1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6065631f6SChris Mason #include <linux/bio.h> 75a0e3ad6STejun Heo #include <linux/slab.h> 8065631f6SChris Mason #include <linux/pagemap.h> 9065631f6SChris Mason #include <linux/highmem.h> 10a3d46aeaSNikolay Borisov #include <linux/sched/mm.h> 11d5178578SJohannes Thumshirn #include <crypto/hash.h> 12*9b569ea0SJosef Bacik #include "messages.h" 13cea62800SJohannes Thumshirn #include "misc.h" 141e1d2701SChris Mason #include "ctree.h" 15dee26a9fSChris Mason #include "disk-io.h" 169f5fae2fSChris Mason #include "transaction.h" 17facc8a22SMiao Xie #include "volumes.h" 181de037a4SChris Mason #include "print-tree.h" 19ebb8765bSAnand Jain #include "compression.h" 20c7f13d42SJosef Bacik #include "fs.h" 211e1d2701SChris Mason 2242049bf6SChris Mason #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \ 2342049bf6SChris Mason sizeof(struct btrfs_item) * 2) / \ 2442049bf6SChris Mason size) - 1)) 2507d400a6SYan Zheng 26221b8318SZach Brown #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \ 2709cbfeafSKirill A. Shutemov PAGE_SIZE)) 287ca4be45SChris Mason 2941a2ee75SJosef Bacik /** 30ca4207aeSNikolay Borisov * Set inode's size according to filesystem options 31ca4207aeSNikolay Borisov * 32ca4207aeSNikolay Borisov * @inode: inode we want to update the disk_i_size for 33ca4207aeSNikolay Borisov * @new_i_size: i_size we want to set to, 0 if we use i_size 3441a2ee75SJosef Bacik * 3541a2ee75SJosef Bacik * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read() 3641a2ee75SJosef Bacik * returns as it is perfectly fine with a file that has holes without hole file 3741a2ee75SJosef Bacik * extent items. 3841a2ee75SJosef Bacik * 3941a2ee75SJosef Bacik * However without NO_HOLES we need to only return the area that is contiguous 4041a2ee75SJosef Bacik * from the 0 offset of the file. Otherwise we could end up adjust i_size up 4141a2ee75SJosef Bacik * to an extent that has a gap in between. 4241a2ee75SJosef Bacik * 4341a2ee75SJosef Bacik * Finally new_i_size should only be set in the case of truncate where we're not 4441a2ee75SJosef Bacik * ready to use i_size_read() as the limiter yet. 4541a2ee75SJosef Bacik */ 4676aea537SNikolay Borisov void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size) 4741a2ee75SJosef Bacik { 4876aea537SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 4941a2ee75SJosef Bacik u64 start, end, i_size; 5041a2ee75SJosef Bacik int ret; 5141a2ee75SJosef Bacik 5276aea537SNikolay Borisov i_size = new_i_size ?: i_size_read(&inode->vfs_inode); 5341a2ee75SJosef Bacik if (btrfs_fs_incompat(fs_info, NO_HOLES)) { 5476aea537SNikolay Borisov inode->disk_i_size = i_size; 5541a2ee75SJosef Bacik return; 5641a2ee75SJosef Bacik } 5741a2ee75SJosef Bacik 5876aea537SNikolay Borisov spin_lock(&inode->lock); 5976aea537SNikolay Borisov ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start, 6076aea537SNikolay Borisov &end, EXTENT_DIRTY); 6141a2ee75SJosef Bacik if (!ret && start == 0) 6241a2ee75SJosef Bacik i_size = min(i_size, end + 1); 6341a2ee75SJosef Bacik else 6441a2ee75SJosef Bacik i_size = 0; 6576aea537SNikolay Borisov inode->disk_i_size = i_size; 6676aea537SNikolay Borisov spin_unlock(&inode->lock); 6741a2ee75SJosef Bacik } 6841a2ee75SJosef Bacik 6941a2ee75SJosef Bacik /** 70ca4207aeSNikolay Borisov * Mark range within a file as having a new extent inserted 71ca4207aeSNikolay Borisov * 72ca4207aeSNikolay Borisov * @inode: inode being modified 73ca4207aeSNikolay Borisov * @start: start file offset of the file extent we've inserted 74ca4207aeSNikolay Borisov * @len: logical length of the file extent item 7541a2ee75SJosef Bacik * 7641a2ee75SJosef Bacik * Call when we are inserting a new file extent where there was none before. 7741a2ee75SJosef Bacik * Does not need to call this in the case where we're replacing an existing file 7841a2ee75SJosef Bacik * extent, however if not sure it's fine to call this multiple times. 7941a2ee75SJosef Bacik * 8041a2ee75SJosef Bacik * The start and len must match the file extent item, so thus must be sectorsize 8141a2ee75SJosef Bacik * aligned. 8241a2ee75SJosef Bacik */ 8341a2ee75SJosef Bacik int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start, 8441a2ee75SJosef Bacik u64 len) 8541a2ee75SJosef Bacik { 8641a2ee75SJosef Bacik if (len == 0) 8741a2ee75SJosef Bacik return 0; 8841a2ee75SJosef Bacik 8941a2ee75SJosef Bacik ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize)); 9041a2ee75SJosef Bacik 9141a2ee75SJosef Bacik if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES)) 9241a2ee75SJosef Bacik return 0; 9341a2ee75SJosef Bacik return set_extent_bits(&inode->file_extent_tree, start, start + len - 1, 9441a2ee75SJosef Bacik EXTENT_DIRTY); 9541a2ee75SJosef Bacik } 9641a2ee75SJosef Bacik 9741a2ee75SJosef Bacik /** 98ca4207aeSNikolay Borisov * Marks an inode range as not having a backing extent 99ca4207aeSNikolay Borisov * 100ca4207aeSNikolay Borisov * @inode: inode being modified 101ca4207aeSNikolay Borisov * @start: start file offset of the file extent we've inserted 102ca4207aeSNikolay Borisov * @len: logical length of the file extent item 10341a2ee75SJosef Bacik * 10441a2ee75SJosef Bacik * Called when we drop a file extent, for example when we truncate. Doesn't 10541a2ee75SJosef Bacik * need to be called for cases where we're replacing a file extent, like when 10641a2ee75SJosef Bacik * we've COWed a file extent. 10741a2ee75SJosef Bacik * 10841a2ee75SJosef Bacik * The start and len must match the file extent item, so thus must be sectorsize 10941a2ee75SJosef Bacik * aligned. 11041a2ee75SJosef Bacik */ 11141a2ee75SJosef Bacik int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start, 11241a2ee75SJosef Bacik u64 len) 11341a2ee75SJosef Bacik { 11441a2ee75SJosef Bacik if (len == 0) 11541a2ee75SJosef Bacik return 0; 11641a2ee75SJosef Bacik 11741a2ee75SJosef Bacik ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) || 11841a2ee75SJosef Bacik len == (u64)-1); 11941a2ee75SJosef Bacik 12041a2ee75SJosef Bacik if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES)) 12141a2ee75SJosef Bacik return 0; 12241a2ee75SJosef Bacik return clear_extent_bit(&inode->file_extent_tree, start, 123bd015294SJosef Bacik start + len - 1, EXTENT_DIRTY, NULL); 12441a2ee75SJosef Bacik } 12541a2ee75SJosef Bacik 1261e25a2e3SJohannes Thumshirn static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info, 1271e25a2e3SJohannes Thumshirn u16 csum_size) 1281e25a2e3SJohannes Thumshirn { 1291e25a2e3SJohannes Thumshirn u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size; 1301e25a2e3SJohannes Thumshirn 1311e25a2e3SJohannes Thumshirn return ncsums * fs_info->sectorsize; 1321e25a2e3SJohannes Thumshirn } 13307d400a6SYan Zheng 1342b6433c7SJosef Bacik /* 1352b6433c7SJosef Bacik * Calculate the total size needed to allocate for an ordered sum structure 1362b6433c7SJosef Bacik * spanning @bytes in the file. 1372b6433c7SJosef Bacik */ 1382b6433c7SJosef Bacik static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes) 1392b6433c7SJosef Bacik { 1402b6433c7SJosef Bacik int num_sectors = (int)DIV_ROUND_UP(bytes, fs_info->sectorsize); 1412b6433c7SJosef Bacik 1422b6433c7SJosef Bacik return sizeof(struct btrfs_ordered_sum) + num_sectors * fs_info->csum_size; 1432b6433c7SJosef Bacik } 1442b6433c7SJosef Bacik 145d1f68ba0SOmar Sandoval int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans, 146dee26a9fSChris Mason struct btrfs_root *root, 147d1f68ba0SOmar Sandoval u64 objectid, u64 pos, u64 num_bytes) 1489f5fae2fSChris Mason { 149dee26a9fSChris Mason int ret = 0; 150dee26a9fSChris Mason struct btrfs_file_extent_item *item; 151dee26a9fSChris Mason struct btrfs_key file_key; 1525caf2a00SChris Mason struct btrfs_path *path; 1535f39d397SChris Mason struct extent_buffer *leaf; 154dee26a9fSChris Mason 1555caf2a00SChris Mason path = btrfs_alloc_path(); 156db5b493aSTsutomu Itoh if (!path) 157db5b493aSTsutomu Itoh return -ENOMEM; 158dee26a9fSChris Mason file_key.objectid = objectid; 159b18c6685SChris Mason file_key.offset = pos; 160962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_DATA_KEY; 161dee26a9fSChris Mason 1625caf2a00SChris Mason ret = btrfs_insert_empty_item(trans, root, path, &file_key, 