btrfs-progs: add btrfs_clear_free_space_tree() from the kernel
[platform/upstream/btrfs-progs.git] / chunk-recover.c
index 439599f..e6b26ac 100644 (file)
@@ -47,7 +47,7 @@ struct recover_control {
 
        u16 csum_size;
        u32 sectorsize;
-       u32 leafsize;
+       u32 nodesize;
        u64 generation;
        u64 chunk_root_generation;
 
@@ -78,19 +78,19 @@ struct device_scan {
        struct recover_control *rc;
        struct btrfs_device *dev;
        int fd;
+       u64 bytenr;
 };
 
 static struct extent_record *btrfs_new_extent_record(struct extent_buffer *eb)
 {
        struct extent_record *rec;
 
-       rec = malloc(sizeof(*rec));
+       rec = calloc(1, sizeof(*rec));
        if (!rec) {
                fprintf(stderr, "Fail to allocate memory for extent record.\n");
                exit(1);
        }
 
-       memset(rec, 0, sizeof(*rec));
        rec->cache.start = btrfs_header_bytenr(eb);
        rec->cache.size = eb->len;
        rec->generation = btrfs_header_generation(eb);
@@ -247,7 +247,7 @@ again:
                                              generation);
                        /*
                         * According to the current kernel code, the following
-                        * case is impossble, or there is something wrong in
+                        * case is impossible, or there is something wrong in
                         * the kernel code.
                         */
                        if (memcmp(((void *)exist) + offset,
@@ -260,7 +260,7 @@ again:
                list_del_init(&exist->list);
                free(exist);
                /*
-                * We must do seach again to avoid the following cache.
+                * We must do search again to avoid the following cache.
                 * /--old bg 1--//--old bg 2--/
                 *        /--new bg--/
                 */
@@ -477,7 +477,7 @@ static void print_scan_result(struct recover_control *rc)
        printf("DEVICE SCAN RESULT:\n");
        printf("Filesystem Information:\n");
        printf("\tsectorsize: %d\n", rc->sectorsize);
-       printf("\tleafsize: %d\n", rc->leafsize);
+       printf("\tnodesize: %d\n", rc->nodesize);
        printf("\ttree root generation: %llu\n", rc->generation);
        printf("\tchunk root generation: %llu\n", rc->chunk_root_generation);
        printf("\n");
@@ -618,7 +618,7 @@ static int check_chunk_by_metadata(struct recover_control *rc,
                    btrfs_dev_extent_chunk_offset(l, dev_extent)) {
                        if (rc->verbose)
                                fprintf(stderr,
-                                       "Device tree unmatch with chunks dev_extent[%llu, %llu], chunk[%llu, %llu]\n",
+                                       "Device tree mismatch with chunks dev_extent[%llu, %llu], chunk[%llu, %llu]\n",
                                        btrfs_dev_extent_chunk_offset(l,
                                                                dev_extent),
                                        btrfs_dev_extent_length(l, dev_extent),
@@ -654,7 +654,7 @@ bg_check:
        if (chunk->type_flags != btrfs_disk_block_group_flags(l, bg_ptr)) {
                if (rc->verbose)
                        fprintf(stderr,
-                               "Chunk[%llu, %llu]'s type(%llu) is differemt with Block Group's type(%llu)\n",
+                               "Chunk[%llu, %llu]'s type(%llu) is different with Block Group's type(%llu)\n",
                                chunk->offset, chunk->length, chunk->type_flags,
                                btrfs_disk_block_group_flags(l, bg_ptr));
                btrfs_release_path(&path);
@@ -761,18 +761,20 @@ static int scan_one_device(void *dev_scan_struct)
        if (ret)
                return 1;
 
-       buf = malloc(sizeof(*buf) + rc->leafsize);
+       buf = malloc(sizeof(*buf) + rc->nodesize);
        if (!buf)
                return -ENOMEM;
-       buf->len = rc->leafsize;
+       buf->len = rc->nodesize;
 
        bytenr = 0;
        while (1) {
+               dev_scan->bytenr = bytenr;
+
                if (is_super_block_address(bytenr))
                        bytenr += rc->sectorsize;
 
