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> 12cea62800SJohannes Thumshirn #include "misc.h" 131e1d2701SChris Mason #include "ctree.h" 14dee26a9fSChris Mason #include "disk-io.h" 159f5fae2fSChris Mason #include "transaction.h" 16facc8a22SMiao Xie #include "volumes.h" 171de037a4SChris Mason #include "print-tree.h" 18ebb8765bSAnand Jain #include "compression.h" 191e1d2701SChris Mason 2042049bf6SChris Mason #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \ 2142049bf6SChris Mason sizeof(struct btrfs_item) * 2) / \ 2242049bf6SChris Mason size) - 1)) 2307d400a6SYan Zheng 24221b8318SZach Brown #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \ 2509cbfeafSKirill A. Shutemov PAGE_SIZE)) 267ca4be45SChris Mason 2741a2ee75SJosef Bacik /** 28ca4207aeSNikolay Borisov * Set inode's size according to filesystem options 29ca4207aeSNikolay Borisov * 30ca4207aeSNikolay Borisov * @inode: inode we want to update the disk_i_size for 31ca4207aeSNikolay Borisov * @new_i_size: i_size we want to set to, 0 if we use i_size 3241a2ee75SJosef Bacik * 3341a2ee75SJosef Bacik * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read() 3441a2ee75SJosef Bacik * returns as it is perfectly fine with a file that has holes without hole file 3541a2ee75SJosef Bacik * extent items. 3641a2ee75SJosef Bacik * 3741a2ee75SJosef Bacik * However without NO_HOLES we need to only return the area that is contiguous 3841a2ee75SJosef Bacik * from the 0 offset of the file. Otherwise we could end up adjust i_size up 3941a2ee75SJosef Bacik * to an extent that has a gap in between. 4041a2ee75SJosef Bacik * 4141a2ee75SJosef Bacik * Finally new_i_size should only be set in the case of truncate where we're not 4241a2ee75SJosef Bacik * ready to use i_size_read() as the limiter yet. 4341a2ee75SJosef Bacik */ 4476aea537SNikolay Borisov void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size) 4541a2ee75SJosef Bacik { 4676aea537SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 4741a2ee75SJosef Bacik u64 start, end, i_size; 4841a2ee75SJosef Bacik int ret; 4941a2ee75SJosef Bacik 5076aea537SNikolay Borisov i_size = new_i_size ?: i_size_read(&inode->vfs_inode); 5141a2ee75SJosef Bacik if (btrfs_fs_incompat(fs_info, NO_HOLES)) { 5276aea537SNikolay Borisov inode->disk_i_size = i_size; 5341a2ee75SJosef Bacik return; 5441a2ee75SJosef Bacik } 5541a2ee75SJosef Bacik 5676aea537SNikolay Borisov spin_lock(&inode->lock); 5776aea537SNikolay Borisov ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start, 5876aea537SNikolay Borisov &end, EXTENT_DIRTY); 5941a2ee75SJosef Bacik if (!ret && start == 0) 6041a2ee75SJosef Bacik i_size = min(i_size, end + 1); 6141a2ee75SJosef Bacik else 6241a2ee75SJosef Bacik i_size = 0; 6376aea537SNikolay Borisov inode->disk_i_size = i_size; 6476aea537SNikolay Borisov spin_unlock(&inode->lock); 6541a2ee75SJosef Bacik } 6641a2ee75SJosef Bacik 6741a2ee75SJosef Bacik /** 68ca4207aeSNikolay Borisov * Mark range within a file as having a new extent inserted 69ca4207aeSNikolay Borisov * 70ca4207aeSNikolay Borisov * @inode: inode being modified 71ca4207aeSNikolay Borisov * @start: start file offset of the file extent we've inserted 72ca4207aeSNikolay Borisov * @len: logical length of the file extent item 7341a2ee75SJosef Bacik * 7441a2ee75SJosef Bacik * Call when we are inserting a new file extent where there was none before. 7541a2ee75SJosef Bacik * Does not need to call this in the case where we're replacing an existing file 7641a2ee75SJosef Bacik * extent, however if not sure it's fine to call this multiple times. 7741a2ee75SJosef Bacik * 7841a2ee75SJosef Bacik * The start and len must match the file extent item, so thus must be sectorsize 7941a2ee75SJosef Bacik * aligned. 8041a2ee75SJosef Bacik */ 8141a2ee75SJosef Bacik int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start, 8241a2ee75SJosef Bacik u64 len) 8341a2ee75SJosef Bacik { 8441a2ee75SJosef Bacik if (len == 0) 8541a2ee75SJosef Bacik return 0; 8641a2ee75SJosef Bacik 8741a2ee75SJosef Bacik ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize)); 8841a2ee75SJosef Bacik 8941a2ee75SJosef Bacik if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES)) 9041a2ee75SJosef Bacik return 0; 9141a2ee75SJosef Bacik return set_extent_bits(&inode->file_extent_tree, start, start + len - 1, 9241a2ee75SJosef Bacik EXTENT_DIRTY); 9341a2ee75SJosef Bacik } 9441a2ee75SJosef Bacik 9541a2ee75SJosef Bacik /** 96ca4207aeSNikolay Borisov * Marks an inode range as not having a backing extent 97ca4207aeSNikolay Borisov * 98ca4207aeSNikolay Borisov * @inode: inode being modified 99ca4207aeSNikolay Borisov * @start: start file offset of the file extent we've inserted 100ca4207aeSNikolay Borisov * @len: logical length of the file extent item 10141a2ee75SJosef Bacik * 10241a2ee75SJosef Bacik * Called when we drop a file extent, for example when we truncate. Doesn't 10341a2ee75SJosef Bacik * need to be called for cases where we're replacing a file extent, like when 10441a2ee75SJosef Bacik * we've COWed a file extent. 10541a2ee75SJosef Bacik * 10641a2ee75SJosef Bacik * The start and len must match the file extent item, so thus must be sectorsize 10741a2ee75SJosef Bacik * aligned. 10841a2ee75SJosef Bacik */ 10941a2ee75SJosef Bacik int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start, 11041a2ee75SJosef Bacik u64 len) 11141a2ee75SJosef Bacik { 11241a2ee75SJosef Bacik if (len == 0) 11341a2ee75SJosef Bacik return 0; 11441a2ee75SJosef Bacik 11541a2ee75SJosef Bacik ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) || 11641a2ee75SJosef Bacik len == (u64)-1); 11741a2ee75SJosef Bacik 11841a2ee75SJosef Bacik if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES)) 11941a2ee75SJosef Bacik return 0; 12041a2ee75SJosef Bacik return clear_extent_bit(&inode->file_extent_tree, start, 121bd015294SJosef Bacik start + len - 1, EXTENT_DIRTY, NULL); 12241a2ee75SJosef Bacik } 12341a2ee75SJosef Bacik 1241e25a2e3SJohannes Thumshirn static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info, 1251e25a2e3SJohannes Thumshirn u16 csum_size) 1261e25a2e3SJohannes Thumshirn { 1271e25a2e3SJohannes Thumshirn u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size; 1281e25a2e3SJohannes Thumshirn 1291e25a2e3SJohannes Thumshirn return ncsums * fs_info->sectorsize; 1301e25a2e3SJohannes Thumshirn } 13107d400a6SYan Zheng 132*2b6433c7SJosef Bacik /* 133*2b6433c7SJosef Bacik * Calculate the total size needed to allocate for an ordered sum structure 134*2b6433c7SJosef Bacik * spanning @bytes in the file. 135*2b6433c7SJosef Bacik */ 136*2b6433c7SJosef Bacik static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes) 137*2b6433c7SJosef Bacik { 138*2b6433c7SJosef Bacik int num_sectors = (int)DIV_ROUND_UP(bytes, fs_info->sectorsize); 139*2b6433c7SJosef Bacik 140*2b6433c7SJosef Bacik return sizeof(struct btrfs_ordered_sum) + num_sectors * fs_info->csum_size; 141*2b6433c7SJosef Bacik } 142*2b6433c7SJosef Bacik 143d1f68ba0SOmar Sandoval int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans, 144dee26a9fSChris Mason struct btrfs_root *root, 145d1f68ba0SOmar Sandoval u64 objectid, u64 pos, u64 num_bytes) 1469f5fae2fSChris Mason { 147dee26a9fSChris Mason int ret = 0; 148dee26a9fSChris Mason struct btrfs_file_extent_item *item; 149dee26a9fSChris Mason struct btrfs_key file_key; 1505caf2a00SChris Mason struct btrfs_path *path; 1515f39d397SChris Mason struct extent_buffer *leaf; 152dee26a9fSChris Mason 1535caf2a00SChris Mason path = btrfs_alloc_path(); 154db5b493aSTsutomu Itoh if (!path) 155db5b493aSTsutomu Itoh return -ENOMEM; 156dee26a9fSChris Mason file_key.objectid = objectid; 157b18c6685SChris Mason file_key.offset = pos; 158962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_DATA_KEY; 159dee26a9fSChris Mason 1605caf2a00SChris Mason ret = btrfs_insert_empty_item(trans, root, path, &file_key, 161dee26a9fSChris Mason sizeof(*item)); 16254aa1f4dSChris Mason if (ret < 0) 16354aa1f4dSChris Mason goto out; 16479787eaaSJeff Mahoney BUG_ON(ret); /* Can't happen */ 1655f39d397SChris Mason leaf = path->nodes[0]; 1665f39d397SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], 167dee26a9fSChris