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> 129b569ea0SJosef 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" 17103c1972SChristoph Hellwig #include "bio.h" 181de037a4SChris Mason #include "print-tree.h" 19ebb8765bSAnand Jain #include "compression.h" 20c7f13d42SJosef Bacik #include "fs.h" 2107e81dc9SJosef Bacik #include "accessors.h" 227c8ede16SJosef Bacik #include "file-item.h" 237f0add25SJosef Bacik #include "super.h" 241e1d2701SChris Mason 2542049bf6SChris Mason #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \ 2642049bf6SChris Mason sizeof(struct btrfs_item) * 2) / \ 2742049bf6SChris Mason size) - 1)) 2807d400a6SYan Zheng 29221b8318SZach Brown #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \ 3009cbfeafSKirill A. Shutemov PAGE_SIZE)) 317ca4be45SChris Mason 3243dd529aSDavid Sterba /* 3343dd529aSDavid Sterba * Set inode's size according to filesystem options. 34ca4207aeSNikolay Borisov * 35ca4207aeSNikolay Borisov * @inode: inode we want to update the disk_i_size for 36ca4207aeSNikolay Borisov * @new_i_size: i_size we want to set to, 0 if we use i_size 3741a2ee75SJosef Bacik * 3841a2ee75SJosef Bacik * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read() 3941a2ee75SJosef Bacik * returns as it is perfectly fine with a file that has holes without hole file 4041a2ee75SJosef Bacik * extent items. 4141a2ee75SJosef Bacik * 4241a2ee75SJosef Bacik * However without NO_HOLES we need to only return the area that is contiguous 4341a2ee75SJosef Bacik * from the 0 offset of the file. Otherwise we could end up adjust i_size up 4441a2ee75SJosef Bacik * to an extent that has a gap in between. 4541a2ee75SJosef Bacik * 4641a2ee75SJosef Bacik * Finally new_i_size should only be set in the case of truncate where we're not 4741a2ee75SJosef Bacik * ready to use i_size_read() as the limiter yet. 4841a2ee75SJosef Bacik */ 4976aea537SNikolay Borisov void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size) 5041a2ee75SJosef Bacik { 5176aea537SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 5241a2ee75SJosef Bacik u64 start, end, i_size; 5341a2ee75SJosef Bacik int ret; 5441a2ee75SJosef Bacik 5576aea537SNikolay Borisov i_size = new_i_size ?: i_size_read(&inode->vfs_inode); 5641a2ee75SJosef Bacik if (btrfs_fs_incompat(fs_info, NO_HOLES)) { 5776aea537SNikolay Borisov inode->disk_i_size = i_size; 5841a2ee75SJosef Bacik return; 5941a2ee75SJosef Bacik } 6041a2ee75SJosef Bacik 6176aea537SNikolay Borisov spin_lock(&inode->lock); 6276aea537SNikolay Borisov ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start, 6376aea537SNikolay Borisov &end, EXTENT_DIRTY); 6441a2ee75SJosef Bacik if (!ret && start == 0) 6541a2ee75SJosef Bacik i_size = min(i_size, end + 1); 6641a2ee75SJosef Bacik else 6741a2ee75SJosef Bacik i_size = 0; 6876aea537SNikolay Borisov inode->disk_i_size = i_size; 6976aea537SNikolay Borisov spin_unlock(&inode->lock); 7041a2ee75SJosef Bacik } 7141a2ee75SJosef Bacik 7243dd529aSDavid Sterba /* 7343dd529aSDavid Sterba * Mark range within a file as having a new extent inserted. 74ca4207aeSNikolay Borisov * 75ca4207aeSNikolay Borisov * @inode: inode being modified 76ca4207aeSNikolay Borisov * @start: start file offset of the file extent we've inserted 77ca4207aeSNikolay Borisov * @len: logical length of the file extent item 7841a2ee75SJosef Bacik * 7941a2ee75SJosef Bacik * Call when we are inserting a new file extent where there was none before. 8041a2ee75SJosef Bacik * Does not need to call this in the case where we're replacing an existing file 8141a2ee75SJosef Bacik * extent, however if not sure it's fine to call this multiple times. 8241a2ee75SJosef Bacik * 8341a2ee75SJosef Bacik * The start and len must match the file extent item, so thus must be sectorsize 8441a2ee75SJosef Bacik * aligned. 8541a2ee75SJosef Bacik */ 8641a2ee75SJosef Bacik int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start, 8741a2ee75SJosef Bacik u64 len) 8841a2ee75SJosef Bacik { 8941a2ee75SJosef Bacik if (len == 0) 9041a2ee75SJosef Bacik return 0; 9141a2ee75SJosef Bacik 9241a2ee75SJosef Bacik ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize)); 9341a2ee75SJosef Bacik 9441a2ee75SJosef Bacik if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES)) 9541a2ee75SJosef Bacik return 0; 9641a2ee75SJosef Bacik return set_extent_bits(&inode->file_extent_tree, start, start + len - 1, 9741a2ee75SJosef Bacik EXTENT_DIRTY); 9841a2ee75SJosef Bacik } 9941a2ee75SJosef Bacik 10043dd529aSDavid Sterba /* 10143dd529aSDavid Sterba * Mark an inode range as not having a backing extent. 102ca4207aeSNikolay Borisov * 103ca4207aeSNikolay Borisov * @inode: inode being modified 104ca4207aeSNikolay Borisov * @start: start file offset of the file extent we've inserted 105ca4207aeSNikolay Borisov * @len: logical length of the file extent item 10641a2ee75SJosef Bacik * 10741a2ee75SJosef Bacik * Called when we drop a file extent, for example when we truncate. Doesn't 10841a2ee75SJosef Bacik * need to be called for cases where we're replacing a file extent, like when 10941a2ee75SJosef Bacik * we've COWed a file extent. 11041a2ee75SJosef Bacik * 11141a2ee75SJosef Bacik * The start and len must match the file extent item, so thus must be sectorsize 11241a2ee75SJosef Bacik * aligned. 11341a2ee75SJosef Bacik */ 11441a2ee75SJosef Bacik int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start, 11541a2ee75SJosef Bacik u64 len) 11641a2ee75SJosef Bacik { 11741a2ee75SJosef Bacik if (len == 0) 11841a2ee75SJosef Bacik return 0; 11941a2ee75SJosef Bacik 12041a2ee75SJosef Bacik ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) || 12141a2ee75SJosef Bacik len == (u64)-1); 12241a2ee75SJosef Bacik 12341a2ee75SJosef Bacik if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES)) 12441a2ee75SJosef Bacik return 0; 12541a2ee75SJosef Bacik return clear_extent_bit(&inode->file_extent_tree, start, 126bd015294SJosef Bacik start + len - 1, EXTENT_DIRTY, NULL); 12741a2ee75SJosef Bacik } 12841a2ee75SJosef Bacik 129cb649e81SQu Wenruo static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes) 1301e25a2e3SJohannes Thumshirn { 131cb649e81SQu Wenruo ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize)); 1321e25a2e3SJohannes Thumshirn 133cb649e81SQu Wenruo return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size; 134cb649e81SQu Wenruo } 135cb649e81SQu Wenruo 136cb649e81SQu Wenruo static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size) 137cb649e81SQu Wenruo { 138cb649e81SQu Wenruo ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size)); 139cb649e81SQu Wenruo 140cb649e81SQu Wenruo return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits; 141cb649e81SQu Wenruo } 142cb649e81SQu Wenruo 143cb649e81SQu Wenruo static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info) 144cb649e81SQu Wenruo { 145cb649e81SQu Wenruo u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum), 146cb649e81SQu Wenruo fs_info->csum_size); 147cb649e81SQu Wenruo 148cb649e81SQu Wenruo return csum_size_to_bytes(fs_info, max_csum_size); 1491e25a2e3SJohannes Thumshirn } 15007d400a6SYan Zheng 1512b6433c7SJosef Bacik /* 1522b6433c7SJosef Bacik * Calculate the total size needed to allocate for an ordered sum structure 1532b6433c7SJosef Bacik * spanning @bytes in the file. 1542b6433c7SJosef Bacik */ 1552b6433c7SJosef Bacik static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes) 1562b6433c7SJosef Bacik { 157cb649e81SQu Wenruo return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes); 1582b6433c7SJosef Bacik } 1592b6433c7SJosef Bacik 160d1f68ba0SOmar Sandoval int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans, 161dee26a9fSChris Mason struct btrfs_root *root, 162d1f68ba0SOmar Sandoval u64 objectid, u64 pos, u64 num_bytes) 1639f5fae2fSChris Mason { 164dee26a9fSChris Mason int ret = 0; 165dee26a9fSChris Mason struct btrfs_file_extent_item *item; 166dee26a9fSChris Mason struct btrfs_key file_key; 1675caf2a00SChris Mason struct btrfs_path *path; 1685f39d397SChris Mason struct extent_buffer *leaf; 169dee26a9fSChris Mason 1705caf2a00SChris Mason path = btrfs_alloc_path(); 171db5b493aSTsutomu Itoh if (!path) 172db5b493aSTsutomu Itoh return -ENOMEM; 173dee26a9fSChris Mason file_key.objectid = objectid; 174b18c6685SChris Mason file_key.offset = pos; 175962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_DATA_KEY; 176dee26a9fSChris Mason 1775caf2a00SChris Mason ret = btrfs_insert_empty_item(trans, root, path, &file_key, 178dee26a9fSChris Mason sizeof(*item)); 