-               if (pread64(fd, buf->data, rc->leafsize, bytenr) <
-                   rc->leafsize)
+               if (pread64(fd, buf->data, rc->nodesize, bytenr) <
+                   rc->nodesize)
                        break;
 
                if (memcmp_extent_buffer(buf, rc->fs_devices->fsid,
@@ -816,7 +818,7 @@ static int scan_one_device(void *dev_scan_struct)
                        break;
                }
 next_node:
-               bytenr += rc->leafsize;
+               bytenr += rc->nodesize;
        }
 out:
        close(fd);
@@ -831,12 +833,11 @@ static int scan_devices(struct recover_control *rc)
        struct btrfs_device *dev;
        struct device_scan *dev_scans;
        pthread_t *t_scans;
-       int *t_rets;
+       long *t_rets;
        int devnr = 0;
        int devidx = 0;
-       int cancel_from = 0;
-       int cancel_to = 0;
        int i;
+       int all_done;
 
        list_for_each_entry(dev, &rc->fs_devices->devices, dev_list)
                devnr++;
@@ -845,11 +846,16 @@ static int scan_devices(struct recover_control *rc)
        if (!dev_scans)
                return -ENOMEM;
        t_scans = (pthread_t *)malloc(sizeof(pthread_t) * devnr);
-       if (!t_scans)
+       if (!t_scans) {
+               free(dev_scans);
                return -ENOMEM;
-       t_rets = (int *)malloc(sizeof(int) * devnr);
-       if (!t_rets)
+       }
+       t_rets = (long *)malloc(sizeof(long) * devnr);
+       if (!t_rets) {
+               free(dev_scans);
+               free(t_scans);
                return -ENOMEM;
+       }
 
        list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
                fd = open(dev->name, O_RDONLY);
@@ -862,32 +868,64 @@ static int scan_devices(struct recover_control *rc)
                dev_scans[devidx].rc = rc;
                dev_scans[devidx].dev = dev;
                dev_scans[devidx].fd = fd;
-               ret = pthread_create(&t_scans[devidx], NULL,
-                                    (void *)scan_one_device,
-                                    (void *)&dev_scans[devidx]);
-               if (ret) {
-                       cancel_from = 0;
-                       cancel_to = devidx - 1;
-                       goto out1;
-               }
+               dev_scans[devidx].bytenr = -1;
                devidx++;
        }
 
-       i = 0;
-       while (i < devidx) {
-               ret = pthread_join(t_scans[i], (void **)&t_rets[i]);
-               if (ret || t_rets[i]) {
-                       ret = 1;
-                       cancel_from = i + 1;
-                       cancel_to = devnr - 1;
+       for (i = 0; i < devidx; i++) {
+               ret = pthread_create(&t_scans[i], NULL,
+                                    (void *)scan_one_device,
+                                    (void *)&dev_scans[i]);
+               if (ret)
                        goto out1;
+
+               dev_scans[i].bytenr = 0;
+       }
+
+       while (1) {
+               all_done = 1;
+               for (i = 0; i < devidx; i++) {
+                       if (dev_scans[i].bytenr == -1)
+                               continue;
+                       ret = pthread_tryjoin_np(t_scans[i],
+                                                (void **)&t_rets[i]);
+                       if (ret == EBUSY) {
+                               all_done = 0;
+                               continue;
+                       }
+                       if (ret || t_rets[i]) {
+                               ret = 1;
+                               goto out1;
+                       }
+                       dev_scans[i].bytenr = -1;
+               }
+
+               printf("\rScanning: ");
+               for (i = 0; i < devidx; i++) {
+                       if (dev_scans[i].bytenr == -1)
+                               printf("%sDONE in dev%d",
+                                      i ? ", " : "", i);
+                       else
+                               printf("%s%llu in dev%d",
+                                      i ? ", " : "", dev_scans[i].bytenr, i);
+               }
+               /* clear chars if exist in tail */
+               printf("                ");
+               printf("\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b");
+               fflush(stdout);
+
+               if (all_done) {
+                       printf("\n");
+                       break;
                }
-               i++;
+
+               sleep(1);
        }
 out1:
-       while (ret && (cancel_from <= cancel_to)) {
-               pthread_cancel(t_scans[cancel_from]);
-               cancel_from++;
+       for (i = 0; i < devidx; i++) {
+               if (dev_scans[i].bytenr == -1)
+                       continue;
+               pthread_cancel(t_scans[i]);
        }
 out2:
        free(dev_scans);
@@ -930,7 +968,7 @@ static int build_device_map_by_chunk_record(struct btrfs_root *root,
                map->stripes[i].dev = btrfs_find_device(root, devid,
                                                        uuid, NULL);
                if (!map->stripes[i].dev) {
-                       kfree(map);
+                       free(map);
                        return -EIO;
                }
        }
@@ -1032,7 +1070,7 @@ again:
                    key.type == BTRFS_METADATA_ITEM_KEY) {
                        old_val = btrfs_super_bytes_used(fs_info->super_copy);
                        if (key.type == BTRFS_METADATA_ITEM_KEY)
-                               old_val += root->leafsize;
+                               old_val += root->nodesize;
                        else
                                old_val += key.offset;
                        btrfs_set_super_bytes_used(fs_info->super_copy,
@@ -1060,8 +1098,7 @@ err:
        return ret;
 }
 