Mason struct btrfs_file_extent_item); 168d1f68ba0SOmar Sandoval btrfs_set_file_extent_disk_bytenr(leaf, item, 0); 169d1f68ba0SOmar Sandoval btrfs_set_file_extent_disk_num_bytes(leaf, item, 0); 170d1f68ba0SOmar Sandoval btrfs_set_file_extent_offset(leaf, item, 0); 171db94535dSChris Mason btrfs_set_file_extent_num_bytes(leaf, item, num_bytes); 172d1f68ba0SOmar Sandoval btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes); 1735f39d397SChris Mason btrfs_set_file_extent_generation(leaf, item, trans->transid); 1745f39d397SChris Mason btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); 175d1f68ba0SOmar Sandoval btrfs_set_file_extent_compression(leaf, item, 0); 176d1f68ba0SOmar Sandoval btrfs_set_file_extent_encryption(leaf, item, 0); 177d1f68ba0SOmar Sandoval btrfs_set_file_extent_other_encoding(leaf, item, 0); 178c8b97818SChris Mason 1795f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 18054aa1f4dSChris Mason out: 1815caf2a00SChris Mason btrfs_free_path(path); 18254aa1f4dSChris Mason return ret; 1839f5fae2fSChris Mason } 184dee26a9fSChris Mason 18548a3b636SEric Sandeen static struct btrfs_csum_item * 18648a3b636SEric Sandeen btrfs_lookup_csum(struct btrfs_trans_handle *trans, 187b18c6685SChris Mason struct btrfs_root *root, 1886567e837SChris Mason struct btrfs_path *path, 189d20f7043SChris Mason u64 bytenr, int cow) 1906567e837SChris Mason { 1910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1926567e837SChris Mason int ret; 1936567e837SChris Mason struct btrfs_key file_key; 1946567e837SChris Mason struct btrfs_key found_key; 1956567e837SChris Mason struct btrfs_csum_item *item; 1965f39d397SChris Mason struct extent_buffer *leaf; 1976567e837SChris Mason u64 csum_offset = 0; 198223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 199a429e513SChris Mason int csums_in_item; 2006567e837SChris Mason 201d20f7043SChris Mason file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 202d20f7043SChris Mason file_key.offset = bytenr; 203962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_CSUM_KEY; 204b18c6685SChris Mason ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow); 2056567e837SChris Mason if (ret < 0) 2066567e837SChris Mason goto fail; 2075f39d397SChris Mason leaf = path->nodes[0]; 2086567e837SChris Mason if (ret > 0) { 2096567e837SChris Mason ret = 1; 21070b2befdSChris Mason if (path->slots[0] == 0) 2116567e837SChris Mason goto fail; 2126567e837SChris Mason path->slots[0]--; 2135f39d397SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 214962a298fSDavid Sterba if (found_key.type != BTRFS_EXTENT_CSUM_KEY) 2156567e837SChris Mason goto fail; 216d20f7043SChris Mason 217d20f7043SChris Mason csum_offset = (bytenr - found_key.offset) >> 218265fdfa6SDavid Sterba fs_info->sectorsize_bits; 2193212fa14SJosef Bacik csums_in_item = btrfs_item_size(leaf, path->slots[0]); 220607d432dSJosef Bacik csums_in_item /= csum_size; 221a429e513SChris Mason 22282d130ffSMiao Xie if (csum_offset == csums_in_item) { 223a429e513SChris Mason ret = -EFBIG; 2246567e837SChris Mason goto fail; 22582d130ffSMiao Xie } else if (csum_offset > csums_in_item) { 22682d130ffSMiao Xie goto fail; 2276567e837SChris Mason } 2286567e837SChris Mason } 2296567e837SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item); 230509659cdSChris Mason item = (struct btrfs_csum_item *)((unsigned char *)item + 231607d432dSJosef Bacik csum_offset * csum_size); 2326567e837SChris Mason return item; 2336567e837SChris Mason fail: 2346567e837SChris Mason if (ret > 0) 235b18c6685SChris Mason ret = -ENOENT; 2366567e837SChris Mason return ERR_PTR(ret); 2376567e837SChris Mason } 2386567e837SChris Mason 239dee26a9fSChris Mason int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans, 240dee26a9fSChris Mason struct btrfs_root *root, 241dee26a9fSChris Mason struct btrfs_path *path, u64 objectid, 2429773a788SChris Mason u64 offset, int mod) 243dee26a9fSChris Mason { 244dee26a9fSChris Mason struct btrfs_key file_key; 245dee26a9fSChris Mason int ins_len = mod < 0 ? -1 : 0; 246dee26a9fSChris Mason int cow = mod != 0; 247dee26a9fSChris Mason 248dee26a9fSChris Mason file_key.objectid = objectid; 24970b2befdSChris Mason file_key.offset = offset; 250962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_DATA_KEY; 251f8ee80deSMarcos Paulo de Souza 252f8ee80deSMarcos Paulo de Souza return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow); 253dee26a9fSChris Mason } 254f254e52cSChris Mason 2556275193eSQu Wenruo /* 2566275193eSQu Wenruo * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and 2576275193eSQu Wenruo * estore the result to @dst. 2586275193eSQu Wenruo * 2596275193eSQu Wenruo * Return >0 for the number of sectors we found. 2606275193eSQu Wenruo * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum 2616275193eSQu Wenruo * for it. Caller may want to try next sector until one range is hit. 2626275193eSQu Wenruo * Return <0 for fatal error. 2636275193eSQu Wenruo */ 2646275193eSQu Wenruo static int search_csum_tree(struct btrfs_fs_info *fs_info, 2656275193eSQu Wenruo struct btrfs_path *path, u64 disk_bytenr, 2666275193eSQu Wenruo u64 len, u8 *dst) 2676275193eSQu Wenruo { 268fc28b25eSJosef Bacik struct btrfs_root *csum_root; 2696275193eSQu Wenruo struct btrfs_csum_item *item = NULL; 2706275193eSQu Wenruo struct btrfs_key key; 2716275193eSQu Wenruo const u32 sectorsize = fs_info->sectorsize; 2726275193eSQu Wenruo const u32 csum_size = fs_info->csum_size; 2736275193eSQu Wenruo u32 itemsize; 2746275193eSQu Wenruo int ret; 2756275193eSQu Wenruo u64 csum_start; 2766275193eSQu Wenruo u64 csum_len; 2776275193eSQu Wenruo 2786275193eSQu Wenruo ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) && 2796275193eSQu Wenruo IS_ALIGNED(len, sectorsize)); 2806275193eSQu Wenruo 2816275193eSQu Wenruo /* Check if the current csum item covers disk_bytenr */ 2826275193eSQu Wenruo if (path->nodes[0]) { 2836275193eSQu Wenruo item = btrfs_item_ptr(path->nodes[0], path->slots[0], 2846275193eSQu Wenruo struct btrfs_csum_item); 2856275193eSQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2863212fa14SJosef Bacik itemsize = btrfs_item_size(path->nodes[0], path->slots[0]); 2876275193eSQu Wenruo 2886275193eSQu Wenruo csum_start = key.offset; 2896275193eSQu Wenruo csum_len = (itemsize / csum_size) * sectorsize; 2906275193eSQu Wenruo 2916275193eSQu Wenruo if (in_range(disk_bytenr, csum_start, csum_len)) 2926275193eSQu Wenruo goto found; 2936275193eSQu Wenruo } 2946275193eSQu Wenruo 2956275193eSQu Wenruo /* Current item doesn't contain the desired range, search again */ 2966275193eSQu Wenruo btrfs_release_path(path); 297fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 298fc28b25eSJosef Bacik item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0); 2996275193eSQu Wenruo if (IS_ERR(item)) { 3006275193eSQu Wenruo ret = PTR_ERR(item); 3016275193eSQu Wenruo goto out; 3026275193eSQu Wenruo } 3036275193eSQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 3043212fa14SJosef Bacik itemsize = btrfs_item_size(path->nodes[0], path->slots[0]); 3056275193eSQu Wenruo 3066275193eSQu Wenruo csum_start = key.offset; 3076275193eSQu Wenruo csum_len = (itemsize / csum_size) * sectorsize; 3086275193eSQu Wenruo ASSERT(in_range(disk_bytenr, csum_start, csum_len)); 3096275193eSQu Wenruo 3106275193eSQu Wenruo found: 3116275193eSQu Wenruo ret = (min(csum_start + csum_len, disk_bytenr + len) - 3126275193eSQu Wenruo disk_bytenr) >> fs_info->sectorsize_bits; 3136275193eSQu Wenruo read_extent_buffer(path->nodes[0], dst, (unsigned long)item, 3146275193eSQu Wenruo ret * csum_size); 3156275193eSQu Wenruo out: 31603ddb19dSJosef Bacik if (ret == -ENOENT || ret == -EFBIG) 3176275193eSQu Wenruo ret = 0; 3186275193eSQu Wenruo return ret; 3196275193eSQu Wenruo } 3206275193eSQu Wenruo 3216275193eSQu Wenruo /* 3226275193eSQu Wenruo * Locate the file_offset of @cur_disk_bytenr of a @bio. 3236275193eSQu Wenruo * 3246275193eSQu Wenruo * Bio of btrfs represents read range of 3256275193eSQu Wenruo * [bi_sector << 9, bi_sector << 9 + bi_size). 3266275193eSQu Wenruo * Knowing this, we can iterate through each bvec to locate the page belong to 3276275193eSQu Wenruo * @cur_disk_bytenr and get the file offset. 