17954aa1f4dSChris Mason if (ret < 0) 18054aa1f4dSChris Mason goto out; 18179787eaaSJeff Mahoney BUG_ON(ret); /* Can't happen */ 1825f39d397SChris Mason leaf = path->nodes[0]; 1835f39d397SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], 184dee26a9fSChris Mason struct btrfs_file_extent_item); 185d1f68ba0SOmar Sandoval btrfs_set_file_extent_disk_bytenr(leaf, item, 0); 186d1f68ba0SOmar Sandoval btrfs_set_file_extent_disk_num_bytes(leaf, item, 0); 187d1f68ba0SOmar Sandoval btrfs_set_file_extent_offset(leaf, item, 0); 188db94535dSChris Mason btrfs_set_file_extent_num_bytes(leaf, item, num_bytes); 189d1f68ba0SOmar Sandoval btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes); 1905f39d397SChris Mason btrfs_set_file_extent_generation(leaf, item, trans->transid); 1915f39d397SChris Mason btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); 192d1f68ba0SOmar Sandoval btrfs_set_file_extent_compression(leaf, item, 0); 193d1f68ba0SOmar Sandoval btrfs_set_file_extent_encryption(leaf, item, 0); 194d1f68ba0SOmar Sandoval btrfs_set_file_extent_other_encoding(leaf, item, 0); 195c8b97818SChris Mason 1965f39d397SChris Mason btrfs_mark_buffer_dirty(leaf); 19754aa1f4dSChris Mason out: 1985caf2a00SChris Mason btrfs_free_path(path); 19954aa1f4dSChris Mason return ret; 2009f5fae2fSChris Mason } 201dee26a9fSChris Mason 20248a3b636SEric Sandeen static struct btrfs_csum_item * 20348a3b636SEric Sandeen btrfs_lookup_csum(struct btrfs_trans_handle *trans, 204b18c6685SChris Mason struct btrfs_root *root, 2056567e837SChris Mason struct btrfs_path *path, 206d20f7043SChris Mason u64 bytenr, int cow) 2076567e837SChris Mason { 2080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2096567e837SChris Mason int ret; 2106567e837SChris Mason struct btrfs_key file_key; 2116567e837SChris Mason struct btrfs_key found_key; 2126567e837SChris Mason struct btrfs_csum_item *item; 2135f39d397SChris Mason struct extent_buffer *leaf; 2146567e837SChris Mason u64 csum_offset = 0; 215223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 216a429e513SChris Mason int csums_in_item; 2176567e837SChris Mason 218d20f7043SChris Mason file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 219d20f7043SChris Mason file_key.offset = bytenr; 220962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_CSUM_KEY; 221b18c6685SChris Mason ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow); 2226567e837SChris Mason if (ret < 0) 2236567e837SChris Mason goto fail; 2245f39d397SChris Mason leaf = path->nodes[0]; 2256567e837SChris Mason if (ret > 0) { 2266567e837SChris Mason ret = 1; 22770b2befdSChris Mason if (path->slots[0] == 0) 2286567e837SChris Mason goto fail; 2296567e837SChris Mason path->slots[0]--; 2305f39d397SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 231962a298fSDavid Sterba if (found_key.type != BTRFS_EXTENT_CSUM_KEY) 2326567e837SChris Mason goto fail; 233d20f7043SChris Mason 234d20f7043SChris Mason csum_offset = (bytenr - found_key.offset) >> 235265fdfa6SDavid Sterba fs_info->sectorsize_bits; 2363212fa14SJosef Bacik csums_in_item = btrfs_item_size(leaf, path->slots[0]); 237607d432dSJosef Bacik csums_in_item /= csum_size; 238a429e513SChris Mason 23982d130ffSMiao Xie if (csum_offset == csums_in_item) { 240a429e513SChris Mason ret = -EFBIG; 2416567e837SChris Mason goto fail; 24282d130ffSMiao Xie } else if (csum_offset > csums_in_item) { 24382d130ffSMiao Xie goto fail; 2446567e837SChris Mason } 2456567e837SChris Mason } 2466567e837SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item); 247509659cdSChris Mason item = (struct btrfs_csum_item *)((unsigned char *)item + 248607d432dSJosef Bacik csum_offset * csum_size); 2496567e837SChris Mason return item; 2506567e837SChris Mason fail: 2516567e837SChris Mason if (ret > 0) 252b18c6685SChris Mason ret = -ENOENT; 2536567e837SChris Mason return ERR_PTR(ret); 2546567e837SChris Mason } 2556567e837SChris Mason 256dee26a9fSChris Mason int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans, 257dee26a9fSChris Mason struct btrfs_root *root, 258dee26a9fSChris Mason struct btrfs_path *path, u64 objectid, 2599773a788SChris Mason u64 offset, int mod) 260dee26a9fSChris Mason { 261dee26a9fSChris Mason struct btrfs_key file_key; 262dee26a9fSChris Mason int ins_len = mod < 0 ? -1 : 0; 263dee26a9fSChris Mason int cow = mod != 0; 264dee26a9fSChris Mason 265dee26a9fSChris Mason file_key.objectid = objectid; 26670b2befdSChris Mason file_key.offset = offset; 267962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_DATA_KEY; 268f8ee80deSMarcos Paulo de Souza 269f8ee80deSMarcos Paulo de Souza return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow); 270dee26a9fSChris Mason } 271f254e52cSChris Mason 2726275193eSQu Wenruo /* 2736275193eSQu Wenruo * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and 27443dd529aSDavid Sterba * store the result to @dst. 2756275193eSQu Wenruo * 2766275193eSQu Wenruo * Return >0 for the number of sectors we found. 2776275193eSQu Wenruo * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum 2786275193eSQu Wenruo * for it. Caller may want to try next sector until one range is hit. 2796275193eSQu Wenruo * Return <0 for fatal error. 2806275193eSQu Wenruo */ 2816275193eSQu Wenruo static int search_csum_tree(struct btrfs_fs_info *fs_info, 2826275193eSQu Wenruo struct btrfs_path *path, u64 disk_bytenr, 2836275193eSQu Wenruo u64 len, u8 *dst) 2846275193eSQu Wenruo { 285fc28b25eSJosef Bacik struct btrfs_root *csum_root; 2866275193eSQu Wenruo struct btrfs_csum_item *item = NULL; 2876275193eSQu Wenruo struct btrfs_key key; 2886275193eSQu Wenruo const u32 sectorsize = fs_info->sectorsize; 2896275193eSQu Wenruo const u32 csum_size = fs_info->csum_size; 2906275193eSQu Wenruo u32 itemsize; 2916275193eSQu Wenruo int ret; 2926275193eSQu Wenruo u64 csum_start; 2936275193eSQu Wenruo u64 csum_len; 2946275193eSQu Wenruo 2956275193eSQu Wenruo ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) && 2966275193eSQu Wenruo IS_ALIGNED(len, sectorsize)); 2976275193eSQu Wenruo 2986275193eSQu Wenruo /* Check if the current csum item covers disk_bytenr */ 2996275193eSQu Wenruo if (path->nodes[0]) { 3006275193eSQu Wenruo item = btrfs_item_ptr(path->nodes[0], path->slots[0], 3016275193eSQu Wenruo struct btrfs_csum_item); 3026275193eSQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 3033212fa14SJosef Bacik itemsize = btrfs_item_size(path->nodes[0], path->slots[0]); 3046275193eSQu Wenruo 3056275193eSQu Wenruo csum_start = key.offset; 3066275193eSQu Wenruo csum_len = (itemsize / csum_size) * sectorsize; 3076275193eSQu Wenruo 3086275193eSQu Wenruo if (in_range(disk_bytenr, csum_start, csum_len)) 3096275193eSQu Wenruo goto found; 3106275193eSQu Wenruo } 3116275193eSQu Wenruo 3126275193eSQu Wenruo /* Current item doesn't contain the desired range, search again */ 3136275193eSQu Wenruo btrfs_release_path(path); 314fc28b25eSJosef Bacik csum_root = btrfs_csum_root(fs_info, disk_bytenr); 315fc28b25eSJosef Bacik item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0); 3166275193eSQu Wenruo if (IS_ERR(item)) { 3176275193eSQu Wenruo ret = PTR_ERR(item); 3186275193eSQu Wenruo goto out; 3196275193eSQu Wenruo } 3206275193eSQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 3213212fa14SJosef Bacik itemsize = btrfs_item_size(path->nodes[0], path->slots[0]); 3226275193eSQu Wenruo 3236275193eSQu Wenruo csum_start = key.offset; 3246275193eSQu Wenruo csum_len = (itemsize / csum_size) * sectorsize; 3256275193eSQu Wenruo ASSERT(in_range(disk_bytenr, csum_start, csum_len)); 3266275193eSQu Wenruo 3276275193eSQu Wenruo found: 3286275193eSQu Wenruo ret = (min(csum_start + csum_len, disk_bytenr + len) - 3296275193eSQu Wenruo disk_bytenr) >> fs_info->sectorsize_bits; 3306275193eSQu Wenruo read_extent_buffer(path->nodes[0], dst, (unsigned long)item, 3316275193eSQu Wenruo ret * csum_size); 3326275193eSQu Wenruo out: 33303ddb19dSJosef Bacik if (ret == -ENOENT || ret == -EFBIG) 3346275193eSQu Wenruo ret = 0; 3356275193eSQu Wenruo return ret; 3366275193eSQu Wenruo } 3376275193eSQu Wenruo 3386275193eSQu Wenruo /* 3396275193eSQu Wenruo * Lookup the checksum for the read bio in csum tree. 3409e46458aSQu Wenruo * 341e62958fcSOmar Sandoval * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise. 342e62958fcSOmar Sandoval */ 3434ae2edf1SChristoph Hellwig blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio) 34461b49440SChris Mason { 3454ae2edf1SChristoph Hellwig struct btrfs_inode *inode = bbio->inode; 3464ae2edf1SChristoph Hellwig