linux-stable/fs/hfsplus/bfind.c
Chao Yu be4edd1642
hfsplus: fix to avoid false alarm of circular locking
Syzbot report potential ABBA deadlock as below:

loop0: detected capacity change from 0 to 1024
======================================================
WARNING: possible circular locking dependency detected
6.9.0-syzkaller-10323-g8f6a15f095a6 #0 Not tainted
------------------------------------------------------
syz-executor171/5344 is trying to acquire lock:
ffff88807cb980b0 (&tree->tree_lock){+.+.}-{3:3}, at: hfsplus_file_truncate+0x811/0xb50 fs/hfsplus/extents.c:595

but task is already holding lock:
ffff88807a930108 (&HFSPLUS_I(inode)->extents_lock){+.+.}-{3:3}, at: hfsplus_file_truncate+0x2da/0xb50 fs/hfsplus/extents.c:576

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&HFSPLUS_I(inode)->extents_lock){+.+.}-{3:3}:
       lock_acquire+0x1ed/0x550 kernel/locking/lockdep.c:5754
       __mutex_lock_common kernel/locking/mutex.c:608 [inline]
       __mutex_lock+0x136/0xd70 kernel/locking/mutex.c:752
       hfsplus_file_extend+0x21b/0x1b70 fs/hfsplus/extents.c:457
       hfsplus_bmap_reserve+0x105/0x4e0 fs/hfsplus/btree.c:358
       hfsplus_rename_cat+0x1d0/0x1050 fs/hfsplus/catalog.c:456
       hfsplus_rename+0x12e/0x1c0 fs/hfsplus/dir.c:552
       vfs_rename+0xbdb/0xf00 fs/namei.c:4887
       do_renameat2+0xd94/0x13f0 fs/namei.c:5044
       __do_sys_rename fs/namei.c:5091 [inline]
       __se_sys_rename fs/namei.c:5089 [inline]
       __x64_sys_rename+0x86/0xa0 fs/namei.c:5089
       do_syscall_x64 arch/x86/entry/common.c:52 [inline]
       do_syscall_64+0xf5/0x240 arch/x86/entry/common.c:83
       entry_SYSCALL_64_after_hwframe+0x77/0x7f

-> #0 (&tree->tree_lock){+.+.}-{3:3}:
       check_prev_add kernel/locking/lockdep.c:3134 [inline]
       check_prevs_add kernel/locking/lockdep.c:3253 [inline]
       validate_chain+0x18cb/0x58e0 kernel/locking/lockdep.c:3869
       __lock_acquire+0x1346/0x1fd0 kernel/locking/lockdep.c:5137
       lock_acquire+0x1ed/0x550 kernel/locking/lockdep.c:5754
       __mutex_lock_common kernel/locking/mutex.c:608 [inline]
       __mutex_lock+0x136/0xd70 kernel/locking/mutex.c:752
       hfsplus_file_truncate+0x811/0xb50 fs/hfsplus/extents.c:595
       hfsplus_setattr+0x1ce/0x280 fs/hfsplus/inode.c:265
       notify_change+0xb9d/0xe70 fs/attr.c:497
       do_truncate+0x220/0x310 fs/open.c:65
       handle_truncate fs/namei.c:3308 [inline]
       do_open fs/namei.c:3654 [inline]
       path_openat+0x2a3d/0x3280 fs/namei.c:3807
       do_filp_open+0x235/0x490 fs/namei.c:3834
       do_sys_openat2+0x13e/0x1d0 fs/open.c:1406
       do_sys_open fs/open.c:1421 [inline]
       __do_sys_creat fs/open.c:1497 [inline]
       __se_sys_creat fs/open.c:1491 [inline]
       __x64_sys_creat+0x123/0x170 fs/open.c:1491
       do_syscall_x64 arch/x86/entry/common.c:52 [inline]
       do_syscall_64+0xf5/0x240 arch/x86/entry/common.c:83
       entry_SYSCALL_64_after_hwframe+0x77/0x7f

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&HFSPLUS_I(inode)->extents_lock);
                               lock(&tree->tree_lock);
                               lock(&HFSPLUS_I(inode)->extents_lock);
  lock(&tree->tree_lock);

This is a false alarm as tree_lock mutex are different, one is
from sbi->cat_tree, and another is from sbi->ext_tree:

Thread A			Thread B
- hfsplus_rename
 - hfsplus_rename_cat
  - hfs_find_init
   - mutext_lock(cat_tree->tree_lock)
				- hfsplus_setattr
				 - hfsplus_file_truncate
				  - mutex_lock(hip->extents_lock)
				  - hfs_find_init
				   - mutext_lock(ext_tree->tree_lock)
  - hfs_bmap_reserve
   - hfsplus_file_extend
    - mutex_lock(hip->extents_lock)

So, let's call mutex_lock_nested for tree_lock mutex lock, and pass
correct lock class for it.

