754 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * Copyright (C) 2007 Oracle. All rights reserved.4 */5 6#include "ctree.h"7#include "fs.h"8#include "messages.h"9#include "inode-item.h"10#include "disk-io.h"11#include "transaction.h"12#include "space-info.h"13#include "accessors.h"14#include "extent-tree.h"15#include "file-item.h"16 17struct btrfs_inode_ref *btrfs_find_name_in_backref(const struct extent_buffer *leaf,18 int slot,19 const struct fscrypt_str *name)20{21 struct btrfs_inode_ref *ref;22 unsigned long ptr;23 unsigned long name_ptr;24 u32 item_size;25 u32 cur_offset = 0;26 int len;27 28 item_size = btrfs_item_size(leaf, slot);29 ptr = btrfs_item_ptr_offset(leaf, slot);30 while (cur_offset < item_size) {31 ref = (struct btrfs_inode_ref *)(ptr + cur_offset);32 len = btrfs_inode_ref_name_len(leaf, ref);33 name_ptr = (unsigned long)(ref + 1);34 cur_offset += len + sizeof(*ref);35 if (len != name->len)36 continue;37 if (memcmp_extent_buffer(leaf, name->name, name_ptr,38 name->len) == 0)39 return ref;40 }41 return NULL;42}43 44struct btrfs_inode_extref *btrfs_find_name_in_ext_backref(45 const struct extent_buffer *leaf, int slot, u64 ref_objectid,46 const struct fscrypt_str *name)47{48 struct btrfs_inode_extref *extref;49 unsigned long ptr;50 unsigned long name_ptr;51 u32 item_size;52 u32 cur_offset = 0;53 int ref_name_len;54 55 item_size = btrfs_item_size(leaf, slot);56 ptr = btrfs_item_ptr_offset(leaf, slot);57 58 /*59 * Search all extended backrefs in this item. We're only60 * looking through any collisions so most of the time this is61 * just going to compare against one buffer. If all is well,62 * we'll return success and the inode ref object.63 */64 while (cur_offset < item_size) {65 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);66 name_ptr = (unsigned long)(&extref->name);67 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);68 69 if (ref_name_len == name->len &&70 btrfs_inode_extref_parent(leaf, extref) == ref_objectid &&71 (memcmp_extent_buffer(leaf, name->name, name_ptr,72 name->len) == 0))73 return extref;74 75 cur_offset += ref_name_len + sizeof(*extref);76 }77 return NULL;78}79 80/* Returns NULL if no extref found */81struct btrfs_inode_extref *82btrfs_lookup_inode_extref(struct btrfs_trans_handle *trans,83 struct btrfs_root *root,84 struct btrfs_path *path,85 const struct fscrypt_str *name,86 u64 inode_objectid, u64 ref_objectid, int ins_len,87 int cow)88{89 int ret;90 struct btrfs_key key;91 92 key.objectid = inode_objectid;93 key.type = BTRFS_INODE_EXTREF_KEY;94 key.offset = btrfs_extref_hash(ref_objectid, name->name, name->len);95 96 ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);97 if (ret < 0)98 return ERR_PTR(ret);99 if (ret > 0)100 return NULL;101 return btrfs_find_name_in_ext_backref(path->nodes[0], path->slots[0],102 ref_objectid, name);103 104}105 106static int btrfs_del_inode_extref(struct btrfs_trans_handle *trans,107 struct btrfs_root *root,108 const struct fscrypt_str *name,109 u64 inode_objectid, u64 ref_objectid,110 u64 *index)111{112 struct btrfs_path *path;113 struct btrfs_key key;114 struct btrfs_inode_extref *extref;115 struct extent_buffer *leaf;116 int ret;117 int del_len = name->len + sizeof(*extref);118 unsigned long ptr;119 unsigned long item_start;120 u32 item_size;121 122 key.objectid = inode_objectid;123 key.type = BTRFS_INODE_EXTREF_KEY;124 key.offset = btrfs_extref_hash(ref_objectid, name->name, name->len);125 126 path = btrfs_alloc_path();127 if (!path)128 