*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/mdd/mdd_device.c
*
{
int rc = -EINVAL;
const char *dev;
+
ENTRY;
/* LU-8040 Set defaults here, before values configs */
mdd->mdd_cl.mc_flags = 0; /* off by default */
- mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
+ /* per-server mask is set via parameters if needed */
+ mdd->mdd_cl.mc_proc_mask = CHANGELOG_MINMASK;
+ /* current mask is calculated from mask above and users masks */
+ mdd->mdd_cl.mc_current_mask = CHANGELOG_MINMASK;
mdd->mdd_cl.mc_deniednext = 60; /* 60 secs by default */
dev = lustre_cfg_string(lcfg, 0);
DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
changelog_rec_name(&rec->cr), PFID(&llh->lgh_id.lgl_oi.oi_fid));
-
+ spin_lock(&mdd->mdd_cl.mc_lock);
mdd->mdd_cl.mc_index = rec->cr.cr_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
return LLOG_PROC_BREAK;
}
+char *mdd_chlg_username(struct llog_changelog_user_rec2 *rec, char *buf,
+ size_t len)
+{
+ if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
+ rec->cur_name[0])
+ snprintf(buf, len, "%s%u-%s", CHANGELOG_USER_PREFIX,
+ rec->cur_id, rec->cur_name);
+ else
+ snprintf(buf, len, "%s%u", CHANGELOG_USER_PREFIX, rec->cur_id);
+ return buf;
+}
+
+__u32 mdd_chlg_usermask(struct llog_changelog_user_rec2 *rec)
+{
+ return rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 ?
+ rec->cur_mask : 0;
+}
+
static int changelog_user_init_cb(const struct lu_env *env,
struct llog_handle *llh,
struct llog_rec_hdr *hdr, void *data)
{
- struct mdd_device *mdd = (struct mdd_device *)data;
- struct llog_changelog_user_rec *rec =
- (struct llog_changelog_user_rec *)hdr;
+ struct mdd_device *mdd = data;
+ struct llog_changelog_user_rec2 *rec;
+ char user_name[CHANGELOG_USER_NAMELEN_FULL];
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
- LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
- CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec=%llu"
- " in log "DFID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
- rec->cur_id, rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
+ rec = container_of(hdr, typeof(*rec), cur_hdr);
+ if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
+ rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
+ CWARN("%s: unknown user type %x at index %u in log "DFID"\n",
+ mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
+ rec->cur_hdr.lrh_type, PFID(&llh->lgh_id.lgl_oi.oi_fid));
+
+ return 0;
+ }
+
+ CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
+ mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
+ sizeof(user_name)),
+ hdr->lrh_index, rec->cur_hdr.lrh_index, rec->cur_endrec,
+ PFID(&llh->lgh_id.lgl_oi.oi_fid));
spin_lock(&mdd->mdd_cl.mc_user_lock);
mdd->mdd_cl.mc_lastuser = rec->cur_id;
mdd->mdd_cl.mc_users++;
+ if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
+ mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
+ else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
+ mdd->mdd_cl.mc_current_mask |= CHANGELOG_DEFMASK;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ spin_lock(&mdd->mdd_cl.mc_lock);
if (rec->cur_endrec > mdd->mdd_cl.mc_index)
mdd->mdd_cl.mc_index = rec->cur_endrec;
- spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ spin_unlock(&mdd->mdd_cl.mc_lock);
return LLOG_PROC_BREAK;
}
struct changelog_orphan_data {
- __u64 index;
- struct mdd_device *mdd;
+ __u64 clod_index;
+ struct mdd_device *clod_mdd;
};
/* find oldest changelog record index */
struct llog_handle *llh,
struct llog_rec_hdr *hdr, void *data)
{
- struct mdd_device *mdd = ((struct changelog_orphan_data *)data)->mdd;