struct btrfs_fs_info *fs_info = inode->root->fs_info; 3474ae2edf1SChristoph Hellwig struct bio *bio = &bbio->bio; 348facc8a22SMiao Xie struct btrfs_path *path; 3496275193eSQu Wenruo const u32 sectorsize = fs_info->sectorsize; 350223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 3516275193eSQu Wenruo u32 orig_len = bio->bi_iter.bi_size; 3526275193eSQu Wenruo u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT; 3536275193eSQu Wenruo const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits; 3541784b7d5SJosef Bacik blk_status_t ret = BLK_STS_OK; 355e2eb0248SChristoph Hellwig u32 bio_offset = 0; 35661b49440SChris Mason 3574ae2edf1SChristoph Hellwig if ((inode->flags & BTRFS_INODE_NODATASUM) || 358056c8311SJosef Bacik test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)) 359334c16d8SJosef Bacik return BLK_STS_OK; 360334c16d8SJosef Bacik 3619e46458aSQu Wenruo /* 3629e46458aSQu Wenruo * This function is only called for read bio. 3639e46458aSQu Wenruo * 3649e46458aSQu Wenruo * This means two things: 3659e46458aSQu Wenruo * - All our csums should only be in csum tree 3669e46458aSQu Wenruo * No ordered extents csums, as ordered extents are only for write 3679e46458aSQu Wenruo * path. 3686275193eSQu Wenruo * - No need to bother any other info from bvec 3696275193eSQu Wenruo * Since we're looking up csums, the only important info is the 3706275193eSQu Wenruo * disk_bytenr and the length, which can be extracted from bi_iter 3716275193eSQu Wenruo * directly. 3729e46458aSQu Wenruo */ 3739e46458aSQu Wenruo ASSERT(bio_op(bio) == REQ_OP_READ); 37461b49440SChris Mason path = btrfs_alloc_path(); 375c2db1073STsutomu Itoh if (!path) 3764e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 377facc8a22SMiao Xie 378facc8a22SMiao Xie if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) { 379c3a3b19bSQu Wenruo bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS); 380c3a3b19bSQu Wenruo if (!bbio->csum) { 381facc8a22SMiao Xie btrfs_free_path(path); 3824e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 383facc8a22SMiao Xie } 384facc8a22SMiao Xie } else { 385c3a3b19bSQu Wenruo bbio->csum = bbio->csum_inline; 386facc8a22SMiao Xie } 387facc8a22SMiao Xie 38835478d05SQu Wenruo /* 38935478d05SQu Wenruo * If requested number of sectors is larger than one leaf can contain, 39035478d05SQu Wenruo * kick the readahead for csum tree. 39135478d05SQu Wenruo */ 39235478d05SQu Wenruo if (nblocks > fs_info->csums_per_leaf) 393e4058b54SDavid Sterba path->reada = READA_FORWARD; 39461b49440SChris Mason 3952cf8572dSChris Mason /* 3962cf8572dSChris Mason * the free space stuff is only read when it hasn't been 3972cf8572dSChris Mason * updated in the current transaction. So, we can safely 3982cf8572dSChris Mason * read from the commit root and sidestep a nasty deadlock 3992cf8572dSChris Mason * between reading the free space cache and updating the csum tree. 4002cf8572dSChris Mason */ 4014ae2edf1SChristoph Hellwig if (btrfs_is_free_space_inode(inode)) { 4022cf8572dSChris Mason path->search_commit_root = 1; 403ddf23b3fSJosef Bacik path->skip_locking = 1; 404ddf23b3fSJosef Bacik } 4052cf8572dSChris Mason 406e2eb0248SChristoph Hellwig while (bio_offset < orig_len) { 407e2eb0248SChristoph Hellwig int count; 408e2eb0248SChristoph Hellwig u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset; 409e2eb0248SChristoph Hellwig u8 *csum_dst = bbio->csum + 410e2eb0248SChristoph Hellwig (bio_offset >> fs_info->sectorsize_bits) * csum_size; 4114989d277SChristoph Hellwig 4126275193eSQu Wenruo count = search_csum_tree(fs_info, path, cur_disk_bytenr, 413e2eb0248SChristoph Hellwig orig_len - bio_offset, csum_dst); 4141784b7d5SJosef Bacik if (count < 0) { 4151784b7d5SJosef Bacik ret = errno_to_blk_status(count); 4164ae2edf1SChristoph Hellwig if (bbio->csum != bbio->csum_inline) 4174ae2edf1SChristoph Hellwig kfree(bbio->csum); 4184ae2edf1SChristoph Hellwig bbio->csum = NULL; 4191784b7d5SJosef Bacik break; 4201784b7d5SJosef Bacik } 4211784b7d5SJosef Bacik 4226275193eSQu Wenruo /* 4231784b7d5SJosef Bacik * We didn't find a csum for this range. We need to make sure 4241784b7d5SJosef Bacik * we complain loudly about this, because we are not NODATASUM. 4251784b7d5SJosef Bacik * 4261784b7d5SJosef Bacik * However for the DATA_RELOC inode we could potentially be 4271784b7d5SJosef Bacik * relocating data extents for a NODATASUM inode, so the inode 4281784b7d5SJosef Bacik * itself won't be marked with NODATASUM, but the extent we're 4291784b7d5SJosef Bacik * copying is in fact NODATASUM. If we don't find a csum we 4301784b7d5SJosef Bacik * assume this is the case. 4316275193eSQu Wenruo */ 4321784b7d5SJosef Bacik if (count == 0) { 4336275193eSQu Wenruo memset(csum_dst, 0, csum_size); 434e4100d98SMiao Xie count = 1; 4356275193eSQu Wenruo 4364ae2edf1SChristoph Hellwig if (inode->root->root_key.objectid == 43717d217feSYan Zheng BTRFS_DATA_RELOC_TREE_OBJECTID) { 438e2eb0248SChristoph Hellwig u64 file_offset = bbio->file_offset + bio_offset; 4396275193eSQu Wenruo 44065886d2bSChristoph Hellwig set_extent_bits(&inode->io_tree, file_offset, 4416275193eSQu Wenruo file_offset + sectorsize - 1, 442ceeb0ae7SDavid Sterba EXTENT_NODATASUM); 44317d217feSYan Zheng } else { 4446275193eSQu Wenruo btrfs_warn_rl(fs_info, 4456275193eSQu Wenruo "csum hole found for disk bytenr range [%llu, %llu)", 4466275193eSQu Wenruo cur_disk_bytenr, cur_disk_bytenr + sectorsize); 44717d217feSYan Zheng } 4484989d277SChristoph Hellwig } 449e2eb0248SChristoph Hellwig bio_offset += count * sectorsize; 4504989d277SChristoph Hellwig } 4514989d277SChristoph Hellwig 45261b49440SChris Mason btrfs_free_path(path); 4531784b7d5SJosef Bacik return ret; 4544b46fce2SJosef Bacik } 4554b46fce2SJosef Bacik 45697e38239SQu Wenruo int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end, 45726ce9114SJosef Bacik struct list_head *list, int search_commit, 45826ce9114SJosef Bacik bool nowait) 45917d217feSYan Zheng { 4600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46117d217feSYan Zheng struct btrfs_key key; 46217d217feSYan Zheng struct btrfs_path *path; 46317d217feSYan Zheng struct extent_buffer *leaf; 46417d217feSYan Zheng struct btrfs_ordered_sum *sums; 46517d217feSYan Zheng struct btrfs_csum_item *item; 4660678b618SMark Fasheh LIST_HEAD(tmplist); 46717d217feSYan Zheng int ret; 46817d217feSYan Zheng 4690b246afaSJeff Mahoney ASSERT(IS_ALIGNED(start, fs_info->sectorsize) && 4700b246afaSJeff Mahoney IS_ALIGNED(end + 1, fs_info->sectorsize)); 4714277a9c3SJosef Bacik 47217d217feSYan Zheng path = btrfs_alloc_path(); 473d8926bb3SMark Fasheh if (!path) 474d8926bb3SMark Fasheh return -ENOMEM; 47517d217feSYan Zheng 47626ce9114SJosef Bacik path->nowait = nowait; 477a2de733cSArne Jansen if (search_commit) { 478a2de733cSArne Jansen path->skip_locking = 1; 479e4058b54SDavid Sterba path->reada = READA_FORWARD; 480a2de733cSArne Jansen path->search_commit_root = 1; 481a2de733cSArne Jansen } 482a2de733cSArne Jansen 48317d217feSYan Zheng key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 48417d217feSYan Zheng key.offset = start; 48517d217feSYan Zheng key.type = BTRFS_EXTENT_CSUM_KEY; 48617d217feSYan Zheng 48707d400a6SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 48817d217feSYan Zheng if (ret < 0) 48917d217feSYan Zheng goto fail; 49017d217feSYan Zheng if (ret > 0 && path->slots[0] > 0) { 49117d217feSYan Zheng leaf = path->nodes[0]; 49217d217feSYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); 493cb649e81SQu Wenruo 494cb649e81SQu Wenruo /* 495cb649e81SQu Wenruo * There are two cases we can hit here for the previous csum 496cb649e81SQu Wenruo * item: 497cb649e81SQu Wenruo * 498cb649e81SQu Wenruo * |<- search range ->| 499cb649e81SQu Wenruo * |<- csum item ->| 500cb649e81SQu Wenruo * 501cb649e81SQu Wenruo * Or 502cb649e81SQu Wenruo * |<- search range ->| 503cb649e81SQu Wenruo * |<- csum item ->| 504cb649e81SQu Wenruo * 505cb649e81SQu Wenruo * Check if the previous csum item covers the leading part of 506cb649e81SQu Wenruo * the search range. If so we have to start from previous csum 507cb649e81SQu Wenruo * item. 508cb649e81SQu Wenruo */ 50917d217feSYan Zheng if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID && 51017d217feSYan Zheng key.type == BTRFS_EXTENT_CSUM_KEY) { 511cb649e81SQu Wenruo if (bytes_to_csum_size(fs_info, start - key.offset) < 5123212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0] - 1)) 51317d217feSYan Zheng path->slots[0]--; 51417d217feSYan Zheng } 51517d217feSYan Zheng } 51617d217feSYan