163dee26a9fSChris Mason sizeof(*item)); 16454aa1f4dSChris Mason if (ret < 0) 16554aa1f4dSChris Mason goto out; 16679787eaaSJeff Mahoney BUG_ON(ret); /* Can't happen */ 1675f39d397SChris Mason leaf = path->nodes[0]; 1685f39d397SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], 169dee26a9fSChris Mason struct btrfs_file_extent_item); 170d1f68ba0SOmar Sandoval btrfs_set_file_extent_disk_bytenr(leaf, item, 0); 171d1f68ba0SOmar Sandoval btrfs_set_file_extent_disk_num_bytes(leaf, item, 0); 172d1f68ba0SOmar Sandoval btrfs_set_file_extent_offset(leaf, item, 0); 173db94535dSChris Mason btrfs_set_file_extent_num_bytes(leaf, item, num_bytes); 174d1f68ba0SOmar Sandoval btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes); 1755f39d397SChris Mason btrfs_set_file_extent_generation(leaf, item, trans->transid); 1765f39d397SChris Mason btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); 177d1f68ba0SOmar Sandoval btrfs_set_file_extent_compression(leaf, item, 0); 178d1f68ba0SOmar Sandoval btrfs_set_file_extent_encryption(leaf, item, 0); 179d1f68ba0SOmar Sandoval btrfs_set_file_extent_other_encoding(leaf, item, 0); 180c8b97818SChris Mason 1815f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 18254aa1f4dSChris Mason out: 1835caf2a00SChris Mason btrfs_free_path(path); 18454aa1f4dSChris Mason return ret; 1859f5fae2fSChris Mason } 186dee26a9fSChris Mason 18748a3b636SEric Sandeen static struct btrfs_csum_item * 18848a3b636SEric Sandeen btrfs_lookup_csum(struct btrfs_trans_handle *trans, 189b18c6685SChris Mason struct btrfs_root *root, 1906567e837SChris Mason struct btrfs_path *path, 191d20f7043SChris Mason u64 bytenr, int cow) 1926567e837SChris Mason { 1930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1946567e837SChris Mason int ret; 1956567e837SChris Mason struct btrfs_key file_key; 1966567e837SChris Mason struct btrfs_key found_key; 1976567e837SChris Mason struct btrfs_csum_item *item; 1985f39d397SChris Mason struct extent_buffer *leaf; 1996567e837SChris Mason u64 csum_offset = 0; 200223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 201a429e513SChris Mason int csums_in_item; 2026567e837SChris Mason 203d20f7043SChris Mason file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 204d20f7043SChris Mason file_key.offset = bytenr; 205962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_CSUM_KEY; 206b18c6685SChris Mason ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow); 2076567e837SChris Mason if (ret < 0) 2086567e837SChris Mason goto fail; 2095f39d397SChris Mason leaf = path->nodes[0]; 2106567e837SChris Mason if (ret > 0) { 2116567e837SChris Mason ret = 1; 21270b2befdSChris Mason if (path->slots[0] == 0) 2136567e837SChris Mason goto fail; 2146567e837SChris Mason path->slots[0]--; 2155f39d397SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 216962a298fSDavid Sterba if (found_key.type != BTRFS_EXTENT_CSUM_KEY) 2176567e837SChris Mason goto fail; 218d20f7043SChris Mason 219d20f7043SChris Mason csum_offset = (bytenr - found_key.offset) >> 220265fdfa6SDavid Sterba fs_info->sectorsize_bits; 2213212fa14SJosef Bacik csums_in_item = btrfs_item_size(leaf, path->slots[0]); 222607d432dSJosef Bacik csums_in_item /= csum_size; 223a429e513SChris Mason 22482d130ffSMiao Xie if (csum_offset == csums_in_item) { 225a429e513SChris Mason ret = -EFBIG; 2266567e837SChris Mason goto fail; 22782d130ffSMiao Xie } else if (csum_offset > csums_in_item) { 22882d130ffSMiao Xie goto fail; 2296567e837SChris Mason } 2306567e837SChris Mason } 2316567e837SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item); 232509659cdSChris Mason item = (struct btrfs_csum_item *)((unsigned char *)item + 233607d432dSJosef Bacik csum_offset * csum_size); 2346567e837SChris Mason return item; 2356567e837SChris Mason fail: 2366567e837SChris Mason if (ret > 0) 237b18c6685SChris Mason ret = -ENOENT; 2386567e837SChris Mason return ERR_PTR(ret); 2396567e837SChris Mason } 2406567e837SChris Mason 241dee26a9fSChris Mason int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans, 242dee26a9fSChris Mason struct btrfs_root *root, 243dee26a9fSChris Mason struct btrfs_path *path, u64 objectid, 2449773a788SChris Mason u64 offset, int mod) 245dee26a9fSChris Mason { 246dee26a9fSChris Mason struct btrfs_key file_key; 247dee26a9fSChris Mason int ins_len = mod < 0 ? -1 : 0; 248dee26a9fSChris Mason int cow = mod != 0; 249dee26a9fSChris Mason 250dee26a9fSChris Mason file_key.objectid = objectid; 25170b2befdSChris Mason file_key.offset = offset; 252962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_DATA_KEY; 253f8ee80deSMarcos Paulo de Souza 254f8ee80deSMarcos Paulo de Souza return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow); 255dee26a9fSChris Mason } 256f254e52cSChris Mason 2576275193eSQu Wenruo /* 2586275193eSQu Wenruo * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and 2596275193eSQu Wenruo * estore the result to @dst. 2606275193eSQu Wenruo * 2616275193eSQu Wenruo * Return >0 for the number of sectors we found. 2626275193eSQu Wenruo * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum 2636275193eSQu Wenruo * for it. Caller may want to try next sector until one range is hit. 2646275193eSQu Wenruo * Return <0 for fatal error. 2656275193eSQu Wenruo */ 2666275193eSQu Wenruo static int search_csum_tree(struct btrfs_fs_info *fs_info, 2676275193eSQu Wenruo struct btrfs_path *path, u64 disk_bytenr, 2686275193eSQu Wenruo u64 len, u8 *dst) 2696275193eSQu Wenruo { 270fc28b25eSJosef Bacik struct btrfs_root *csum_root; 2716275193eSQu Wenruo struct btrfs_csum_item *item = NULL; 2726275193eSQu Wenruo struct btrfs_key key; 2736275193eSQu Wenruo const u32 sectorsize = fs_info->sectorsize; 2746275193eSQu Wenruo const u32 csum_size = fs_info->csum_size; 2756275193eSQu Wenruo u32 itemsize; 2766275193eSQu Wenruo int ret; 2776275193eSQu Wenruo u64 csum_start; 2786275193eSQu Wenruo u64 csum_len; 2796275193eSQu Wenruo 2806275193eSQu Wenruo ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) && 2816275193eSQu Wenruo IS_ALIGNED(len, sectorsize)); 2826275193eSQu Wenruo 2836275193eSQu Wenruo /* Check if the current csum item covers disk_bytenr */ 2846275193eSQu Wenruo if (path->nodes[0]) { 2856275193eSQu Wenruo item = btrfs_item_ptr(path->nodes[0], path->slots[0], 2866275193eSQu Wenruo struct btrfs_csum_item); 2876275193eSQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2883212fa14SJosef Bacik itemsize = btrfs_item_size(path->nodes[0], path->slots[0]); 2896275193eSQu Wenruo 2906275193eSQu Wenruo csum_start = key.offset; 2916275193eSQu Wenruo csum_len = (itemsize / csum_size) * sectorsize; 2926275193eSQu Wenruo 2936275193eSQu Wenruo if (in_range(disk_bytenr, csum_start, csum_len)) 2946275193eSQu Wenruo goto found; 2956275193eSQu Wenruo } 2966275193eSQu Wenruo 2976275193eSQu Wenruo /* Current item doesn't contain the desired range, search again */ 2986275193eSQu Wenruo btrfs_release_path(path); 299fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 300fc28b25eSJosef Bacik item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0); 3016275193eSQu Wenruo if (IS_ERR(item)) { 3026275193eSQu Wenruo ret = PTR_ERR(item); 3036275193eSQu Wenruo goto out; 3046275193eSQu Wenruo } 3056275193eSQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 3063212fa14SJosef Bacik itemsize = btrfs_item_size(path->nodes[0], path->slots[0]); 3076275193eSQu Wenruo 3086275193eSQu Wenruo csum_start = key.offset; 3096275193eSQu Wenruo csum_len = (itemsize / csum_size) * sectorsize; 3106275193eSQu Wenruo ASSERT(in_range(disk_bytenr, csum_start, csum_len)); 3116275193eSQu Wenruo 3126275193eSQu Wenruo found: 3136275193eSQu Wenruo ret = (min(csum_start + csum_len, disk_bytenr + len) - 3146275193eSQu Wenruo disk_bytenr) >> fs_info->sectorsize_bits; 3156275193eSQu Wenruo read_extent_buffer(path->nodes[0], dst, (unsigned long)item, 3166275193eSQu Wenruo ret * csum_size); 3176275193eSQu Wenruo out: 31803ddb19dSJosef Bacik if (ret == -ENOENT || ret == -EFBIG) 3196275193eSQu Wenruo ret = 0; 3206275193eSQu Wenruo return ret; 3216275193eSQu Wenruo } 3226275193eSQu Wenruo 3236275193eSQu Wenruo /* 3246275193eSQu Wenruo * Locate the file_offset of @cur_disk_bytenr of a @bio. 3256275193eSQu Wenruo * 3266275193eSQu Wenruo * Bio of btrfs represents read range of 3276275193eSQu Wenruo * [bi_sector << 9, bi_sector << 9 + bi_size). 