Skip to contentSkip to conversation

ext4: cache all of an extent tree's leaf block upon reading

DISCUSSION Dave Jones 2 participants 4 messages Sep 5, 2013 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Dave Jones

author
#1 lore raw
Quoted 16 lines · Linux Kernel
Gitweb:     http://git.kernel.org/linus/;a=commit;h=107a7bd31ac003e42c0f966aa8e5b26947de6024
Commit:     107a7bd31ac003e42c0f966aa8e5b26947de6024
Parent:     3be78c73179c9347bdc0a92b2898063bd2300ff7
Author:     Theodore Ts'o <[email protected]>
AuthorDate: Fri Aug 16 21:23:41 2013 -0400
Committer:  Theodore Ts'o <[email protected]>
CommitDate: Fri Aug 16 21:23:41 2013 -0400

ext4: cache all of an extent tree's leaf block upon reading

+ * ext4_es_cache_extent() inserts information into the extent status
+ * tree if and only if there isn't information about the range in
+ * question already.
+ */

(quoted diff) +1 -0
+void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,

Review

Quoted 10 lines
(quoted diff) +10 -0
+			  ext4_lblk_t len, ext4_fsblk_t pblk,
+			  unsigned int status)
+{
+	struct extent_status *es;
+	struct extent_status newes;
+	ext4_lblk_t end = lblk + len - 1;
+
+	newes.es_lblk = lblk;
+	newes.es_len = len;
+	ext4_es_store_pblock(&newes, pblk);

ext4_es_store_pblock or's the pblk with the existing contents of the struct member. (albeit masked with ES_MASK)

Should there be a

	newes.es_pblk = 0;

up there too ?

It seems like if the stack happened to contain any of ES_WRITTEN | ES_UNWRITTEN | ES_DELAYED | ES_HOLE then it could leak through into the new extent status.

Dave

#2 lore raw
Quoted 16 lines · Dave Jones

Review

Quoted 11 lines
(quoted diff) +11 -0
+void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
+			  ext4_lblk_t len, ext4_fsblk_t pblk,
+			  unsigned int status)
+{
+	struct extent_status *es;
+	struct extent_status newes;
+	ext4_lblk_t end = lblk + len - 1;
+
+	newes.es_lblk = lblk;
+	newes.es_len = len;
+	ext4_es_store_pblock(&newes, pblk);

ext4_es_store_pblock or's the pblk with the existing contents of the struct member. (albeit masked with ES_MASK)

Quoted 5 lines

Should there be a

	newes.es_pblk = 0;

up there too ?

The next line after ext4_es_store_pblock() is:

        ext4_es_store_status(&newes, status);

This will set remaining ES_WRITTEN | ES_UNWRITTEN... bits.

So the only reason to add a line explicitly setting es_pblk to zero would be to suppress a warning from some insufficiently smart static code analysis tool. I didn't see a warning from gcc, but it's possible that this is something which is causing Coverity or some other code scanner heartburn.

Cheers,

- Ted

author
#3 lore raw
Quoted 13 lines · Theodore Ts'o
Quoted 8 lines

ext4_es_store_pblock or's the pblk with the existing contents of the struct member. (albeit masked with ES_MASK)

Should there be a

	newes.es_pblk = 0;

up there too ?

The next line after ext4_es_store_pblock() is:

        ext4_es_store_status(&newes, status);
Quoted 7 lines

This will set remaining ES_WRITTEN | ES_UNWRITTEN... bits.

So the only reason to add a line explicitly setting es_pblk to zero would be to suppress a warning from some insufficiently smart static code analysis tool. I didn't see a warning from gcc, but it's possible that this is something which is causing Coverity or some other code scanner heartburn.

Yep, that's what picked it up. I'll add a 'not a bug' annotation to stop it getting flagged again. This was the only ext* issue that Coverity picked up from yesterdays merge btw, which I guess is good news ;)

Dave

#4 lore raw
Quoted 9 lines · Dave Jones
Quoted 5 lines

So the only reason to add a line explicitly setting es_pblk to zero would be to suppress a warning from some insufficiently smart static code analysis tool. I didn't see a warning from gcc, but it's possible that this is something which is causing Coverity or some other code scanner heartburn.

Yep, that's what picked it up. I'll add a 'not a bug' annotation to stop it getting flagged again. This was the only ext* issue that Coverity picked up from yesterdays merge btw, which I guess is good news ;)

Indeed.

Hmm... I could add a new inline function "ext4_es_store_pblock_status()" which sets both parts of the es_pblk word at once, and which doesn't depend looking at its original value at all. I doubt we would never measure a difference in performance, but in theory it would be more efficient. And if it eliminates a potential static code analysis complaint, maybe the two justifications is good enough to add the new function.

Thanks for checking the coverity results!!

- Ted