return -ENOMEM;129 130 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);131 if (ret > 0)132 ret = -ENOENT;133 if (ret < 0)134 goto out;135 136 /*137 * Sanity check - did we find the right item for this name?138 * This should always succeed so error here will make the FS139 * readonly.140 */141 extref = btrfs_find_name_in_ext_backref(path->nodes[0], path->slots[0],142 ref_objectid, name);143 if (!extref) {144 btrfs_abort_transaction(trans, -ENOENT);145 ret = -ENOENT;146 goto out;147 }148 149 leaf = path->nodes[0];150 item_size = btrfs_item_size(leaf, path->slots[0]);151 if (index)152 *index = btrfs_inode_extref_index(leaf, extref);153 154 if (del_len == item_size) {155 /*156 * Common case only one ref in the item, remove the157 * whole item.158 */159 ret = btrfs_del_item(trans, root, path);160 goto out;161 }162 163 ptr = (unsigned long)extref;164 item_start = btrfs_item_ptr_offset(leaf, path->slots[0]);165 166 memmove_extent_buffer(leaf, ptr, ptr + del_len,167 item_size - (ptr + del_len - item_start));168 169 btrfs_truncate_item(trans, path, item_size - del_len, 1);170 171out:172 btrfs_free_path(path);173 174 return ret;175}176 177int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,178 struct btrfs_root *root, const struct fscrypt_str *name,179 u64 inode_objectid, u64 ref_objectid, u64 *index)180{181 struct btrfs_path *path;182 struct btrfs_key key;183 struct btrfs_inode_ref *ref;184 struct extent_buffer *leaf;185 unsigned long ptr;186 unsigned long item_start;187 u32 item_size;188 u32 sub_item_len;189 int ret;190 int search_ext_refs = 0;191 int del_len = name->len + sizeof(*ref);192 193 key.objectid = inode_objectid;194 key.offset = ref_objectid;195 key.type = BTRFS_INODE_REF_KEY;196 197 path = btrfs_alloc_path();198 if (!path)199 return -ENOMEM;200 201 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);202 if (ret > 0) {203 ret = -ENOENT;204 search_ext_refs = 1;205 goto out;206 } else if (ret < 0) {207 goto out;208 }209 210 ref = btrfs_find_name_in_backref(path->nodes[0], path->slots[0], name);211 if (!ref) {212 ret = -ENOENT;213 search_ext_refs = 1;214 goto out;215 }216 leaf = path->nodes[0];217 item_size = btrfs_item_size(leaf, path->slots[0]);218 219 if (index)220 *index = btrfs_inode_ref_index(leaf, ref);221 222 if (del_len == item_size) {223 ret = btrfs_del_item(trans, root, path);224 goto out;225 }226 ptr = (unsigned long)ref;227 sub_item_len = name->len + sizeof(*ref);228 item_start = btrfs_item_ptr_offset(leaf, path->slots[0]);229 memmove_extent_buffer(leaf, ptr, ptr + sub_item_len,230 item_size - (ptr + sub_item_len - item_start));231 btrfs_truncate_item(trans, path, item_size - sub_item_len, 1);232out:233 btrfs_free_path(path);234 235 if (search_ext_refs) {236 /*237 * No refs were found, or we could not find the238 * name in our ref array. Find and remove the extended239 * inode ref then.240 */241 return btrfs_del_inode_extref(trans, root, name,242 inode_objectid, ref_objectid, index);243 }244 245 return ret;246}247 248/*249 * Insert an extended inode ref into a tree.250 *251 * The caller must have checked against BTRFS_LINK_MAX already.252 */253static int btrfs_insert_inode_extref(struct btrfs_trans_handle *trans,254 struct btrfs_root *root,255 const struct fscrypt_str *name,256 u64 inode_objectid, u64 ref_objectid,257 u64 index)258{259 struct btrfs_inode_extref *extref;260 int ret;261 int ins_len = name->len + sizeof(*extref);262 unsigned long ptr;263 struct btrfs_path *path;264 struct btrfs_key key;265 