3286275193eSQu Wenruo * 3296275193eSQu Wenruo * @inode is used to determine if the bvec page really belongs to @inode. 3306275193eSQu Wenruo * 3316275193eSQu Wenruo * Return 0 if we can't find the file offset 3326275193eSQu Wenruo * Return >0 if we find the file offset and restore it to @file_offset_ret 3336275193eSQu Wenruo */ 3346275193eSQu Wenruo static int search_file_offset_in_bio(struct bio *bio, struct inode *inode, 3356275193eSQu Wenruo u64 disk_bytenr, u64 *file_offset_ret) 3366275193eSQu Wenruo { 3376275193eSQu Wenruo struct bvec_iter iter; 3386275193eSQu Wenruo struct bio_vec bvec; 3396275193eSQu Wenruo u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT; 3406275193eSQu Wenruo int ret = 0; 3416275193eSQu Wenruo 3426275193eSQu Wenruo bio_for_each_segment(bvec, bio, iter) { 3436275193eSQu Wenruo struct page *page = bvec.bv_page; 3446275193eSQu Wenruo 3456275193eSQu Wenruo if (cur > disk_bytenr) 3466275193eSQu Wenruo break; 3476275193eSQu Wenruo if (cur + bvec.bv_len <= disk_bytenr) { 3486275193eSQu Wenruo cur += bvec.bv_len; 3496275193eSQu Wenruo continue; 3506275193eSQu Wenruo } 3516275193eSQu Wenruo ASSERT(in_range(disk_bytenr, cur, bvec.bv_len)); 3526275193eSQu Wenruo if (page->mapping && page->mapping->host && 3536275193eSQu Wenruo page->mapping->host == inode) { 3546275193eSQu Wenruo ret = 1; 3556275193eSQu Wenruo *file_offset_ret = page_offset(page) + bvec.bv_offset + 3566275193eSQu Wenruo disk_bytenr - cur; 3576275193eSQu Wenruo break; 3586275193eSQu Wenruo } 3596275193eSQu Wenruo } 3606275193eSQu Wenruo return ret; 3616275193eSQu Wenruo } 3626275193eSQu Wenruo 363e62958fcSOmar Sandoval /** 3646275193eSQu Wenruo * Lookup the checksum for the read bio in csum tree. 3659e46458aSQu Wenruo * 366e62958fcSOmar Sandoval * @inode: inode that the bio is for. 367fb30f470SOmar Sandoval * @bio: bio to look up. 368fb30f470SOmar Sandoval * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return 369fb30f470SOmar Sandoval * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If 370fb30f470SOmar Sandoval * NULL, the checksum buffer is allocated and returned in 371c3a3b19bSQu Wenruo * btrfs_bio(bio)->csum instead. 372e62958fcSOmar Sandoval * 373e62958fcSOmar Sandoval * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise. 374e62958fcSOmar Sandoval */ 3756275193eSQu Wenruo blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst) 37661b49440SChris Mason { 3770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 378facc8a22SMiao Xie struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 3791784b7d5SJosef Bacik struct btrfs_bio *bbio = NULL; 380facc8a22SMiao Xie struct btrfs_path *path; 3816275193eSQu Wenruo const u32 sectorsize = fs_info->sectorsize; 382223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 3836275193eSQu Wenruo u32 orig_len = bio->bi_iter.bi_size; 3846275193eSQu Wenruo u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT; 3856275193eSQu Wenruo u64 cur_disk_bytenr; 3866275193eSQu Wenruo u8 *csum; 3876275193eSQu Wenruo const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits; 3886275193eSQu Wenruo int count = 0; 3891784b7d5SJosef Bacik blk_status_t ret = BLK_STS_OK; 39061b49440SChris Mason 391056c8311SJosef Bacik if ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) || 392056c8311SJosef Bacik test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)) 393334c16d8SJosef Bacik return BLK_STS_OK; 394334c16d8SJosef Bacik 3959e46458aSQu Wenruo /* 3969e46458aSQu Wenruo * This function is only called for read bio. 3979e46458aSQu Wenruo * 3989e46458aSQu Wenruo * This means two things: 3999e46458aSQu Wenruo * - All our csums should only be in csum tree 4009e46458aSQu Wenruo * No ordered extents csums, as ordered extents are only for write 4019e46458aSQu Wenruo * path. 4026275193eSQu Wenruo * - No need to bother any other info from bvec 4036275193eSQu Wenruo * Since we're looking up csums, the only important info is the 4046275193eSQu Wenruo * disk_bytenr and the length, which can be extracted from bi_iter 4056275193eSQu Wenruo * directly. 4069e46458aSQu Wenruo */ 4079e46458aSQu Wenruo ASSERT(bio_op(bio) == REQ_OP_READ); 40861b49440SChris Mason path = btrfs_alloc_path(); 409c2db1073STsutomu Itoh if (!path) 4104e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 411facc8a22SMiao Xie 412facc8a22SMiao Xie if (!dst) { 4131784b7d5SJosef Bacik bbio = btrfs_bio(bio); 414fb30f470SOmar Sandoval 415facc8a22SMiao Xie if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) { 416c3a3b19bSQu Wenruo bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS); 417c3a3b19bSQu Wenruo if (!bbio->csum) { 418facc8a22SMiao Xie btrfs_free_path(path); 4194e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 420facc8a22SMiao Xie } 421facc8a22SMiao Xie } else { 422c3a3b19bSQu Wenruo bbio->csum = bbio->csum_inline; 423facc8a22SMiao Xie } 424c3a3b19bSQu Wenruo csum = bbio->csum; 425facc8a22SMiao Xie } else { 42610fe6ca8SJohannes Thumshirn csum = dst; 427facc8a22SMiao Xie } 428facc8a22SMiao Xie 42935478d05SQu Wenruo /* 43035478d05SQu Wenruo * If requested number of sectors is larger than one leaf can contain, 43135478d05SQu Wenruo * kick the readahead for csum tree. 43235478d05SQu Wenruo */ 43335478d05SQu Wenruo if (nblocks > fs_info->csums_per_leaf) 434e4058b54SDavid Sterba path->reada = READA_FORWARD; 43561b49440SChris Mason 4362cf8572dSChris Mason /* 4372cf8572dSChris Mason * the free space stuff is only read when it hasn't been 4382cf8572dSChris Mason * updated in the current transaction. So, we can safely 4392cf8572dSChris Mason * read from the commit root and sidestep a nasty deadlock 4402cf8572dSChris Mason * between reading the free space cache and updating the csum tree. 4412cf8572dSChris Mason */ 44270ddc553SNikolay Borisov if (btrfs_is_free_space_inode(BTRFS_I(inode))) { 4432cf8572dSChris Mason path->search_commit_root = 1; 444ddf23b3fSJosef Bacik path->skip_locking = 1; 445ddf23b3fSJosef Bacik } 4462cf8572dSChris Mason 4476275193eSQu Wenruo for (cur_disk_bytenr = orig_disk_bytenr; 4486275193eSQu Wenruo cur_disk_bytenr < orig_disk_bytenr + orig_len; 4496275193eSQu Wenruo cur_disk_bytenr += (count * sectorsize)) { 4506275193eSQu Wenruo u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr; 4516275193eSQu Wenruo unsigned int sector_offset; 4526275193eSQu Wenruo u8 *csum_dst; 453c40a3d38SChandan Rajendra 4546275193eSQu Wenruo /* 4556275193eSQu Wenruo * Although both cur_disk_bytenr and orig_disk_bytenr is u64, 4566275193eSQu Wenruo * we're calculating the offset to the bio start. 4576275193eSQu Wenruo * 4586275193eSQu Wenruo * Bio size is limited to UINT_MAX, thus unsigned int is large 4596275193eSQu Wenruo * enough to contain the raw result, not to mention the right 4606275193eSQu Wenruo * shifted result. 4616275193eSQu Wenruo */ 4626275193eSQu Wenruo ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX); 4636275193eSQu Wenruo sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >> 4646275193eSQu Wenruo fs_info->sectorsize_bits; 4656275193eSQu Wenruo csum_dst = csum + sector_offset * csum_size; 4664989d277SChristoph Hellwig 4676275193eSQu Wenruo count = search_csum_tree(fs_info, path, cur_disk_bytenr, 4686275193eSQu Wenruo search_len, csum_dst); 4691784b7d5SJosef Bacik if (count < 0) { 4701784b7d5SJosef Bacik ret = errno_to_blk_status(count); 4711784b7d5SJosef Bacik if (bbio) 4721784b7d5SJosef Bacik btrfs_bio_free_csum(bbio); 4731784b7d5SJosef Bacik break; 4741784b7d5SJosef Bacik } 4751784b7d5SJosef Bacik 4766275193eSQu Wenruo /* 4771784b7d5SJosef Bacik * We didn't find a csum for this range. We need to make sure 4781784b7d5SJosef Bacik * we complain loudly about this, because we are not NODATASUM. 