Zheng 51717d217feSYan Zheng while (start <= end) { 518cb649e81SQu Wenruo u64 csum_end; 519cb649e81SQu Wenruo 52017d217feSYan Zheng leaf = path->nodes[0]; 52117d217feSYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 52207d400a6SYan Zheng ret = btrfs_next_leaf(root, path); 52317d217feSYan Zheng if (ret < 0) 52417d217feSYan Zheng goto fail; 52517d217feSYan Zheng if (ret > 0) 52617d217feSYan Zheng break; 52717d217feSYan Zheng leaf = path->nodes[0]; 52817d217feSYan Zheng } 52917d217feSYan Zheng 53017d217feSYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 53117d217feSYan Zheng if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 532628c8282SZhi Yong Wu key.type != BTRFS_EXTENT_CSUM_KEY || 533628c8282SZhi Yong Wu key.offset > end) 53417d217feSYan Zheng break; 53517d217feSYan Zheng 53617d217feSYan Zheng if (key.offset > start) 53717d217feSYan Zheng start = key.offset; 53817d217feSYan Zheng 539cb649e81SQu Wenruo csum_end = key.offset + csum_size_to_bytes(fs_info, 540cb649e81SQu Wenruo btrfs_item_size(leaf, path->slots[0])); 54187b29b20SYan Zheng if (csum_end <= start) { 54287b29b20SYan Zheng path->slots[0]++; 54387b29b20SYan Zheng continue; 54487b29b20SYan Zheng } 54517d217feSYan Zheng 54607d400a6SYan Zheng csum_end = min(csum_end, end + 1); 54707d400a6SYan Zheng item = btrfs_item_ptr(path->nodes[0], path->slots[0], 54807d400a6SYan Zheng struct btrfs_csum_item); 54907d400a6SYan Zheng while (start < csum_end) { 550cb649e81SQu Wenruo unsigned long offset; 551cb649e81SQu Wenruo size_t size; 552cb649e81SQu Wenruo 55307d400a6SYan Zheng size = min_t(size_t, csum_end - start, 554cb649e81SQu Wenruo max_ordered_sum_bytes(fs_info)); 5550b246afaSJeff Mahoney sums = kzalloc(btrfs_ordered_sum_size(fs_info, size), 55607d400a6SYan Zheng GFP_NOFS); 5570678b618SMark Fasheh if (!sums) { 5580678b618SMark Fasheh ret = -ENOMEM; 5590678b618SMark Fasheh goto fail; 5600678b618SMark Fasheh } 56117d217feSYan Zheng 56217d217feSYan Zheng sums->bytenr = start; 563f51a4a18SMiao Xie sums->len = (int)size; 56417d217feSYan Zheng 565cb649e81SQu Wenruo offset = bytes_to_csum_size(fs_info, start - key.offset); 56617d217feSYan Zheng 56707d400a6SYan Zheng read_extent_buffer(path->nodes[0], 568f51a4a18SMiao Xie sums->sums, 569f51a4a18SMiao Xie ((unsigned long)item) + offset, 570cb649e81SQu Wenruo bytes_to_csum_size(fs_info, size)); 57117d217feSYan Zheng 572cb649e81SQu Wenruo start += size; 5730678b618SMark Fasheh list_add_tail(&sums->list, &tmplist); 57407d400a6SYan Zheng } 57517d217feSYan Zheng path->slots[0]++; 57617d217feSYan Zheng } 57717d217feSYan Zheng ret = 0; 57817d217feSYan Zheng fail: 5790678b618SMark Fasheh while (ret < 0 && !list_empty(&tmplist)) { 5806e5aafb2SChris Mason sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list); 5810678b618SMark Fasheh list_del(&sums->list); 5820678b618SMark Fasheh kfree(sums); 5830678b618SMark Fasheh } 5840678b618SMark Fasheh list_splice_tail(&tmplist, list); 5850678b618SMark Fasheh 58617d217feSYan Zheng btrfs_free_path(path); 58717d217feSYan Zheng return ret; 58817d217feSYan Zheng } 58917d217feSYan Zheng 59043dd529aSDavid Sterba /* 59197e38239SQu Wenruo * Do the same work as btrfs_lookup_csums_list(), the difference is in how 59297e38239SQu Wenruo * we return the result. 59397e38239SQu Wenruo * 59497e38239SQu Wenruo * This version will set the corresponding bits in @csum_bitmap to represent 59597e38239SQu Wenruo * that there is a csum found. 59697e38239SQu Wenruo * Each bit represents a sector. Thus caller should ensure @csum_buf passed 59797e38239SQu Wenruo * in is large enough to contain all csums. 59897e38239SQu Wenruo */ 59997e38239SQu Wenruo int btrfs_lookup_csums_bitmap(struct btrfs_root *root, u64 start, u64 end, 600*b9795475SQu Wenruo u8 *csum_buf, unsigned long *csum_bitmap, 601*b9795475SQu Wenruo bool search_commit) 60297e38239SQu Wenruo { 60397e38239SQu Wenruo struct btrfs_fs_info *fs_info = root->fs_info; 60497e38239SQu Wenruo struct btrfs_key key; 60597e38239SQu Wenruo struct btrfs_path *path; 60697e38239SQu Wenruo struct extent_buffer *leaf; 60797e38239SQu Wenruo struct btrfs_csum_item *item; 60897e38239SQu Wenruo const u64 orig_start = start; 60997e38239SQu Wenruo int ret; 61097e38239SQu Wenruo 61197e38239SQu Wenruo ASSERT(IS_ALIGNED(start, fs_info->sectorsize) && 61297e38239SQu Wenruo IS_ALIGNED(end + 1, fs_info->sectorsize)); 61397e38239SQu Wenruo 61497e38239SQu Wenruo path = btrfs_alloc_path(); 61597e38239SQu Wenruo if (!path) 61697e38239SQu Wenruo return -ENOMEM; 61797e38239SQu Wenruo 618*b9795475SQu Wenruo if (search_commit) { 619*b9795475SQu Wenruo path->skip_locking = 1; 620*b9795475SQu Wenruo path->reada = READA_FORWARD; 621*b9795475SQu Wenruo path->search_commit_root = 1; 622*b9795475SQu Wenruo } 623*b9795475SQu Wenruo 62497e38239SQu Wenruo key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 62597e38239SQu Wenruo key.type = BTRFS_EXTENT_CSUM_KEY; 62697e38239SQu Wenruo key.offset = start; 62797e38239SQu Wenruo 62897e38239SQu Wenruo ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 62997e38239SQu Wenruo if (ret < 0) 63097e38239SQu Wenruo goto fail; 63197e38239SQu Wenruo if (ret > 0 && path->slots[0] > 0) { 63297e38239SQu Wenruo leaf = path->nodes[0]; 63397e38239SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); 63497e38239SQu Wenruo 63597e38239SQu Wenruo /* 63697e38239SQu Wenruo * There are two cases we can hit here for the previous csum 63797e38239SQu Wenruo * item: 63897e38239SQu Wenruo * 63997e38239SQu Wenruo * |<- search range ->| 64097e38239SQu Wenruo * |<- csum item ->| 64197e38239SQu Wenruo * 64297e38239SQu Wenruo * Or 64397e38239SQu Wenruo * |<- search range ->| 64497e38239SQu Wenruo * |<- csum item ->| 64597e38239SQu Wenruo * 64697e38239SQu Wenruo * Check if the previous csum item covers the leading part of 64797e38239SQu Wenruo * the search range. If so we have to start from previous csum 64897e38239SQu Wenruo * item. 64997e38239SQu Wenruo */ 65097e38239SQu Wenruo if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID && 65197e38239SQu Wenruo key.type == BTRFS_EXTENT_CSUM_KEY) { 65297e38239SQu Wenruo if (bytes_to_csum_size(fs_info, start - key.offset) < 65397e38239SQu Wenruo btrfs_item_size(leaf, path->slots[0] - 1)) 65497e38239SQu Wenruo path->slots[0]--; 65597e38239SQu Wenruo } 65697e38239SQu Wenruo } 65797e38239SQu Wenruo 65897e38239SQu Wenruo while (start <= end) { 65997e38239SQu Wenruo u64 csum_end; 66097e38239SQu Wenruo 66197e38239SQu Wenruo leaf = path->nodes[0]; 66297e38239SQu Wenruo if (path->slots[0] >= btrfs_header_nritems(leaf)) { 66397e38239SQu Wenruo ret = btrfs_next_leaf(root, path); 66497e38239SQu Wenruo if (ret < 0) 66597e38239SQu Wenruo goto fail; 66697e38239SQu Wenruo if (ret > 0) 66797e38239SQu Wenruo break; 66897e38239SQu Wenruo leaf = path->nodes[0]; 66997e38239SQu Wenruo } 67097e38239SQu Wenruo 67197e38239SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 67297e38239SQu Wenruo if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 67397e38239SQu Wenruo key.type != BTRFS_EXTENT_CSUM_KEY || 67497e38239SQu Wenruo key.offset > end) 67597e38239SQu Wenruo break; 67697e38239SQu Wenruo 67797e38239SQu Wenruo if (key.offset > start) 67897e38239SQu Wenruo start = key.offset; 67997e38239SQu Wenruo 68097e38239SQu Wenruo csum_end = key.offset + csum_size_to_bytes(fs_info, 68197e38239SQu Wenruo btrfs_item_size(leaf, path->slots[0])); 68297e38239SQu Wenruo if (csum_end <= start) { 68397e38239SQu Wenruo path->slots[0]++; 68497e38239SQu Wenruo continue; 68597e38239SQu Wenruo } 68697e38239SQu Wenruo 68797e38239SQu Wenruo csum_end = min(csum_end, end + 1); 68897e38239SQu Wenruo item = btrfs_item_ptr(path->nodes[0], path->slots[0], 68997e38239SQu Wenruo struct btrfs_csum_item); 69097e38239SQu Wenruo while (start < csum_end) { 69197e38239SQu Wenruo unsigned long offset; 69297e38239SQu Wenruo size_t size; 69397e38239SQu Wenruo u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info, 69497e38239SQu Wenruo start - orig_start); 69597e38239SQu Wenruo 69697e38239SQu Wenruo size = min_t(size_t, csum_end - start, end + 1 - start); 69797e38239SQu Wenruo 69897e38239SQu Wenruo offset = bytes_to_csum_size(fs_info, start - key.offset); 69997e38239SQu Wenruo 70097e38239SQu Wenruo read_extent_buffer(path->nodes[0], csum_dest, 70197e38239SQu Wenruo ((unsigned long)item) + offset, 70297e38239SQu Wenruo bytes_to_csum_size(fs_info, size)); 70397e38239SQu Wenruo 70497e38239SQu Wenruo bitmap_set(csum_bitmap, 70597e38239SQu Wenruo (start - orig_start) >> fs_info->sectorsize_bits, 