struct extent_buffer *leaf;266 267 key.objectid = inode_objectid;268 key.type = BTRFS_INODE_EXTREF_KEY;269 key.offset = btrfs_extref_hash(ref_objectid, name->name, name->len);270 271 path = btrfs_alloc_path();272 if (!path)273 return -ENOMEM;274 275 ret = btrfs_insert_empty_item(trans, root, path, &key,276 ins_len);277 if (ret == -EEXIST) {278 if (btrfs_find_name_in_ext_backref(path->nodes[0],279 path->slots[0],280 ref_objectid,281 name))282 goto out;283 284 btrfs_extend_item(trans, path, ins_len);285 ret = 0;286 }287 if (ret < 0)288 goto out;289 290 leaf = path->nodes[0];291 ptr = (unsigned long)btrfs_item_ptr(leaf, path->slots[0], char);292 ptr += btrfs_item_size(leaf, path->slots[0]) - ins_len;293 extref = (struct btrfs_inode_extref *)ptr;294 295 btrfs_set_inode_extref_name_len(path->nodes[0], extref, name->len);296 btrfs_set_inode_extref_index(path->nodes[0], extref, index);297 btrfs_set_inode_extref_parent(path->nodes[0], extref, ref_objectid);298 299 ptr = (unsigned long)&extref->name;300 write_extent_buffer(path->nodes[0], name->name, ptr, name->len);301 btrfs_mark_buffer_dirty(trans, path->nodes[0]);302 303out:304 btrfs_free_path(path);305 return ret;306}307 308/* Will return 0, -ENOMEM, -EMLINK, or -EEXIST or anything from the CoW path */309int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,310 struct btrfs_root *root, const struct fscrypt_str *name,311 u64 inode_objectid, u64 ref_objectid, u64 index)312{313 struct btrfs_fs_info *fs_info = root->fs_info;314 struct btrfs_path *path;315 struct btrfs_key key;316 struct btrfs_inode_ref *ref;317 unsigned long ptr;318 int ret;319 int ins_len = name->len + sizeof(*ref);320 321 key.objectid = inode_objectid;322 key.offset = ref_objectid;323 key.type = BTRFS_INODE_REF_KEY;324 325 path = btrfs_alloc_path();326 if (!path)327 return -ENOMEM;328 329 path->skip_release_on_error = 1;330 ret = btrfs_insert_empty_item(trans, root, path, &key,331 ins_len);332 if (ret == -EEXIST) {333 u32 old_size;334 ref = btrfs_find_name_in_backref(path->nodes[0], path->slots[0],335 name);336 if (ref)337 goto out;338 339 old_size = btrfs_item_size(path->nodes[0], path->slots[0]);340 btrfs_extend_item(trans, path, ins_len);341 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],342 struct btrfs_inode_ref);343 ref = (struct btrfs_inode_ref *)((unsigned long)ref + old_size);344 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name->len);345 btrfs_set_inode_ref_index(path->nodes[0], ref, index);346 ptr = (unsigned long)(ref + 1);347 ret = 0;348 } else if (ret < 0) {349 if (ret == -EOVERFLOW) {350 if (btrfs_find_name_in_backref(path->nodes[0],351 path->slots[0],352 name))353 ret = -EEXIST;354 else355 ret = -EMLINK;356 }357 goto out;358 } else {359 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],360 struct btrfs_inode_ref);361 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name->len);362 btrfs_set_inode_ref_index(path->nodes[0], ref, index);363 ptr = (unsigned long)(ref + 1);364 }365 write_extent_buffer(path->nodes[0], name->name, ptr, name->len);366 btrfs_mark_buffer_dirty(trans, path->nodes[0]);367 368out:369 btrfs_free_path(path);370 371 if (ret == -EMLINK) {372 struct btrfs_super_block *disk_super = fs_info->super_copy;373 /* We ran out of space in the ref array. Need to374 * add an extended ref. */375 if (btrfs_super_incompat_flags(disk_super)376 & BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF)377 ret = btrfs_insert_inode_extref(trans, root, name,378 inode_objectid,379 