4791784b7d5SJosef Bacik * 4801784b7d5SJosef Bacik * However for the DATA_RELOC inode we could potentially be 4811784b7d5SJosef Bacik * relocating data extents for a NODATASUM inode, so the inode 4821784b7d5SJosef Bacik * itself won't be marked with NODATASUM, but the extent we're 4831784b7d5SJosef Bacik * copying is in fact NODATASUM. If we don't find a csum we 4841784b7d5SJosef Bacik * assume this is the case. 4856275193eSQu Wenruo */ 4861784b7d5SJosef Bacik if (count == 0) { 4876275193eSQu Wenruo memset(csum_dst, 0, csum_size); 488e4100d98SMiao Xie count = 1; 4896275193eSQu Wenruo 49017d217feSYan Zheng if (BTRFS_I(inode)->root->root_key.objectid == 49117d217feSYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID) { 4926275193eSQu Wenruo u64 file_offset; 4936275193eSQu Wenruo int ret; 4946275193eSQu Wenruo 4956275193eSQu Wenruo ret = search_file_offset_in_bio(bio, inode, 4966275193eSQu Wenruo cur_disk_bytenr, &file_offset); 4976275193eSQu Wenruo if (ret) 4986275193eSQu Wenruo set_extent_bits(io_tree, file_offset, 4996275193eSQu Wenruo file_offset + sectorsize - 1, 500ceeb0ae7SDavid Sterba EXTENT_NODATASUM); 50117d217feSYan Zheng } else { 5026275193eSQu Wenruo btrfs_warn_rl(fs_info, 5036275193eSQu Wenruo "csum hole found for disk bytenr range [%llu, %llu)", 5046275193eSQu Wenruo cur_disk_bytenr, cur_disk_bytenr + sectorsize); 50517d217feSYan Zheng } 5064989d277SChristoph Hellwig } 5074989d277SChristoph Hellwig } 5084989d277SChristoph Hellwig 50961b49440SChris Mason btrfs_free_path(path); 5101784b7d5SJosef Bacik return ret; 5114b46fce2SJosef Bacik } 5124b46fce2SJosef Bacik 51317d217feSYan Zheng int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end, 514a2de733cSArne Jansen struct list_head *list, int search_commit) 51517d217feSYan Zheng { 5160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 51717d217feSYan Zheng struct btrfs_key key; 51817d217feSYan Zheng struct btrfs_path *path; 51917d217feSYan Zheng struct extent_buffer *leaf; 52017d217feSYan Zheng struct btrfs_ordered_sum *sums; 52117d217feSYan Zheng struct btrfs_csum_item *item; 5220678b618SMark Fasheh LIST_HEAD(tmplist); 52317d217feSYan Zheng unsigned long offset; 52417d217feSYan Zheng int ret; 52517d217feSYan Zheng size_t size; 52617d217feSYan Zheng u64 csum_end; 527223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 52817d217feSYan Zheng 5290b246afaSJeff Mahoney ASSERT(IS_ALIGNED(start, fs_info->sectorsize) && 5300b246afaSJeff Mahoney IS_ALIGNED(end + 1, fs_info->sectorsize)); 5314277a9c3SJosef Bacik 53217d217feSYan Zheng path = btrfs_alloc_path(); 533d8926bb3SMark Fasheh if (!path) 534d8926bb3SMark Fasheh return -ENOMEM; 53517d217feSYan Zheng 536a2de733cSArne Jansen if (search_commit) { 537a2de733cSArne Jansen path->skip_locking = 1; 538e4058b54SDavid Sterba path->reada = READA_FORWARD; 539a2de733cSArne Jansen path->search_commit_root = 1; 540a2de733cSArne Jansen } 541a2de733cSArne Jansen 54217d217feSYan Zheng key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 54317d217feSYan Zheng key.offset = start; 54417d217feSYan Zheng key.type = BTRFS_EXTENT_CSUM_KEY; 54517d217feSYan Zheng 54607d400a6SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 54717d217feSYan Zheng if (ret < 0) 54817d217feSYan Zheng goto fail; 54917d217feSYan Zheng if (ret > 0 && path->slots[0] > 0) { 55017d217feSYan Zheng leaf = path->nodes[0]; 55117d217feSYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); 55217d217feSYan Zheng if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID && 55317d217feSYan Zheng key.type == BTRFS_EXTENT_CSUM_KEY) { 554265fdfa6SDavid Sterba offset = (start - key.offset) >> fs_info->sectorsize_bits; 55517d217feSYan Zheng if (offset * csum_size < 5563212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0] - 1)) 55717d217feSYan Zheng path->slots[0]--; 55817d217feSYan Zheng } 55917d217feSYan Zheng } 56017d217feSYan Zheng 56117d217feSYan Zheng while (start <= end) { 56217d217feSYan Zheng leaf = path->nodes[0]; 56317d217feSYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 56407d400a6SYan Zheng ret = btrfs_next_leaf(root, path); 56517d217feSYan Zheng if (ret < 0) 56617d217feSYan Zheng goto fail; 56717d217feSYan Zheng if (ret > 0) 56817d217feSYan Zheng break; 56917d217feSYan Zheng leaf = path->nodes[0]; 57017d217feSYan Zheng } 57117d217feSYan Zheng 57217d217feSYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 57317d217feSYan Zheng if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 574628c8282SZhi Yong Wu key.type != BTRFS_EXTENT_CSUM_KEY || 575628c8282SZhi Yong Wu key.offset > end) 57617d217feSYan Zheng break; 57717d217feSYan Zheng 57817d217feSYan Zheng if (key.offset > start) 57917d217feSYan Zheng start = key.offset; 58017d217feSYan Zheng 5813212fa14SJosef Bacik size = btrfs_item_size(leaf, path->slots[0]); 5820b246afaSJeff Mahoney csum_end = key.offset + (size / csum_size) * fs_info->sectorsize; 58387b29b20SYan Zheng if (csum_end <= start) { 58487b29b20SYan Zheng path->slots[0]++; 58587b29b20SYan Zheng continue; 58687b29b20SYan Zheng } 58717d217feSYan Zheng 58807d400a6SYan Zheng csum_end = min(csum_end, end + 1); 58907d400a6SYan Zheng item = btrfs_item_ptr(path->nodes[0], path->slots[0], 59007d400a6SYan Zheng struct btrfs_csum_item); 59107d400a6SYan Zheng while (start < csum_end) { 59207d400a6SYan Zheng size = min_t(size_t, csum_end - start, 5931e25a2e3SJohannes Thumshirn max_ordered_sum_bytes(fs_info, csum_size)); 5940b246afaSJeff Mahoney sums = kzalloc(btrfs_ordered_sum_size(fs_info, size), 59507d400a6SYan Zheng GFP_NOFS); 5960678b618SMark Fasheh if (!sums) { 5970678b618SMark Fasheh ret = -ENOMEM; 5980678b618SMark Fasheh goto fail; 5990678b618SMark Fasheh } 60017d217feSYan Zheng 60117d217feSYan Zheng sums->bytenr = start; 602f51a4a18SMiao Xie sums->len = (int)size; 60317d217feSYan Zheng 604265fdfa6SDavid Sterba offset = (start - key.offset) >> fs_info->sectorsize_bits; 60517d217feSYan Zheng offset *= csum_size; 606265fdfa6SDavid Sterba size >>= fs_info->sectorsize_bits; 60717d217feSYan Zheng 60807d400a6SYan Zheng read_extent_buffer(path->nodes[0], 609f51a4a18SMiao Xie sums->sums, 610f51a4a18SMiao Xie ((unsigned long)item) + offset, 611f51a4a18SMiao Xie csum_size * size); 61217d217feSYan Zheng 6130b246afaSJeff Mahoney start += fs_info->sectorsize * size; 6140678b618SMark Fasheh list_add_tail(&sums->list, &tmplist); 61507d400a6SYan Zheng } 61617d217feSYan Zheng path->slots[0]++; 61717d217feSYan Zheng } 61817d217feSYan Zheng ret = 0; 61917d217feSYan Zheng fail: 6200678b618SMark Fasheh while (ret < 0 && !list_empty(&tmplist)) { 6216e5aafb2SChris Mason sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list); 6220678b618SMark Fasheh list_del(&sums->list); 6230678b618SMark Fasheh kfree(sums); 6240678b618SMark Fasheh } 6250678b618SMark Fasheh list_splice_tail(&tmplist, list); 6260678b618SMark Fasheh 62717d217feSYan Zheng btrfs_free_path(path); 62817d217feSYan Zheng return ret; 62917d217feSYan Zheng } 63017d217feSYan Zheng 631e331f6b1SOmar Sandoval /** 632e331f6b1SOmar Sandoval * Calculate checksums of the data contained inside a bio 633e331f6b1SOmar Sandoval * 63451d470aeSNikolay Borisov * @inode: Owner of the data inside the bio 63551d470aeSNikolay Borisov * @bio: Contains the data to be checksummed 636e331f6b1SOmar Sandoval * @offset: If (u64)-1, @bio may contain discontiguous bio vecs, so the 637e331f6b1SOmar Sandoval * file offsets are determined from the page offsets in the bio. 638e331f6b1SOmar Sandoval * Otherwise, this is the starting file offset of the bio vecs in 639e331f6b1SOmar Sandoval * @bio, which must be contiguous. 640e331f6b1SOmar Sandoval * @one_ordered: If true, @bio only refers to one ordered extent. 