Fixes: 31651c6071 ("hfsplus: avoid deadlock on file truncation")
Reported-by: syzbot+6030b3b1b9bf70e538c4@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/linux-fsdevel/000000000000e37a4005ef129563@google.com
Cc: Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
Signed-off-by: Chao Yu <chao@kernel.org>
Link: https://lore.kernel.org/r/20240607142304.455441-1-chao@kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
2024-06-21 11:41:04 +02:00

284 lines
5.8 KiB
C

// SPDX-License-Identifier: GPL-2.0
/*
* linux/fs/hfsplus/bfind.c
*
* Copyright (C) 2001
* Brad Boyer (flar@allandria.com)
* (C) 2003 Ardis Technologies <roman@ardistech.com>
*
* Search routines for btrees
*/
#include <linux/slab.h>
#include "hfsplus_fs.h"
int hfs_find_init(struct hfs_btree *tree, struct hfs_find_data *fd)
{
void *ptr;
fd->tree = tree;
fd->bnode = NULL;
ptr = kmalloc(tree->max_key_len * 2 + 4, GFP_KERNEL);
if (!ptr)
return -ENOMEM;
fd->search_key = ptr;
fd->key = ptr + tree->max_key_len + 2;
hfs_dbg(BNODE_REFS, "find_init: %d (%p)\n",
tree->cnid, __builtin_return_address(0));
mutex_lock_nested(&tree->tree_lock,
hfsplus_btree_lock_class(tree));
return 0;
}
void hfs_find_exit(struct hfs_find_data *fd)
{
hfs_bnode_put(fd->bnode);
kfree(fd->search_key);
hfs_dbg(BNODE_REFS, "find_exit: %d (%p)\n",
fd->tree->cnid, __builtin_return_address(0));
mutex_unlock(&fd->tree->tree_lock);
fd->tree = NULL;
}
int hfs_find_1st_rec_by_cnid(struct hfs_bnode *bnode,
struct hfs_find_data *fd,
int *begin,
int *end,
int *cur_rec)
{
__be32 cur_cnid;
__be32 search_cnid;
if (bnode->tree->cnid == HFSPLUS_EXT_CNID) {
cur_cnid = fd->key->ext.cnid;
search_cnid = fd->search_key->ext.cnid;
} else if (bnode->tree->cnid == HFSPLUS_CAT_CNID) {
cur_cnid = fd->key->cat.parent;
search_cnid = fd->search_key->cat.parent;
} else if (bnode->tree->cnid == HFSPLUS_ATTR_CNID) {
cur_cnid = fd->key->attr.cnid;
search_cnid = fd->search_key->attr.cnid;
} else {
cur_cnid = 0; /* used-uninitialized warning */
search_cnid = 0;
BUG();
}
if (cur_cnid == search_cnid) {
(*end) = (*cur_rec);
if ((*begin) == (*end))
return 1;
} else {
if (be32_to_cpu(cur_cnid) < be32_to_cpu(search_cnid))
(*begin) = (*cur_rec) + 1;
else
(*end) = (*cur_rec) - 1;
}
return 0;
}
int hfs_find_rec_by_key(struct hfs_bnode *bnode,
struct hfs_find_data *fd,
int *begin,
int *end,
int *cur_rec)
{
int cmpval;
cmpval = bnode->tree->keycmp(fd->key, fd->search_key);
if (!cmpval) {
(*end) = (*cur_rec);
return 1;
}
if (cmpval < 0)
(*begin) = (*cur_rec) + 1;
else
*(end) = (*cur_rec) - 1;
return 0;
}
/* Find the record in bnode that best matches key (not greater than...)*/
int __hfs_brec_find(struct hfs_bnode *bnode, struct hfs_find_data *fd,
search_strategy_t rec_found)
{
u16 off, len, keylen;
int rec;
int b, e;
int res;
BUG_ON(!rec_found);
b = 0;
e = bnode->num_recs - 1;
res = -ENOENT;
do {
rec = (e + b) / 2;
len = hfs_brec_lenoff(bnode, rec, &off);
keylen = hfs_brec_keylen(bnode, rec);
if (keylen == 0) {
res = -EINVAL;
goto fail;
}
hfs_bnode_read(bnode, fd->key, off, keylen);