ref_objectid, index);380 }381 382 return ret;383}384 385int btrfs_insert_empty_inode(struct btrfs_trans_handle *trans,386 struct btrfs_root *root,387 struct btrfs_path *path, u64 objectid)388{389 struct btrfs_key key;390 int ret;391 key.objectid = objectid;392 key.type = BTRFS_INODE_ITEM_KEY;393 key.offset = 0;394 395 ret = btrfs_insert_empty_item(trans, root, path, &key,396 sizeof(struct btrfs_inode_item));397 return ret;398}399 400int btrfs_lookup_inode(struct btrfs_trans_handle *trans, struct btrfs_root401 *root, struct btrfs_path *path,402 struct btrfs_key *location, int mod)403{404 int ins_len = mod < 0 ? -1 : 0;405 int cow = mod != 0;406 int ret;407 int slot;408 struct extent_buffer *leaf;409 struct btrfs_key found_key;410 411 ret = btrfs_search_slot(trans, root, location, path, ins_len, cow);412 if (ret > 0 && location->type == BTRFS_ROOT_ITEM_KEY &&413 location->offset == (u64)-1 && path->slots[0] != 0) {414 slot = path->slots[0] - 1;415 leaf = path->nodes[0];416 btrfs_item_key_to_cpu(leaf, &found_key, slot);417 if (found_key.objectid == location->objectid &&418 found_key.type == location->type) {419 path->slots[0]--;420 return 0;421 }422 }423 return ret;424}425 426static inline void btrfs_trace_truncate(const struct btrfs_inode *inode,427 const struct extent_buffer *leaf,428 const struct btrfs_file_extent_item *fi,429 u64 offset, int extent_type, int slot)430{431 if (!inode)432 return;433 if (extent_type == BTRFS_FILE_EXTENT_INLINE)434 trace_btrfs_truncate_show_fi_inline(inode, leaf, fi, slot,435 offset);436 else437 trace_btrfs_truncate_show_fi_regular(inode, leaf, fi, offset);438}439 440/*441 * Remove inode items from a given root.442 *443 * @trans: A transaction handle.444 * @root: The root from which to remove items.445 * @inode: The inode whose items we want to remove.446 * @control: The btrfs_truncate_control to control how and what we447 * are truncating.448 *449 * Remove all keys associated with the inode from the given root that have a key450 * with a type greater than or equals to @min_type. When @min_type has a value of451 * BTRFS_EXTENT_DATA_KEY, only remove file extent items that have an offset value452 * greater than or equals to @new_size. If a file extent item that starts before453 * @new_size and ends after it is found, its length is adjusted.454 *455 * Returns: 0 on success, < 0 on error and NEED_TRUNCATE_BLOCK when @min_type is456 * BTRFS_EXTENT_DATA_KEY and the caller must truncate the last block.457 */458int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,459 struct btrfs_root *root,460 struct btrfs_truncate_control *control)461{462 struct btrfs_fs_info *fs_info = root->fs_info;463 struct btrfs_path *path;464 struct extent_buffer *leaf;465 struct btrfs_file_extent_item *fi;466 struct btrfs_key key;467 struct btrfs_key found_key;468 u64 new_size = control->new_size;469 u64 extent_num_bytes = 0;470 u64 extent_offset = 0;471 u64 item_end = 0;472 u32 found_type = (u8)-1;473 int del_item;474 int pending_del_nr = 0;475 int pending_del_slot = 0;476 int extent_type = -1;477 int ret;478 u64 bytes_deleted = 0;479 bool be_nice = false;480 481 ASSERT(control->inode || !control->clear_extent_range);482 ASSERT(new_size == 0 || control->min_type == BTRFS_EXTENT_DATA_KEY);483 484 control->last_size = new_size;485 control->sub_bytes = 0;486 487 /*488 * For shareable roots we want to back off from time to time, this turns489 * out to be subvolume roots, reloc roots, and data reloc roots.490 */491 