- struct llog_changelog_rec *rec = container_of(hdr,
- struct llog_changelog_rec,
- cr_hdr);
+ struct changelog_orphan_data *clod = data;
+ struct mdd_device *mdd = clod->clod_mdd;
+ struct llog_changelog_rec *rec;
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
+ rec = container_of(hdr, typeof(*rec), cr_hdr);
if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
CWARN("%s: invalid record at index %d in log "DFID"\n",
mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
return 0;
}
- CDEBUG(D_INFO, "%s: seeing record at index %d/%d/%llu t=%x %.*s in log "
- DFID"\n", mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
+ CDEBUG(D_INFO,
+ "%s: record at index %d/%d/%llu t=%x %.*s in log "DFID"\n",
+ mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
PFID(&llh->lgh_id.lgl_oi.oi_fid));
- ((struct changelog_orphan_data *)data)->index = rec->cr.cr_index;
+ clod->clod_index = rec->cr.cr_index;
+
return LLOG_PROC_BREAK;
}
struct llog_handle *llh,
struct llog_rec_hdr *hdr, void *data)
{
- struct mdd_device *mdd = ((struct changelog_orphan_data *)data)->mdd;
- struct llog_changelog_user_rec *rec = container_of(hdr,
- struct llog_changelog_user_rec,
- cur_hdr);
+ struct changelog_orphan_data *clod = data;
+ struct mdd_device *mdd = clod->clod_mdd;
+ struct llog_changelog_user_rec2 *rec;
+ char user_name[CHANGELOG_USER_NAMELEN_FULL];
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
- if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC) {
- CWARN("%s: invalid user at index %d in log "DFID"\n",
+ rec = container_of(hdr, typeof(*rec), cur_hdr);
+ if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
+ rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
+ CWARN("%s: unknown user type %u at index %u in log "DFID"\n",
mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
- PFID(&llh->lgh_id.lgl_oi.oi_fid));
+ rec->cur_hdr.lrh_type, PFID(&llh->lgh_id.lgl_oi.oi_fid));
/* try to find some next valid record and thus allow to recover
* from a corrupted LLOG, instead to assert and force a crash
*/
return 0;
}
- CDEBUG(D_INFO, "%s: seeing user at index %d/%d id=%d endrec=%llu in "
- "log "DFID"\n", mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
- rec->cur_hdr.lrh_index, rec->cur_id, rec->cur_endrec,
- PFID(&llh->lgh_id.lgl_oi.oi_fid));
+ CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
+ mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
+ sizeof(user_name)),
+ hdr->lrh_index, rec->cur_hdr.lrh_index,
+ rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
- if (((struct changelog_orphan_data *)data)->index == 0 ||
- rec->cur_endrec < ((struct changelog_orphan_data *)data)->index)
- ((struct changelog_orphan_data *)data)->index = rec->cur_endrec;
+ clod->clod_index = min_t(__u64, clod->clod_index, rec->cur_endrec);
return 0;
}
struct llog_rec_hdr *hdr, void *data)
{
struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
- struct llog_cookie cookie;
struct changelog_cancel_cookie *cl_cookie =
(struct changelog_cancel_cookie *)data;
- int rc;
ENTRY;
/* records are in order, so we're done */
RETURN(LLOG_PROC_BREAK);
- cookie.lgc_lgl = llh->lgh_id;
- cookie.lgc_index = hdr->lrh_index;
+ if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
+ if (cfs_fail_val == 0)
+ cfs_fail_val = hdr->lrh_index;
+ if (cfs_fail_val == hdr->lrh_index)
+ OBD_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
+ }
+
+ /* Records folow one by one, cr_index++. We could calculate the
+ * last cr_index at this plain llog. And if it less then cookie endrec
+ * cancel the whole file.