64151d470aeSNikolay Borisov */ 642bd242a08SNikolay Borisov blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio, 643e331f6b1SOmar Sandoval u64 offset, bool one_ordered) 644e015640fSChris Mason { 645c3504372SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 646d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 647e6dcd2dcSChris Mason struct btrfs_ordered_sum *sums; 6486cd7ce49SChristoph Hellwig struct btrfs_ordered_extent *ordered = NULL; 649e331f6b1SOmar Sandoval const bool use_page_offsets = (offset == (u64)-1); 650e015640fSChris Mason char *data; 65117347cecSLiu Bo struct bvec_iter iter; 65217347cecSLiu Bo struct bio_vec bvec; 653f51a4a18SMiao Xie int index; 654e331f6b1SOmar Sandoval unsigned int blockcount; 6553edf7d33SChris Mason unsigned long total_bytes = 0; 6563edf7d33SChris Mason unsigned long this_sum_bytes = 0; 65717347cecSLiu Bo int i; 658a3d46aeaSNikolay Borisov unsigned nofs_flag; 659e015640fSChris Mason 660a3d46aeaSNikolay Borisov nofs_flag = memalloc_nofs_save(); 661a3d46aeaSNikolay Borisov sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size), 662a3d46aeaSNikolay Borisov GFP_KERNEL); 663a3d46aeaSNikolay Borisov memalloc_nofs_restore(nofs_flag); 664a3d46aeaSNikolay Borisov 665e015640fSChris Mason if (!sums) 6664e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 6673edf7d33SChris Mason 6684f024f37SKent Overstreet sums->len = bio->bi_iter.bi_size; 669e6dcd2dcSChris Mason INIT_LIST_HEAD(&sums->list); 670d20f7043SChris Mason 6711201b58bSDavid Sterba sums->bytenr = bio->bi_iter.bi_sector << 9; 672f51a4a18SMiao Xie index = 0; 673e015640fSChris Mason 674d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 675d5178578SJohannes Thumshirn 67617347cecSLiu Bo bio_for_each_segment(bvec, bio, iter) { 677e331f6b1SOmar Sandoval if (use_page_offsets) 67817347cecSLiu Bo offset = page_offset(bvec.bv_page) + bvec.bv_offset; 679d20f7043SChris Mason 6806cd7ce49SChristoph Hellwig if (!ordered) { 6816cd7ce49SChristoph Hellwig ordered = btrfs_lookup_ordered_extent(inode, offset); 682bbc9a6ebSQu Wenruo /* 683bbc9a6ebSQu Wenruo * The bio range is not covered by any ordered extent, 684bbc9a6ebSQu Wenruo * must be a code logic error. 685bbc9a6ebSQu Wenruo */ 686bbc9a6ebSQu Wenruo if (unlikely(!ordered)) { 687bbc9a6ebSQu Wenruo WARN(1, KERN_WARNING 688bbc9a6ebSQu Wenruo "no ordered extent for root %llu ino %llu offset %llu\n", 689bbc9a6ebSQu Wenruo inode->root->root_key.objectid, 690bbc9a6ebSQu Wenruo btrfs_ino(inode), offset); 691bbc9a6ebSQu Wenruo kvfree(sums); 692bbc9a6ebSQu Wenruo return BLK_STS_IOERR; 693bbc9a6ebSQu Wenruo } 6946cd7ce49SChristoph Hellwig } 6956cd7ce49SChristoph Hellwig 696e331f6b1SOmar Sandoval blockcount = BTRFS_BYTES_TO_BLKS(fs_info, 69717347cecSLiu Bo bvec.bv_len + fs_info->sectorsize 698c40a3d38SChandan Rajendra - 1); 699c40a3d38SChandan Rajendra 700e331f6b1SOmar Sandoval for (i = 0; i < blockcount; i++) { 701e331f6b1SOmar Sandoval if (!one_ordered && 702e331f6b1SOmar Sandoval !in_range(offset, ordered->file_offset, 703e331f6b1SOmar Sandoval ordered->num_bytes)) { 7043edf7d33SChris Mason unsigned long bytes_left; 705c40a3d38SChandan Rajendra 7063edf7d33SChris Mason sums->len = this_sum_bytes; 7073edf7d33SChris Mason this_sum_bytes = 0; 708f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 7093edf7d33SChris Mason btrfs_put_ordered_extent(ordered); 7103edf7d33SChris Mason 7114f024f37SKent Overstreet bytes_left = bio->bi_iter.bi_size - total_bytes; 7123edf7d33SChris Mason 713a3d46aeaSNikolay Borisov nofs_flag = memalloc_nofs_save(); 714a3d46aeaSNikolay Borisov sums = kvzalloc(btrfs_ordered_sum_size(fs_info, 715a3d46aeaSNikolay Borisov bytes_left), GFP_KERNEL); 716a3d46aeaSNikolay Borisov memalloc_nofs_restore(nofs_flag); 71779787eaaSJeff Mahoney BUG_ON(!sums); /* -ENOMEM */ 7183edf7d33SChris Mason sums->len = bytes_left; 719c40a3d38SChandan Rajendra ordered = btrfs_lookup_ordered_extent(inode, 720c40a3d38SChandan Rajendra offset); 721c40a3d38SChandan Rajendra ASSERT(ordered); /* Logic error */ 7221201b58bSDavid Sterba sums->bytenr = (bio->bi_iter.bi_sector << 9) 723c40a3d38SChandan Rajendra + total_bytes; 724f51a4a18SMiao Xie index = 0; 725c40a3d38SChandan Rajendra } 726c40a3d38SChandan Rajendra 7273dcfbcceSChristoph Hellwig data = bvec_kmap_local(&bvec); 7283dcfbcceSChristoph Hellwig crypto_shash_digest(shash, 7293dcfbcceSChristoph Hellwig data + (i * fs_info->sectorsize), 730fd08001fSEric Biggers fs_info->sectorsize, 731fd08001fSEric Biggers sums->sums + index); 7323dcfbcceSChristoph Hellwig kunmap_local(data); 733713cebfbSDavid Sterba index += fs_info->csum_size; 7340b246afaSJeff Mahoney offset += fs_info->sectorsize; 7350b246afaSJeff Mahoney this_sum_bytes += fs_info->sectorsize; 7360b246afaSJeff Mahoney total_bytes += fs_info->sectorsize; 737c40a3d38SChandan Rajendra } 738c40a3d38SChandan Rajendra 739e015640fSChris Mason } 740ed98b56aSChris Mason this_sum_bytes = 0; 741f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 7423edf7d33SChris Mason btrfs_put_ordered_extent(ordered); 743e015640fSChris Mason return 0; 744e015640fSChris Mason } 745e015640fSChris Mason 746459931ecSChris Mason /* 747459931ecSChris Mason * helper function for csum removal, this expects the 748459931ecSChris Mason * key to describe the csum pointed to by the path, and it expects 749459931ecSChris Mason * the csum to overlap the range [bytenr, len] 750459931ecSChris Mason * 751459931ecSChris Mason * The csum should not be entirely contained in the range and the 752459931ecSChris Mason * range should not be entirely contained in the csum. 753459931ecSChris Mason * 754459931ecSChris Mason * This calls btrfs_truncate_item with the correct args based on the 755459931ecSChris Mason * overlap, and fixes up the key as required. 756459931ecSChris Mason */ 7572ff7e61eSJeff Mahoney static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info, 758459931ecSChris Mason struct btrfs_path *path, 759459931ecSChris Mason struct btrfs_key *key, 760459931ecSChris Mason u64 bytenr, u64 len) 761459931ecSChris Mason { 762459931ecSChris Mason struct extent_buffer *leaf; 763223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 764459931ecSChris Mason u64 csum_end; 765459931ecSChris Mason u64 end_byte = bytenr + len; 766265fdfa6SDavid Sterba u32 blocksize_bits = fs_info->sectorsize_bits; 767459931ecSChris Mason 768459931ecSChris Mason leaf = path->nodes[0]; 7693212fa14SJosef Bacik csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size; 770265fdfa6SDavid Sterba csum_end <<= blocksize_bits; 771459931ecSChris Mason csum_end += key->offset; 772459931ecSChris Mason 773459931ecSChris Mason if (key->offset < bytenr && csum_end <= end_byte) { 774459931ecSChris Mason /* 775459931ecSChris Mason * [ bytenr - len ] 776459931ecSChris Mason * [ ] 777459931ecSChris Mason * [csum ] 778459931ecSChris Mason * A simple truncate off the end of the item 779459931ecSChris Mason */ 780459931ecSChris Mason u32 new_size = (bytenr - key->offset) >> blocksize_bits; 781459931ecSChris Mason new_size *= csum_size; 78278ac4f9eSDavid Sterba btrfs_truncate_item(path, new_size, 1); 783459931ecSChris Mason } else if (key->offset >= bytenr && csum_end > end_byte && 784459931ecSChris Mason end_byte > key->offset) { 785459931ecSChris Mason /* 786459931ecSChris Mason * [ bytenr - len ] 787459931ecSChris Mason * [ ] 788459931ecSChris Mason * [csum ] 789459931ecSChris Mason * we need to truncate from the beginning of the csum 790459931ecSChris Mason */ 791459931ecSChris Mason u32 new_size = (csum_end - end_byte) >> blocksize_bits; 792459931ecSChris Mason new_size *= csum_size; 793459931ecSChris Mason 79478ac4f9eSDavid Sterba btrfs_truncate_item(path, new_size, 0); 795459931ecSChris Mason 796459931ecSChris Mason key->offset = end_byte; 7970b246afaSJeff Mahoney btrfs_set_item_key_safe(fs_info, path, key); 798459931ecSChris Mason } else { 799459931ecSChris Mason BUG(); 800459931ecSChris Mason } 801459931ecSChris Mason } 802459931ecSChris Mason 803459931ecSChris Mason /* 804459931ecSChris Mason * deletes the csum items from the csum tree for a given 805459931ecSChris Mason * range of bytes. 