[LU-5770] wrong tx credit calculations in mdd_declare_links_add()

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.6.0, Lustre 2.5.1
• Fix Version/s: Lustre 2.8.0
• Type: Bug
• Priority: Critical
• Reporter: Alexander Zarochentsev
• Assignee: Alex Zhuravlev
• Resolution: Fixed
• Labels: patch
• Severity: 3

Description

looking for a reason of

 kernel BUG at fs/jbd2/transaction.c:1033! 

in sanityN test 12, I ran a simple debugging code

diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c
index 0da56c0..59efd5b 100644
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -2988,6 +2988,8 @@ static int osd_declare_xattr_set(const struct lu_env *env,
                        credits *= (buf->lb_len + sb->s_blocksize - 1) >>
                                        sb->s_blocksize_bits;
                }
+               if (buf) 
+                       printk("osd_declare_xattr_set: buflen = %ld\n", buf->lb_len);
                /*
                 * xattr set may involve inode quota change, reserve credits for
                 * dquot_initialize()
@@ -3071,6 +3073,8 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
                if (rc != 0)
                        RETURN(rc);
        }
+       if (buf) 
+               printk("osd_xattr_set: buflen = %ld\n", buf->lb_len);
 
        rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
                               fs_flags);

got the following output:

osd_xattr_set: buflen = 256280
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256310
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256340
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256370
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256400
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256430
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256460
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256490
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256520
osd_declare_xattr_set: buflen = 4096
osd_xattr_set: buflen = 256550

from

REFORMAT=y MDSSIZE=400000 ONLY=12 LDISKFS_MKFS_OPTS="-O large_xattr" sh sanityn.sh

mdd_declare_links_add() always uses DEFAULT_LINKEA_SIZE (4096) because linkea_data not yet initialised at the moment of mdd_declare_links_add() call. With really large LinkEA the difference between allocated and used transaction credits can cause that kernel BUG() in jbd2 code.

Issue Links

• is related to: LU-7332 LustreError: 201113:0:(osd_internal.h...
• is related to: LU-5794 improve credit checks in osd-ldiskfs

Comments

[Alexander Zarochentsev]

patch http://review.whamcloud.com/#/c/12412/

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/12412/
Subject: LU-5770 osd: find bufsize in declare_xattr_set
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: f6e7cc6d3a640e02192096584f4c6625b99e4678

[Joseph Gmitter (Inactive)]

Landed for 2.8