70697e38239SQu Wenruo size >> fs_info->sectorsize_bits); 70797e38239SQu Wenruo 70897e38239SQu Wenruo start += size; 70997e38239SQu Wenruo } 71097e38239SQu Wenruo path->slots[0]++; 71197e38239SQu Wenruo } 71297e38239SQu Wenruo ret = 0; 71397e38239SQu Wenruo fail: 71497e38239SQu Wenruo btrfs_free_path(path); 71597e38239SQu Wenruo return ret; 71697e38239SQu Wenruo } 71797e38239SQu Wenruo 71897e38239SQu Wenruo /* 71943dd529aSDavid Sterba * Calculate checksums of the data contained inside a bio. 72051d470aeSNikolay Borisov */ 721f8c44673SChristoph Hellwig blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio) 722e015640fSChris Mason { 723f8c44673SChristoph Hellwig struct btrfs_inode *inode = bbio->inode; 724c3504372SNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 725d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 726f8c44673SChristoph Hellwig struct bio *bio = &bbio->bio; 727f8c44673SChristoph Hellwig u64 offset = bbio->file_offset; 728e6dcd2dcSChris Mason struct btrfs_ordered_sum *sums; 7296cd7ce49SChristoph Hellwig struct btrfs_ordered_extent *ordered = NULL; 730e015640fSChris Mason char *data; 73117347cecSLiu Bo struct bvec_iter iter; 73217347cecSLiu Bo struct bio_vec bvec; 733f51a4a18SMiao Xie int index; 734e331f6b1SOmar Sandoval unsigned int blockcount; 7353edf7d33SChris Mason unsigned long total_bytes = 0; 7363edf7d33SChris Mason unsigned long this_sum_bytes = 0; 73717347cecSLiu Bo int i; 738a3d46aeaSNikolay Borisov unsigned nofs_flag; 739e015640fSChris Mason 740a3d46aeaSNikolay Borisov nofs_flag = memalloc_nofs_save(); 741a3d46aeaSNikolay Borisov sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size), 742a3d46aeaSNikolay Borisov GFP_KERNEL); 743a3d46aeaSNikolay Borisov memalloc_nofs_restore(nofs_flag); 744a3d46aeaSNikolay Borisov 745e015640fSChris Mason if (!sums) 7464e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 7473edf7d33SChris Mason 7484f024f37SKent Overstreet sums->len = bio->bi_iter.bi_size; 749e6dcd2dcSChris Mason INIT_LIST_HEAD(&sums->list); 750d20f7043SChris Mason 7511201b58bSDavid Sterba sums->bytenr = bio->bi_iter.bi_sector << 9; 752f51a4a18SMiao Xie index = 0; 753e015640fSChris Mason 754d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 755d5178578SJohannes Thumshirn 75617347cecSLiu Bo bio_for_each_segment(bvec, bio, iter) { 7576cd7ce49SChristoph Hellwig if (!ordered) { 7586cd7ce49SChristoph Hellwig ordered = btrfs_lookup_ordered_extent(inode, offset); 759bbc9a6ebSQu Wenruo /* 760bbc9a6ebSQu Wenruo * The bio range is not covered by any ordered extent, 761bbc9a6ebSQu Wenruo * must be a code logic error. 762bbc9a6ebSQu Wenruo */ 763bbc9a6ebSQu Wenruo if (unlikely(!ordered)) { 764bbc9a6ebSQu Wenruo WARN(1, KERN_WARNING 765bbc9a6ebSQu Wenruo "no ordered extent for root %llu ino %llu offset %llu\n", 766bbc9a6ebSQu Wenruo inode->root->root_key.objectid, 767bbc9a6ebSQu Wenruo btrfs_ino(inode), offset); 768bbc9a6ebSQu Wenruo kvfree(sums); 769bbc9a6ebSQu Wenruo return BLK_STS_IOERR; 770bbc9a6ebSQu Wenruo } 7716cd7ce49SChristoph Hellwig } 7726cd7ce49SChristoph Hellwig 773e331f6b1SOmar Sandoval blockcount = BTRFS_BYTES_TO_BLKS(fs_info, 77417347cecSLiu Bo bvec.bv_len + fs_info->sectorsize 775c40a3d38SChandan Rajendra - 1); 776c40a3d38SChandan Rajendra 777e331f6b1SOmar Sandoval for (i = 0; i < blockcount; i++) { 778f8c44673SChristoph Hellwig if (!(bio->bi_opf & REQ_BTRFS_ONE_ORDERED) && 779e331f6b1SOmar Sandoval !in_range(offset, ordered->file_offset, 780e331f6b1SOmar Sandoval ordered->num_bytes)) { 7813edf7d33SChris Mason unsigned long bytes_left; 782c40a3d38SChandan Rajendra 7833edf7d33SChris Mason sums->len = this_sum_bytes; 7843edf7d33SChris Mason this_sum_bytes = 0; 785f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 7863edf7d33SChris Mason btrfs_put_ordered_extent(ordered); 7873edf7d33SChris Mason 7884f024f37SKent Overstreet bytes_left = bio->bi_iter.bi_size - total_bytes; 7893edf7d33SChris Mason 790a3d46aeaSNikolay Borisov nofs_flag = memalloc_nofs_save(); 791a3d46aeaSNikolay Borisov sums = kvzalloc(btrfs_ordered_sum_size(fs_info, 792a3d46aeaSNikolay Borisov bytes_left), GFP_KERNEL); 793a3d46aeaSNikolay Borisov memalloc_nofs_restore(nofs_flag); 79479787eaaSJeff Mahoney BUG_ON(!sums); /* -ENOMEM */ 7953edf7d33SChris Mason sums->len = bytes_left; 796c40a3d38SChandan Rajendra ordered = btrfs_lookup_ordered_extent(inode, 797c40a3d38SChandan Rajendra offset); 798c40a3d38SChandan Rajendra ASSERT(ordered); /* Logic error */ 7991201b58bSDavid Sterba sums->bytenr = (bio->bi_iter.bi_sector << 9) 800c40a3d38SChandan Rajendra + total_bytes; 801f51a4a18SMiao Xie index = 0; 802c40a3d38SChandan Rajendra } 803c40a3d38SChandan Rajendra 8043dcfbcceSChristoph Hellwig data = bvec_kmap_local(&bvec); 8053dcfbcceSChristoph Hellwig crypto_shash_digest(shash, 8063dcfbcceSChristoph Hellwig data + (i * fs_info->sectorsize), 807fd08001fSEric Biggers fs_info->sectorsize, 808fd08001fSEric Biggers sums->sums + index); 8093dcfbcceSChristoph Hellwig kunmap_local(data); 810713cebfbSDavid Sterba index += fs_info->csum_size; 8110b246afaSJeff Mahoney offset += fs_info->sectorsize; 8120b246afaSJeff Mahoney this_sum_bytes += fs_info->sectorsize; 8130b246afaSJeff Mahoney total_bytes += fs_info->sectorsize; 814c40a3d38SChandan Rajendra } 815c40a3d38SChandan Rajendra 816e015640fSChris Mason } 817ed98b56aSChris Mason this_sum_bytes = 0; 818f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 8193edf7d33SChris Mason btrfs_put_ordered_extent(ordered); 820e015640fSChris Mason return 0; 821e015640fSChris Mason } 822e015640fSChris Mason 823459931ecSChris Mason /* 82443dd529aSDavid Sterba * Remove one checksum overlapping a range. 825459931ecSChris Mason * 82643dd529aSDavid Sterba * This expects the key to describe the csum pointed to by the path, and it 82743dd529aSDavid Sterba * expects the csum to overlap the range [bytenr, len] 828459931ecSChris Mason * 82943dd529aSDavid Sterba * The csum should not be entirely contained in the range and the range should 83043dd529aSDavid Sterba * not be entirely contained in the csum. 83143dd529aSDavid Sterba * 83243dd529aSDavid Sterba * This calls btrfs_truncate_item with the correct args based on the overlap, 83343dd529aSDavid Sterba * and fixes up the key as required. 834459931ecSChris Mason */ 8352ff7e61eSJeff Mahoney static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info, 836459931ecSChris Mason struct btrfs_path *path, 837459931ecSChris Mason struct btrfs_key *key, 838459931ecSChris Mason u64 bytenr, u64 len) 839459931ecSChris Mason { 840459931ecSChris Mason struct extent_buffer *leaf; 841223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 842459931ecSChris Mason u64 csum_end; 843459931ecSChris Mason u64 end_byte = bytenr + len; 844265fdfa6SDavid Sterba u32 blocksize_bits = fs_info->sectorsize_bits; 845459931ecSChris Mason 846459931ecSChris Mason leaf = path->nodes[0]; 8473212fa14SJosef Bacik csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size; 848265fdfa6SDavid Sterba csum_end <<= blocksize_bits; 849459931ecSChris Mason csum_end += key->offset; 850459931ecSChris Mason 851459931ecSChris Mason if (key->offset < bytenr && csum_end <= end_byte) { 852459931ecSChris Mason /* 853459931ecSChris Mason * [ bytenr - len ] 854459931ecSChris Mason * [ ] 855459931ecSChris Mason * [csum ] 856459931ecSChris Mason * A simple truncate off the end of the item 857459931ecSChris Mason */ 858459931ecSChris Mason u32 new_size = (bytenr - key->offset) >> blocksize_bits; 859459931ecSChris Mason new_size *= csum_size; 86078ac4f9eSDavid Sterba btrfs_truncate_item(path, new_size, 1); 861459931ecSChris Mason } else if (key->offset >= bytenr && csum_end > end_byte && 862459931ecSChris Mason end_byte > key->offset) { 863459931ecSChris Mason /* 864459931ecSChris Mason * [ bytenr - len ] 865459931ecSChris Mason * [ ] 866459931ecSChris Mason * [csum ] 867459931ecSChris Mason * we need to truncate from the beginning of the csum 868459931ecSChris Mason */ 869459931ecSChris Mason u32 new_size = (csum_end - end_byte) >> blocksize_bits; 870459931ecSChris Mason new_size *= csum_size; 871459931ecSChris Mason 87278ac4f9eSDavid Sterba btrfs_truncate_item(path, new_size, 0); 873459931ecSChris Mason 874459931ecSChris Mason key->offset = end_byte; 8750b246afaSJeff Mahoney btrfs_set_item_key_safe(fs_info, path, key); 876459931ecSChris Mason } else { 877459931ecSChris Mason BUG(); 878459931ecSChris Mason } 879459931ecSChris Mason } 880459931ecSChris Mason 881459931ecSChris Mason /* 88243dd529aSDavid Sterba * Delete the csum items from the csum tree for a given range of bytes. 