+ */
+ if ((LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr) - hdr->lrh_index +
+ rec->cr.cr_index) < cl_cookie->endrec) {
+ int rc;
+
+ if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
+ if (cfs_fail_val == 0) {
+ cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
+ OBD_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
+ }
+ }
+ rc = llog_destroy(env, llh);
+ if (!rc) {
+ CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
+ PFID(&llh->lgh_id.lgl_oi.oi_fid));
+ RETURN(LLOG_DEL_PLAIN);
+ }
+ }
/* cancel them one at a time. I suppose we could store up the cookies
* and cancel them all at once; probably more efficient, but this is
* done as a user call, so who cares... */
- rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
- &cookie);
- RETURN(rc < 0 ? rc : 0);
+
+ RETURN(LLOG_DEL_RECORD);
}
static int llog_changelog_cancel(const struct lu_env *env,
static int mdd_changelog_llog_init(const struct lu_env *env,
struct mdd_device *mdd)
{
- struct obd_device *obd = mdd2obd_dev(mdd);
- struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
- struct changelog_orphan_data changelog_orphan = { .index = 0,
- .mdd = mdd },
- user_orphan = { .index = 0,
- .mdd = mdd };
- int rc;
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
+ struct changelog_orphan_data clod = {
+ .clod_mdd = mdd,
+ .clod_index = -1,
+ }, user_orphan = {
+ .clod_mdd = mdd,
+ .clod_index = -1,
+ };
+ int rc;
ENTRY;
GOTO(out_uclose, rc);
}
+ /* Finally apply per-server mask */
+ mdd->mdd_cl.mc_current_mask |= mdd->mdd_cl.mc_proc_mask;
+
/* If we have registered users, assume we want changelogs on */
if (mdd->mdd_cl.mc_lastuser > 0) {
rc = mdd_changelog_on(env, mdd);
* XXX we may need to run end of purge as a separate thread
*/
rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
- &changelog_orphan, 0, 0);
+ &clod, 0, 0);
if (rc < 0) {
CERROR("%s: changelog detect orphan failed: rc = %d\n",
obd->obd_name, rc);
obd->obd_name, rc);
GOTO(out_uclose, rc);
}
- if (unlikely(changelog_orphan.index < user_orphan.index)) {
+ if (unlikely(clod.clod_index < user_orphan.clod_index)) {
struct changelog_cancel_cookie cl_cookie = {
- .endrec = user_orphan.index,
+ .endrec = user_orphan.clod_index,
.mdd = mdd,
};
CWARN("%s : orphan changelog records found, starting from "
"index %llu to index %llu, being cleared now\n",
- obd->obd_name, changelog_orphan.index, user_orphan.index);
+ obd->obd_name, clod.clod_index, user_orphan.clod_index);
/* XXX we may need to run end of purge as a separate thread */
rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
struct obd_device *obd = mdd2obd_dev(mdd);
struct llog_ctxt *ctxt;
- mdd->mdd_cl.mc_flags = 0;
+ if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
+ return;
+ mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
again:
/* stop GC-thread if running */
ENTRY;
- if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
+ if (mdd->mdd_cl.mc_current_mask & BIT(CL_MARK)) {
mdd->mdd_cl.mc_starttime = ktime_get();
RETURN(0);
}
reclen = llog_data_len(sizeof(*rec) + len);
- buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
if (buf->lb_buf == NULL)
RETURN(-ENOMEM);
rec = buf->lb_buf;
name++;
sscanf(name, SFID, RFID(f));
- if (!fid_is_sane(f)) {
- CWARN("%s: Trying to lookup invalid FID [%s] in %s/%s, FID "
- "format should be "DFID"\n", mdd2obd_dev(mdd)->obd_name,
- lname->ln_name, dot_lustre_name, mdd_obf_dir_name,
- (__u64)FID_SEQ_NORMAL, 1, 0);
- GOTO(out, rc = -EINVAL);
- }
+ if (!fid_is_sane(f))
+ GOTO(out, rc = -ENOENT);
if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
- !fid_seq_is_dot(f->f_seq)) {
- CWARN("%s: Trying to lookup invalid FID "DFID" in %s/%s, "
- "sequence should be >= %#llx or within [%#llx,"
- "%#llx].\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
- dot_lustre_name, mdd_obf_dir_name, (__u64)FID_SEQ_NORMAL,
- (__u64)FID_SEQ_IGIF, (__u64)FID_SEQ_IGIF_MAX);
- GOTO(out, rc = -EINVAL);
- }
+ !fid_seq_is_dot(f->f_seq))
+ GOTO(out, rc = -ENOENT);
/* Check if object with this fid exists */
child = mdd_object_find(env, mdd, f);
RETURN(-EPERM);
}
-static struct md_dir_operations mdd_obf_dir_ops = {
+static const struct md_dir_operations mdd_obf_dir_ops = {
.mdo_lookup = obf_lookup,
.mdo_create = mdd_obf_create,
.mdo_rename = mdd_dummy_rename,
.mdo_unlink = mdd_dummy_unlink
};
-static struct md_dir_operations mdd_lpf_dir_ops = {
+static const struct md_dir_operations mdd_lpf_dir_ops = {
.mdo_lookup = mdd_lookup,
.mdo_create = mdd_dummy_create,
.mdo_rename = mdd_dummy_rename,
LASSERT(obj != NULL);
mdo = lu2md(obj);
} else {
- mdo = ERR_PTR(PTR_ERR(obj));
+ mdo = ERR_CAST(obj);
}
return mdo;
}
struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
struct nm_config_file *nodemap_config;
struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
+ struct dt_object *root = NULL;
+ struct thandle *th = NULL;
struct lu_fid fid;
int rc;
GOTO(out_los, rc);
}
+ /* store a default directory layout on the root directory if
+ * it doesn't already exist to improve MDT space balance.