806459931ecSChris Mason */ 807459931ecSChris Mason int btrfs_del_csums(struct btrfs_trans_handle *trans, 80840e046acSFilipe Manana struct btrfs_root *root, u64 bytenr, u64 len) 809459931ecSChris Mason { 81040e046acSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 811459931ecSChris Mason struct btrfs_path *path; 812459931ecSChris Mason struct btrfs_key key; 813459931ecSChris Mason u64 end_byte = bytenr + len; 814459931ecSChris Mason u64 csum_end; 815459931ecSChris Mason struct extent_buffer *leaf; 816b86652beSJosef Bacik int ret = 0; 817223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 818265fdfa6SDavid Sterba u32 blocksize_bits = fs_info->sectorsize_bits; 819459931ecSChris Mason 82084d2d6c7SJosef Bacik ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID || 82140e046acSFilipe Manana root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 82240e046acSFilipe Manana 823459931ecSChris Mason path = btrfs_alloc_path(); 8242a29edc6Sliubo if (!path) 8252a29edc6Sliubo return -ENOMEM; 826459931ecSChris Mason 827459931ecSChris Mason while (1) { 828459931ecSChris Mason key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 829459931ecSChris Mason key.offset = end_byte - 1; 830459931ecSChris Mason key.type = BTRFS_EXTENT_CSUM_KEY; 831459931ecSChris Mason 832459931ecSChris Mason ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 833459931ecSChris Mason if (ret > 0) { 834b86652beSJosef Bacik ret = 0; 835459931ecSChris Mason if (path->slots[0] == 0) 83665a246c5STsutomu Itoh break; 837459931ecSChris Mason path->slots[0]--; 838ad0397a7SJosef Bacik } else if (ret < 0) { 83965a246c5STsutomu Itoh break; 840459931ecSChris Mason } 841ad0397a7SJosef Bacik 842459931ecSChris Mason leaf = path->nodes[0]; 843459931ecSChris Mason btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 844459931ecSChris Mason 845459931ecSChris Mason if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 846459931ecSChris Mason key.type != BTRFS_EXTENT_CSUM_KEY) { 847459931ecSChris Mason break; 848459931ecSChris Mason } 849459931ecSChris Mason 850459931ecSChris Mason if (key.offset >= end_byte) 851459931ecSChris Mason break; 852459931ecSChris Mason 8533212fa14SJosef Bacik csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size; 854459931ecSChris Mason csum_end <<= blocksize_bits; 855459931ecSChris Mason csum_end += key.offset; 856459931ecSChris Mason 857459931ecSChris Mason /* this csum ends before we start, we're done */ 858459931ecSChris Mason if (csum_end <= bytenr) 859459931ecSChris Mason break; 860459931ecSChris Mason 861459931ecSChris Mason /* delete the entire item, it is inside our range */ 862459931ecSChris Mason if (key.offset >= bytenr && csum_end <= end_byte) { 8636f546216SFilipe Manana int del_nr = 1; 8646f546216SFilipe Manana 8656f546216SFilipe Manana /* 8666f546216SFilipe Manana * Check how many csum items preceding this one in this 8676f546216SFilipe Manana * leaf correspond to our range and then delete them all 8686f546216SFilipe Manana * at once. 8696f546216SFilipe Manana */ 8706f546216SFilipe Manana if (key.offset > bytenr && path->slots[0] > 0) { 8716f546216SFilipe Manana int slot = path->slots[0] - 1; 8726f546216SFilipe Manana 8736f546216SFilipe Manana while (slot >= 0) { 8746f546216SFilipe Manana struct btrfs_key pk; 8756f546216SFilipe Manana 8766f546216SFilipe Manana btrfs_item_key_to_cpu(leaf, &pk, slot); 8776f546216SFilipe Manana if (pk.offset < bytenr || 8786f546216SFilipe Manana pk.type != BTRFS_EXTENT_CSUM_KEY || 8796f546216SFilipe Manana pk.objectid != 8806f546216SFilipe Manana BTRFS_EXTENT_CSUM_OBJECTID) 8816f546216SFilipe Manana break; 8826f546216SFilipe Manana path->slots[0] = slot; 8836f546216SFilipe Manana del_nr++; 8846f546216SFilipe Manana key.offset = pk.offset; 8856f546216SFilipe Manana slot--; 8866f546216SFilipe Manana } 8876f546216SFilipe Manana } 8886f546216SFilipe Manana ret = btrfs_del_items(trans, root, path, 8896f546216SFilipe Manana path->slots[0], del_nr); 89065a246c5STsutomu Itoh if (ret) 891b86652beSJosef Bacik break; 892dcbdd4dcSChris Mason if (key.offset == bytenr) 893dcbdd4dcSChris Mason break; 894459931ecSChris Mason } else if (key.offset < bytenr && csum_end > end_byte) { 895459931ecSChris Mason unsigned long offset; 896459931ecSChris Mason unsigned long shift_len; 897459931ecSChris Mason unsigned long item_offset; 898459931ecSChris Mason /* 899459931ecSChris Mason * [ bytenr - len ] 900459931ecSChris Mason * [csum ] 901459931ecSChris Mason * 902459931ecSChris Mason * Our bytes are in the middle of the csum, 903459931ecSChris Mason * we need to split this item and insert a new one. 904459931ecSChris Mason * 905459931ecSChris Mason * But we can't drop the path because the 906459931ecSChris Mason * csum could change, get removed, extended etc. 907459931ecSChris Mason * 908459931ecSChris Mason * The trick here is the max size of a csum item leaves 909459931ecSChris Mason * enough room in the tree block for a single 910459931ecSChris Mason * item header. So, we split the item in place, 911459931ecSChris Mason * adding a new header pointing to the existing 912459931ecSChris Mason * bytes. Then we loop around again and we have 913459931ecSChris Mason * a nicely formed csum item that we can neatly 914459931ecSChris Mason * truncate. 915459931ecSChris Mason */ 916459931ecSChris Mason offset = (bytenr - key.offset) >> blocksize_bits; 917459931ecSChris Mason offset *= csum_size; 918459931ecSChris Mason 919459931ecSChris Mason shift_len = (len >> blocksize_bits) * csum_size; 920459931ecSChris Mason 921459931ecSChris Mason item_offset = btrfs_item_ptr_offset(leaf, 922459931ecSChris Mason path->slots[0]); 923459931ecSChris Mason 924b159fa28SDavid Sterba memzero_extent_buffer(leaf, item_offset + offset, 925459931ecSChris Mason shift_len); 926459931ecSChris Mason key.offset = bytenr; 927459931ecSChris Mason 928459931ecSChris Mason /* 929459931ecSChris Mason * btrfs_split_item returns -EAGAIN when the 930459931ecSChris Mason * item changed size or key 931459931ecSChris Mason */ 932459931ecSChris Mason ret = btrfs_split_item(trans, root, path, &key, offset); 93379787eaaSJeff Mahoney if (ret && ret != -EAGAIN) { 93466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 935b86652beSJosef Bacik break; 93679787eaaSJeff Mahoney } 937b86652beSJosef Bacik ret = 0; 938459931ecSChris Mason 939459931ecSChris Mason key.offset = end_byte - 1; 940459931ecSChris Mason } else { 9412ff7e61eSJeff Mahoney truncate_one_csum(fs_info, path, &key, bytenr, len); 942dcbdd4dcSChris Mason if (key.offset < bytenr) 943dcbdd4dcSChris Mason break; 944459931ecSChris Mason } 945b3b4aa74SDavid Sterba btrfs_release_path(path); 946459931ecSChris Mason } 947459931ecSChris Mason btrfs_free_path(path); 94865a246c5STsutomu Itoh return ret; 949459931ecSChris Mason } 950459931ecSChris Mason 951ea7036deSFilipe Manana static int find_next_csum_offset(struct btrfs_root *root, 952ea7036deSFilipe Manana struct btrfs_path *path, 953ea7036deSFilipe Manana u64 *next_offset) 954ea7036deSFilipe Manana { 955ea7036deSFilipe Manana const u32 nritems = btrfs_header_nritems(path->nodes[0]); 956ea7036deSFilipe Manana struct btrfs_key found_key; 957ea7036deSFilipe Manana int slot = path->slots[0] + 1; 958ea7036deSFilipe Manana int ret; 959ea7036deSFilipe Manana 960ea7036deSFilipe Manana if (nritems == 0 || slot >= nritems) { 961ea7036deSFilipe Manana ret = btrfs_next_leaf(root, path); 962ea7036deSFilipe Manana if (ret < 0) { 963ea7036deSFilipe Manana return ret; 964ea7036deSFilipe Manana } else if (ret > 0) { 965ea7036deSFilipe Manana *next_offset = (u64)-1; 966ea7036deSFilipe Manana return 0; 967ea7036deSFilipe Manana } 968ea7036deSFilipe Manana slot = path->slots[0]; 969ea7036deSFilipe Manana } 970ea7036deSFilipe Manana 971ea7036deSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot); 972ea7036deSFilipe Manana 973ea7036deSFilipe Manana if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 974ea7036deSFilipe Manana found_key.type != BTRFS_EXTENT_CSUM_KEY) 975ea7036deSFilipe Manana *next_offset = (u64)-1; 976ea7036deSFilipe Manana else 977ea7036deSFilipe Manana *next_offset = found_key.offset; 978ea7036deSFilipe Manana 979ea7036deSFilipe Manana return 0; 980ea7036deSFilipe Manana } 981ea7036deSFilipe Manana 982065631f6SChris Mason int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans, 983d20f7043SChris Mason struct btrfs_root *root, 984e6dcd2dcSChris Mason struct btrfs_ordered_sum *sums) 985f254e52cSChris Mason { 9860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 987f254e52cSChris Mason struct btrfs_key file_key; 9886567e837SChris Mason struct btrfs_key found_key; 9895caf2a00SChris Mason struct btrfs_path *path; 990f254e52cSChris Mason struct btrfs_csum_item *item; 991065631f6SChris Mason struct btrfs_csum_item *item_end; 992ff79f819SChris Mason struct extent_buffer *leaf = NULL; 993f51a4a18SMiao Xie u64 next_offset; 994f51a4a18SMiao Xie u64 total_bytes = 0; 9956567e837SChris Mason u64 csum_offset; 996f51a4a18SMiao Xie u64 bytenr; 997f578d4bdSChris Mason u32 ins_size; 998f51a4a18SMiao Xie int