883459931ecSChris Mason */ 884459931ecSChris Mason int btrfs_del_csums(struct btrfs_trans_handle *trans, 88540e046acSFilipe Manana struct btrfs_root *root, u64 bytenr, u64 len) 886459931ecSChris Mason { 88740e046acSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 888459931ecSChris Mason struct btrfs_path *path; 889459931ecSChris Mason struct btrfs_key key; 890459931ecSChris Mason u64 end_byte = bytenr + len; 891459931ecSChris Mason u64 csum_end; 892459931ecSChris Mason struct extent_buffer *leaf; 893b86652beSJosef Bacik int ret = 0; 894223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 895265fdfa6SDavid Sterba u32 blocksize_bits = fs_info->sectorsize_bits; 896459931ecSChris Mason 89784d2d6c7SJosef Bacik ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID || 89840e046acSFilipe Manana root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 89940e046acSFilipe Manana 900459931ecSChris Mason path = btrfs_alloc_path(); 9012a29edc6Sliubo if (!path) 9022a29edc6Sliubo return -ENOMEM; 903459931ecSChris Mason 904459931ecSChris Mason while (1) { 905459931ecSChris Mason key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 906459931ecSChris Mason key.offset = end_byte - 1; 907459931ecSChris Mason key.type = BTRFS_EXTENT_CSUM_KEY; 908459931ecSChris Mason 909459931ecSChris Mason ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 910459931ecSChris Mason if (ret > 0) { 911b86652beSJosef Bacik ret = 0; 912459931ecSChris Mason if (path->slots[0] == 0) 91365a246c5STsutomu Itoh break; 914459931ecSChris Mason path->slots[0]--; 915ad0397a7SJosef Bacik } else if (ret < 0) { 91665a246c5STsutomu Itoh break; 917459931ecSChris Mason } 918ad0397a7SJosef Bacik 919459931ecSChris Mason leaf = path->nodes[0]; 920459931ecSChris Mason btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 921459931ecSChris Mason 922459931ecSChris Mason if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 923459931ecSChris Mason key.type != BTRFS_EXTENT_CSUM_KEY) { 924459931ecSChris Mason break; 925459931ecSChris Mason } 926459931ecSChris Mason 927459931ecSChris Mason if (key.offset >= end_byte) 928459931ecSChris Mason break; 929459931ecSChris Mason 9303212fa14SJosef Bacik csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size; 931459931ecSChris Mason csum_end <<= blocksize_bits; 932459931ecSChris Mason csum_end += key.offset; 933459931ecSChris Mason 934459931ecSChris Mason /* this csum ends before we start, we're done */ 935459931ecSChris Mason if (csum_end <= bytenr) 936459931ecSChris Mason break; 937459931ecSChris Mason 938459931ecSChris Mason /* delete the entire item, it is inside our range */ 939459931ecSChris Mason if (key.offset >= bytenr && csum_end <= end_byte) { 9406f546216SFilipe Manana int del_nr = 1; 9416f546216SFilipe Manana 9426f546216SFilipe Manana /* 9436f546216SFilipe Manana * Check how many csum items preceding this one in this 9446f546216SFilipe Manana * leaf correspond to our range and then delete them all 9456f546216SFilipe Manana * at once. 9466f546216SFilipe Manana */ 9476f546216SFilipe Manana if (key.offset > bytenr && path->slots[0] > 0) { 9486f546216SFilipe Manana int slot = path->slots[0] - 1; 9496f546216SFilipe Manana 9506f546216SFilipe Manana while (slot >= 0) { 9516f546216SFilipe Manana struct btrfs_key pk; 9526f546216SFilipe Manana 9536f546216SFilipe Manana btrfs_item_key_to_cpu(leaf, &pk, slot); 9546f546216SFilipe Manana if (pk.offset < bytenr || 9556f546216SFilipe Manana pk.type != BTRFS_EXTENT_CSUM_KEY || 9566f546216SFilipe Manana pk.objectid != 9576f546216SFilipe Manana BTRFS_EXTENT_CSUM_OBJECTID) 9586f546216SFilipe Manana break; 9596f546216SFilipe Manana path->slots[0] = slot; 9606f546216SFilipe Manana del_nr++; 9616f546216SFilipe Manana key.offset = pk.offset; 9626f546216SFilipe Manana slot--; 9636f546216SFilipe Manana } 9646f546216SFilipe Manana } 9656f546216SFilipe Manana ret = btrfs_del_items(trans, root, path, 9666f546216SFilipe Manana path->slots[0], del_nr); 96765a246c5STsutomu Itoh if (ret) 968b86652beSJosef Bacik break; 969dcbdd4dcSChris Mason if (key.offset == bytenr) 970dcbdd4dcSChris Mason break; 971459931ecSChris Mason } else if (key.offset < bytenr && csum_end > end_byte) { 972459931ecSChris Mason unsigned long offset; 973459931ecSChris Mason unsigned long shift_len; 974459931ecSChris Mason unsigned long item_offset; 975459931ecSChris Mason /* 976459931ecSChris Mason * [ bytenr - len ] 977459931ecSChris Mason * [csum ] 978459931ecSChris Mason * 979459931ecSChris Mason * Our bytes are in the middle of the csum, 980459931ecSChris Mason * we need to split this item and insert a new one. 981459931ecSChris Mason * 982459931ecSChris Mason * But we can't drop the path because the 983459931ecSChris Mason * csum could change, get removed, extended etc. 984459931ecSChris Mason * 985459931ecSChris Mason * The trick here is the max size of a csum item leaves 986459931ecSChris Mason * enough room in the tree block for a single 987459931ecSChris Mason * item header. So, we split the item in place, 988459931ecSChris Mason * adding a new header pointing to the existing 989459931ecSChris Mason * bytes. Then we loop around again and we have 990459931ecSChris Mason * a nicely formed csum item that we can neatly 991459931ecSChris Mason * truncate. 992459931ecSChris Mason */ 993459931ecSChris Mason offset = (bytenr - key.offset) >> blocksize_bits; 994459931ecSChris Mason offset *= csum_size; 995459931ecSChris Mason 996459931ecSChris Mason shift_len = (len >> blocksize_bits) * csum_size; 997459931ecSChris Mason 998459931ecSChris Mason item_offset = btrfs_item_ptr_offset(leaf, 999459931ecSChris Mason path->slots[0]); 1000459931ecSChris Mason 1001b159fa28SDavid Sterba memzero_extent_buffer(leaf, item_offset + offset, 1002459931ecSChris Mason shift_len); 1003459931ecSChris Mason key.offset = bytenr; 1004459931ecSChris Mason 1005459931ecSChris Mason /* 1006459931ecSChris Mason * btrfs_split_item returns -EAGAIN when the 1007459931ecSChris Mason * item changed size or key 1008459931ecSChris Mason */ 1009459931ecSChris Mason ret = btrfs_split_item(trans, root, path, &key, offset); 101079787eaaSJeff Mahoney if (ret && ret != -EAGAIN) { 101166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1012b86652beSJosef Bacik break; 101379787eaaSJeff Mahoney } 1014b86652beSJosef Bacik ret = 0; 1015459931ecSChris Mason 1016459931ecSChris Mason key.offset = end_byte - 1; 1017459931ecSChris Mason } else { 10182ff7e61eSJeff Mahoney truncate_one_csum(fs_info, path, &key, bytenr, len); 1019dcbdd4dcSChris Mason if (key.offset < bytenr) 1020dcbdd4dcSChris Mason break; 1021459931ecSChris Mason } 1022b3b4aa74SDavid Sterba btrfs_release_path(path); 1023459931ecSChris Mason } 1024459931ecSChris Mason btrfs_free_path(path); 102565a246c5STsutomu Itoh return ret; 1026459931ecSChris Mason } 1027459931ecSChris Mason 1028ea7036deSFilipe Manana static int find_next_csum_offset(struct btrfs_root *root, 1029ea7036deSFilipe Manana struct btrfs_path *path, 1030ea7036deSFilipe Manana u64 *next_offset) 1031ea7036deSFilipe Manana { 1032ea7036deSFilipe Manana const u32 nritems = btrfs_header_nritems(path->nodes[0]); 1033ea7036deSFilipe Manana struct btrfs_key found_key; 1034ea7036deSFilipe Manana int slot = path->slots[0] + 1; 1035ea7036deSFilipe Manana int ret; 1036ea7036deSFilipe Manana 1037ea7036deSFilipe Manana if (nritems == 0 || slot >= nritems) { 1038ea7036deSFilipe Manana ret = btrfs_next_leaf(root, path); 1039ea7036deSFilipe Manana if (ret < 0) { 1040ea7036deSFilipe Manana return ret; 1041ea7036deSFilipe Manana } else if (ret > 0) { 1042ea7036deSFilipe Manana *next_offset = (u64)-1; 1043ea7036deSFilipe Manana return 0; 1044ea7036deSFilipe Manana } 1045ea7036deSFilipe Manana slot = path->slots[0]; 1046ea7036deSFilipe Manana } 1047ea7036deSFilipe Manana 1048ea7036deSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot); 1049ea7036deSFilipe Manana 1050ea7036deSFilipe Manana if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 1051ea7036deSFilipe Manana found_key.type != BTRFS_EXTENT_CSUM_KEY) 1052ea7036deSFilipe Manana *next_offset = (u64)-1; 1053ea7036deSFilipe Manana else 1054ea7036deSFilipe Manana *next_offset = found_key.offset; 1055ea7036deSFilipe Manana 1056ea7036deSFilipe Manana return 0; 1057ea7036deSFilipe Manana } 1058ea7036deSFilipe Manana 1059065631f6SChris Mason int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans, 1060d20f7043SChris