3286275193eSQu Wenruo * Knowing this, we can iterate through each bvec to locate the page belong to 3296275193eSQu Wenruo * @cur_disk_bytenr and get the file offset. 3306275193eSQu Wenruo * 3316275193eSQu Wenruo * @inode is used to determine if the bvec page really belongs to @inode. 3326275193eSQu Wenruo * 3336275193eSQu Wenruo * Return 0 if we can't find the file offset 3346275193eSQu Wenruo * Return >0 if we find the file offset and restore it to @file_offset_ret 3356275193eSQu Wenruo */ 3366275193eSQu Wenruo static int search_file_offset_in_bio(struct bio *bio, struct inode *inode, 3376275193eSQu Wenruo u64 disk_bytenr, u64 *file_offset_ret) 3386275193eSQu Wenruo { 3396275193eSQu Wenruo struct bvec_iter iter; 3406275193eSQu Wenruo struct bio_vec bvec; 3416275193eSQu Wenruo u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT; 3426275193eSQu Wenruo int ret = 0; 3436275193eSQu Wenruo 3446275193eSQu Wenruo bio_for_each_segment(bvec, bio, iter) { 3456275193eSQu Wenruo struct page *page = bvec.bv_page; 3466275193eSQu Wenruo 3476275193eSQu Wenruo if (cur > disk_bytenr) 3486275193eSQu Wenruo break; 3496275193eSQu Wenruo if (cur + bvec.bv_len <= disk_bytenr) { 3506275193eSQu Wenruo cur += bvec.bv_len; 3516275193eSQu Wenruo continue; 3526275193eSQu Wenruo } 3536275193eSQu Wenruo ASSERT(in_range(disk_bytenr, cur, bvec.bv_len)); 3546275193eSQu Wenruo if (page->mapping && page->mapping->host && 3556275193eSQu Wenruo page->mapping->host == inode) { 3566275193eSQu Wenruo ret = 1; 3576275193eSQu Wenruo *file_offset_ret = page_offset(page) + bvec.bv_offset + 3586275193eSQu Wenruo disk_bytenr - cur; 3596275193eSQu Wenruo break; 3606275193eSQu Wenruo } 3616275193eSQu Wenruo } 3626275193eSQu Wenruo return ret; 3636275193eSQu Wenruo } 3646275193eSQu Wenruo 365e62958fcSOmar Sandoval /** 3666275193eSQu Wenruo * Lookup the checksum for the read bio in csum tree. 3679e46458aSQu Wenruo * 368e62958fcSOmar Sandoval * @inode: inode that the bio is for. 369fb30f470SOmar Sandoval * @bio: bio to look up. 370fb30f470SOmar Sandoval * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return 371fb30f470SOmar Sandoval * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If 372fb30f470SOmar Sandoval * NULL, the checksum buffer is allocated and returned in 373c3a3b19bSQu Wenruo * btrfs_bio(bio)->csum instead. 374e62958fcSOmar Sandoval * 375e62958fcSOmar Sandoval * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise. 376e62958fcSOmar Sandoval */ 3776275193eSQu Wenruo blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst) 37861b49440SChris Mason { 3790b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 380facc8a22SMiao Xie struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 3811784b7d5SJosef Bacik struct btrfs_bio *bbio = NULL; 382facc8a22SMiao Xie struct btrfs_path *path; 3836275193eSQu Wenruo const u32 sectorsize = fs_info->sectorsize; 384223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 3856275193eSQu Wenruo u32 orig_len = bio->bi_iter.bi_size; 3866275193eSQu Wenruo u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT; 3876275193eSQu Wenruo u64 cur_disk_bytenr; 3886275193eSQu Wenruo u8 *csum; 3896275193eSQu Wenruo const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits; 3906275193eSQu Wenruo int count = 0; 3911784b7d5SJosef Bacik blk_status_t ret = BLK_STS_OK; 39261b49440SChris Mason 393056c8311SJosef Bacik if ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) || 394056c8311SJosef Bacik test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)) 395334c16d8SJosef Bacik return BLK_STS_OK; 396334c16d8SJosef Bacik 3979e46458aSQu Wenruo /* 3989e46458aSQu Wenruo * This function is only called for read bio. 3999e46458aSQu Wenruo * 4009e46458aSQu Wenruo * This means two things: 4019e46458aSQu Wenruo * - All our csums should only be in csum tree 4029e46458aSQu Wenruo * No ordered extents csums, as ordered extents are only for write 4039e46458aSQu Wenruo * path. 4046275193eSQu Wenruo * - No need to bother any other info from bvec 4056275193eSQu Wenruo * Since we're looking up csums, the only important info is the 4066275193eSQu Wenruo * disk_bytenr and the length, which can be extracted from bi_iter 4076275193eSQu Wenruo * directly. 4089e46458aSQu Wenruo */ 4099e46458aSQu Wenruo ASSERT(bio_op(bio) == REQ_OP_READ); 41061b49440SChris Mason path = btrfs_alloc_path(); 411c2db1073STsutomu Itoh if (!path) 4124e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 413facc8a22SMiao Xie 414facc8a22SMiao Xie if (!dst) { 4151784b7d5SJosef Bacik bbio = btrfs_bio(bio); 416fb30f470SOmar Sandoval 417facc8a22SMiao Xie if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) { 418c3a3b19bSQu Wenruo bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS); 419c3a3b19bSQu Wenruo if (!bbio->csum) { 420facc8a22SMiao Xie btrfs_free_path(path); 4214e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 422facc8a22SMiao Xie } 423facc8a22SMiao Xie } else { 424c3a3b19bSQu Wenruo bbio->csum = bbio->csum_inline; 425facc8a22SMiao Xie } 426c3a3b19bSQu Wenruo csum = bbio->csum; 427facc8a22SMiao Xie } else { 42810fe6ca8SJohannes Thumshirn csum = dst; 429facc8a22SMiao Xie } 430facc8a22SMiao Xie 43135478d05SQu Wenruo /* 43235478d05SQu Wenruo * If requested number of sectors is larger than one leaf can contain, 43335478d05SQu Wenruo * kick the readahead for csum tree. 43435478d05SQu Wenruo */ 43535478d05SQu Wenruo if (nblocks > fs_info->csums_per_leaf) 436e4058b54SDavid Sterba path->reada = READA_FORWARD; 43761b49440SChris Mason 4382cf8572dSChris Mason /* 4392cf8572dSChris Mason * the free space stuff is only read when it hasn't been 4402cf8572dSChris Mason * updated in the current transaction. So, we can safely 4412cf8572dSChris Mason * read from the commit root and sidestep a nasty deadlock 4422cf8572dSChris Mason * between reading the free space cache and updating the csum tree. 4432cf8572dSChris Mason */ 44470ddc553SNikolay Borisov if (btrfs_is_free_space_inode(BTRFS_I(inode))) { 4452cf8572dSChris Mason path->search_commit_root = 1; 446ddf23b3fSJosef Bacik path->skip_locking = 1; 447ddf23b3fSJosef Bacik } 4482cf8572dSChris Mason 4496275193eSQu Wenruo for (cur_disk_bytenr = orig_disk_bytenr; 4506275193eSQu Wenruo cur_disk_bytenr < orig_disk_bytenr + orig_len; 4516275193eSQu Wenruo cur_disk_bytenr += (count * sectorsize)) { 4526275193eSQu Wenruo u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr; 4536275193eSQu Wenruo unsigned int sector_offset; 4546275193eSQu Wenruo u8 *csum_dst; 455c40a3d38SChandan Rajendra 4566275193eSQu Wenruo /* 4576275193eSQu Wenruo * Although both cur_disk_bytenr and orig_disk_bytenr is u64, 4586275193eSQu Wenruo * we're calculating the offset to the bio start. 4596275193eSQu Wenruo * 4606275193eSQu Wenruo * Bio size is limited to UINT_MAX, thus unsigned int is large 4616275193eSQu Wenruo * enough to contain the raw result, not to mention the right 4626275193eSQu Wenruo * shifted result. 4636275193eSQu Wenruo */ 4646275193eSQu Wenruo ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX); 4656275193eSQu Wenruo sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >> 4666275193eSQu Wenruo fs_info->sectorsize_bits; 4676275193eSQu Wenruo csum_dst = csum + sector_offset * csum_size; 4684989d277SChristoph Hellwig 4696275193eSQu Wenruo count = search_csum_tree(fs_info, path, cur_disk_bytenr, 4706275193eSQu Wenruo search_len, csum_dst); 4711784b7d5SJosef Bacik if (count < 0) { 4721784b7d5SJosef Bacik ret = errno_to_blk_status(count); 4731784b7d5SJosef Bacik if (bbio) 4741784b7d5SJosef Bacik btrfs_bio_free_csum(bbio); 4751784b7d5SJosef Bacik break; 4761784b7d5SJosef Bacik } 4771784b7d5SJosef Bacik 4786275193eSQu Wenruo /* 4791784b7d5SJosef Bacik * We didn't find a csum for this range. We need to make sure 4801784b7d5SJosef Bacik * we complain loudly about this, because we are not NODATASUM. 