From: Gao Xiang gaoxiang25@huawei.com
Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).
After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().
Let's fix it now.
[1] https://lore.kernel.org/r/1163995781.68824.1566084358245.JavaMail.zimbra@nod...
Reported-by: Richard Weinberger richard@nod.at Fixes: 3aa8ec716e52 ("staging: erofs: add directory operations") Cc: stable@vger.kernel.org # 4.19+ Reviewed-by: Chao Yu yuchao0@huawei.com Signed-off-by: Gao Xiang gaoxiang25@huawei.com --- changelog from v3: - kill message when memory allocation fails as suggested by Matthew;
[RESEND] --> add the missing v3 version in subject, no logic change.
changelog from v2: - transform EIO to EFSCORRUPTED as suggested by Matthew;
changelog from v1: - fix the incorrect external link in commit message.
This patch is based on staging-testing tree and https://lore.kernel.org/r/20190817082313.21040-1-hsiangkao@aol.com/ can still be properly applied after this patch.
drivers/staging/erofs/dir.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/erofs/dir.c b/drivers/staging/erofs/dir.c index 5f38382637e6..77ef856df9f3 100644 --- a/drivers/staging/erofs/dir.c +++ b/drivers/staging/erofs/dir.c @@ -82,8 +82,15 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx) unsigned int nameoff, maxsize;
dentry_page = read_mapping_page(mapping, i, NULL); - if (IS_ERR(dentry_page)) - continue; + if (dentry_page == ERR_PTR(-ENOMEM)) { + err = -ENOMEM; + break; + } else if (IS_ERR(dentry_page)) { + errln("fail to readdir of logical block %u of nid %llu", + i, EROFS_V(dir)->nid); + err = -EFSCORRUPTED; + break; + }
de = (struct erofs_dirent *)kmap(dentry_page);