-static int block_group_free_all_extent(struct btrfs_trans_handle *trans,
-                                      struct btrfs_root *root,
+static int block_group_free_all_extent(struct btrfs_root *root,
                                       struct block_group_record *bg)
 {
        struct btrfs_block_group_cache *cache;
@@ -1101,7 +1138,7 @@ static int remove_chunk_extent_item(struct btrfs_trans_handle *trans,
                if (ret)
                        return ret;
 
-               ret = block_group_free_all_extent(trans, root, chunk->bg_rec);
+               ret = block_group_free_all_extent(root, chunk->bg_rec);
                if (ret)
                        return ret;
        }
@@ -1122,9 +1159,9 @@ static int __rebuild_chunk_root(struct btrfs_trans_handle *trans,
                if (min_devid > dev->devid)
                        min_devid = dev->devid;
        }
-       disk_key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
-       disk_key.type = BTRFS_DEV_ITEM_KEY;
-       disk_key.offset = min_devid;
+       btrfs_set_disk_key_objectid(&disk_key, BTRFS_DEV_ITEMS_OBJECTID);
+       btrfs_set_disk_key_type(&disk_key, BTRFS_DEV_ITEM_KEY);
+       btrfs_set_disk_key_offset(&disk_key, min_devid);
 
        cow = btrfs_alloc_free_block(trans, root, root->nodesize,
                                     BTRFS_CHUNK_TREE_OBJECTID,
@@ -1154,13 +1191,10 @@ static int __rebuild_device_items(struct btrfs_trans_handle *trans,
 {
        struct btrfs_device *dev;
        struct btrfs_key key;
-       struct btrfs_dev_item *dev_item;
+       struct btrfs_dev_item dev_item_tmp;
+       struct btrfs_dev_item *dev_item = &dev_item_tmp;
        int ret = 0;
 
-       dev_item = malloc(sizeof(struct btrfs_dev_item));
-       if (!dev_item)
-               return -ENOMEM;
-
        list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
                key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
                key.type = BTRFS_DEV_ITEM_KEY;
@@ -1181,7 +1215,6 @@ static int __rebuild_device_items(struct btrfs_trans_handle *trans,
                                        dev_item, sizeof(*dev_item));
        }
 
-       free(dev_item);
        return ret;
 }
 
@@ -1201,7 +1234,7 @@ static int __insert_chunk_item(struct btrfs_trans_handle *trans,
        key.offset = chunk_rec->offset;
 
        ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
-                               btrfs_chunk_item_size(chunk->num_stripes));
+                               btrfs_chunk_item_size(chunk_rec->num_stripes));
        free(chunk);
        return ret;
 }
@@ -1365,26 +1398,24 @@ static int rebuild_block_group(struct btrfs_trans_handle *trans,
 {
        struct chunk_record *chunk_rec;
        struct btrfs_key search_key;
-       struct btrfs_path *path;
+       struct btrfs_path path;
        u64 used = 0;
        int ret = 0;
 
        if (list_empty(&rc->rebuild_chunks))
                return 0;
 