4811784b7d5SJosef Bacik * 4821784b7d5SJosef Bacik * However for the DATA_RELOC inode we could potentially be 4831784b7d5SJosef Bacik * relocating data extents for a NODATASUM inode, so the inode 4841784b7d5SJosef Bacik * itself won't be marked with NODATASUM, but the extent we're 4851784b7d5SJosef Bacik * copying is in fact NODATASUM. If we don't find a csum we 4861784b7d5SJosef Bacik * assume this is the case. 4876275193eSQu Wenruo */ 4881784b7d5SJosef Bacik if (count == 0) { 4896275193eSQu Wenruo memset(csum_dst, 0, csum_size); 490e4100d98SMiao Xie count = 1; 4916275193eSQu Wenruo 49217d217feSYan Zheng if (BTRFS_I(inode)->root->root_key.objectid == 49317d217feSYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID) { 4946275193eSQu Wenruo u64 file_offset; 4956275193eSQu Wenruo int ret; 4966275193eSQu Wenruo 4976275193eSQu Wenruo ret = search_file_offset_in_bio(bio, inode, 4986275193eSQu Wenruo cur_disk_bytenr, &file_offset); 4996275193eSQu Wenruo if (ret) 5006275193eSQu Wenruo set_extent_bits(io_tree, file_offset, 5016275193eSQu Wenruo file_offset + sectorsize - 1, 502ceeb0ae7SDavid Sterba EXTENT_NODATASUM); 50317d217feSYan Zheng } else { 5046275193eSQu Wenruo btrfs_warn_rl(fs_info, 5056275193eSQu Wenruo "csum hole found for disk bytenr range [%llu, %llu)", 5066275193eSQu Wenruo cur_disk_bytenr, cur_disk_bytenr + sectorsize); 50717d217feSYan Zheng } 5084989d277SChristoph Hellwig } 5094989d277SChristoph Hellwig } 5104989d277SChristoph Hellwig 51161b49440SChris Mason btrfs_free_path(path); 5121784b7d5SJosef Bacik return ret; 5134b46fce2SJosef Bacik } 5144b46fce2SJosef Bacik 51517d217feSYan Zheng int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end, 51626ce9114SJosef Bacik struct list_head *list, int search_commit, 51726ce9114SJosef Bacik bool nowait) 51817d217feSYan Zheng { 5190b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 52017d217feSYan Zheng struct btrfs_key key; 52117d217feSYan Zheng struct btrfs_path *path; 52217d217feSYan Zheng struct extent_buffer *leaf; 52317d217feSYan Zheng struct btrfs_ordered_sum *sums; 52417d217feSYan Zheng struct btrfs_csum_item *item; 5250678b618SMark Fasheh LIST_HEAD(tmplist); 52617d217feSYan Zheng unsigned long offset; 52717d217feSYan Zheng int ret; 52817d217feSYan Zheng size_t size; 52917d217feSYan Zheng u64 csum_end; 530223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 53117d217feSYan Zheng 5320b246afaSJeff Mahoney ASSERT(IS_ALIGNED(start, fs_info->sectorsize) && 5330b246afaSJeff Mahoney IS_ALIGNED(end + 1, fs_info->sectorsize)); 5344277a9c3SJosef Bacik 53517d217feSYan Zheng path = btrfs_alloc_path(); 536d8926bb3SMark Fasheh if (!path) 537d8926bb3SMark Fasheh return -ENOMEM; 53817d217feSYan Zheng 53926ce9114SJosef Bacik path->nowait = nowait; 540a2de733cSArne Jansen if (search_commit) { 541a2de733cSArne Jansen path->skip_locking = 1; 542e4058b54SDavid Sterba path->reada = READA_FORWARD; 543a2de733cSArne Jansen path->search_commit_root = 1; 544a2de733cSArne Jansen } 545a2de733cSArne Jansen 54617d217feSYan Zheng key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 54717d217feSYan Zheng key.offset = start; 54817d217feSYan Zheng key.type = BTRFS_EXTENT_CSUM_KEY; 54917d217feSYan Zheng 55007d400a6SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 55117d217feSYan Zheng if (ret < 0) 55217d217feSYan Zheng goto fail; 55317d217feSYan Zheng if (ret > 0 && path->slots[0] > 0) { 55417d217feSYan Zheng leaf = path->nodes[0]; 55517d217feSYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); 55617d217feSYan Zheng if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID && 55717d217feSYan Zheng key.type == BTRFS_EXTENT_CSUM_KEY) { 558265fdfa6SDavid Sterba offset = (start - key.offset) >> fs_info->sectorsize_bits; 55917d217feSYan Zheng if (offset * csum_size < 5603212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0] - 1)) 56117d217feSYan Zheng path->slots[0]--; 56217d217feSYan Zheng } 56317d217feSYan Zheng } 56417d217feSYan Zheng 56517d217feSYan Zheng while (start <= end) { 56617d217feSYan Zheng leaf = path->nodes[0]; 56717d217feSYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 56807d400a6SYan Zheng ret = btrfs_next_leaf(root, path); 56917d217feSYan Zheng if (ret < 0) 57017d217feSYan Zheng goto fail; 57117d217feSYan Zheng if (ret > 0) 57217d217feSYan Zheng break; 57317d217feSYan Zheng leaf = path->nodes[0]; 57417d217feSYan Zheng } 57517d217feSYan Zheng 57617d217feSYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 57717d217feSYan Zheng if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 578628c8282SZhi Yong Wu key.type != BTRFS_EXTENT_CSUM_KEY || 579628c8282SZhi Yong Wu key.offset > end) 58017d217feSYan Zheng break; 58117d217feSYan Zheng 58217d217feSYan Zheng if (key.offset > start) 58317d217feSYan Zheng start = key.offset; 58417d217feSYan Zheng 5853212fa14SJosef Bacik size = btrfs_item_size(leaf, path->slots[0]); 5860b246afaSJeff Mahoney csum_end = key.offset + (size / csum_size) * fs_info->sectorsize; 58787b29b20SYan Zheng if (csum_end <= start) { 58887b29b20SYan Zheng path->slots[0]++; 58987b29b20SYan Zheng continue; 59087b29b20SYan Zheng } 59117d217feSYan Zheng 59207d400a6SYan Zheng csum_end = min(csum_end, end + 1); 59307d400a6SYan Zheng item = btrfs_item_ptr(path->nodes[0], path->slots[0], 59407d400a6SYan Zheng struct btrfs_csum_item); 59507d400a6SYan Zheng while (start < csum_end) { 59607d400a6SYan Zheng size = min_t(size_t, csum_end - start, 5971e25a2e3SJohannes Thumshirn max_ordered_sum_bytes(fs_info, csum_size)); 5980b246afaSJeff Mahoney sums = kzalloc(btrfs_ordered_sum_size(fs_info, size), 59907d400a6SYan Zheng GFP_NOFS); 6000678b618SMark Fasheh if (!sums) { 6010678b618SMark Fasheh ret = -ENOMEM; 6020678b618SMark Fasheh goto fail; 6030678b618SMark Fasheh } 60417d217feSYan Zheng 60517d217feSYan Zheng sums->bytenr = start; 606f51a4a18SMiao Xie sums->len = (int)size; 60717d217feSYan Zheng 608265fdfa6SDavid Sterba offset = (start - key.offset) >> fs_info->sectorsize_bits; 60917d217feSYan Zheng offset *= csum_size; 610265fdfa6SDavid Sterba size >>= fs_info->sectorsize_bits; 61117d217feSYan Zheng 61207d400a6SYan Zheng read_extent_buffer(path->nodes[0], 613f51a4a18SMiao Xie sums->sums, 614f51a4a18SMiao Xie ((unsigned long)item) + offset, 615f51a4a18SMiao Xie csum_size * size); 61617d217feSYan Zheng 6170b246afaSJeff Mahoney start += fs_info->sectorsize * size; 6180678b618SMark Fasheh list_add_tail(&sums->list, &tmplist); 61907d400a6SYan Zheng } 62017d217feSYan Zheng path->slots[0]++; 62117d217feSYan Zheng } 62217d217feSYan Zheng ret = 0; 62317d217feSYan Zheng fail: 6240678b618SMark Fasheh while (ret < 0 && !list_empty(&tmplist)) { 6256e5aafb2SChris Mason sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list); 6260678b618SMark Fasheh list_del(&sums->list); 6270678b618SMark Fasheh kfree(sums); 6280678b618SMark Fasheh } 6290678b618SMark Fasheh list_splice_tail(&tmplist, list); 6300678b618SMark Fasheh 63117d217feSYan Zheng btrfs_free_path(path); 63217d217feSYan Zheng return ret; 63317d217feSYan Zheng } 63417d217feSYan Zheng 635e331f6b1SOmar Sandoval /** 636e331f6b1SOmar Sandoval * Calculate checksums of the data contained inside a bio 637e331f6b1SOmar Sandoval * 63851d470aeSNikolay Borisov * @inode: Owner of the data inside the bio 63951d470aeSNikolay Borisov * @bio: Contains the data to be checksummed 640e331f6b1SOmar Sandoval * @offset: If (u64)-1, @bio may contain discontiguous bio vecs, so the 641e331f6b1SOmar Sandoval * file offsets are determined from the page offsets in the bio. 642e331f6b1SOmar Sandoval * Otherwise, this is the starting file offset of the bio vecs in 643e331f6b1SOmar Sandoval * @bio, which must be contiguous. 644e331f6b1SOmar Sandoval * @one_ordered: If true, @bio only refers to one ordered extent. 