if (rec_found(bnode, fd, &b, &e, &rec)) {
res = 0;
goto done;
}
} while (b <= e);
if (rec != e && e >= 0) {
len = hfs_brec_lenoff(bnode, e, &off);
keylen = hfs_brec_keylen(bnode, e);
if (keylen == 0) {
res = -EINVAL;
goto fail;
}
hfs_bnode_read(bnode, fd->key, off, keylen);
}
done:
fd->record = e;
fd->keyoffset = off;
fd->keylength = keylen;
fd->entryoffset = off + keylen;
fd->entrylength = len - keylen;
fail:
return res;
}
/* Traverse a B*Tree from the root to a leaf finding best fit to key */
/* Return allocated copy of node found, set recnum to best record */
int hfs_brec_find(struct hfs_find_data *fd, search_strategy_t do_key_compare)
{
struct hfs_btree *tree;
struct hfs_bnode *bnode;
u32 nidx, parent;
__be32 data;
int height, res;
tree = fd->tree;
if (fd->bnode)
hfs_bnode_put(fd->bnode);
fd->bnode = NULL;
nidx = tree->root;
if (!nidx)
return -ENOENT;
height = tree->depth;
res = 0;
parent = 0;
for (;;) {
bnode = hfs_bnode_find(tree, nidx);
if (IS_ERR(bnode)) {
res = PTR_ERR(bnode);
bnode = NULL;
break;
}
if (bnode->height != height)
goto invalid;
if (bnode->type != (--height ? HFS_NODE_INDEX : HFS_NODE_LEAF))
goto invalid;
bnode->parent = parent;
res = __hfs_brec_find(bnode, fd, do_key_compare);
if (!height)
break;
if (fd->record < 0)
goto release;
parent = nidx;
hfs_bnode_read(bnode, &data, fd->entryoffset, 4);
nidx = be32_to_cpu(data);
hfs_bnode_put(bnode);
}
fd->bnode = bnode;
return res;
invalid:
pr_err("inconsistency in B*Tree (%d,%d,%d,%u,%u)\n",
height, bnode->height, bnode->type, nidx, parent);
res = -EIO;
release:
hfs_bnode_put(bnode);
return res;
}
int hfs_brec_read(struct hfs_find_data *fd, void *rec, int rec_len)
{
int res;
res = hfs_brec_find(fd, hfs_find_rec_by_key);
if (res)
return res;
if (fd->entrylength > rec_len)
return -EINVAL;
hfs_bnode_read(fd->bnode, rec, fd->entryoffset, fd->entrylength);
return 0;
}
int hfs_brec_goto(struct hfs_find_data *fd, int cnt)
{
struct hfs_btree *tree;
struct hfs_bnode *bnode;
int idx, res = 0;
u16 off, len, keylen;
bnode = fd->bnode;
tree = bnode->tree;
if (cnt < 0) {
cnt = -cnt;
while (cnt > fd->record) {
cnt -= fd->record + 1;
fd->record = bnode->num_recs - 1;
idx = bnode->prev;
if (!idx) {
res = -ENOENT;
goto out;
}
hfs_bnode_put(bnode);
bnode = hfs_bnode_find(tree, idx);
if (IS_ERR(bnode)) {
res = PTR_ERR(bnode);
bnode = NULL;
goto out;
}
}
fd->record -= cnt;
} else {
while (cnt >= bnode->num_recs - fd->record) {
cnt -= bnode->num_recs - fd->record;
fd->record = 0;
idx = bnode->next;
if (!idx) {
res = -ENOENT;
goto out;
}
hfs_bnode_put(bnode);
bnode = hfs_bnode_find(tree, idx);
if (IS_ERR(bnode)) {
res = PTR_ERR(bnode);
bnode = NULL;
goto out;
}
}
fd->record += cnt;
}
len = hfs_brec_lenoff(bnode, fd->record, &off);
keylen = hfs_brec_keylen(bnode, fd->record);
if (keylen == 0) {
res = -EINVAL;
goto out;
}
fd->keyoffset = off;
fd->keylength = keylen;
fd->entryoffset = off + keylen;
fd->entrylength = len - keylen;
hfs_bnode_read(bnode, fd->key, off, keylen);
out:
fd->bnode = bnode;
return res;
}