_dev(root, leaf, dev_item);
3315                                 if (ret)
3316                                         goto error;
3317                         }
3318                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3319                         struct btrfs_chunk *chunk;
3320                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3321                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3322                         if (ret)
3323                                 goto error;
3324                 }
3325                 path->slots[0]++;
3326         }
3327         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3328                 key.objectid = 0;
3329                 btrfs_release_path(root, path);
3330                 goto again;
3331         }
3332         ret = 0;
3333 error:
3334         btrfs_free_path(path);
3335         return ret;
3336 }