64551d470aeSNikolay Borisov */ 646bd242a08SNikolay Borisov blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio, 647e331f6b1SOmar Sandoval u64 offset, bool one_ordered) 648e015640fSChris Mason { 649c3504372SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 650d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 651e6dcd2dcSChris Mason struct btrfs_ordered_sum *sums; 6526cd7ce49SChristoph Hellwig struct btrfs_ordered_extent *ordered = NULL; 653e331f6b1SOmar Sandoval const bool use_page_offsets = (offset == (u64)-1); 654e015640fSChris Mason char *data; 65517347cecSLiu Bo struct bvec_iter iter; 65617347cecSLiu Bo struct bio_vec bvec; 657f51a4a18SMiao Xie int index; 658e331f6b1SOmar Sandoval unsigned int blockcount; 6593edf7d33SChris Mason unsigned long total_bytes = 0; 6603edf7d33SChris Mason unsigned long this_sum_bytes = 0; 66117347cecSLiu Bo int i; 662a3d46aeaSNikolay Borisov unsigned nofs_flag; 663e015640fSChris Mason 664a3d46aeaSNikolay Borisov nofs_flag = memalloc_nofs_save(); 665a3d46aeaSNikolay Borisov sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size), 666a3d46aeaSNikolay Borisov GFP_KERNEL); 667a3d46aeaSNikolay Borisov memalloc_nofs_restore(nofs_flag); 668a3d46aeaSNikolay Borisov 669e015640fSChris Mason if (!sums) 6704e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 6713edf7d33SChris Mason 6724f024f37SKent Overstreet sums->len = bio->bi_iter.bi_size; 673e6dcd2dcSChris Mason INIT_LIST_HEAD(&sums->list); 674d20f7043SChris Mason 6751201b58bSDavid Sterba sums->bytenr = bio->bi_iter.bi_sector << 9; 676f51a4a18SMiao Xie index = 0; 677e015640fSChris Mason 678d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 679d5178578SJohannes Thumshirn 68017347cecSLiu Bo bio_for_each_segment(bvec, bio, iter) { 681e331f6b1SOmar Sandoval if (use_page_offsets) 68217347cecSLiu Bo offset = page_offset(bvec.bv_page) + bvec.bv_offset; 683d20f7043SChris Mason 6846cd7ce49SChristoph Hellwig if (!ordered) { 6856cd7ce49SChristoph Hellwig ordered = btrfs_lookup_ordered_extent(inode, offset); 686bbc9a6ebSQu Wenruo /* 687bbc9a6ebSQu Wenruo * The bio range is not covered by any ordered extent, 688bbc9a6ebSQu Wenruo * must be a code logic error. 689bbc9a6ebSQu Wenruo */ 690bbc9a6ebSQu Wenruo if (unlikely(!ordered)) { 691bbc9a6ebSQu Wenruo WARN(1, KERN_WARNING 692bbc9a6ebSQu Wenruo "no ordered extent for root %llu ino %llu offset %llu\n", 693bbc9a6ebSQu Wenruo inode->root->root_key.objectid, 694bbc9a6ebSQu Wenruo btrfs_ino(inode), offset); 695bbc9a6ebSQu Wenruo kvfree(sums); 696bbc9a6ebSQu Wenruo return BLK_STS_IOERR; 697bbc9a6ebSQu Wenruo } 6986cd7ce49SChristoph Hellwig } 6996cd7ce49SChristoph Hellwig 700e331f6b1SOmar Sandoval blockcount = BTRFS_BYTES_TO_BLKS(fs_info, 70117347cecSLiu Bo bvec.bv_len + fs_info->sectorsize 702c40a3d38SChandan Rajendra - 1); 703c40a3d38SChandan Rajendra 704e331f6b1SOmar Sandoval for (i = 0; i < blockcount; i++) { 705e331f6b1SOmar Sandoval if (!one_ordered && 706e331f6b1SOmar Sandoval !in_range(offset, ordered->file_offset, 707e331f6b1SOmar Sandoval ordered->num_bytes)) { 7083edf7d33SChris Mason unsigned long bytes_left; 709c40a3d38SChandan Rajendra 7103edf7d33SChris Mason sums->len = this_sum_bytes; 7113edf7d33SChris Mason this_sum_bytes = 0; 712f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 7133edf7d33SChris Mason btrfs_put_ordered_extent(ordered); 7143edf7d33SChris Mason 7154f024f37SKent Overstreet bytes_left = bio->bi_iter.bi_size - total_bytes; 7163edf7d33SChris Mason 717a3d46aeaSNikolay Borisov nofs_flag = memalloc_nofs_save(); 718a3d46aeaSNikolay Borisov sums = kvzalloc(btrfs_ordered_sum_size(fs_info, 719a3d46aeaSNikolay Borisov bytes_left), GFP_KERNEL); 720a3d46aeaSNikolay Borisov memalloc_nofs_restore(nofs_flag); 72179787eaaSJeff Mahoney BUG_ON(!sums); /* -ENOMEM */ 7223edf7d33SChris Mason sums->len = bytes_left; 723c40a3d38SChandan Rajendra ordered = btrfs_lookup_ordered_extent(inode, 724c40a3d38SChandan Rajendra offset); 725c40a3d38SChandan Rajendra ASSERT(ordered); /* Logic error */ 7261201b58bSDavid Sterba sums->bytenr = (bio->bi_iter.bi_sector << 9) 727c40a3d38SChandan Rajendra + total_bytes; 728f51a4a18SMiao Xie index = 0; 729c40a3d38SChandan Rajendra } 730c40a3d38SChandan Rajendra 7313dcfbcceSChristoph Hellwig data = bvec_kmap_local(&bvec); 7323dcfbcceSChristoph Hellwig crypto_shash_digest(shash, 7333dcfbcceSChristoph Hellwig data + (i * fs_info->sectorsize), 734fd08001fSEric Biggers fs_info->sectorsize, 735fd08001fSEric Biggers sums->sums + index); 7363dcfbcceSChristoph Hellwig kunmap_local(data); 737713cebfbSDavid Sterba index += fs_info->csum_size; 7380b246afaSJeff Mahoney offset += fs_info->sectorsize; 7390b246afaSJeff Mahoney this_sum_bytes += fs_info->sectorsize; 7400b246afaSJeff Mahoney total_bytes += fs_info->sectorsize; 741c40a3d38SChandan Rajendra } 742c40a3d38SChandan Rajendra 743e015640fSChris Mason } 744ed98b56aSChris Mason this_sum_bytes = 0; 745f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 7463edf7d33SChris Mason btrfs_put_ordered_extent(ordered); 747e015640fSChris Mason return 0; 748e015640fSChris Mason } 749e015640fSChris Mason 750459931ecSChris Mason /* 751459931ecSChris Mason * helper function for csum removal, this expects the 752459931ecSChris Mason * key to describe the csum pointed to by the path, and it expects 753459931ecSChris Mason * the csum to overlap the range [bytenr, len] 754459931ecSChris Mason * 755459931ecSChris Mason * The csum should not be entirely contained in the range and the 756459931ecSChris Mason * range should not be entirely contained in the csum. 757459931ecSChris Mason * 758459931ecSChris Mason * This calls btrfs_truncate_item with the correct args based on the 759459931ecSChris Mason * overlap, and fixes up the key as required. 760459931ecSChris Mason */ 7612ff7e61eSJeff Mahoney static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info, 762459931ecSChris Mason struct btrfs_path *path, 763459931ecSChris Mason struct btrfs_key *key, 764459931ecSChris Mason u64 bytenr, u64 len) 765459931ecSChris Mason { 766459931ecSChris Mason struct extent_buffer *leaf; 767223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 768459931ecSChris Mason u64 csum_end; 769459931ecSChris Mason u64 end_byte = bytenr + len; 770265fdfa6SDavid Sterba u32 blocksize_bits = fs_info->sectorsize_bits; 771459931ecSChris Mason 772459931ecSChris Mason leaf = path->nodes[0]; 7733212fa14SJosef Bacik csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size; 774265fdfa6SDavid Sterba csum_end <<= blocksize_bits; 775459931ecSChris Mason csum_end += key->offset; 776459931ecSChris Mason 777459931ecSChris Mason if (key->offset < bytenr && csum_end <= end_byte) { 778459931ecSChris Mason /* 779459931ecSChris Mason * [ bytenr - len ] 780459931ecSChris Mason * [ ] 781459931ecSChris Mason * [csum ] 782459931ecSChris Mason * A simple truncate off the end of the item 783459931ecSChris Mason */ 784459931ecSChris Mason u32 new_size = (bytenr - key->offset) >> blocksize_bits; 785459931ecSChris Mason new_size *= csum_size; 78678ac4f9eSDavid Sterba btrfs_truncate_item(path, new_size, 1); 787459931ecSChris Mason } else if (key->offset >= bytenr && csum_end > end_byte && 788459931ecSChris Mason end_byte > key->offset) { 789459931ecSChris Mason /* 790459931ecSChris Mason * [ bytenr - len ] 791459931ecSChris Mason * [ ] 792459931ecSChris Mason * [csum ] 793459931ecSChris Mason * we need to truncate from the beginning of the csum 794459931ecSChris Mason */ 795459931ecSChris Mason u32 new_size = (csum_end - end_byte) >> blocksize_bits; 796459931ecSChris Mason new_size *= csum_size; 797459931ecSChris Mason 79878ac4f9eSDavid Sterba btrfs_truncate_item(path, new_size, 0); 799459931ecSChris Mason 800459931ecSChris Mason key->offset = end_byte; 8010b246afaSJeff Mahoney btrfs_set_item_key_safe(fs_info, path, key); 802459931ecSChris Mason } else { 803459931ecSChris Mason BUG(); 804459931ecSChris Mason } 805459931ecSChris Mason } 806459931ecSChris Mason 807459931ecSChris Mason /* 808459931ecSChris Mason * deletes the csum items from the csum tree for a given 809459931ecSChris Mason * range of bytes. 