Mason struct btrfs_root *root, 1061e6dcd2dcSChris Mason struct btrfs_ordered_sum *sums) 1062f254e52cSChris Mason { 10630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1064f254e52cSChris Mason struct btrfs_key file_key; 10656567e837SChris Mason struct btrfs_key found_key; 10665caf2a00SChris Mason struct btrfs_path *path; 1067f254e52cSChris Mason struct btrfs_csum_item *item; 1068065631f6SChris Mason struct btrfs_csum_item *item_end; 1069ff79f819SChris Mason struct extent_buffer *leaf = NULL; 1070f51a4a18SMiao Xie u64 next_offset; 1071f51a4a18SMiao Xie u64 total_bytes = 0; 10726567e837SChris Mason u64 csum_offset; 1073f51a4a18SMiao Xie u64 bytenr; 1074f578d4bdSChris Mason u32 ins_size; 1075f51a4a18SMiao Xie int index = 0; 1076f51a4a18SMiao Xie int found_next; 1077f51a4a18SMiao Xie int ret; 1078223486c2SDavid Sterba const u32 csum_size = fs_info->csum_size; 10796e92f5e6SChris Mason 10805caf2a00SChris Mason path = btrfs_alloc_path(); 1081d8926bb3SMark Fasheh if (!path) 1082d8926bb3SMark Fasheh return -ENOMEM; 1083065631f6SChris Mason again: 1084065631f6SChris Mason next_offset = (u64)-1; 1085065631f6SChris Mason found_next = 0; 1086f51a4a18SMiao Xie bytenr = sums->bytenr + total_bytes; 1087d20f7043SChris Mason file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 1088f51a4a18SMiao Xie file_key.offset = bytenr; 1089962a298fSDavid Sterba file_key.type = BTRFS_EXTENT_CSUM_KEY; 1090a429e513SChris Mason 1091f51a4a18SMiao Xie item = btrfs_lookup_csum(trans, root, path, bytenr, 1); 1092ff79f819SChris Mason if (!IS_ERR(item)) { 1093639cb586SChris Mason ret = 0; 1094f51a4a18SMiao Xie leaf = path->nodes[0]; 1095f51a4a18SMiao Xie item_end = btrfs_item_ptr(leaf, path->slots[0], 1096f51a4a18SMiao Xie struct btrfs_csum_item); 1097f51a4a18SMiao Xie item_end = (struct btrfs_csum_item *)((char *)item_end + 10983212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0])); 1099a429e513SChris Mason goto found; 1100ff79f819SChris Mason } 1101a429e513SChris Mason ret = PTR_ERR(item); 11024a500fd1SYan, Zheng if (ret != -EFBIG && ret != -ENOENT) 1103918cdf44SFilipe Manana goto out; 11044a500fd1SYan, Zheng 1105a429e513SChris Mason if (ret == -EFBIG) { 1106a429e513SChris Mason u32 item_size; 1107a429e513SChris Mason /* we found one, but it isn't big enough yet */ 11085f39d397SChris Mason leaf = path->nodes[0]; 11093212fa14SJosef Bacik item_size = btrfs_item_size(leaf, path->slots[0]); 1110607d432dSJosef Bacik if ((item_size / csum_size) >= 11110b246afaSJeff Mahoney MAX_CSUM_ITEMS(fs_info, csum_size)) { 1112a429e513SChris Mason /* already at max size, make a new one */ 1113a429e513SChris Mason goto insert; 1114a429e513SChris Mason } 1115a429e513SChris Mason } else { 1116ea7036deSFilipe Manana /* We didn't find a csum item, insert one. */ 1117ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1118ea7036deSFilipe Manana if (ret < 0) 11197e4a3f7eSFilipe Manana goto out; 1120f578d4bdSChris Mason found_next = 1; 1121a429e513SChris Mason goto insert; 1122a429e513SChris Mason } 1123a429e513SChris Mason 1124a429e513SChris Mason /* 1125cc14600cSFilipe Manana * At this point, we know the tree has a checksum item that ends at an 1126cc14600cSFilipe Manana * offset matching the start of the checksum range we want to insert. 1127cc14600cSFilipe Manana * We try to extend that item as much as possible and then add as many 1128cc14600cSFilipe Manana * checksums to it as they fit. 1129cc14600cSFilipe Manana * 1130cc14600cSFilipe Manana * First check if the leaf has enough free space for at least one 1131cc14600cSFilipe Manana * checksum. If it has go directly to the item extension code, otherwise 1132cc14600cSFilipe Manana * release the path and do a search for insertion before the extension. 1133a429e513SChris Mason */ 1134cc14600cSFilipe Manana if (btrfs_leaf_free_space(leaf) >= csum_size) { 1135cc14600cSFilipe Manana btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1136cc14600cSFilipe Manana csum_offset = (bytenr - found_key.offset) >> 1137265fdfa6SDavid Sterba fs_info->sectorsize_bits; 1138cc14600cSFilipe Manana goto extend_csum; 1139cc14600cSFilipe Manana } 1140cc14600cSFilipe Manana 1141b3b4aa74SDavid Sterba btrfs_release_path(path); 11429a664971Sethanwu path->search_for_extension = 1; 11436567e837SChris Mason ret = btrfs_search_slot(trans, root, &file_key, path, 1144607d432dSJosef Bacik csum_size, 1); 11459a664971Sethanwu path->search_for_extension = 0; 11466567e837SChris Mason if (ret < 0) 1147918cdf44SFilipe Manana goto out; 1148459931ecSChris Mason 1149459931ecSChris Mason if (ret > 0) { 1150459931ecSChris Mason if (path->slots[0] == 0) 11516567e837SChris Mason goto insert; 11526567e837SChris Mason path->slots[0]--; 1153459931ecSChris Mason } 1154459931ecSChris Mason 11555f39d397SChris Mason leaf = path->nodes[0]; 11565f39d397SChris Mason btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1157265fdfa6SDavid Sterba csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits; 1158459931ecSChris Mason 1159962a298fSDavid Sterba if (found_key.type != BTRFS_EXTENT_CSUM_KEY || 1160d20f7043SChris Mason found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID || 11610b246afaSJeff Mahoney csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) { 11626567e837SChris Mason goto insert; 11636567e837SChris Mason } 1164459931ecSChris Mason 1165cc14600cSFilipe Manana extend_csum: 11663212fa14SJosef Bacik if (csum_offset == btrfs_item_size(leaf, path->slots[0]) / 1167607d432dSJosef Bacik csum_size) { 11682f697dc6SLiu Bo int extend_nr; 11692f697dc6SLiu Bo u64 tmp; 11702f697dc6SLiu Bo u32 diff; 1171459931ecSChris Mason 1172f51a4a18SMiao Xie tmp = sums->len - total_bytes; 1173265fdfa6SDavid Sterba tmp >>= fs_info->sectorsize_bits; 11742f697dc6SLiu Bo WARN_ON(tmp < 1); 1175ea7036deSFilipe Manana extend_nr = max_t(int, 1, tmp); 11762f697dc6SLiu Bo 1177ea7036deSFilipe Manana /* 1178ea7036deSFilipe Manana * A log tree can already have checksum items with a subset of 1179ea7036deSFilipe Manana * the checksums we are trying to log. This can happen after 1180ea7036deSFilipe Manana * doing a sequence of partial writes into prealloc extents and 1181ea7036deSFilipe Manana * fsyncs in between, with a full fsync logging a larger subrange 1182ea7036deSFilipe Manana * of an extent for which a previous fast fsync logged a smaller 1183ea7036deSFilipe Manana * subrange. And this happens in particular due to merging file 1184ea7036deSFilipe Manana * extent items when we complete an ordered extent for a range 1185ea7036deSFilipe Manana * covered by a prealloc extent - this is done at 1186ea7036deSFilipe Manana * btrfs_mark_extent_written(). 1187ea7036deSFilipe Manana * 1188ea7036deSFilipe Manana * So if we try to extend the previous checksum item, which has 1189ea7036deSFilipe Manana * a range that ends at the start of the range we want to insert, 1190ea7036deSFilipe Manana * make sure we don't extend beyond the start offset of the next 1191ea7036deSFilipe Manana * checksum item. If we are at the last item in the leaf, then 1192ea7036deSFilipe Manana * forget the optimization of extending and add a new checksum 1193ea7036deSFilipe Manana * item - it is not worth the complexity of releasing the path, 1194ea7036deSFilipe Manana * getting the first key for the next leaf, repeat the btree 1195ea7036deSFilipe Manana * search, etc, because log trees are temporary anyway and it 1196ea7036deSFilipe Manana * would only save a few bytes of leaf space. 