From: Alex Zhuravlev <bzzz@whamcloud.com>
Date: Mon, 16 Oct 2023 13:59:08 +0000 (+0300)
Subject: LU-16966 osd: take trunc_lock for fallocate
X-Git-Tag: 2.15.4-RC1~14
X-Git-Url: https://git.whamcloud.com/?a=commitdiff_plain;h=refs%2Fchanges%2F10%2F52710%2F2;p=fs%2Flustre-release.git

LU-16966 osd: take trunc_lock for fallocate

as fallocate may need few transactions (or transaction restarted)
we have to avoid any concurrent writes/truncates on this object
until fallocate supports 'restart-from-beginning' - first stop the
transaction, then release the lock, then repeat again (like
the write path does).

Lustre-change: https://review.whamcloud.com/52264
Lustre-commit: 51529fb57f85210e292a15c882cf25a4689ea77d

Signed-off-by: Alex Zhuravlev <bzzz@whamcloud.com>
Change-Id: I0bf38b1886fbf24656b45fe0f87fcbad2227672a
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/52710
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Arshad Hussain <arshad.hussain@aeoncomputing.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
---

diff --git a/lustre/osd-ldiskfs/osd_io.c b/lustre/osd-ldiskfs/osd_io.c
index 5a21f0f..f74862c 100644
--- a/lustre/osd-ldiskfs/osd_io.c
+++ b/lustre/osd-ldiskfs/osd_io.c
@@ -2262,33 +2262,39 @@ static int osd_declare_fallocate(const struct lu_env *env,
 	LASSERT(th);
 	LASSERT(inode);
 
-	if (mode & FALLOC_FL_PUNCH_HOLE) {
-		rc = osd_declare_inode_qid(env, i_uid_read(inode),
-					   i_gid_read(inode),
-					   i_projid_read(inode), 0, oh,
-					   osd_dt_obj(dt), NULL, OSD_QID_BLK);
-		if (rc == 0)
-			rc = osd_trunc_lock(osd_dt_obj(dt), oh, false);
-		RETURN(rc);
-	}
-
-	/* quota space for metadata blocks
-	 * approximate metadata estimate should be good enough.
-	 */
-	quota_space += PAGE_SIZE;
-	quota_space += depth * LDISKFS_BLOCK_SIZE(osd_sb(osd));
+	if ((mode & FALLOC_FL_PUNCH_HOLE) == 0) {
+		/* quota space for metadata blocks
+		 * approximate metadata estimate should be good enough.
+		 */
+		quota_space += PAGE_SIZE;
+		quota_space += depth * LDISKFS_BLOCK_SIZE(osd_sb(osd));
 
-	/* quota space should be reported in 1K blocks */
-	quota_space = toqb(quota_space) + toqb(end - start) +
-		      LDISKFS_META_TRANS_BLOCKS(inode->i_sb);
+		/* quota space should be reported in 1K blocks */
+		quota_space = toqb(quota_space) + toqb(end - start) +
+			LDISKFS_META_TRANS_BLOCKS(inode->i_sb);
 
-	/* We don't need to reserve credits for whole fallocate here.
-	 * We reserve space only for metadata. Fallocate credits are
-	 * extended as required
-	 */
+		/*
+		 * We don't need to reserve credits for whole fallocate here.
+		 * We reserve space only for metadata. Fallocate credits are
+		 * extended as required
+		 */
+	}
 	rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
 				   i_projid_read(inode), quota_space, oh,
 				   osd_dt_obj(dt), NULL, OSD_QID_BLK);
+	if (rc)
+		RETURN(rc);
+
+	/*
+	 * The both hole punch and allocation may need few transactions
+	 * to complete, so we have to avoid concurrent writes/truncates
+	 * as we can't release object lock from within ldiskfs.
+	 * Notice locking order: transaction start, then lock object
+	 * (don't confuse object lock dt_{read|write}_lock() with the
+	 * trunc lock.
+	 */
+	rc = osd_trunc_lock(osd_dt_obj(dt), oh, false);
+
 	RETURN(rc);
 }