810459931ecSChris Mason */ 811459931ecSChris Mason int btrfs_del_csums(struct btrfs_trans_handle *trans, 81240e046acSFilipe Manana struct btrfs_root *root, u64 bytenr, u64 len) 813459931ecSChris Mason { 81440e046acSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 815459931ecSChris Mason struct btrfs_path *path; 816459931ecSChris Mason struct btrfs_key key; 817459931ecSChris Mason u64 end_byte = bytenr + len; 818459931ecSChris Mason u64 csum_end; 819459931ecSChris Mason struct extent_buffer *leaf; 820b86652beSJosef Bacik int ret = 0; 821223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 822265fdfa6SDavid Sterba u32 blocksize_bits = fs_info->sectorsize_bits; 823459931ecSChris Mason 82484d2d6c7SJosef Bacik ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID || 82540e046acSFilipe Manana root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 82640e046acSFilipe Manana 827459931ecSChris Mason path = btrfs_alloc_path(); 8282a29edc6Sliubo if (!path) 8292a29edc6Sliubo return -ENOMEM; 830459931ecSChris Mason 831459931ecSChris Mason while (1) { 832459931ecSChris Mason key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 833459931ecSChris Mason key.offset = end_byte - 1; 834459931ecSChris Mason key.type = BTRFS_EXTENT_CSUM_KEY; 835459931ecSChris Mason 836459931ecSChris Mason ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 837459931ecSChris Mason if (ret > 0) { 838b86652beSJosef Bacik ret = 0; 839459931ecSChris Mason if (path->slots[0] == 0) 84065a246c5STsutomu Itoh break; 841459931ecSChris Mason path->slots[0]--; 842ad0397a7SJosef Bacik } else if (ret < 0) { 84365a246c5STsutomu Itoh break; 844459931ecSChris Mason } 845ad0397a7SJosef Bacik 846459931ecSChris Mason leaf = path->nodes[0]; 847459931ecSChris Mason btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 848459931ecSChris Mason 849459931ecSChris Mason if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 850459931ecSChris Mason key.type != BTRFS_EXTENT_CSUM_KEY) { 851459931ecSChris Mason break; 852459931ecSChris Mason } 853459931ecSChris Mason 854459931ecSChris Mason if (key.offset >= end_byte) 855459931ecSChris Mason break; 856459931ecSChris Mason 8573212fa14SJosef Bacik csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size; 858459931ecSChris Mason csum_end <<= blocksize_bits; 859459931ecSChris Mason csum_end += key.offset; 860459931ecSChris Mason 861459931ecSChris Mason /* this csum ends before we start, we're done */ 862459931ecSChris Mason if (csum_end <= bytenr) 863459931ecSChris Mason break; 864459931ecSChris Mason 865459931ecSChris Mason /* delete the entire item, it is inside our range */ 866459931ecSChris Mason if (key.offset >= bytenr && csum_end <= end_byte) { 8676f546216SFilipe Manana int del_nr = 1; 8686f546216SFilipe Manana 8696f546216SFilipe Manana /* 8706f546216SFilipe Manana * Check how many csum items preceding this one in this 8716f546216SFilipe Manana * leaf correspond to our range and then delete them all 8726f546216SFilipe Manana * at once. 8736f546216SFilipe Manana */ 8746f546216SFilipe Manana if (key.offset > bytenr && path->slots[0] > 0) { 8756f546216SFilipe Manana int slot = path->slots[0] - 1; 8766f546216SFilipe Manana 8776f546216SFilipe Manana while (slot >= 0) { 8786f546216SFilipe Manana struct btrfs_key pk; 8796f546216SFilipe Manana 8806f546216SFilipe Manana btrfs_item_key_to_cpu(leaf, &pk, slot); 8816f546216SFilipe Manana if (pk.offset < bytenr || 8826f546216SFilipe Manana pk.type != BTRFS_EXTENT_CSUM_KEY || 8836f546216SFilipe Manana pk.objectid != 8846f546216SFilipe Manana BTRFS_EXTENT_CSUM_OBJECTID) 8856f546216SFilipe Manana break; 8866f546216SFilipe Manana path->slots[0] = slot; 8876f546216SFilipe Manana del_nr++; 8886f546216SFilipe Manana key.offset = pk.offset; 8896f546216SFilipe Manana slot--; 8906f546216SFilipe Manana } 8916f546216SFilipe Manana } 8926f546216SFilipe Manana ret = btrfs_del_items(trans, root, path, 8936f546216SFilipe Manana path->slots[0], del_nr); 89465a246c5STsutomu Itoh if (ret) 895b86652beSJosef Bacik break; 896dcbdd4dcSChris Mason if (key.offset == bytenr) 897dcbdd4dcSChris Mason break; 898459931ecSChris Mason } else if (key.offset < bytenr && csum_end > end_byte) { 899459931ecSChris Mason unsigned long offset; 900459931ecSChris Mason unsigned long shift_len; 901459931ecSChris Mason unsigned long item_offset; 902459931ecSChris Mason /* 903459931ecSChris Mason * [ bytenr - len ] 904459931ecSChris Mason * [csum ] 905459931ecSChris Mason * 906459931ecSChris Mason * Our bytes are in the middle of the csum, 907459931ecSChris Mason * we need to split this item and insert a new one. 908459931ecSChris Mason * 909459931ecSChris Mason * But we can't drop the path because the 910459931ecSChris Mason * csum could change, get removed, extended etc. 911459931ecSChris Mason * 912459931ecSChris Mason * The trick here is the max size of a csum item leaves 913459931ecSChris Mason * enough room in the tree block for a single 914459931ecSChris Mason * item header. So, we split the item in place, 915459931ecSChris Mason * adding a new header pointing to the existing 916459931ecSChris Mason * bytes. Then we loop around again and we have 917459931ecSChris Mason * a nicely formed csum item that we can neatly 918459931ecSChris Mason * truncate. 919459931ecSChris Mason */ 920459931ecSChris Mason offset = (bytenr - key.offset) >> blocksize_bits; 921459931ecSChris Mason offset *= csum_size; 922459931ecSChris Mason 923459931ecSChris Mason shift_len = (len >> blocksize_bits) * csum_size; 924459931ecSChris Mason 925459931ecSChris Mason item_offset = btrfs_item_ptr_offset(leaf, 926459931ecSChris Mason path->slots[0]); 927459931ecSChris Mason 928b159fa28SDavid Sterba memzero_extent_buffer(leaf, item_offset + offset, 929459931ecSChris Mason shift_len); 930459931ecSChris Mason key.offset = bytenr; 931459931ecSChris Mason 932459931ecSChris Mason /* 933459931ecSChris Mason * btrfs_split_item returns -EAGAIN when the 934459931ecSChris Mason * item changed size or key 935459931ecSChris Mason */ 936459931ecSChris Mason ret = btrfs_split_item(trans, root, path, &key, offset); 93779787eaaSJeff Mahoney if (ret && ret != -EAGAIN) { 93866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 939b86652beSJosef Bacik break; 94079787eaaSJeff Mahoney } 941b86652beSJosef Bacik ret = 0; 942459931ecSChris Mason 943459931ecSChris Mason key.offset = end_byte - 1; 944459931ecSChris Mason } else { 9452ff7e61eSJeff Mahoney truncate_one_csum(fs_info, path, &key, bytenr, len); 946dcbdd4dcSChris Mason if (key.offset < bytenr) 947dcbdd4dcSChris Mason break; 948459931ecSChris Mason } 949b3b4aa74SDavid Sterba btrfs_release_path(path); 950459931ecSChris Mason } 951459931ecSChris Mason btrfs_free_path(path); 95265a246c5STsutomu Itoh return ret; 953459931ecSChris Mason } 954459931ecSChris Mason 955ea7036deSFilipe Manana static int find_next_csum_offset(struct btrfs_root *root, 956ea7036deSFilipe Manana struct btrfs_path *path, 957ea7036deSFilipe Manana u64 *next_offset) 958ea7036deSFilipe Manana { 959ea7036deSFilipe Manana const u32 nritems = btrfs_header_nritems(path->nodes[0]); 960ea7036deSFilipe Manana struct btrfs_key found_key; 961ea7036deSFilipe Manana int slot = path->slots[0] + 1; 962ea7036deSFilipe Manana int ret; 963ea7036deSFilipe Manana 964ea7036deSFilipe Manana if (nritems == 0 || slot >= nritems) { 965ea7036deSFilipe Manana ret = btrfs_next_leaf(root, path); 966ea7036deSFilipe Manana if (ret < 0) { 967ea7036deSFilipe Manana return ret; 968ea7036deSFilipe Manana } else if (ret > 0) { 969ea7036deSFilipe Manana *next_offset = (u64)-1; 970ea7036deSFilipe Manana return 0; 971ea7036deSFilipe Manana } 972ea7036deSFilipe Manana slot = path->slots[0]; 973ea7036deSFilipe Manana } 974ea7036deSFilipe Manana 975ea7036deSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot); 976ea7036deSFilipe Manana 977ea7036deSFilipe Manana if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 978ea7036deSFilipe Manana found_key.type != BTRFS_EXTENT_CSUM_KEY) 979ea7036deSFilipe Manana *next_offset = (u64)-1; 980ea7036deSFilipe Manana else 981ea7036deSFilipe Manana *next_offset = found_key.offset; 982ea7036deSFilipe Manana 983ea7036deSFilipe Manana return 0; 984ea7036deSFilipe Manana } 985ea7036deSFilipe Manana 986065631f6SChris Mason int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans, 987d20f7043SChris Mason struct btrfs_root *root, 988e6dcd2dcSChris Mason struct btrfs_ordered_sum *sums) 989f254e52cSChris Mason { 9900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 991f254e52cSChris Mason struct btrfs_key file_key; 9926567e837SChris Mason struct btrfs_key found_key; 9935caf2a00SChris Mason struct btrfs_path *path; 994f254e52cSChris Mason struct btrfs_csum_item *item; 995065631f6SChris Mason struct btrfs_csum_item *item_end; 996ff79f819SChris Mason struct extent_buffer *leaf = NULL; 997f51a4a18SMiao Xie u64 next_offset; 998f51a4a18SMiao Xie u64 total_bytes = 0; 9996567e837SChris Mason u64 