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))492 be_nice = true;493 494 path = btrfs_alloc_path();495 if (!path)496 return -ENOMEM;497 path->reada = READA_BACK;498 499 key.objectid = control->ino;500 key.offset = (u64)-1;501 key.type = (u8)-1;502 503search_again:504 /*505 * With a 16K leaf size and 128MiB extents, you can actually queue up a506 * huge file in a single leaf. Most of the time that bytes_deleted is507 * > 0, it will be huge by the time we get here508 */509 if (be_nice && bytes_deleted > SZ_32M &&510 btrfs_should_end_transaction(trans)) {511 ret = -EAGAIN;512 goto out;513 }514 515 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);516 if (ret < 0)517 goto out;518 519 if (ret > 0) {520 ret = 0;521 /* There are no items in the tree for us to truncate, we're done */522 if (path->slots[0] == 0)523 goto out;524 path->slots[0]--;525 }526 527 while (1) {528 u64 clear_start = 0, clear_len = 0, extent_start = 0;529 bool refill_delayed_refs_rsv = false;530 531 fi = NULL;532 leaf = path->nodes[0];533 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);534 found_type = found_key.type;535 536 if (found_key.objectid != control->ino)537 break;538 539 if (found_type < control->min_type)540 break;541 542 item_end = found_key.offset;543 if (found_type == BTRFS_EXTENT_DATA_KEY) {544 fi = btrfs_item_ptr(leaf, path->slots[0],545 struct btrfs_file_extent_item);546 extent_type = btrfs_file_extent_type(leaf, fi);547 if (extent_type != BTRFS_FILE_EXTENT_INLINE)548 item_end +=549 btrfs_file_extent_num_bytes(leaf, fi);550 else if (extent_type == BTRFS_FILE_EXTENT_INLINE)551 item_end += btrfs_file_extent_ram_bytes(leaf, fi);552 553 btrfs_trace_truncate(control->inode, leaf, fi,554 found_key.offset, extent_type,555 path->slots[0]);556 item_end--;557 }558 if (found_type > control->min_type) {559 del_item = 1;560 } else {561 if (item_end < new_size)562 break;563 if (found_key.offset >= new_size)564 del_item = 1;565 else566 del_item = 0;567 }568 569 /* FIXME, shrink the extent if the ref count is only 1 */570 if (found_type != BTRFS_EXTENT_DATA_KEY)571 goto delete;572 573 control->extents_found++;574 575 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {576 u64 num_dec;577 578 clear_start = found_key.offset;579 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);580 if (!del_item) {581 u64 orig_num_bytes =582 btrfs_file_extent_num_bytes(leaf, fi);583 extent_num_bytes = ALIGN(new_size -584 found_key.offset,585 fs_info->sectorsize);586 clear_start = ALIGN(new_size, fs_info->sectorsize);587 588 btrfs_set_file_extent_num_bytes(leaf, fi,589 extent_num_bytes);590 num_dec = (orig_num_bytes - extent_num_bytes);591 if (extent_start != 0)592 control->sub_bytes += num_dec;593 btrfs_mark_buffer_dirty(trans, leaf);594 } else {595 extent_num_bytes =596 btrfs_file_extent_disk_num_bytes(leaf, fi);597 extent_offset = found_key.offset -598 btrfs_file_extent_offset(leaf, fi);599 600 /* FIXME blocksize != 4096 */601 num_dec = btrfs_file_extent_num_bytes(leaf, fi);602 if (extent_start != 0)603 control->sub_bytes += num_dec;604 }605 clear_len = num_dec;606 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {607 /*608 * We can't truncate inline items that have had609 * special encodings610 */611 if (!del_item &&612 btrfs_file_extent_encryption(leaf, fi) == 0 &&613 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&614 btrfs_file_extent_compression(leaf, fi) == 0) {615 u32 size = (u32)(new_size - found_key.offset);616 617 btrfs_set_file_extent_ram_bytes(leaf, fi, size);618 