+ */
+ root = dt_locate(env, mdd->mdd_bottom, &fid);
+ if (unlikely(IS_ERR(root)))
+ GOTO(out_los, rc = PTR_ERR(root));
+
+ rc = dt_xattr_get(env, root, &LU_BUF_NULL,
+ XATTR_NAME_DEFAULT_LMV);
+ if (rc == -ENODATA) {
+ struct lu_buf buf;
+ struct lmv_user_md lmv_default = {
+ .lum_magic = LMV_USER_MAGIC,
+ .lum_stripe_count = 1,
+ .lum_stripe_offset = LMV_OFFSET_DEFAULT,
+ .lum_max_inherit = LMV_INHERIT_UNLIMITED,
+ .lum_max_inherit_rr = LMV_INHERIT_RR_ROOT,
+ };
+
+ th = dt_trans_create(env, mdd->mdd_bottom);
+ if (IS_ERR(th))
+ GOTO(out_root_put, rc = PTR_ERR(th));
+
+ buf.lb_buf = &lmv_default;
+ buf.lb_len = sizeof(lmv_default);
+ rc = dt_declare_xattr_set(env, root, &buf,
+ XATTR_NAME_DEFAULT_LMV, 0,
+ th);
+ if (rc)
+ GOTO(out_trans_stop, rc);
+
+ rc = dt_trans_start_local(env, mdd->mdd_bottom, th);
+ if (rc)
+ GOTO(out_trans_stop, rc);
+
+ rc = dt_xattr_set(env, root, &buf,
+ XATTR_NAME_DEFAULT_LMV, 0, th);
+ if (rc)
+ GOTO(out_trans_stop, rc);
+
+ dt_trans_stop(env, mdd->mdd_bottom, th);
+ th = NULL;
+ } else if (rc < 0 && rc != -ERANGE) {
+ CERROR("%s: get default LMV of root failed: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
+
+ GOTO(out_root_put, rc);
+ }
+
+ dt_object_put(env, root);
+ root = NULL;
+
mdd->mdd_root_fid = fid;
rc = mdd_dot_lustre_setup(env, mdd);
if (rc != 0) {
out_dot:
if (mdd_seq_site(mdd)->ss_node_id == 0)
mdd_dot_lustre_cleanup(env, mdd);
+out_trans_stop:
+ if (th != NULL)
+ dt_trans_stop(env, mdd->mdd_bottom, th);
+out_root_put:
+ if (root != NULL)
+ dt_object_put(env, root);
out_los:
local_oid_storage_fini(env, mdd->mdd_los);
mdd->mdd_los = NULL;
return rc;
}
+/**
+ * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
+ *
+ * Find corresponding device by passed parent and name, and allocate FID from
+ * there.
+ *
+ * see include/lu_object.h for the details.
+ */
+static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
+ struct lu_fid *fid, struct lu_object *parent,
+ const struct lu_name *name)
+{
+ struct mdd_device *mdd = lu2mdd_dev(d);
+ struct lu_object *o = lu_object_next(parent);
+
+ return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
+}
+
const struct lu_device_operations mdd_lu_ops = {
.ldo_object_alloc = mdd_object_alloc,
.ldo_process_config = mdd_process_config,
.ldo_recovery_complete = mdd_recovery_complete,
.ldo_prepare = mdd_prepare,
+ .ldo_fid_alloc = mdd_fid_alloc,
};
static int mdd_root_get(const struct lu_env *env,
.o_set_info_async = mdd_obd_set_info_async,
};
+struct mdd_changelog_name_check_data {
+ const char *mcnc_name;
+ __u32 mcnc_id;
+};
+
+/**
+ * changelog_recalc_mask callback
+ *
+ * Is is called per each registered user and calculates combined mask of
+ * all registered users.