csum_offset; 1000f51a4a18SMiao Xie u64 bytenr; 1001f578d4bdSChris Mason u32 ins_size; 1002f51a4a18SMiao Xie int index = 0; 1003f51a4a18SMiao Xie int found_next; 1004f51a4a18SMiao Xie int ret; 1005223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 10066e92f5e6SChris Mason 10075caf2a00SChris Mason path = btrfs_alloc_path(); 1008d8926bb3SMark Fasheh if (!path) 1009d8926bb3SMark Fasheh return -ENOMEM; 1010065631f6SChris Mason again: 1011065631f6SChris Mason next_offset = (u64)-1; 1012065631f6SChris Mason found_next = 0; 1013f51a4a18SMiao Xie bytenr = sums->bytenr + total_bytes; 1014d20f7043SChris Mason file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 1015f51a4a18SMiao Xie file_key.offset = bytenr; 1016962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_CSUM_KEY; 1017a429e513SChris Mason 1018f51a4a18SMiao Xie item = btrfs_lookup_csum(trans, root, path, bytenr, 1); 1019ff79f819SChris Mason if (!IS_ERR(item)) { 1020639cb586SChris Mason ret = 0; 1021f51a4a18SMiao Xie leaf = path->nodes[0]; 1022f51a4a18SMiao Xie item_end = btrfs_item_ptr(leaf, path->slots[0], 1023f51a4a18SMiao Xie struct btrfs_csum_item); 1024f51a4a18SMiao Xie item_end = (struct btrfs_csum_item *)((char *)item_end + 10253212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0])); 1026a429e513SChris Mason goto found; 1027ff79f819SChris Mason } 1028a429e513SChris Mason ret = PTR_ERR(item); 10294a500fd1SYan, Zheng if (ret != -EFBIG && ret != -ENOENT) 1030918cdf44SFilipe Manana goto out; 10314a500fd1SYan, Zheng 1032a429e513SChris Mason if (ret == -EFBIG) { 1033a429e513SChris Mason u32 item_size; 1034a429e513SChris Mason /* we found one, but it isn't big enough yet */ 10355f39d397SChris Mason leaf = path->nodes[0]; 10363212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 1037607d432dSJosef Bacik if ((item_size / csum_size) >= 10380b246afaSJeff Mahoney MAX_CSUM_ITEMS(fs_info, csum_size)) { 1039a429e513SChris Mason /* already at max size, make a new one */ 1040a429e513SChris Mason goto insert; 1041a429e513SChris Mason } 1042a429e513SChris Mason } else { 1043ea7036deSFilipe Manana /* We didn't find a csum item, insert one. */ 1044ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1045ea7036deSFilipe Manana if (ret < 0) 10467e4a3f7eSFilipe Manana goto out; 1047f578d4bdSChris Mason found_next = 1; 1048a429e513SChris Mason goto insert; 1049a429e513SChris Mason } 1050a429e513SChris Mason 1051a429e513SChris Mason /* 1052cc14600cSFilipe Manana * At this point, we know the tree has a checksum item that ends at an 1053cc14600cSFilipe Manana * offset matching the start of the checksum range we want to insert. 1054cc14600cSFilipe Manana * We try to extend that item as much as possible and then add as many 1055cc14600cSFilipe Manana * checksums to it as they fit. 1056cc14600cSFilipe Manana * 1057cc14600cSFilipe Manana * First check if the leaf has enough free space for at least one 1058cc14600cSFilipe Manana * checksum. If it has go directly to the item extension code, otherwise 1059cc14600cSFilipe Manana * release the path and do a search for insertion before the extension. 1060a429e513SChris Mason */ 1061cc14600cSFilipe Manana if (btrfs_leaf_free_space(leaf) >= csum_size) { 1062cc14600cSFilipe Manana btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1063cc14600cSFilipe Manana csum_offset = (bytenr - found_key.offset) >> 1064265fdfa6SDavid Sterba fs_info->sectorsize_bits; 1065cc14600cSFilipe Manana goto extend_csum; 1066cc14600cSFilipe Manana } 1067cc14600cSFilipe Manana 1068b3b4aa74SDavid Sterba btrfs_release_path(path); 10699a664971Sethanwu path->search_for_extension = 1; 10706567e837SChris Mason ret = btrfs_search_slot(trans, root, &file_key, path, 1071607d432dSJosef Bacik csum_size, 1); 10729a664971Sethanwu path->search_for_extension = 0; 10736567e837SChris Mason if (ret < 0) 1074918cdf44SFilipe Manana goto out; 1075459931ecSChris Mason 1076459931ecSChris Mason if (ret > 0) { 1077459931ecSChris Mason if (path->slots[0] == 0) 10786567e837SChris Mason goto insert; 10796567e837SChris Mason path->slots[0]--; 1080459931ecSChris Mason } 1081459931ecSChris Mason 10825f39d397SChris Mason leaf = path->nodes[0]; 10835f39d397SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1084265fdfa6SDavid Sterba csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits; 1085459931ecSChris Mason 1086962a298fSDavid Sterba if (found_key.type != BTRFS_EXTENT_CSUM_KEY || 1087d20f7043SChris Mason found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 10880b246afaSJeff Mahoney csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) { 10896567e837SChris Mason goto insert; 10906567e837SChris Mason } 1091459931ecSChris Mason 1092cc14600cSFilipe Manana extend_csum: 10933212fa14SJosef Bacik if (csum_offset == btrfs_item_size(leaf, path->slots[0]) / 1094607d432dSJosef Bacik csum_size) { 10952f697dc6SLiu Bo int extend_nr; 10962f697dc6SLiu Bo u64 tmp; 10972f697dc6SLiu Bo u32 diff; 1098459931ecSChris Mason 1099f51a4a18SMiao Xie tmp = sums->len - total_bytes; 1100265fdfa6SDavid Sterba tmp >>= fs_info->sectorsize_bits; 11012f697dc6SLiu Bo WARN_ON(tmp < 1); 1102ea7036deSFilipe Manana extend_nr = max_t(int, 1, tmp); 11032f697dc6SLiu Bo 1104ea7036deSFilipe Manana /* 1105ea7036deSFilipe Manana * A log tree can already have checksum items with a subset of 1106ea7036deSFilipe Manana * the checksums we are trying to log. This can happen after 1107ea7036deSFilipe Manana * doing a sequence of partial writes into prealloc extents and 1108ea7036deSFilipe Manana * fsyncs in between, with a full fsync logging a larger subrange 1109ea7036deSFilipe Manana * of an extent for which a previous fast fsync logged a smaller 1110ea7036deSFilipe Manana * subrange. And this happens in particular due to merging file 1111ea7036deSFilipe Manana * extent items when we complete an ordered extent for a range 1112ea7036deSFilipe Manana * covered by a prealloc extent - this is done at 1113ea7036deSFilipe Manana * btrfs_mark_extent_written(). 1114ea7036deSFilipe Manana * 1115ea7036deSFilipe Manana * So if we try to extend the previous checksum item, which has 1116ea7036deSFilipe Manana * a range that ends at the start of the range we want to insert, 1117ea7036deSFilipe Manana * make sure we don't extend beyond the start offset of the next 1118ea7036deSFilipe Manana * checksum item. If we are at the last item in the leaf, then 1119ea7036deSFilipe Manana * forget the optimization of extending and add a new checksum 1120ea7036deSFilipe Manana * item - it is not worth the complexity of releasing the path, 1121ea7036deSFilipe Manana * getting the first key for the next leaf, repeat the btree 1122ea7036deSFilipe Manana * search, etc, because log trees are temporary anyway and it 1123ea7036deSFilipe Manana * would only save a few bytes of leaf space. 1124ea7036deSFilipe Manana */ 1125ea7036deSFilipe Manana if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { 1126ea7036deSFilipe Manana if (path->slots[0] + 1 >= 1127ea7036deSFilipe Manana btrfs_header_nritems(path->nodes[0])) { 1128ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1129ea7036deSFilipe Manana if (ret < 0) 1130ea7036deSFilipe Manana goto out; 1131ea7036deSFilipe Manana found_next = 1; 1132ea7036deSFilipe Manana goto insert; 1133ea7036deSFilipe Manana } 1134ea7036deSFilipe Manana 1135ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1136ea7036deSFilipe Manana if (ret < 0) 1137ea7036deSFilipe Manana goto out; 1138ea7036deSFilipe Manana 1139ea7036deSFilipe Manana tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits; 1140ea7036deSFilipe Manana if (tmp <= INT_MAX) 1141ea7036deSFilipe Manana extend_nr = min_t(int, extend_nr, tmp); 1142ea7036deSFilipe Manana } 1143ea7036deSFilipe Manana 11442f697dc6SLiu Bo diff = (csum_offset + extend_nr) * csum_size; 11450b246afaSJeff Mahoney diff = min(diff, 11460b246afaSJeff Mahoney MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size); 1147459931ecSChris Mason 11483212fa14SJosef Bacik diff = diff - btrfs_item_size(leaf, path->slots[0]); 1149cc14600cSFilipe Manana diff = min_t(u32, btrfs_leaf_free_space(leaf), diff); 11502f697dc6SLiu Bo diff /= csum_size; 11512f697dc6SLiu Bo diff *= csum_size; 1152459931ecSChris