LU-14098 obdclass: try to skip corrupted llog records

[fs/lustre-release.git] / lustre / obdclass / llog_osd.c

diff --git a/lustre/obdclass/llog_osd.c b/lustre/obdclass/llog_osd.c
index 798451f..63abf66 100644 (file)
--- a/lustre/obdclass/llog_osd.c
+++ b/lustre/obdclass/llog_osd.c
@@ -437,7 +437,7 @@ static int llog_osd_write_rec(const struct lu_env *env,
                /* llog can be empty only when first record is being written */
                LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
 
-               if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) {
+               if (!test_bit_le(idx, LLOG_HDR_BITMAP(llh))) {
                        CERROR("%s: modify unset record %u\n",
                               o->do_lu.lo_dev->ld_obd->obd_name, idx);
                        RETURN(-ENOENT);
@@ -611,9 +611,10 @@ static int llog_osd_write_rec(const struct lu_env *env,
        /* the lgh_hdr_mutex protects llog header data from concurrent
         * update/cancel, the llh_count and llh_bitmap are protected */
        mutex_lock(&loghandle->lgh_hdr_mutex);
-       if (ext2_set_bit(index, LLOG_HDR_BITMAP(llh))) {
-               CERROR("%s: index %u already set in log bitmap\n",
-                      o->do_lu.lo_dev->ld_obd->obd_name, index);
+       if (__test_and_set_bit_le(index, LLOG_HDR_BITMAP(llh))) {
+               CERROR("%s: index %u already set in llog bitmap "DFID"\n",
+                      o->do_lu.lo_dev->ld_obd->obd_name, index,
+                      PFID(lu_object_fid(&o->do_lu)));
                mutex_unlock(&loghandle->lgh_hdr_mutex);
                LBUG(); /* should never happen */
        }
@@ -722,7 +723,7 @@ out_unlock:
 out:
        /* cleanup llog for error case */
        mutex_lock(&loghandle->lgh_hdr_mutex);
-       ext2_clear_bit(index, LLOG_HDR_BITMAP(llh));
+       clear_bit_le(index, LLOG_HDR_BITMAP(llh));
        llh->llh_count--;
        mutex_unlock(&loghandle->lgh_hdr_mutex);
 
@@ -910,8 +911,9 @@ static int llog_osd_next_block(const struct lu_env *env,
        if (rc)
                GOTO(out, rc);
 
-       CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off"
-              "%llu), size %llu\n", next_idx, *cur_idx,
+       CDEBUG(D_OTHER,
+              "looking for log index %u (cur idx %u off %llu), size %llu\n",
+              next_idx, *cur_idx,
               *cur_offset, lgi->lgi_attr.la_size);
 
        while (*cur_offset < lgi->lgi_attr.la_size) {
@@ -966,9 +968,25 @@ static int llog_osd_next_block(const struct lu_env *env,
                rec = buf;
                if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
                        lustre_swab_llog_rec(rec);
-
                tail = (struct llog_rec_tail *)((char *)buf + rc -
                                                sizeof(struct llog_rec_tail));
+
+               if (llog_verify_record(loghandle, rec)) {
+                       /*
+                        * the block seems corrupted. make a pad record so the
+                        * caller can skip the block and try with the next one
+                        */
+                       rec->lrh_len = rc;
+                       rec->lrh_index = next_idx;
+                       rec->lrh_type = LLOG_PAD_MAGIC;
+
+                       tail = rec_tail(rec);
+                       tail->lrt_len = rc;
+                       tail->lrt_index = next_idx;
+
+                       GOTO(out, rc = 0);
+               }
+
                /* get the last record in block */
                last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
                                                   tail->lrt_len);
@@ -1005,7 +1023,7 @@ static int llog_osd_next_block(const struct lu_env *env,
 
                /* sanity check that the start of the new buffer is no farther
                 * than the record that we wanted.  This shouldn't happen. */
-               if (rec->lrh_index > next_idx) {
+               if (next_idx && rec->lrh_index > next_idx) {
                        if (!force_mini_rec && next_idx > last_idx)
                                goto retry;
 
@@ -1246,8 +1264,7 @@ static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
                } else {
                        /* If logid == NULL, then it means the caller needs
                         * to allocate new FID (llog_cat_declare_add_rec()). */
-                       rc = obd_fid_alloc(env, ctxt->loc_exp,
-                                          &lgi->lgi_fid, NULL);
+                       rc = dt_fid_alloc(env, dt, &lgi->lgi_fid, NULL, NULL);
                        if (rc < 0)
                                RETURN(rc);
                        rc = 0;
@@ -1308,7 +1325,12 @@ generate:
                        GOTO(out, rc);
                new_id = true;
        }
-
+       if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
+           cfs_fail_val == 1) {
+               cfs_fail_val = 2;
+               OBD_RACE(OBD_FAIL_MDS_LLOG_UMOUNT_RACE);
+               msleep(MSEC_PER_SEC);
+       }
        o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
        if (IS_ERR(o))
                GOTO(out_name, rc = PTR_ERR(o));
@@ -2009,7 +2031,7 @@ int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
                if (IS_ERR(th))
                        GOTO(out, rc = PTR_ERR(th));
 
-               lgi->lgi_attr.la_valid = LA_MODE;
+               lgi->lgi_attr.la_valid = LA_MODE | LA_TYPE;
                lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
                lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);