+ */
+static int mdd_changelog_name_check_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct llog_changelog_user_rec2 *rec;
+ struct mdd_changelog_name_check_data *mcnc = data;
+
+ rec = container_of(hdr, typeof(*rec), cur_hdr);
+ if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
+ !strncmp(rec->cur_name, mcnc->mcnc_name, sizeof(rec->cur_name))) {
+ mcnc->mcnc_id = rec->cur_id;
+ return -EEXIST;
+ }
+ return 0;
+}
+
+static int mdd_changelog_name_check(const struct lu_env *env,
+ struct llog_ctxt *ctxt,
+ struct mdd_device *mdd, const char *name)
+{
+ struct mdd_changelog_name_check_data mcnc = { .mcnc_name = name, };
+ int chr = 0;
+ int rc;
+
+ ENTRY;
+
+ /* first symbol is a letter */
+ if (!isalpha(name[0])) {
+ rc = -EINVAL;
+ CERROR("%s: first char '%c' in '%s' is not letter: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, name[0], name, rc);
+ RETURN(rc);
+ }
+
+ /* name is valid: contains letters, numbers and '-', '_' only */
+ while (name[++chr]) {
+ if (!(isalnum(name[chr]) || name[chr] == '_' ||
+ name[chr] == '-')) {
+ rc = -EINVAL;
+ CERROR("%s: wrong char '%c' in name '%s': rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, name[chr], name, rc);
+ RETURN(rc);
+ }
+ }
+
+ if (chr > CHANGELOG_USER_NAMELEN) {
+ rc = -ENAMETOOLONG;
+ CERROR("%s: name '%s' is over %d symbols limit: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, name,
+ CHANGELOG_USER_NAMELEN, rc);
+ RETURN(rc);
+ }
+
+ rc = llog_cat_process(env, ctxt->loc_handle,
+ mdd_changelog_name_check_cb, &mcnc, 0, 0);
+ if (rc == -EEXIST)
+ CWARN("%s: changelog name %s exists already: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, name, rc);
+ else if (rc < 0)
+ CWARN("%s: failed user changelog processing: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
+ RETURN(rc);
+}
+
static int mdd_changelog_user_register(const struct lu_env *env,
- struct mdd_device *mdd, int *id)
+ struct mdd_device *mdd, int *id,
+ const char *name, const char *mask)
{
- struct llog_ctxt *ctxt;
- struct llog_changelog_user_rec *rec;
- int rc;
- ENTRY;
+ struct llog_ctxt *ctxt;
+ struct llog_changelog_user_rec2 *rec;
+ char user_name[CHANGELOG_USER_NAMELEN_FULL];
+ int rc;
- ctxt = llog_get_context(mdd2obd_dev(mdd),
+ ENTRY;
+
+ ctxt = llog_get_context(mdd2obd_dev(mdd),
LLOG_CHANGELOG_USER_ORIG_CTXT);
- if (ctxt == NULL)
- RETURN(-ENXIO);
+ if (ctxt == NULL)
+ RETURN(-ENXIO);
- OBD_ALLOC_PTR(rec);
- if (rec == NULL) {
- llog_ctxt_put(ctxt);
- RETURN(-ENOMEM);
- }
+ OBD_ALLOC_PTR(rec);
+ if (rec == NULL) {
+ llog_ctxt_put(ctxt);
+ RETURN(-ENOMEM);
+ }
CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
- rec->cur_hdr.lrh_len = sizeof(*rec);
- rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
+ rec->cur_hdr.lrh_len = sizeof(*rec);
+ /* keep old record type for users without mask/name for
+ * compatibility needs
+ */
+ if (mask || (name && name[0]))
+ rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
+ else
+ rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
spin_lock(&mdd->mdd_cl.mc_user_lock);
if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
spin_unlock(&mdd->mdd_cl.mc_user_lock);
- CERROR("Maximum number of changelog users exceeded!\n");
- GOTO(out, rc = -EOVERFLOW);
+ rc = -EOVERFLOW;
+ CERROR("%s: registering %s user: max ID is exceeded: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name,
+ (name && name[0]) ? name : "new", rc);
+ GOTO(out, rc);
}
*id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
mdd->mdd_cl.mc_users++;