1197ea7036deSFilipe Manana */ 1198ea7036deSFilipe Manana if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { 1199ea7036deSFilipe Manana if (path->slots[0] + 1 >= 1200ea7036deSFilipe Manana btrfs_header_nritems(path->nodes[0])) { 1201ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1202ea7036deSFilipe Manana if (ret < 0) 1203ea7036deSFilipe Manana goto out; 1204ea7036deSFilipe Manana found_next = 1; 1205ea7036deSFilipe Manana goto insert; 1206ea7036deSFilipe Manana } 1207ea7036deSFilipe Manana 1208ea7036deSFilipe Manana ret = find_next_csum_offset(root, path, &next_offset); 1209ea7036deSFilipe Manana if (ret < 0) 1210ea7036deSFilipe Manana goto out; 1211ea7036deSFilipe Manana 1212ea7036deSFilipe Manana tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits; 1213ea7036deSFilipe Manana if (tmp <= INT_MAX) 1214ea7036deSFilipe Manana extend_nr = min_t(int, extend_nr, tmp); 1215ea7036deSFilipe Manana } 1216ea7036deSFilipe Manana 12172f697dc6SLiu Bo diff = (csum_offset + extend_nr) * csum_size; 12180b246afaSJeff Mahoney diff = min(diff, 12190b246afaSJeff Mahoney MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size); 1220459931ecSChris Mason 12213212fa14SJosef Bacik diff = diff - btrfs_item_size(leaf, path->slots[0]); 1222cc14600cSFilipe Manana diff = min_t(u32, btrfs_leaf_free_space(leaf), diff); 12232f697dc6SLiu Bo diff /= csum_size; 12242f697dc6SLiu Bo diff *= csum_size; 1225459931ecSChris Mason 1226c71dd880SDavid Sterba btrfs_extend_item(path, diff); 1227f51a4a18SMiao Xie ret = 0; 12286567e837SChris Mason goto csum; 12296567e837SChris Mason } 12306567e837SChris Mason 12316567e837SChris Mason insert: 1232b3b4aa74SDavid Sterba btrfs_release_path(path); 12336567e837SChris Mason csum_offset = 0; 1234f578d4bdSChris Mason if (found_next) { 12352f697dc6SLiu Bo u64 tmp; 1236d20f7043SChris Mason 1237f51a4a18SMiao Xie tmp = sums->len - total_bytes; 1238265fdfa6SDavid Sterba tmp >>= fs_info->sectorsize_bits; 12392f697dc6SLiu Bo tmp = min(tmp, (next_offset - file_key.offset) >> 1240265fdfa6SDavid Sterba fs_info->sectorsize_bits); 12412f697dc6SLiu Bo 124250d0446eSSeraphime Kirkovski tmp = max_t(u64, 1, tmp); 124350d0446eSSeraphime Kirkovski tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size)); 1244607d432dSJosef Bacik ins_size = csum_size * tmp; 1245f578d4bdSChris Mason } else { 1246607d432dSJosef Bacik ins_size = csum_size; 1247f578d4bdSChris Mason } 12485caf2a00SChris Mason ret = btrfs_insert_empty_item(trans, root, path, &file_key, 1249f578d4bdSChris Mason ins_size); 125054aa1f4dSChris Mason if (ret < 0) 1251918cdf44SFilipe Manana goto out; 1252fae7f21cSDulshani Gunawardhana if (WARN_ON(ret != 0)) 1253918cdf44SFilipe Manana goto out; 12545f39d397SChris Mason leaf = path->nodes[0]; 1255f51a4a18SMiao Xie csum: 12565f39d397SChris Mason item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item); 1257f51a4a18SMiao Xie item_end = (struct btrfs_csum_item *)((unsigned char *)item + 12583212fa14SJosef Bacik btrfs_item_size(leaf, path->slots[0])); 1259509659cdSChris Mason item = (struct btrfs_csum_item *)((unsigned char *)item + 1260607d432dSJosef Bacik csum_offset * csum_size); 1261b18c6685SChris Mason found: 1262265fdfa6SDavid Sterba ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits; 1263f51a4a18SMiao Xie ins_size *= csum_size; 1264f51a4a18SMiao Xie ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item, 1265f51a4a18SMiao Xie ins_size); 1266f51a4a18SMiao Xie write_extent_buffer(leaf, sums->sums + index, (unsigned long)item, 1267f51a4a18SMiao Xie ins_size); 1268aadfeb6eSChris Mason 12691e25a2e3SJohannes Thumshirn index += ins_size; 1270f51a4a18SMiao Xie ins_size /= csum_size; 12710b246afaSJeff Mahoney total_bytes += ins_size * fs_info->sectorsize; 1272a6591715SChris Mason 12735caf2a00SChris Mason btrfs_mark_buffer_dirty(path->nodes[0]); 1274e6dcd2dcSChris Mason if (total_bytes < sums->len) { 1275b3b4aa74SDavid Sterba btrfs_release_path(path); 1276b9473439SChris Mason cond_resched(); 1277065631f6SChris Mason goto again; 1278065631f6SChris Mason } 127953863232SChris Mason out: 12805caf2a00SChris Mason btrfs_free_path(path); 1281f254e52cSChris Mason return ret; 1282f254e52cSChris Mason } 12837ffbb598SFilipe Manana 12849cdc5124SNikolay Borisov void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode, 12857ffbb598SFilipe Manana const struct btrfs_path *path, 12867ffbb598SFilipe Manana struct btrfs_file_extent_item *fi, 12877ffbb598SFilipe Manana struct extent_map *em) 12887ffbb598SFilipe Manana { 12893ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 12909cdc5124SNikolay Borisov struct btrfs_root *root = inode->root; 12917ffbb598SFilipe Manana struct extent_buffer *leaf = path->nodes[0]; 12927ffbb598SFilipe Manana const int slot = path->slots[0]; 12937ffbb598SFilipe Manana struct btrfs_key key; 12947ffbb598SFilipe Manana u64 extent_start, extent_end; 12957ffbb598SFilipe Manana u64 bytenr; 12967ffbb598SFilipe Manana u8 type = btrfs_file_extent_type(leaf, fi); 12977ffbb598SFilipe Manana int compress_type = btrfs_file_extent_compression(leaf, fi); 12987ffbb598SFilipe Manana 12997ffbb598SFilipe Manana btrfs_item_key_to_cpu(leaf, &key, slot); 13007ffbb598SFilipe Manana extent_start = key.offset; 1301a5eeb3d1SFilipe Manana extent_end = btrfs_file_extent_end(path); 13027ffbb598SFilipe Manana em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); 130340e7efe0SQu Wenruo em->generation = btrfs_file_extent_generation(leaf, fi); 13047ffbb598SFilipe Manana if (type == BTRFS_FILE_EXTENT_REG || 13057ffbb598SFilipe Manana type == BTRFS_FILE_EXTENT_PREALLOC) { 13067ffbb598SFilipe Manana em->start = extent_start; 13077ffbb598SFilipe Manana em->len = extent_end - extent_start; 13087ffbb598SFilipe Manana em->orig_start = extent_start - 13097ffbb598SFilipe Manana btrfs_file_extent_offset(leaf, fi); 13107ffbb598SFilipe Manana em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); 13117ffbb598SFilipe Manana bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 13127ffbb598SFilipe Manana if (bytenr == 0) { 13137ffbb598SFilipe Manana em->block_start = EXTENT_MAP_HOLE; 13147ffbb598SFilipe Manana return; 13157ffbb598SFilipe Manana } 13167ffbb598SFilipe Manana if (compress_type != BTRFS_COMPRESS_NONE) { 13177ffbb598SFilipe Manana set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 13187ffbb598SFilipe Manana em->compress_type = compress_type; 13197ffbb598SFilipe Manana em->block_start = bytenr; 13207ffbb598SFilipe Manana em->block_len = em->orig_block_len; 13217ffbb598SFilipe Manana } else { 13227ffbb598SFilipe Manana bytenr += btrfs_file_extent_offset(leaf, fi); 13237ffbb598SFilipe Manana em->block_start = bytenr; 13247ffbb598SFilipe Manana em->block_len = em->len; 13257ffbb598SFilipe Manana if (type == BTRFS_FILE_EXTENT_PREALLOC) 13267ffbb598SFilipe Manana set_bit(EXTENT_FLAG_PREALLOC, &em->flags); 13277ffbb598SFilipe Manana } 13287ffbb598SFilipe Manana } else if (type == BTRFS_FILE_EXTENT_INLINE) { 13297ffbb598SFilipe Manana em->block_start = EXTENT_MAP_INLINE; 13307ffbb598SFilipe Manana em->start = extent_start; 13317ffbb598SFilipe Manana em->len = extent_end - extent_start; 13327ffbb598SFilipe Manana /* 13337ffbb598SFilipe Manana * Initialize orig_start and block_len with the same values 13347ffbb598SFilipe Manana * as in inode.c:btrfs_get_extent(). 13357ffbb598SFilipe Manana */ 13367ffbb598SFilipe Manana em->orig_start = EXTENT_MAP_HOLE; 13377ffbb598SFilipe Manana em->block_len = (u64)-1; 13387ffbb598SFilipe Manana em->compress_type = compress_type; 1339280f15cbSQu Wenruo if (compress_type != BTRFS_COMPRESS_NONE) 1340280f15cbSQu Wenruo set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 13417ffbb598SFilipe Manana } else { 13420b246afaSJeff Mahoney btrfs_err(fs_info, 13439cdc5124SNikolay Borisov "unknown file extent item type %d, inode %llu, offset %llu, " 13449cdc5124SNikolay Borisov "root %llu", type, btrfs_ino(inode), extent_start, 13457ffbb598SFilipe Manana root->root_key.objectid); 13467ffbb598SFilipe Manana } 13477ffbb598SFilipe Manana } 1348a5eeb3d1SFilipe Manana 1349a5eeb3d1SFilipe Manana /* 1350a5eeb3d1SFilipe Manana * Returns the end offset (non inclusive) of the file extent item the given path 1351a5eeb3d1SFilipe Manana * points to. If it points to an inline extent, the returned offset is rounded 1352a5eeb3d1SFilipe Manana * up to the sector size. 1353a5eeb3d1SFilipe Manana */ 1354a5eeb3d1SFilipe Manana u64 btrfs_file_extent_end(const struct btrfs_path *path) 1355a5eeb3d1SFilipe Manana { 1356a5eeb3d1SFilipe Manana const struct extent_buffer *leaf = path->nodes[0]; 1357a5eeb3d1SFilipe Manana const int slot = path->slots[0]; 1358a5eeb3d1SFilipe Manana struct btrfs_file_extent_item *fi; 1359a5eeb3d1SFilipe Manana struct btrfs_key key; 1360a5eeb3d1SFilipe Manana u64 end; 1361a5eeb3d1SFilipe Manana 1362a5eeb3d1SFilipe Manana btrfs_item_key_to_cpu(leaf, &key, slot); 1363a5eeb3d1SFilipe Manana ASSERT(key.type == BTRFS_EXTENT_DATA_KEY); 1364a5eeb3d1SFilipe Manana fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 1365a5eeb3d1SFilipe Manana 1366a5eeb3d1SFilipe Manana if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) { 1367a5eeb3d1SFilipe Manana end = btrfs_file_extent_ram_bytes(leaf, fi); 1368a5eeb3d1SFilipe Manana end = ALIGN(key.offset + end, leaf->fs_info->sectorsize); 1369a5eeb3d1SFilipe Manana } else { 1370a5eeb3d1SFilipe Manana end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); 1371a5eeb3d1SFilipe Manana } 1372a5eeb3d1SFilipe Manana 1373a5eeb3d1SFilipe Manana return end; 1374a5eeb3d1SFilipe Manana } 1375