index = 0; 999f51a4a18SMiao Xie int found_next; 1000f51a4a18SMiao Xie int ret; 1001223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 10026e92f5e6SChris Mason 10035caf2a00SChris Mason path = btrfs_alloc_path(); 1004d8926bb3SMark Fasheh if (!path) 1005d8926bb3SMark Fasheh return -ENOMEM; 1006065631f6SChris Mason again: 1007065631f6SChris Mason next_offset = (u64)-1; 1008065631f6SChris Mason found_next = 0; 1009f51a4a18SMiao Xie bytenr = sums->bytenr + total_bytes; 1010d20f7043SChris Mason file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 1011f51a4a18SMiao Xie file_key.offset = bytenr; 1012962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_CSUM_KEY; 1013a429e513SChris Mason 1014f51a4a18SMiao Xie item = btrfs_lookup_csum(trans, root, path, bytenr, 1); 1015ff79f819SChris Mason if (!IS_ERR(item)) { 1016639cb586SChris Mason ret = 0; 1017f51a4a18SMiao Xie leaf = path->nodes[0]; 1018f51a4a18SMiao Xie item_end = btrfs_item_ptr(leaf, path->slots[0], 1019f51a4a18SMiao Xie struct btrfs_csum_item); 1020f51a4a18SMiao Xie item_end = (struct btrfs_csum_item *)((char *)item_end + 10213212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0])); 1022a429e513SChris Mason goto found; 1023ff79f819SChris Mason } 1024a429e513SChris Mason ret = PTR_ERR(item); 10254a500fd1SYan, Zheng if (ret != -EFBIG && ret != -ENOENT) 1026918cdf44SFilipe Manana goto out; 10274a500fd1SYan, Zheng 1028a429e513SChris Mason if (ret == -EFBIG) { 1029a429e513SChris Mason u32 item_size; 1030a429e513SChris Mason /* we found one, but it isn't big enough yet */ 10315f39d397SChris Mason leaf = path->nodes[0]; 10323212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 1033607d432dSJosef Bacik if ((item_size / csum_size) >= 10340b246afaSJeff Mahoney MAX_CSUM_ITEMS(fs_info, csum_size)) { 1035a429e513SChris Mason /* already at max size, make a new one */ 1036a429e513SChris Mason goto insert; 1037a429e513SChris Mason } 1038a429e513SChris Mason } else { 1039ea7036deSFilipe Manana /* We didn't find a csum item, insert one. */ 1040ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1041ea7036deSFilipe Manana if (ret < 0) 10427e4a3f7eSFilipe Manana goto out; 1043f578d4bdSChris Mason found_next = 1; 1044a429e513SChris Mason goto insert; 1045a429e513SChris Mason } 1046a429e513SChris Mason 1047a429e513SChris Mason /* 1048cc14600cSFilipe Manana * At this point, we know the tree has a checksum item that ends at an 1049cc14600cSFilipe Manana * offset matching the start of the checksum range we want to insert. 1050cc14600cSFilipe Manana * We try to extend that item as much as possible and then add as many 1051cc14600cSFilipe Manana * checksums to it as they fit. 1052cc14600cSFilipe Manana * 1053cc14600cSFilipe Manana * First check if the leaf has enough free space for at least one 1054cc14600cSFilipe Manana * checksum. If it has go directly to the item extension code, otherwise 1055cc14600cSFilipe Manana * release the path and do a search for insertion before the extension. 1056a429e513SChris Mason */ 1057cc14600cSFilipe Manana if (btrfs_leaf_free_space(leaf) >= csum_size) { 1058cc14600cSFilipe Manana btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1059cc14600cSFilipe Manana csum_offset = (bytenr - found_key.offset) >> 1060265fdfa6SDavid Sterba fs_info->sectorsize_bits; 1061cc14600cSFilipe Manana goto extend_csum; 1062cc14600cSFilipe Manana } 1063cc14600cSFilipe Manana 1064b3b4aa74SDavid Sterba btrfs_release_path(path); 10659a664971Sethanwu path->search_for_extension = 1; 10666567e837SChris Mason ret = btrfs_search_slot(trans, root, &file_key, path, 1067607d432dSJosef Bacik csum_size, 1); 10689a664971Sethanwu path->search_for_extension = 0; 10696567e837SChris Mason if (ret < 0) 1070918cdf44SFilipe Manana goto out; 1071459931ecSChris Mason 1072459931ecSChris Mason if (ret > 0) { 1073459931ecSChris Mason if (path->slots[0] == 0) 10746567e837SChris Mason goto insert; 10756567e837SChris Mason path->slots[0]--; 1076459931ecSChris Mason } 1077459931ecSChris Mason 10785f39d397SChris Mason leaf = path->nodes[0]; 10795f39d397SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1080265fdfa6SDavid Sterba csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits; 1081459931ecSChris Mason 1082962a298fSDavid Sterba if (found_key.type != BTRFS_EXTENT_CSUM_KEY || 1083d20f7043SChris Mason found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 10840b246afaSJeff Mahoney csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) { 10856567e837SChris Mason goto insert; 10866567e837SChris Mason } 1087459931ecSChris Mason 1088cc14600cSFilipe Manana extend_csum: 10893212fa14SJosef Bacik if (csum_offset == btrfs_item_size(leaf, path->slots[0]) / 1090607d432dSJosef Bacik csum_size) { 10912f697dc6SLiu Bo int extend_nr; 10922f697dc6SLiu Bo u64 tmp; 10932f697dc6SLiu Bo u32 diff; 1094459931ecSChris Mason 1095f51a4a18SMiao Xie tmp = sums->len - total_bytes; 1096265fdfa6SDavid Sterba tmp >>= fs_info->sectorsize_bits; 10972f697dc6SLiu Bo WARN_ON(tmp < 1); 1098ea7036deSFilipe Manana extend_nr = max_t(int, 1, tmp); 10992f697dc6SLiu Bo 1100ea7036deSFilipe Manana /* 1101ea7036deSFilipe Manana * A log tree can already have checksum items with a subset of 1102ea7036deSFilipe Manana * the checksums we are trying to log. This can happen after 1103ea7036deSFilipe Manana * doing a sequence of partial writes into prealloc extents and 1104ea7036deSFilipe Manana * fsyncs in between, with a full fsync logging a larger subrange 1105ea7036deSFilipe Manana * of an extent for which a previous fast fsync logged a smaller 1106ea7036deSFilipe Manana * subrange. And this happens in particular due to merging file 1107ea7036deSFilipe Manana * extent items when we complete an ordered extent for a range 1108ea7036deSFilipe Manana * covered by a prealloc extent - this is done at 1109ea7036deSFilipe Manana * btrfs_mark_extent_written(). 1110ea7036deSFilipe Manana * 1111ea7036deSFilipe Manana * So if we try to extend the previous checksum item, which has 1112ea7036deSFilipe Manana * a range that ends at the start of the range we want to insert, 1113ea7036deSFilipe Manana * make sure we don't extend beyond the start offset of the next 1114ea7036deSFilipe Manana * checksum item. If we are at the last item in the leaf, then 1115ea7036deSFilipe Manana * forget the optimization of extending and add a new checksum 1116ea7036deSFilipe Manana * item - it is not worth the complexity of releasing the path, 1117ea7036deSFilipe Manana * getting the first key for the next leaf, repeat the btree 1118ea7036deSFilipe Manana * search, etc, because log trees are temporary anyway and it 1119ea7036deSFilipe Manana * would only save a few bytes of leaf space. 1120ea7036deSFilipe Manana */ 1121ea7036deSFilipe Manana if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { 1122ea7036deSFilipe Manana if (path->slots[0] + 1 >= 1123ea7036deSFilipe Manana btrfs_header_nritems(path->nodes[0])) { 1124ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1125ea7036deSFilipe Manana if (ret < 0) 1126ea7036deSFilipe Manana goto out; 1127ea7036deSFilipe Manana found_next = 1; 1128ea7036deSFilipe Manana goto insert; 1129ea7036deSFilipe Manana } 1130ea7036deSFilipe Manana 1131ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1132ea7036deSFilipe Manana if (ret < 0) 1133ea7036deSFilipe Manana goto out; 1134ea7036deSFilipe Manana 1135ea7036deSFilipe Manana tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits; 1136ea7036deSFilipe Manana if (tmp <= INT_MAX) 1137ea7036deSFilipe Manana extend_nr = min_t(int, extend_nr, tmp); 1138ea7036deSFilipe Manana } 1139ea7036deSFilipe Manana 11402f697dc6SLiu Bo diff = (csum_offset + extend_nr) * csum_size; 11410b246afaSJeff Mahoney diff = min(diff, 11420b246afaSJeff Mahoney MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size); 1143459931ecSChris Mason 11443212fa14SJosef Bacik diff = diff - btrfs_item_size(leaf, path->slots[0]); 1145cc14600cSFilipe Manana diff = min_t(u32, btrfs_leaf_free_space(leaf), diff); 11462f697dc6SLiu Bo diff /= csum_size; 11472f697dc6SLiu Bo diff *= csum_size; 1148459931ecSChris Mason 1149c71dd880SDavid Sterba btrfs_extend_item(path, diff); 