- rec->cur_endrec = mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
rec->cur_time = (__u32)ktime_get_real_seconds();
if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
rec->cur_time = 0;
- spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ rec->cur_endrec = mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+
+ if (mask) {
+ /* if user will use relative mask apply it on default one */
+ rec->cur_mask = CHANGELOG_DEFMASK;
+ rc = cfs_str2mask(mask, changelog_type2str, &rec->cur_mask,
+ CHANGELOG_MINMASK, CHANGELOG_ALLMASK);
+ if (rc)
+ GOTO(out_users, rc);
+ } else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK) {
+ /* a maskless users means default mask but only if server has
+ * no specific mask set
+ */
+ rec->cur_mask = CHANGELOG_DEFMASK;
+ }
+
+ if (name && name[0]) {
+ rc = mdd_changelog_name_check(env, ctxt, mdd, name);
+ if (rc)
+ GOTO(out_users, rc);
+ strlcpy(rec->cur_name, name, sizeof(rec->cur_name));
+ }
+ mdd_chlg_username(rec, user_name, sizeof(user_name));
rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
if (rc) {
- CWARN("%s: Failed to register changelog user %d: rc=%d\n",
- mdd2obd_dev(mdd)->obd_name, *id, rc);
- spin_lock(&mdd->mdd_cl.mc_user_lock);
- mdd->mdd_cl.mc_users--;
- spin_unlock(&mdd->mdd_cl.mc_user_lock);
- GOTO(out, rc);
+ CWARN("%s: failed to register changelog user %s: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, user_name, rc);
+ GOTO(out_users, rc);
}
- CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
+ /* apply user mask finally */
+ spin_lock(&mdd->mdd_cl.mc_user_lock);
+ mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
+
+ CDEBUG(D_IOCTL, "%s: registered changelog user '%s', mask %#x\n",
+ mdd2obd_dev(mdd)->obd_name, user_name, rec->cur_mask);
/* Assume we want it on since somebody registered */
rc = mdd_changelog_on(env, mdd);
if (rc)
+ /* record is added, so don't decrement users on error */
GOTO(out, rc);
+out_users:
+ if (rc) {
+ spin_lock(&mdd->mdd_cl.mc_user_lock);
+ mdd->mdd_cl.mc_users--;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ }
+out:
+ OBD_FREE_PTR(rec);
+ llog_ctxt_put(ctxt);
+ RETURN(rc);
+}
+
+struct mdd_changelog_recalc_mask_data {
+ struct mdd_device *mcrm_mdd;
+ __u32 mcrm_mask;
+};
+
+/**
+ * changelog_recalc_mask callback
+ *
+ * Is is called per each registered user and calculates combined mask of
+ * all registered users.
+ */
+static int mdd_changelog_recalc_mask_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct llog_changelog_user_rec2 *rec;
+ struct mdd_changelog_recalc_mask_data *mcrm = data;
+
+ rec = container_of(hdr, typeof(*rec), cur_hdr);
+ if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
+ mcrm->mcrm_mask |= rec->cur_mask;
+ else if (mcrm->mcrm_mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
+ mcrm->mcrm_mask |= CHANGELOG_DEFMASK;
+
+ return 0;
+}
+
+int mdd_changelog_recalc_mask(const struct lu_env *env, struct mdd_device *mdd)
+{
+ struct llog_ctxt *ctxt;
+ struct mdd_changelog_recalc_mask_data mcrm = {
+ .mcrm_mdd = mdd,
+ .mcrm_mask = mdd->mdd_cl.mc_proc_mask,
+ };
+ int rc;
+
+ ENTRY;
+
+ ctxt = llog_get_context(mdd2obd_dev(mdd),
+ LLOG_CHANGELOG_USER_ORIG_CTXT);
+ if (!ctxt)
+ RETURN(-ENXIO);
+
+ if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
+ GOTO(out, rc = -ENXIO);
+
+ rc = llog_cat_process(env, ctxt->loc_handle,
+ mdd_changelog_recalc_mask_cb, &mcrm, 0, 0);
+ if (rc < 0)
+ CWARN("%s: failed user changelog processing: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
+