Mason 1153c71dd880SDavid Sterba btrfs_extend_item(path, diff); 1154f51a4a18SMiao Xie ret = 0; 11556567e837SChris Mason goto csum; 11566567e837SChris Mason } 11576567e837SChris Mason 11586567e837SChris Mason insert: 1159b3b4aa74SDavid Sterba btrfs_release_path(path); 11606567e837SChris Mason csum_offset = 0; 1161f578d4bdSChris Mason if (found_next) { 11622f697dc6SLiu Bo u64 tmp; 1163d20f7043SChris Mason 1164f51a4a18SMiao Xie tmp = sums->len - total_bytes; 1165265fdfa6SDavid Sterba tmp >>= fs_info->sectorsize_bits; 11662f697dc6SLiu Bo tmp = min(tmp, (next_offset - file_key.offset) >> 1167265fdfa6SDavid Sterba fs_info->sectorsize_bits); 11682f697dc6SLiu Bo 116950d0446eSSeraphime Kirkovski tmp = max_t(u64, 1, tmp); 117050d0446eSSeraphime Kirkovski tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size)); 1171607d432dSJosef Bacik ins_size = csum_size * tmp; 1172f578d4bdSChris Mason } else { 1173607d432dSJosef Bacik ins_size = csum_size; 1174f578d4bdSChris Mason } 11755caf2a00SChris Mason ret = btrfs_insert_empty_item(trans, root, path, &file_key, 1176f578d4bdSChris Mason ins_size); 117754aa1f4dSChris Mason if (ret < 0) 1178918cdf44SFilipe Manana goto out; 1179fae7f21cSDulshani Gunawardhana if (WARN_ON(ret != 0)) 1180918cdf44SFilipe Manana goto out; 11815f39d397SChris Mason leaf = path->nodes[0]; 1182f51a4a18SMiao Xie csum: 11835f39d397SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item); 1184f51a4a18SMiao Xie item_end = (struct btrfs_csum_item *)((unsigned char *)item + 11853212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0])); 1186509659cdSChris Mason item = (struct btrfs_csum_item *)((unsigned char *)item + 1187607d432dSJosef Bacik csum_offset * csum_size); 1188b18c6685SChris Mason found: 1189265fdfa6SDavid Sterba ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits; 1190f51a4a18SMiao Xie ins_size *= csum_size; 1191f51a4a18SMiao Xie ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item, 1192f51a4a18SMiao Xie ins_size); 1193f51a4a18SMiao Xie write_extent_buffer(leaf, sums->sums + index, (unsigned long)item, 1194f51a4a18SMiao Xie ins_size); 1195aadfeb6eSChris Mason 11961e25a2e3SJohannes Thumshirn index += ins_size; 1197f51a4a18SMiao Xie ins_size /= csum_size; 11980b246afaSJeff Mahoney total_bytes += ins_size * fs_info->sectorsize; 1199a6591715SChris Mason 12005caf2a00SChris Mason btrfs_mark_buffer_dirty(path->nodes[0]); 1201e6dcd2dcSChris Mason if (total_bytes < sums->len) { 1202b3b4aa74SDavid Sterba btrfs_release_path(path); 1203b9473439SChris Mason cond_resched(); 1204065631f6SChris Mason goto again; 1205065631f6SChris Mason } 120653863232SChris Mason out: 12075caf2a00SChris Mason btrfs_free_path(path); 1208f254e52cSChris Mason return ret; 1209f254e52cSChris Mason } 12107ffbb598SFilipe Manana 12119cdc5124SNikolay Borisov void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode, 12127ffbb598SFilipe Manana const struct btrfs_path *path, 12137ffbb598SFilipe Manana struct btrfs_file_extent_item *fi, 12147ffbb598SFilipe Manana const bool new_inline, 12157ffbb598SFilipe Manana struct extent_map *em) 12167ffbb598SFilipe Manana { 12173ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 12189cdc5124SNikolay Borisov struct btrfs_root *root = inode->root; 12197ffbb598SFilipe Manana struct extent_buffer *leaf = path->nodes[0]; 12207ffbb598SFilipe Manana const int slot = path->slots[0]; 12217ffbb598SFilipe Manana struct btrfs_key key; 12227ffbb598SFilipe Manana u64 extent_start, extent_end; 12237ffbb598SFilipe Manana u64 bytenr; 12247ffbb598SFilipe Manana u8 type = btrfs_file_extent_type(leaf, fi); 12257ffbb598SFilipe Manana int compress_type = btrfs_file_extent_compression(leaf, fi); 12267ffbb598SFilipe Manana 12277ffbb598SFilipe Manana btrfs_item_key_to_cpu(leaf, &key, slot); 12287ffbb598SFilipe Manana extent_start = key.offset; 1229a5eeb3d1SFilipe Manana extent_end = btrfs_file_extent_end(path); 12307ffbb598SFilipe Manana em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); 123140e7efe0SQu Wenruo em->generation = btrfs_file_extent_generation(leaf, fi); 12327ffbb598SFilipe Manana if (type == BTRFS_FILE_EXTENT_REG || 12337ffbb598SFilipe Manana type == BTRFS_FILE_EXTENT_PREALLOC) { 12347ffbb598SFilipe Manana em->start = extent_start; 12357ffbb598SFilipe Manana em->len = extent_end - extent_start; 12367ffbb598SFilipe Manana em->orig_start = extent_start - 12377ffbb598SFilipe Manana btrfs_file_extent_offset(leaf, fi); 12387ffbb598SFilipe Manana em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); 12397ffbb598SFilipe Manana bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 12407ffbb598SFilipe Manana if (bytenr == 0) { 12417ffbb598SFilipe Manana em->block_start = EXTENT_MAP_HOLE; 12427ffbb598SFilipe Manana return; 12437ffbb598SFilipe Manana } 12447ffbb598SFilipe Manana if (compress_type != BTRFS_COMPRESS_NONE) { 12457ffbb598SFilipe Manana set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 12467ffbb598SFilipe Manana em->compress_type = compress_type; 12477ffbb598SFilipe Manana em->block_start = bytenr; 12487ffbb598SFilipe Manana em->block_len = em->orig_block_len; 12497ffbb598SFilipe Manana } else { 12507ffbb598SFilipe Manana bytenr += btrfs_file_extent_offset(leaf, fi); 12517ffbb598SFilipe Manana em->block_start = bytenr; 12527ffbb598SFilipe Manana em->block_len = em->len; 12537ffbb598SFilipe Manana if (type == BTRFS_FILE_EXTENT_PREALLOC) 12547ffbb598SFilipe Manana set_bit(EXTENT_FLAG_PREALLOC, &em->flags); 12557ffbb598SFilipe Manana } 12567ffbb598SFilipe Manana } else if (type == BTRFS_FILE_EXTENT_INLINE) { 12577ffbb598SFilipe Manana em->block_start = EXTENT_MAP_INLINE; 12587ffbb598SFilipe Manana em->start = extent_start; 12597ffbb598SFilipe Manana em->len = extent_end - extent_start; 12607ffbb598SFilipe Manana /* 12617ffbb598SFilipe Manana * Initialize orig_start and block_len with the same values 12627ffbb598SFilipe Manana * as in inode.c:btrfs_get_extent(). 12637ffbb598SFilipe Manana */ 12647ffbb598SFilipe Manana em->orig_start = EXTENT_MAP_HOLE; 12657ffbb598SFilipe Manana em->block_len = (u64)-1; 12667ffbb598SFilipe Manana if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) { 12677ffbb598SFilipe Manana set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 12687ffbb598SFilipe Manana em->compress_type = compress_type; 12697ffbb598SFilipe Manana } 12707ffbb598SFilipe Manana } else { 12710b246afaSJeff Mahoney btrfs_err(fs_info, 12729cdc5124SNikolay Borisov "unknown file extent item type %d, inode %llu, offset %llu, " 12739cdc5124SNikolay Borisov "root %llu", type, btrfs_ino(inode), extent_start, 12747ffbb598SFilipe Manana root->root_key.objectid); 12757ffbb598SFilipe Manana } 12767ffbb598SFilipe Manana } 1277a5eeb3d1SFilipe Manana 1278a5eeb3d1SFilipe Manana /* 1279a5eeb3d1SFilipe Manana * Returns the end offset (non inclusive) of the file extent item the given path 1280a5eeb3d1SFilipe Manana * points to. If it points to an inline extent, the returned offset is rounded 1281a5eeb3d1SFilipe Manana * up to the sector size. 1282a5eeb3d1SFilipe Manana */ 1283a5eeb3d1SFilipe Manana u64 btrfs_file_extent_end(const struct btrfs_path *path) 1284a5eeb3d1SFilipe Manana { 1285a5eeb3d1SFilipe Manana const struct extent_buffer *leaf = path->nodes[0]; 1286a5eeb3d1SFilipe Manana const int slot = path->slots[0]; 1287a5eeb3d1SFilipe Manana struct btrfs_file_extent_item *fi; 1288a5eeb3d1SFilipe Manana struct btrfs_key key; 1289a5eeb3d1SFilipe Manana u64 end; 1290a5eeb3d1SFilipe Manana 1291a5eeb3d1SFilipe Manana btrfs_item_key_to_cpu(leaf, &key, slot); 1292a5eeb3d1SFilipe Manana ASSERT(key.type == BTRFS_EXTENT_DATA_KEY); 1293a5eeb3d1SFilipe Manana fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 1294a5eeb3d1SFilipe Manana 1295a5eeb3d1SFilipe Manana if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) { 1296a5eeb3d1SFilipe Manana end = btrfs_file_extent_ram_bytes(leaf, fi); 1297a5eeb3d1SFilipe Manana end = ALIGN(key.offset + end, leaf->fs_info->sectorsize); 1298a5eeb3d1SFilipe Manana } else { 1299a5eeb3d1SFilipe Manana end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); 1300a5eeb3d1SFilipe Manana } 1301a5eeb3d1SFilipe Manana 1302a5eeb3d1SFilipe Manana return end; 1303a5eeb3d1SFilipe Manana } 1304