1150f51a4a18SMiao Xie ret = 0; 11516567e837SChris Mason goto csum; 11526567e837SChris Mason } 11536567e837SChris Mason 11546567e837SChris Mason insert: 1155b3b4aa74SDavid Sterba btrfs_release_path(path); 11566567e837SChris Mason csum_offset = 0; 1157f578d4bdSChris Mason if (found_next) { 11582f697dc6SLiu Bo u64 tmp; 1159d20f7043SChris Mason 1160f51a4a18SMiao Xie tmp = sums->len - total_bytes; 1161265fdfa6SDavid Sterba tmp >>= fs_info->sectorsize_bits; 11622f697dc6SLiu Bo tmp = min(tmp, (next_offset - file_key.offset) >> 1163265fdfa6SDavid Sterba fs_info->sectorsize_bits); 11642f697dc6SLiu Bo 116550d0446eSSeraphime Kirkovski tmp = max_t(u64, 1, tmp); 116650d0446eSSeraphime Kirkovski tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size)); 1167607d432dSJosef Bacik ins_size = csum_size * tmp; 1168f578d4bdSChris Mason } else { 1169607d432dSJosef Bacik ins_size = csum_size; 1170f578d4bdSChris Mason } 11715caf2a00SChris Mason ret = btrfs_insert_empty_item(trans, root, path, &file_key, 1172f578d4bdSChris Mason ins_size); 117354aa1f4dSChris Mason if (ret < 0) 1174918cdf44SFilipe Manana goto out; 1175fae7f21cSDulshani Gunawardhana if (WARN_ON(ret != 0)) 1176918cdf44SFilipe Manana goto out; 11775f39d397SChris Mason leaf = path->nodes[0]; 1178f51a4a18SMiao Xie csum: 11795f39d397SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item); 1180f51a4a18SMiao Xie item_end = (struct btrfs_csum_item *)((unsigned char *)item + 11813212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0])); 1182509659cdSChris Mason item = (struct btrfs_csum_item *)((unsigned char *)item + 1183607d432dSJosef Bacik csum_offset * csum_size); 1184b18c6685SChris Mason found: 1185265fdfa6SDavid Sterba ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits; 1186f51a4a18SMiao Xie ins_size *= csum_size; 1187f51a4a18SMiao Xie ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item, 1188f51a4a18SMiao Xie ins_size); 1189f51a4a18SMiao Xie write_extent_buffer(leaf, sums->sums + index, (unsigned long)item, 1190f51a4a18SMiao Xie ins_size); 1191aadfeb6eSChris Mason 11921e25a2e3SJohannes Thumshirn index += ins_size; 1193f51a4a18SMiao Xie ins_size /= csum_size; 11940b246afaSJeff Mahoney total_bytes += ins_size * fs_info->sectorsize; 1195a6591715SChris Mason 11965caf2a00SChris Mason btrfs_mark_buffer_dirty(path->nodes[0]); 1197e6dcd2dcSChris Mason if (total_bytes < sums->len) { 1198b3b4aa74SDavid Sterba btrfs_release_path(path); 1199b9473439SChris Mason cond_resched(); 1200065631f6SChris Mason goto again; 1201065631f6SChris Mason } 120253863232SChris Mason out: 12035caf2a00SChris Mason btrfs_free_path(path); 1204f254e52cSChris Mason return ret; 1205f254e52cSChris Mason } 12067ffbb598SFilipe Manana 12079cdc5124SNikolay Borisov void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode, 12087ffbb598SFilipe Manana const struct btrfs_path *path, 12097ffbb598SFilipe Manana struct btrfs_file_extent_item *fi, 12107ffbb598SFilipe Manana const bool new_inline, 12117ffbb598SFilipe Manana struct extent_map *em) 12127ffbb598SFilipe Manana { 12133ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 12149cdc5124SNikolay Borisov struct btrfs_root *root = inode->root; 12157ffbb598SFilipe Manana struct extent_buffer *leaf = path->nodes[0]; 12167ffbb598SFilipe Manana const int slot = path->slots[0]; 12177ffbb598SFilipe Manana struct btrfs_key key; 12187ffbb598SFilipe Manana u64 extent_start, extent_end; 12197ffbb598SFilipe Manana u64 bytenr; 12207ffbb598SFilipe Manana u8 type = btrfs_file_extent_type(leaf, fi); 12217ffbb598SFilipe Manana int compress_type = btrfs_file_extent_compression(leaf, fi); 12227ffbb598SFilipe Manana 12237ffbb598SFilipe Manana btrfs_item_key_to_cpu(leaf, &key, slot); 12247ffbb598SFilipe Manana extent_start = key.offset; 1225a5eeb3d1SFilipe Manana extent_end = btrfs_file_extent_end(path); 12267ffbb598SFilipe Manana em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); 122740e7efe0SQu Wenruo em->generation = btrfs_file_extent_generation(leaf, fi); 12287ffbb598SFilipe Manana if (type == BTRFS_FILE_EXTENT_REG || 12297ffbb598SFilipe Manana type == BTRFS_FILE_EXTENT_PREALLOC) { 12307ffbb598SFilipe Manana em->start = extent_start; 12317ffbb598SFilipe Manana em->len = extent_end - extent_start; 12327ffbb598SFilipe Manana em->orig_start = extent_start - 12337ffbb598SFilipe Manana btrfs_file_extent_offset(leaf, fi); 12347ffbb598SFilipe Manana em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); 12357ffbb598SFilipe Manana bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 12367ffbb598SFilipe Manana if (bytenr == 0) { 12377ffbb598SFilipe Manana em->block_start = EXTENT_MAP_HOLE; 12387ffbb598SFilipe Manana return; 12397ffbb598SFilipe Manana } 12407ffbb598SFilipe Manana if (compress_type != BTRFS_COMPRESS_NONE) { 12417ffbb598SFilipe Manana set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 12427ffbb598SFilipe Manana em->compress_type = compress_type; 12437ffbb598SFilipe Manana em->block_start = bytenr; 12447ffbb598SFilipe Manana em->block_len = em->orig_block_len; 12457ffbb598SFilipe Manana } else { 12467ffbb598SFilipe Manana bytenr += btrfs_file_extent_offset(leaf, fi); 12477ffbb598SFilipe Manana em->block_start = bytenr; 12487ffbb598SFilipe Manana em->block_len = em->len; 12497ffbb598SFilipe Manana if (type == BTRFS_FILE_EXTENT_PREALLOC) 12507ffbb598SFilipe Manana set_bit(EXTENT_FLAG_PREALLOC, &em->flags); 12517ffbb598SFilipe Manana } 12527ffbb598SFilipe Manana } else if (type == BTRFS_FILE_EXTENT_INLINE) { 12537ffbb598SFilipe Manana em->block_start = EXTENT_MAP_INLINE; 12547ffbb598SFilipe Manana em->start = extent_start; 12557ffbb598SFilipe Manana em->len = extent_end - extent_start; 12567ffbb598SFilipe Manana /* 12577ffbb598SFilipe Manana * Initialize orig_start and block_len with the same values 12587ffbb598SFilipe Manana * as in inode.c:btrfs_get_extent(). 12597ffbb598SFilipe Manana */ 12607ffbb598SFilipe Manana em->orig_start = EXTENT_MAP_HOLE; 12617ffbb598SFilipe Manana em->block_len = (u64)-1; 12627ffbb598SFilipe Manana if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) { 12637ffbb598SFilipe Manana set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 12647ffbb598SFilipe Manana em->compress_type = compress_type; 12657ffbb598SFilipe Manana } 12667ffbb598SFilipe Manana } else { 12670b246afaSJeff Mahoney btrfs_err(fs_info, 12689cdc5124SNikolay Borisov "unknown file extent item type %d, inode %llu, offset %llu, " 12699cdc5124SNikolay Borisov "root %llu", type, btrfs_ino(inode), extent_start, 12707ffbb598SFilipe Manana root->root_key.objectid); 12717ffbb598SFilipe Manana } 12727ffbb598SFilipe Manana } 1273a5eeb3d1SFilipe Manana 1274a5eeb3d1SFilipe Manana /* 1275a5eeb3d1SFilipe Manana * Returns the end offset (non inclusive) of the file extent item the given path 1276a5eeb3d1SFilipe Manana * points to. If it points to an inline extent, the returned offset is rounded 1277a5eeb3d1SFilipe Manana * up to the sector size. 1278a5eeb3d1SFilipe Manana */ 1279a5eeb3d1SFilipe Manana u64 btrfs_file_extent_end(const struct btrfs_path *path) 1280a5eeb3d1SFilipe Manana { 1281a5eeb3d1SFilipe Manana const struct extent_buffer *leaf = path->nodes[0]; 1282a5eeb3d1SFilipe Manana const int slot = path->slots[0]; 1283a5eeb3d1SFilipe Manana struct btrfs_file_extent_item *fi; 1284a5eeb3d1SFilipe Manana struct btrfs_key key; 1285a5eeb3d1SFilipe Manana u64 end; 1286a5eeb3d1SFilipe Manana 1287a5eeb3d1SFilipe Manana btrfs_item_key_to_cpu(leaf, &key, slot); 1288a5eeb3d1SFilipe Manana ASSERT(key.type == BTRFS_EXTENT_DATA_KEY); 1289a5eeb3d1SFilipe Manana fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 1290a5eeb3d1SFilipe Manana 1291a5eeb3d1SFilipe Manana if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) { 1292a5eeb3d1SFilipe Manana end = btrfs_file_extent_ram_bytes(leaf, fi); 1293a5eeb3d1SFilipe Manana end = ALIGN(key.offset + end, leaf->fs_info->sectorsize); 1294a5eeb3d1SFilipe Manana } else { 1295a5eeb3d1SFilipe Manana end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); 1296a5eeb3d1SFilipe Manana } 1297a5eeb3d1SFilipe Manana 1298a5eeb3d1SFilipe Manana return end; 1299a5eeb3d1SFilipe Manana } 1300