+ spin_lock(&mdd->mdd_cl.mc_user_lock);
+ CDEBUG(D_INFO, "%s: recalc changelog mask: %#x -> %#x\n",
+ mdd2obd_dev(mdd)->obd_name, mdd->mdd_cl.mc_current_mask,
+ mcrm.mcrm_mask);
+ mdd->mdd_cl.mc_current_mask = mcrm.mcrm_mask;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
+ EXIT;
out:
- OBD_FREE_PTR(rec);
- llog_ctxt_put(ctxt);
- RETURN(rc);
+ llog_ctxt_put(ctxt);
+
+ return rc;
}
struct mdd_changelog_user_purge {
struct llog_handle *llh,
struct llog_rec_hdr *hdr, void *data)
{
- struct llog_changelog_user_rec *rec;
- struct mdd_changelog_user_purge *mcup = data;
- struct llog_cookie cookie;
- int rc;
+ struct llog_changelog_user_rec2 *rec;
+ struct mdd_changelog_user_purge *mcup = data;
+ struct llog_cookie cookie;
+ int rc;
ENTRY;
if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
RETURN(-ENXIO);
- rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
+ rec = container_of(hdr, typeof(*rec), cur_hdr);
mcup->mcup_usercount++;
if (rec->cur_id != mcup->mcup_id) {
/* truncate to the lowest endrec that is not this user */
- mcup->mcup_minrec = min(mcup->mcup_minrec,
- rec->cur_endrec);
+ mcup->mcup_minrec = min(mcup->mcup_minrec, rec->cur_endrec);
RETURN(0);
}
__u64 mcuc_minrec;
__u32 mcuc_id;
bool mcuc_flush;
+ struct mdd_device *mcuc_mdd;
};
/**
struct llog_rec_hdr *hdr,
void *data)
{
- struct llog_changelog_user_rec *rec;
+ struct llog_changelog_user_rec2 *rec;
struct mdd_changelog_user_clear *mcuc = data;
+ char user_name[CHANGELOG_USER_NAMELEN_FULL];
+ struct mdd_device *mdd = mcuc->mcuc_mdd;
int rc;
ENTRY;
if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
RETURN(-ENXIO);
- rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
-
+ rec = container_of(hdr, typeof(*rec), cur_hdr);
/* Does the changelog id match the requested id? */
if (rec->cur_id != mcuc->mcuc_id) {
mcuc->mcuc_minrec = min(mcuc->mcuc_minrec,
/* cur_endrec is the oldest purgeable record, make sure we're newer */
if (rec->cur_endrec > mcuc->mcuc_endrec) {
- CDEBUG(D_IOCTL, "Request %llu out of range: %llu\n",
- mcuc->mcuc_endrec, rec->cur_endrec);
- RETURN(-EINVAL);
+ rc = -EINVAL;
+ CDEBUG(D_IOCTL,
+ "%s: request %llu > endrec %llu for user %s: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name,
+ mcuc->mcuc_endrec, rec->cur_endrec,
+ mdd_chlg_username(rec, user_name, sizeof(user_name)),
+ rc);
+ RETURN(rc);
}
/* Flag that we've met all the range and user checks.
mcuc->mcuc_flush = true;
- CDEBUG(D_IOCTL, "Rewriting changelog user %u endrec to %llu\n",
- mcuc->mcuc_id, rec->cur_endrec);
+ CDEBUG(D_IOCTL, "%s: rewriting changelog user %s endrec = %llu\n",
+ mdd2obd_dev(mdd)->obd_name,
+ mdd_chlg_username(rec, user_name, sizeof(user_name)),
+ rec->cur_endrec);
/* Update the endrec */
rc = llog_write(env, llh, hdr, hdr->lrh_index);
.mcuc_id = id,
.mcuc_minrec = endrec,
.mcuc_flush = false,
+ .mcuc_mdd = mdd,
};
struct llog_ctxt *ctxt;
__u64 start_rec;
mdd_changelog_clear_cb, (void *)&mcuc,
0, 0);
- if (rc < 0) {
+ if (rc == -EINVAL) {
+ CDEBUG(D_IOCTL, "%s: No changelog recnum <= %llu to clear\n",
+ mdd2obd_dev(mdd)->obd_name, (unsigned long long) endrec);
+ RETURN(-EINVAL);
+ } else if (rc < 0) {
CWARN("%s: Failure to clear the changelog for user %d: %d\n",
mdd2obd_dev(mdd)->obd_name, id, rc);
} else if (mcuc.mcuc_flush) {
mcuc.mcuc_minrec);
}
} else {
- CWARN("%s: No entry for user %d\n",
+ CDEBUG(D_IOCTL, "%s: No entry for user %d\n",
mdd2obd_dev(mdd)->obd_name, id);
rc = -ENOENT;
}
return rc;
}
+static int mdd_changelog_user_deregister(const struct lu_env *env,