-       path = btrfs_alloc_path();
-       if (!path)
-               return -ENOMEM;
+       btrfs_init_path(&path);
        list_for_each_entry(chunk_rec, &rc->rebuild_chunks, list) {
                search_key.objectid = chunk_rec->offset;
                search_key.type = BTRFS_EXTENT_ITEM_KEY;
                search_key.offset = 0;
                ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
-                                       &search_key, path, 0, 0);
+                                       &search_key, &path, 0, 0);
                if (ret < 0)
                        goto out;
                ret = calculate_bg_used(root->fs_info->extent_root,
-                                       chunk_rec, path, &used);
+                                       chunk_rec, &path, &used);
                /*
                 * Extent tree is damaged, better to rebuild the whole extent
                 * tree. Currently, change the used to chunk's len to prevent
@@ -1399,7 +1430,7 @@ static int rebuild_block_group(struct btrfs_trans_handle *trans,
                                "Mark the block group full to prevent block rsv problems\n");
                        used = chunk_rec->length;
                }
-               btrfs_release_path(path);
+               btrfs_release_path(&path);
                ret = __insert_block_group(trans, chunk_rec,
                                           root->fs_info->extent_root,
                                           used);
@@ -1407,7 +1438,7 @@ static int rebuild_block_group(struct btrfs_trans_handle *trans,
                        goto out;
        }
 out:
-       btrfs_free_path(path);
+       btrfs_release_path(&path);
        return ret;
 }
 
@@ -1437,7 +1468,8 @@ open_ctree_with_broken_chunk(struct recover_control *rc)
 
        disk_super = fs_info->super_copy;
        ret = btrfs_read_dev_super(fs_info->fs_devices->latest_bdev,
-                                  disk_super, fs_info->super_bytenr, 1);
+                                  disk_super, fs_info->super_bytenr,
+                                  SBREAD_RECOVER);
        if (ret) {
                fprintf(stderr, "No valid btrfs found\n");
                goto out_devices;
@@ -1445,7 +1477,7 @@ open_ctree_with_broken_chunk(struct recover_control *rc)
 
        memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
 
-       ret = btrfs_check_fs_compatibility(disk_super, 1);
+       ret = btrfs_check_fs_compatibility(disk_super, OPEN_CTREE_WRITES);
        if (ret)
                goto out_devices;
 
@@ -1454,7 +1486,7 @@ open_ctree_with_broken_chunk(struct recover_control *rc)
        sectorsize = btrfs_super_sectorsize(disk_super);
        stripesize = btrfs_super_stripesize(disk_super);
 
-       __setup_root(nodesize, leafsize, sectorsize, stripesize,
+       btrfs_setup_root(nodesize, leafsize, sectorsize, stripesize,
                     fs_info->chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
 
        ret = build_device_maps_by_chunk_records(rc, fs_info->chunk_root);
@@ -1487,6 +1519,7 @@ static int recover_prepare(struct recover_control *rc, char *path)
        int ret;
        int fd;
        struct btrfs_super_block *sb;
+       char buf[BTRFS_SUPER_INFO_SIZE];
        struct btrfs_fs_devices *fs_devices;
 
        ret = 0;
@@ -1496,21 +1529,16 @@ static int recover_prepare(struct recover_control *rc, char *path)
                return -1;
        }
 
-       sb = malloc(BTRFS_SUPER_INFO_SIZE);
-       if (!sb) {
-               fprintf(stderr, "allocating memory for sb failed.\n");
-               ret = -ENOMEM;
-               goto fail_close_fd;
-       }
-
-       ret = btrfs_read_dev_super(fd, sb, BTRFS_SUPER_INFO_OFFSET, 1);
+       sb = (struct btrfs_super_block*)buf;
+       ret = btrfs_read_dev_super(fd, sb, BTRFS_SUPER_INFO_OFFSET,
+                       SBREAD_RECOVER);
        if (ret) {
                fprintf(stderr, "read super block error\n");
-               goto fail_free_sb;
+               goto out_close_fd;
        }
 
        rc->sectorsize = btrfs_super_sectorsize(sb);
-       rc->leafsize = btrfs_super_leafsize(sb);
+       rc->nodesize = btrfs_super_nodesize(sb);
        rc->generation = btrfs_super_generation(sb);
        rc->chunk_root_generation = btrfs_super_chunk_root_generation(sb);
        rc->csum_size = btrfs_super_csum_size(sb);
@@ -1519,21 +1547,19 @@ static int recover_prepare(struct recover_control *rc, char *path)
        if (btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_SEEDING) {
                fprintf(stderr, "this device is seed device\n");
                ret = -1;
-               goto fail_free_sb;
+               goto out_close_fd;
        }
 