size = btrfs_file_extent_calc_inline_size(size);619 btrfs_truncate_item(trans, path, size, 1);620 } else if (!del_item) {621 /*622 * We have to bail so the last_size is set to623 * just before this extent.624 */625 ret = BTRFS_NEED_TRUNCATE_BLOCK;626 break;627 } else {628 /*629 * Inline extents are special, we just treat630 * them as a full sector worth in the file631 * extent tree just for simplicity sake.632 */633 clear_len = fs_info->sectorsize;634 }635 636 control->sub_bytes += item_end + 1 - new_size;637 }638delete:639 /*640 * We only want to clear the file extent range if we're641 * modifying the actual inode's mapping, which is just the642 * normal truncate path.643 */644 if (control->clear_extent_range) {645 ret = btrfs_inode_clear_file_extent_range(control->inode,646 clear_start, clear_len);647 if (ret) {648 btrfs_abort_transaction(trans, ret);649 break;650 }651 }652 653 if (del_item) {654 ASSERT(!pending_del_nr ||655 ((path->slots[0] + 1) == pending_del_slot));656 657 control->last_size = found_key.offset;658 if (!pending_del_nr) {659 /* No pending yet, add ourselves */660 pending_del_slot = path->slots[0];661 pending_del_nr = 1;662 } else if (path->slots[0] + 1 == pending_del_slot) {663 /* Hop on the pending chunk */664 pending_del_nr++;665 pending_del_slot = path->slots[0];666 }667 } else {668 control->last_size = new_size;669 break;670 }671 672 if (del_item && extent_start != 0 && !control->skip_ref_updates) {673 struct btrfs_ref ref = {674 .action = BTRFS_DROP_DELAYED_REF,675 .bytenr = extent_start,676 .num_bytes = extent_num_bytes,677 .owning_root = btrfs_root_id(root),678 .ref_root = btrfs_header_owner(leaf),679 };680 681 bytes_deleted += extent_num_bytes;682 683 btrfs_init_data_ref(&ref, control->ino, extent_offset,684 btrfs_root_id(root), false);685 ret = btrfs_free_extent(trans, &ref);686 if (ret) {687 btrfs_abort_transaction(trans, ret);688 break;689 }690 if (be_nice && btrfs_check_space_for_delayed_refs(fs_info))691 refill_delayed_refs_rsv = true;692 }693 694 if (found_type == BTRFS_INODE_ITEM_KEY)695 break;696 697 if (path->slots[0] == 0 ||698 path->slots[0] != pending_del_slot ||699 refill_delayed_refs_rsv) {700 if (pending_del_nr) {701 ret = btrfs_del_items(trans, root, path,702 pending_del_slot,703 pending_del_nr);704 if (ret) {705 btrfs_abort_transaction(trans, ret);706 break;707 }708 pending_del_nr = 0;709 }710 btrfs_release_path(path);711 712 /*713 * We can generate a lot of delayed refs, so we need to714 * throttle every once and a while and make sure we're715 * adding enough space to keep up with the work we are716 * generating. Since we hold a transaction here we717 * can't flush, and we don't want to FLUSH_LIMIT because718 * we could have generated too many delayed refs to719 * actually allocate, so just bail if we're short and720 * let the normal reservation dance happen higher up.721 */722 if (refill_delayed_refs_rsv) {723 ret = btrfs_delayed_refs_rsv_refill(fs_info,724 BTRFS_RESERVE_NO_FLUSH);725 if (ret) {726 ret = -EAGAIN;727 break;728 }729 }730 goto search_again;731 } else {732 path->slots[0]--;733 }734 }735out:736 if (ret >= 0 && pending_del_nr) {737 int err;738 739 err = btrfs_del_items(trans, root, path, pending_del_slot,740 pending_del_nr);741 if (err) {742 btrfs_abort_transaction(trans, err);743 ret = err;744 }745 }746 747 ASSERT(control->last_size >= new_size);748 if (!ret && control->last_size > new_size)749 control->last_size = new_size;750 751 btrfs_free_path(path);752 return ret;753}754