From: Jan Kara <jack@suse.cz>
To: 常凤楠 <changfengnan@hikvision.com>
Cc: Jan Kara <jack@suse.cz>, changfengnan <changfengnan@qq.com>,
"adilger@dilger.ca" <adilger@dilger.ca>,
"darrick.wong@oracle.com" <darrick.wong@oracle.com>,
"jack@suse.com" <jack@suse.com>,
"linux-ext4@vger.kernel.org" <linux-ext4@vger.kernel.org>,
"tytso@mit.edu" <tytso@mit.edu>
Subject: Re: 答复: 答复: [PATCH] jbd2: avoid transaction reuse after reformatting
Date: Fri, 18 Sep 2020 15:02:52 +0200 [thread overview]
Message-ID: <20200918130252.GG18920@quack2.suse.cz> (raw)
In-Reply-To: <6a08086d98c64f97bcaed1edd38861f6@hikvision.com>
Hello,
On Fri 18-09-20 01:49:09, 常凤楠 wrote:
> Sorry about my mailer, the patch is in the attachment.
Thanks for the patch. Functionally the patch looks mostly OK now. The only
concern I have is that it handles checksum failures only in
JBD2_DESCRIPTOR_BLOCK. This is the most likely case but it could also
happen that JBD2_REVOKE_BLOCK or JBD2_COMMIT_BLOCK is the first one you see
with mismatching checksum. So I think you need to handle these cases as
well. I think your ri_commit_block logic below is an attempt to deal with
these cases (but it's difficult to be sure because of complete lack of
comments) but it is not reliable. A valid transaction can begin both with a
descriptor or with a revoke block.
A few other comments mostly about coding style below:
diff --git a/fs/jbd2/recovery.c b/fs/jbd2/recovery.c
index a4967b27ffb6..f7702e14077f 100644
--- a/fs/jbd2/recovery.c
+++ b/fs/jbd2/recovery.c
@@ -417,7 +417,7 @@ static int do_one_pass(journal_t *journal,
struct recovery_info *info, enum passtype pass)
{
unsigned int first_commit_ID, next_commit_ID;
- unsigned long next_log_block;
+ unsigned long next_log_block, ri_commit_block = 0;
int err, success = 0;
journal_superblock_t * sb;
journal_header_t * tmp;
@@ -428,7 +428,9 @@ static int do_one_pass(journal_t *journal,
__u32 crc32_sum = ~0; /* Transactional Checksums */
int descr_csum_size = 0;
int block_error = 0;
-
+ bool need_check_commit_time = false;
+ __be64 last_trans_commit_time;
All variable names in this function seem to be indented by one more
column. Please keep the indentation.
+
This empty line has whitespace on it. Please delete.
/*
* First thing is to establish what we expect to find in the log
* (in terms of transaction IDs), and where (in terms of log
@@ -514,18 +516,18 @@ static int do_one_pass(journal_t *journal,
switch(blocktype) {
case JBD2_DESCRIPTOR_BLOCK:
/* Verify checksum first */
+ if(pass == PASS_SCAN)
^ Coding style requires space before opening (.
You have this problem at multiple places.
+ ri_commit_block = 0;
+
if (jbd2_journal_has_csum_v2or3(journal))
descr_csum_size =
sizeof(struct jbd2_journal_block_tail);
if (descr_csum_size > 0 &&
!jbd2_descriptor_block_csum_verify(journal,
bh->b_data)) {
- printk(KERN_ERR "JBD2: Invalid checksum "
- "recovering block %lu in log\n",
- next_log_block);
- err = -EFSBADCRC;
- brelse(bh);
- goto failed;
+ need_check_commit_time = true;
+ jbd_debug(1, "invalid descriptor block found in %lu, continue recovery first.\n",next_log_block);
+
}
/* If it is a valid descriptor block, replay it
@@ -535,6 +537,7 @@ static int do_one_pass(journal_t *journal,
if (pass != PASS_REPLAY) {
if (pass == PASS_SCAN &&
jbd2_has_feature_checksum(journal) &&
+ !need_check_commit_time &&
!info->end_transaction) {
if (calc_chksums(journal, bh,
&next_log_block,
@@ -688,6 +691,36 @@ static int do_one_pass(journal_t *journal,
* are present verify them in PASS_SCAN; else not
* much to do other than move on to the next sequence
* number. */
+ if(pass == PASS_SCAN) {
+ struct commit_header *cbh =
+ (struct commit_header *)bh->b_data;
+ if(need_check_commit_time) {
+ __be64 commit_time = be64_to_cpu(cbh->h_commit_sec);
+ if(commit_time >= last_trans_commit_time) {
+ printk(KERN_ERR "JBD2: Invalid checksum found in log, %d\n",
+ next_commit_ID);
+ err = -EFSBADCRC;
+ brelse(bh);
+ goto failed;
+ }
+ else
+ {
Coding style requires to put opening { on the same line as 'else'. Like:
else {
+ /*it's not belong to same journal, just end this recovery with success*/
+ jbd_debug(1, "JBD2: Invalid checksum found in block in log, but not same journal %d\n",
+ next_commit_ID);
+ err = 0;
+ brelse(bh);
+ goto done;
+ }
+ }
+ if(ri_commit_block) {
+ jbd_debug(1, "invalid commit block found in %lu, stop here.\n",next_log_block);
+ brelse(bh);
+ goto done;
+ }
+ ri_commit_block = next_log_block;
Why does the ri_commit_block logic exist? I don't see it bringing any
benefit...
+ last_trans_commit_time = be64_to_cpu(cbh->h_commit_sec);
+ }
if (pass == PASS_SCAN &&
jbd2_has_feature_checksum(journal)) {
int chksum_err, chksum_seen;
@@ -755,6 +788,12 @@ static int do_one_pass(journal_t *journal,
continue;
case JBD2_REVOKE_BLOCK:
+ if (pass == PASS_SCAN &&
+ ri_commit_block) {
+ jbd_debug(1, "invalid revoke block found in %lu, stop here.\n",next_log_block);
+ brelse(bh);
+ goto done;
+ }
This is wrong. A valid transaction can start with a revoke block...
/* If we aren't in the REVOKE pass, then we can
* just skip over this block. */
if (pass != PASS_REVOKE) {
Honza
--
Jan Kara <jack@suse.com>
SUSE Labs, CR
next prev parent reply other threads:[~2020-09-18 13:02 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <tencent_2341B065211F204FA07C3ADDA1AE07706405@qq.com>
2020-09-10 12:06 ` [PATCH] jbd2: avoid transaction reuse after reformatting 常凤楠
2020-09-10 18:45 ` Andreas Dilger
2020-09-11 10:06 ` Jan Kara
2020-09-14 11:50 ` 答复: " 常凤楠
2020-09-17 10:44 ` Jan Kara
2020-09-18 1:49 ` 答复: " 常凤楠
2020-09-18 13:02 ` Jan Kara [this message]
2020-09-23 6:29 ` 答复: " 常凤楠
2020-09-23 12:24 ` Jan Kara
2020-10-03 4:24 ` Theodore Y. Ts'o
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200918130252.GG18920@quack2.suse.cz \
--to=jack@suse.cz \
--cc=adilger@dilger.ca \
--cc=changfengnan@hikvision.com \
--cc=changfengnan@qq.com \
--cc=darrick.wong@oracle.com \
--cc=jack@suse.com \
--cc=linux-ext4@vger.kernel.org \
--cc=tytso@mit.edu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).