-       ret = btrfs_scan_fs_devices(fd, path, &fs_devices, 0, 1, 0);
+       ret = btrfs_scan_fs_devices(fd, path, &fs_devices, 0, SBREAD_RECOVER, 0);
        if (ret)
-               goto fail_free_sb;
+               goto out_close_fd;
 
        rc->fs_devices = fs_devices;
 
        if (rc->verbose)
                print_all_devices(&rc->fs_devices->devices);
 
-fail_free_sb:
-       free(sb);
-fail_close_fd:
+out_close_fd:
        close(fd);
        return ret;
 }
@@ -1581,16 +1607,19 @@ static int btrfs_verify_device_extents(struct block_group_record *bg,
                                       struct list_head *devexts, int ndevexts)
 {
        struct device_extent_record *devext;
-       u64 strpie_length;
+       u64 stripe_length;
        int expected_num_stripes;
 
        expected_num_stripes = calc_num_stripes(bg->flags);
        if (expected_num_stripes && expected_num_stripes != ndevexts)
                return 1;
 
-       strpie_length = calc_stripe_length(bg->flags, bg->offset, ndevexts);
+       if (check_num_stripes(bg->flags, ndevexts) < 0)
+               return 1;
+
+       stripe_length = calc_stripe_length(bg->flags, bg->offset, ndevexts);
        list_for_each_entry(devext, devexts, chunk_list) {
-               if (devext->length != strpie_length)
+               if (devext->length != stripe_length)
                        return 1;
        }
        return 0;
@@ -1883,7 +1912,7 @@ static int check_one_csum(int fd, u64 start, u32 len, u32 tree_csum)
        }
        ret = 0;
        csum_result = btrfs_csum_data(NULL, data, csum_result, len);
-       btrfs_csum_final(csum_result, (char *)&csum_result);
+       btrfs_csum_final(csum_result, (u8 *)&csum_result);
        if (csum_result != tree_csum)
                ret = 1;
 out:
@@ -1916,9 +1945,12 @@ static int insert_stripe(struct list_head *devexts,
        dev = btrfs_find_device_by_devid(rc->fs_devices, devext->objectid,
                                        0);
        if (!dev)
-               return 1;
-       BUG_ON(btrfs_find_device_by_devid(rc->fs_devices, devext->objectid,
-                                       1));
+               return -ENOENT;
+       if (btrfs_find_device_by_devid(rc->fs_devices, devext->objectid, 1)) {
+               error("unexpected: found another device with id %llu",
+                               (unsigned long long)devext->objectid);
+               return -EINVAL;
+       }
 
        chunk->stripes[index].devid = devext->objectid;
        chunk->stripes[index].offset = devext->offset;
@@ -2195,10 +2227,9 @@ static int btrfs_recover_chunks(struct recover_control *rc)
                nstripes = btrfs_get_device_extents(bg->objectid,
                                                    &rc->devext.no_chunk_orphans,
                                                    &devexts);
-               chunk = malloc(btrfs_chunk_record_size(nstripes));
+               chunk = calloc(1, btrfs_chunk_record_size(nstripes));
                if (!chunk)
                        return -ENOMEM;
-               memset(chunk, 0, btrfs_chunk_record_size(nstripes));
                INIT_LIST_HEAD(&chunk->dextents);
                chunk->bg_rec = bg;
                chunk->cache.start = bg->objectid;
@@ -2217,6 +2248,13 @@ static int btrfs_recover_chunks(struct recover_control *rc)
                chunk->sub_stripes = calc_sub_nstripes(bg->flags);
 
                ret = insert_cache_extent(&rc->chunk, &chunk->cache);
+               if (ret == -EEXIST) {
+                       error("duplicate entry in cache start %llu size %llu",
+                                       (unsigned long long)chunk->cache.start,
+                                       (unsigned long long)chunk->cache.size);
+                       free(chunk);
+                       return ret;
+               }
                BUG_ON(ret);
 
                list_del_init(&bg->list);
@@ -2278,7 +2316,7 @@ static void validate_rebuild_chunks(struct recover_control *rc)
 }
 
 /*
- * Return 0 when succesful, < 0 on error and > 0 if aborted by user
+ * Return 0 when successful, < 0 on error and > 0 if aborted by user
  */
 int btrfs_recover_chunk_tree(char *path, int verbose, int yes)
 {