+ struct mdd_device *mdd, int *id,
+ const char *name)
+{
+ struct llog_ctxt *ctxt;
+ struct mdd_changelog_name_check_data mcnc = {
+ .mcnc_name = name,
+ .mcnc_id = 0,
+ };
+ int rc;
+
+ ENTRY;
+
+ if (name) {
+ ctxt = llog_get_context(mdd2obd_dev(mdd),
+ LLOG_CHANGELOG_USER_ORIG_CTXT);
+ if (!ctxt)
+ RETURN(-ENXIO);
+
+ rc = llog_cat_process(env, ctxt->loc_handle,
+ mdd_changelog_name_check_cb, &mcnc, 0, 0);
+ llog_ctxt_put(ctxt);
+
+ if (rc != -EEXIST) {
+ CDEBUG(D_IOCTL, "%s: no entry for username %s\n",
+ mdd2obd_dev(mdd)->obd_name, name);
+ RETURN(-ENOENT);
+ }
+ *id = mcnc.mcnc_id;
+ }
+
+ /* explicitly clear changelog first, to protect from crash in
+ * the middle of purge that would lead to unregistered consumer
+ * but pending changelog entries
+ */
+ rc = mdd_changelog_clear(env, mdd, *id, 0);
+ if (!rc)
+ rc = mdd_changelog_user_purge(env, mdd, *id);
+
+ /* recalc changelog current mask */
+ mdd_changelog_recalc_mask(env, mdd);
+
+ RETURN(rc);
+}
+
/** mdd_iocontrol
* May be called remotely from mdt_iocontrol_handle or locally from
* mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
}
/* Below ioctls use obd_ioctl_data */
- if (len != sizeof(*data)) {
- CERROR("Bad ioctl size %d\n", len);
- RETURN(-EINVAL);
- }
if (data->ioc_version != OBD_IOCTL_VERSION) {
CERROR("Bad magic %x != %x\n", data->ioc_version,
OBD_IOCTL_VERSION);
if (unlikely(!barrier_entry(mdd->mdd_bottom)))
RETURN(-EINPROGRESS);
- rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
+ rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1,
+ data->ioc_inlbuf1,
+ data->ioc_inlbuf2);
barrier_exit(mdd->mdd_bottom);
break;
case OBD_IOC_CHANGELOG_DEREG:
if (unlikely(!barrier_entry(mdd->mdd_bottom)))
RETURN(-EINPROGRESS);
- /* explicitly clear changelog first, to protect from crash in
- * the middle of purge that would lead to unregistered consumer
- * but pending changelog entries
- */
- rc = mdd_changelog_clear(env, mdd, data->ioc_u32_1, 0);
- if (!rc)
- rc = mdd_changelog_user_purge(env,
- mdd, data->ioc_u32_1);
+ rc = mdd_changelog_user_deregister(env, mdd, &data->ioc_u32_1,
+ data->ioc_inlbuf1);
barrier_exit(mdd->mdd_bottom);
break;
.mdo_dtconf_get = mdd_dtconf_get,
};
-static struct lu_device_type_operations mdd_device_type_ops = {
- .ldto_init = mdd_type_init,
- .ldto_fini = mdd_type_fini,
+static const struct lu_device_type_operations mdd_device_type_ops = {
+ .ldto_init = mdd_type_init,
+ .ldto_fini = mdd_type_fini,
- .ldto_start = mdd_type_start,
- .ldto_stop = mdd_type_stop,
+ .ldto_start = mdd_type_start,
+ .ldto_stop = mdd_type_stop,
- .ldto_device_alloc = mdd_device_alloc,
- .ldto_device_free = mdd_device_free,
+ .ldto_device_alloc = mdd_device_alloc,
+ .ldto_device_free = mdd_device_free,
- .ldto_device_fini = mdd_device_fini
+ .ldto_device_fini = mdd_device_fini
};
static struct lu_device_type mdd_device_type = {
{
struct mdd_thread_info *info = data;
- lu_buf_free(&info->mti_big_buf);
- lu_buf_free(&info->mti_link_buf);
- lu_buf_free(&info->mti_xattr_buf);
+ lu_buf_free(&info->mdi_big_buf);
+ lu_buf_free(&info->mdi_link_buf);
+ lu_buf_free(&info->mdi_xattr_buf);
+ lu_buf_free(&info->mdi_chlg_buf);
OBD_FREE_PTR(info);
}
changelog_orig_logops = llog_common_cat_ops;
changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
- rc = class_register_type(&mdd_obd_device_ops, NULL, false, NULL,
+ rc = class_register_type(&mdd_obd_device_ops, NULL, false,
LUSTRE_MDD_NAME, &mdd_device_type);
if (rc)
lu_kmem_fini(mdd_caches);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff