/*
* Set the attributes of an object
*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, oid) called and then assigned
* to a transaction group.
*/
static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_attr *la, struct thandle *handle)
{
struct osd_thread_info *info = osd_oti_get(env);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
struct osa_attr *osa = &info->oti_osa;
sa_bulk_attr_t *bulk;
__u64 valid = la->la_valid;
int cnt;
int rc = 0;
ENTRY;
down_read(&obj->oo_guard);
if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
GOTO(out, rc = -ENOENT);
LASSERT(handle != NULL);
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_sa_hdl);
oh = container_of0(handle, struct osd_thandle, ot_super);
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(oh->ot_tx->tx_txg != 0);
/* Only allow set size for regular file */
if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
valid &= ~(LA_SIZE | LA_BLOCKS);
if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
valid &= ~LA_CTIME;
if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
valid &= ~LA_MTIME;
if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
valid &= ~LA_ATIME;
if (valid == 0)
GOTO(out, rc = 0);
if (valid & LA_FLAGS) {
struct lustre_mdt_attrs *lma;
struct lu_buf buf;
if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
lma = (struct lustre_mdt_attrs *)&info->oti_buf;
buf.lb_buf = lma;
buf.lb_len = sizeof(info->oti_buf);
rc = osd_xattr_get(env, &obj->oo_dt, &buf,
XATTR_NAME_LMA);
if (rc > 0) {
lma->lma_incompat =
le32_to_cpu(lma->lma_incompat);
lma->lma_incompat |=
lustre_to_lma_flags(la->la_flags);
lma->lma_incompat =
cpu_to_le32(lma->lma_incompat);
buf.lb_buf = lma;
buf.lb_len = sizeof(*lma);
rc = osd_xattr_set_internal(env, obj, &buf,
XATTR_NAME_LMA,
LU_XATTR_REPLACE,
oh);
}
if (rc < 0) {
CWARN("%s: failed to set LMA flags: rc = %d\n",
osd->od_svname, rc);
RETURN(rc);
}
}
}
OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
if (bulk == NULL)
GOTO(out, rc = -ENOMEM);
/* do both accounting updates outside oo_attr_lock below */
if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
/* Update user accounting. Failure isn't fatal, but we still
* log an error message */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
la->la_uid, 1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname, la->la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
obj->oo_attr.la_uid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname,
obj->oo_attr.la_uid, rc);
}
if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
/* Update group accounting. Failure isn't fatal, but we still
* log an error message */
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
la->la_gid, 1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname, la->la_gid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
obj->oo_attr.la_gid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname,
obj->oo_attr.la_gid, rc);
}
write_lock(&obj->oo_attr_lock);
cnt = 0;
if (valid & LA_ATIME) {
osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
osa->atime, 16);
}
if (valid & LA_MTIME) {
osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
osa->mtime, 16);
}
if (valid & LA_CTIME) {
osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
osa->ctime, 16);
}
if (valid & LA_MODE) {
/* mode is stored along with type, so read it first */
obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
(la->la_mode & ~S_IFMT);
osa->mode = obj->oo_attr.la_mode;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
&osa->mode, 8);
}
if (valid & LA_SIZE) {
osa->size = obj->oo_attr.la_size = la->la_size;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
&osa->size, 8);
}
if (valid & LA_NLINK) {
osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
&osa->nlink, 8);
}
if (valid & LA_RDEV) {
osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
&osa->rdev, 8);
}
if (valid & LA_FLAGS) {
osa->flags = attrs_fs2zfs(la->la_flags);
/* many flags are not supported by zfs, so ensure a good cached
* copy */
obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
&osa->flags, 8);
}
if (valid & LA_UID) {
osa->uid = obj->oo_attr.la_uid = la->la_uid;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
&osa->uid, 8);
}
if (valid & LA_GID) {
osa->gid = obj->oo_attr.la_gid = la->la_gid;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
&osa->gid, 8);
}
obj->oo_attr.la_valid |= valid;
write_unlock(&obj->oo_attr_lock);
rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
out:
up_read(&obj->oo_guard);
RETURN(rc);
}
int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
uint64_t parent)
{
sa_bulk_attr_t *bulk;
sa_handle_t *sa_hdl;
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
uint64_t gen;
uint64_t crtime[2];
timestruc_t now;
int cnt;
int rc;
gethrestime(&now);
gen = dmu_tx_get_txg(tx);
ZFS_TIME_ENCODE(&now, crtime);
osa->atime[0] = la->la_atime;
osa->ctime[0] = la->la_ctime;
osa->mtime[0] = la->la_mtime;
osa->mode = la->la_mode;
osa->uid = la->la_uid;
osa->gid = la->la_gid;
osa->rdev = la->la_rdev;
osa->nlink = la->la_nlink;
osa->flags = attrs_fs2zfs(la->la_flags);
osa->size = la->la_size;
/* Now add in all of the "SA" attributes */
rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
if (rc)
return rc;
OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
if (bulk == NULL) {
rc = -ENOMEM;
goto out;
}
/*
* we need to create all SA below upon object create.
*
* XXX The attribute order matters since the accounting callback relies
* on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
* look up the UID/GID attributes. Moreover, the callback does not seem
* to support the spill block.
* We define attributes in the same order as SA_*_OFFSET in order to
* work around the problem. See ORI-610.
*/
cnt = 0;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
out:
sa_handle_destroy(sa_hdl);
return rc;
}
/*
* Retrieve the attributes of a DMU object
*/
int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
struct osd_object *obj, struct lu_attr *la)
{
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
sa_handle_t *sa_hdl;
sa_bulk_attr_t *bulk;
int cnt = 0;
int rc;
ENTRY;
LASSERT(obj->oo_db != NULL);
rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
SA_HDL_PRIVATE, &sa_hdl);
if (rc)
RETURN(rc);
OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
if (bulk == NULL)
GOTO(out_sa, rc = -ENOMEM);
la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
if (rc)
GOTO(out_bulk, rc);
la->la_atime = osa->atime[0];
la->la_mtime = osa->mtime[0];
la->la_ctime = osa->ctime[0];
la->la_mode = osa->mode;
la->la_uid = osa->uid;
la->la_gid = osa->gid;
la->la_nlink = osa->nlink;
la->la_flags = attrs_zfs2fs(osa->flags);
la->la_size = osa->size;
/* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
* flags is stored in LMA, and it is only for orphan directory */
if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
struct osd_thread_info *info = osd_oti_get(env);
struct lustre_mdt_attrs *lma;
struct lu_buf buf;
lma = (struct lustre_mdt_attrs *)info->oti_buf;
buf.lb_buf = lma;
buf.lb_len = sizeof(info->oti_buf);
rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
if (rc > 0) {
rc = 0;
lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
obj->oo_lma_flags =
lma_to_lustre_flags(lma->lma_incompat);
} else if (rc == -ENODATA) {
rc = 0;
}
}
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
if (rc)
GOTO(out_bulk, rc);
la->la_rdev = osa->rdev;
la->la_valid |= LA_RDEV;
}
out_bulk:
OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
out_sa:
sa_handle_destroy(sa_hdl);
RETURN(rc);
}
int llog_reverse_process(const struct lu_env *env,
struct llog_handle *loghandle, llog_cb_t cb,
void *data, void *catdata)
{
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = catdata;
void *buf;
int rc = 0, first_index = 1, index, idx;
OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
if (!buf)
return -ENOMEM;
if (cd != NULL)
first_index = cd->lpcd_first_idx + 1;
if (cd != NULL && cd->lpcd_last_idx)
index = cd->lpcd_last_idx;
else
index = LLOG_BITMAP_BYTES * 8 - 1;
while (rc == 0) {
struct llog_rec_hdr *rec;
struct llog_rec_tail *tail;
/* skip records not set in bitmap */
while (index >= first_index &&
!ext2_test_bit(index, llh->llh_bitmap))
--index;
LASSERT(index >= first_index - 1);
if (index == first_index - 1)
break;
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, LLOG_CHUNK_SIZE);
rc = llog_prev_block(env, loghandle, index, buf,
LLOG_CHUNK_SIZE);
if (rc)
goto out;
rec = buf;
idx = rec->lrh_index;
CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
while (idx < index) {
rec = (void *)rec + rec->lrh_len;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
idx ++;
}
LASSERT(idx == index);
tail = (void *)rec + rec->lrh_len - sizeof(*tail);
/* process records in buffer, starting where we found one */
while ((void *)tail > buf) {
if (tail->lrt_index == 0) {
/* no more records */
rc = 0;
goto out;
}
/* if set, process the callback on this record */
if (ext2_test_bit(index, llh->llh_bitmap)) {
rec = (void *)tail - tail->lrt_len +
sizeof(*tail);
rc = cb(env, loghandle, rec, data);
if (rc == LLOG_PROC_BREAK) {
goto out;
} else if (rc == LLOG_DEL_RECORD) {
llog_cancel_rec(env, loghandle,
tail->lrt_index);
rc = 0;
}
if (rc)
goto out;
}
/* previous record, still in buffer? */
--index;
if (index < first_index) {
rc = 0;
goto out;
}
tail = (void *)tail - tail->lrt_len;
}
}
out:
if (buf)
OBD_FREE(buf, LLOG_CHUNK_SIZE);
return rc;
}
/*
* Iterate over all slave index files associated with global index \glb_fid and
* invoke a callback function for each slave index file.
*
* \param env - is the environment passed by the caller
* \param parent - is the parent directory where the slave index files are
* stored
* \param glb_fid - is the fid of the global index file associated with the
* slave indexes to scan
* \param func - is the callback function to call each time a slave index
* file is found
* \param arg - is an opaq argument passed to the callback function \func
*/
int lquota_disk_for_each_slv(const struct lu_env *env, struct dt_object *parent,
struct lu_fid *glb_fid, lquota_disk_slv_cb_t func,
void *arg)
{
struct lquota_thread_info *qti = lquota_info(env);
struct dt_it *it;
const struct dt_it_ops *iops;
char *name;
int rc;
ENTRY;
OBD_ALLOC(name, LQUOTA_NAME_MAX);
if (name == NULL)
RETURN(-ENOMEM);
/* filename associated with slave index files are prefixed with the most
* signicant bits of the global FID */
sprintf(name, "0x%x-", glb_fid->f_oid);
iops = &parent->do_index_ops->dio_it;
it = iops->init(env, parent, 0);
if (IS_ERR(it)) {
OBD_FREE(name, LQUOTA_NAME_MAX);
RETURN(PTR_ERR(it));
}
rc = iops->load(env, it, 0);
if (rc == 0) {
/*
* Iterator didn't find record with exactly the key requested.
*
* It is currently either
*
* - positioned above record with key less than
* requested---skip it.
*
* - or not positioned at all (is in IAM_IT_SKEWED
* state)---position it on the next item.
*/
rc = iops->next(env, it);
} else if (rc > 0)
rc = 0;
while (rc == 0) {
struct dt_key *key;
int len;
len = iops->key_size(env, it);
/* IAM iterator can return record with zero len. */
if (len == 0 || len <= strlen(name) || len >= LQUOTA_NAME_MAX)
goto next;
key = iops->key(env, it);
if (IS_ERR(key)) {
rc = PTR_ERR(key);
break;
}
if (strncmp((char *)key, name, strlen(name)) != 0)
goto next;
/* ldiskfs OSD returns filename as stored in directory entry
* which does not end up with '\0' */
memcpy(&qti->qti_buf, key, len);
qti->qti_buf[len] = '\0';
/* lookup fid associated with this slave index file */
rc = dt_lookup_dir(env, parent, qti->qti_buf, &qti->qti_fid);
if (rc)
break;
if (qti->qti_fid.f_seq != FID_SEQ_QUOTA)
goto next;
rc = func(env, glb_fid, (char *)key, &qti->qti_fid, arg);
if (rc)
break;
next:
do {
rc = iops->next(env, it);
} while (rc == -ESTALE);
}
iops->put(env, it);
iops->fini(env, it);
OBD_FREE(name, LQUOTA_NAME_MAX);
if (rc > 0)
rc = 0;
RETURN(rc);
}
static int lustre_stop_mgc(struct super_block *sb)
{
struct lustre_sb_info *lsi = s2lsi(sb);
struct obd_device *obd;
char *niduuid = NULL, *ptr = NULL;
int i, rc = 0, len = 0;
if (!lsi)
return -ENOENT;
obd = lsi->lsi_mgc;
if (!obd)
return -ENOENT;
lsi->lsi_mgc = NULL;
mutex_lock(&mgc_start_lock);
LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
/* This is not fatal, every client that stops
will call in here. */
CDEBUG(D_MOUNT, "mgc still has %d references.\n",
atomic_read(&obd->u.cli.cl_mgc_refcount));
GOTO(out, rc = -EBUSY);
}
/* The MGC has no recoverable data in any case.
* force shutdown set in umount_begin */
obd->obd_no_recov = 1;
if (obd->u.cli.cl_mgc_mgsexp) {
/* An error is not fatal, if we are unable to send the
disconnect mgs ping evictor cleans up the export */
rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
if (rc)
CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
}
/* Save the obdname for cleaning the nid uuids, which are
obdname_XX */
len = strlen(obd->obd_name) + 6;
OBD_ALLOC(niduuid, len);
if (niduuid) {
strcpy(niduuid, obd->obd_name);
ptr = niduuid + strlen(niduuid);
}
rc = class_manual_cleanup(obd);
if (rc)
GOTO(out, rc);
/* Clean the nid uuids */
if (!niduuid)
GOTO(out, rc = -ENOMEM);
for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
sprintf(ptr, "_%x", i);
rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
niduuid, NULL, NULL, NULL);
if (rc)
CERROR("del MDC UUID %s failed: rc = %d\n",
niduuid, rc);
}
out:
if (niduuid)
OBD_FREE(niduuid, len);
/* class_import_put will get rid of the additional connections */
mutex_unlock(&mgc_start_lock);
return rc;
}
int lfsck_namespace_setup(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct lfsck_component *com;
struct lfsck_namespace *ns;
struct dt_object *root = NULL;
struct dt_object *obj;
int rc;
ENTRY;
LASSERT(lfsck->li_master);
OBD_ALLOC_PTR(com);
if (com == NULL)
RETURN(-ENOMEM);
CFS_INIT_LIST_HEAD(&com->lc_link);
CFS_INIT_LIST_HEAD(&com->lc_link_dir);
init_rwsem(&com->lc_sem);
atomic_set(&com->lc_ref, 1);
com->lc_lfsck = lfsck;
com->lc_type = LT_NAMESPACE;
com->lc_ops = &lfsck_namespace_ops;
com->lc_file_size = sizeof(struct lfsck_namespace);
OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
if (com->lc_file_ram == NULL)
GOTO(out, rc = -ENOMEM);
OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
if (com->lc_file_disk == NULL)
GOTO(out, rc = -ENOMEM);
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
GOTO(out, rc = PTR_ERR(root));
if (unlikely(!dt_try_as_dir(env, root)))
GOTO(out, rc = -ENOTDIR);
obj = local_index_find_or_create(env, lfsck->li_los, root,
lfsck_namespace_name,
S_IFREG | S_IRUGO | S_IWUSR,
&dt_lfsck_features);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
com->lc_obj = obj;
rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
if (rc != 0)
GOTO(out, rc);
rc = lfsck_namespace_load(env, com);
if (rc > 0)
rc = lfsck_namespace_reset(env, com, true);
else if (rc == -ENODATA)
rc = lfsck_namespace_init(env, com);
if (rc != 0)
GOTO(out, rc);
ns = com->lc_file_ram;
switch (ns->ln_status) {
case LS_INIT:
case LS_COMPLETED:
case LS_FAILED:
case LS_STOPPED:
spin_lock(&lfsck->li_lock);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_idle);
spin_unlock(&lfsck->li_lock);
break;
default:
CERROR("%s: unknown lfsck_namespace status: rc = %u\n",
lfsck_lfsck2name(lfsck), ns->ln_status);
/* fall through */
case LS_SCANNING_PHASE1:
case LS_SCANNING_PHASE2:
/* No need to store the status to disk right now.
* If the system crashed before the status stored,
* it will be loaded back when next time. */
ns->ln_status = LS_CRASHED;
/* fall through */
case LS_PAUSED:
case LS_CRASHED:
spin_lock(&lfsck->li_lock);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_scan);
cfs_list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
spin_unlock(&lfsck->li_lock);
break;
}
GOTO(out, rc = 0);
out:
if (root != NULL && !IS_ERR(root))
lu_object_put(env, &root->do_lu);
if (rc != 0)
lfsck_component_cleanup(env, com);
return rc;
}
/* temporary for testing */
static ssize_t mdc_kuc_write(struct file *file,
const char __user *buffer,
size_t count, loff_t *off)
{
struct obd_device *obd =
((struct seq_file *)file->private_data)->private;
struct kuc_hdr *lh;
struct hsm_action_list *hal;
struct hsm_action_item *hai;
int len;
int fd, rc;
rc = lprocfs_write_helper(buffer, count, &fd);
if (rc)
return rc;
if (fd < 0)
return -ERANGE;
CWARN("message to fd %d\n", fd);
len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
/* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
OBD_ALLOC(lh, len);
if (!lh)
return -ENOMEM;
lh->kuc_magic = KUC_MAGIC;
lh->kuc_transport = KUC_TRANSPORT_HSM;
lh->kuc_msgtype = HMT_ACTION_LIST;
lh->kuc_msglen = len;
hal = (struct hsm_action_list *)(lh + 1);
hal->hal_version = HAL_VERSION;
hal->hal_archive_id = 1;
hal->hal_flags = 0;
obd_uuid2fsname(hal->hal_fsname, obd->obd_name, MTI_NAME_MAXLEN);
/* mock up an action list */
hal->hal_count = 2;
hai = hai_zero(hal);
hai->hai_action = HSMA_ARCHIVE;
hai->hai_fid.f_oid = 5;
hai->hai_len = sizeof(*hai);
hai = hai_next(hai);
hai->hai_action = HSMA_RESTORE;
hai->hai_fid.f_oid = 10;
hai->hai_len = sizeof(*hai);
/* This works for either broadcast or unicast to a single fd */
if (fd == 0) {
rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
} else {
struct file *fp = fget(fd);
rc = libcfs_kkuc_msg_put(fp, lh);
fput(fp);
}
OBD_FREE(lh, len);
if (rc < 0)
return rc;
return count;
}
/* Test catalogue additions */
static int llog_test_4(struct obd_device *obd)
{
struct llog_handle *cath;
char name[10];
int rc, i, buflen;
struct llog_mini_rec lmr;
struct llog_cookie cookie;
struct llog_ctxt *ctxt = llog_get_context(obd, LLOG_TEST_ORIG_CTXT);
int num_recs = 0;
char *buf;
struct llog_rec_hdr rec;
ENTRY;
lmr.lmr_hdr.lrh_len = lmr.lmr_tail.lrt_len = LLOG_MIN_REC_SIZE;
lmr.lmr_hdr.lrh_type = 0xf00f00;
sprintf(name, "%x", llog_test_rand+1);
CWARN("4a: create a catalog log with name: %s\n", name);
rc = llog_create(ctxt, &cath, NULL, name);
if (rc) {
CERROR("1a: llog_create with name %s failed: %d\n", name, rc);
GOTO(out, rc);
}
llog_init_handle(cath, LLOG_F_IS_CAT, &uuid);
num_recs++;
cat_logid = cath->lgh_id;
CWARN("4b: write 1 record into the catalog\n");
rc = llog_cat_add_rec(cath, &lmr.lmr_hdr, &cookie, NULL);
if (rc != 1) {
CERROR("4b: write 1 catalog record failed at: %d\n", rc);
GOTO(out, rc);
}
num_recs++;
if ((rc = verify_handle("4b", cath, 2)))
GOTO(ctxt_release, rc);
if ((rc = verify_handle("4b", cath->u.chd.chd_current_log, num_recs)))
GOTO(ctxt_release, rc);
CWARN("4c: cancel 1 log record\n");
rc = llog_cat_cancel_records(cath, 1, &cookie);
if (rc) {
CERROR("4c: cancel 1 catalog based record failed: %d\n", rc);
GOTO(out, rc);
}
num_recs--;
if ((rc = verify_handle("4c", cath->u.chd.chd_current_log, num_recs)))
GOTO(ctxt_release, rc);
CWARN("4d: write 40,000 more log records\n");
for (i = 0; i < 40000; i++) {
rc = llog_cat_add_rec(cath, &lmr.lmr_hdr, NULL, NULL);
if (rc) {
CERROR("4d: write 40000 records failed at #%d: %d\n",
i + 1, rc);
GOTO(out, rc);
}
num_recs++;
}
CWARN("4e: add 5 large records, one record per block\n");
buflen = LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr)
- sizeof(struct llog_rec_tail);
OBD_ALLOC(buf, buflen);
if (buf == NULL)
GOTO(out, rc = -ENOMEM);
for (i = 0; i < 5; i++) {
rec.lrh_len = buflen;
rec.lrh_type = OBD_CFG_REC;
rc = llog_cat_add_rec(cath, &rec, NULL, buf);
if (rc) {
CERROR("4e: write 5 records failed at #%d: %d\n",
i + 1, rc);
OBD_FREE(buf, buflen);
GOTO(out, rc);
}
num_recs++;
}
OBD_FREE(buf, buflen);
out:
CWARN("4f: put newly-created catalog\n");
rc = llog_cat_put(cath);
ctxt_release:
llog_ctxt_put(ctxt);
if (rc)
CERROR("1b: close log %s failed: %d\n", name, rc);
RETURN(rc);
}
/* Test catalogue additions */
static int llog_test_4(const struct lu_env *env, struct obd_device *obd)
{
struct llog_handle *cath;
char name[10];
int rc, rc2, i, buflen;
struct llog_mini_rec lmr;
struct llog_cookie cookie;
struct llog_ctxt *ctxt;
int num_recs = 0;
char *buf;
struct llog_rec_hdr *rec;
ENTRY;
ctxt = llog_get_context(obd, LLOG_TEST_ORIG_CTXT);
LASSERT(ctxt);
lmr.lmr_hdr.lrh_len = lmr.lmr_tail.lrt_len = LLOG_MIN_REC_SIZE;
lmr.lmr_hdr.lrh_type = 0xf00f00;
sprintf(name, "%x", llog_test_rand + 1);
CWARN("4a: create a catalog log with name: %s\n", name);
rc = llog_open_create(env, ctxt, &cath, NULL, name);
if (rc) {
CERROR("4a: llog_create with name %s failed: %d\n", name, rc);
GOTO(ctxt_release, rc);
}
rc = llog_init_handle(env, cath, LLOG_F_IS_CAT, &uuid);
if (rc) {
CERROR("4a: can't init llog handle: %d\n", rc);
GOTO(out, rc);
}
num_recs++;
cat_logid = cath->lgh_id;
CWARN("4b: write 1 record into the catalog\n");
rc = llog_cat_add(env, cath, &lmr.lmr_hdr, &cookie);
if (rc != 1) {
CERROR("4b: write 1 catalog record failed at: %d\n", rc);
GOTO(out, rc);
}
num_recs++;
rc = verify_handle("4b", cath, 2);
if (rc)
GOTO(out, rc);
rc = verify_handle("4b", cath->u.chd.chd_current_log, num_recs);
if (rc)
GOTO(out, rc);
CWARN("4c: cancel 1 log record\n");
rc = llog_cat_cancel_records(env, cath, 1, &cookie);
if (rc) {
CERROR("4c: cancel 1 catalog based record failed: %d\n", rc);
GOTO(out, rc);
}
num_recs--;
rc = verify_handle("4c", cath->u.chd.chd_current_log, num_recs);
if (rc)
GOTO(out, rc);
CWARN("4d: write %d more log records\n", LLOG_TEST_RECNUM);
for (i = 0; i < LLOG_TEST_RECNUM; i++) {
rc = llog_cat_add(env, cath, &lmr.lmr_hdr, NULL);
if (rc) {
CERROR("4d: write %d records failed at #%d: %d\n",
LLOG_TEST_RECNUM, i + 1, rc);
GOTO(out, rc);
}
num_recs++;
}
/* make sure new plain llog appears */
rc = verify_handle("4d", cath, 3);
if (rc)
GOTO(out, rc);
CWARN("4e: add 5 large records, one record per block\n");
buflen = LLOG_CHUNK_SIZE;
OBD_ALLOC(buf, buflen);
if (buf == NULL)
GOTO(out, rc = -ENOMEM);
for (i = 0; i < 5; i++) {
rec = (void *)buf;
rec->lrh_len = buflen;
rec->lrh_type = OBD_CFG_REC;
rc = llog_cat_add(env, cath, rec, NULL);
if (rc) {
CERROR("4e: write 5 records failed at #%d: %d\n",
i + 1, rc);
GOTO(out_free, rc);
}
num_recs++;
}
out_free:
OBD_FREE(buf, buflen);
out:
CWARN("4f: put newly-created catalog\n");
rc2 = llog_cat_close(env, cath);
if (rc2) {
CERROR("4: close log %s failed: %d\n", name, rc2);
if (rc == 0)
rc = rc2;
}
ctxt_release:
llog_ctxt_put(ctxt);
RETURN(rc);
}
static int llog_lvfs_open(const struct lu_env *env, struct llog_handle *handle,
struct llog_logid *logid, char *name,
enum llog_open_param open_param)
{
struct llog_ctxt *ctxt = handle->lgh_ctxt;
struct dentry *dchild = NULL;
struct obd_device *obd;
int rc = 0;
LASSERT(ctxt);
LASSERT(ctxt->loc_exp);
LASSERT(ctxt->loc_exp->exp_obd);
obd = ctxt->loc_exp->exp_obd;
LASSERT(handle);
if (logid != NULL) {
dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, &logid->lgl_oi,
logid->lgl_ogen);
if (IS_ERR(dchild)) {
rc = PTR_ERR(dchild);
CERROR("%s: error looking up logfile #"DOSTID "#%08x:"
" rc = %d\n", ctxt->loc_obd->obd_name,
POSTID(&logid->lgl_oi), logid->lgl_ogen, rc);
GOTO(out, rc);
}
if (dchild->d_inode == NULL) {
l_dput(dchild);
rc = -ENOENT;
CERROR("%s: nonexistent llog #"DOSTID"#%08x:"
"rc = %d\n", ctxt->loc_obd->obd_name,
POSTID(&logid->lgl_oi), logid->lgl_ogen, rc);
GOTO(out, rc);
}
handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
O_RDWR | O_LARGEFILE);
l_dput(dchild);
if (IS_ERR(handle->lgh_file)) {
rc = PTR_ERR(handle->lgh_file);
handle->lgh_file = NULL;
CERROR("%s: error opening llog #"DOSTID"#%08x:"
"rc = %d\n", ctxt->loc_obd->obd_name,
POSTID(&logid->lgl_oi), logid->lgl_ogen, rc);
GOTO(out, rc);
}
handle->lgh_id = *logid;
} else if (name) {
handle->lgh_file = llog_filp_open(MOUNT_CONFIGS_DIR, name,
O_RDWR | O_LARGEFILE, 0644);
if (IS_ERR(handle->lgh_file)) {
rc = PTR_ERR(handle->lgh_file);
handle->lgh_file = NULL;
if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
if (handle->lgh_name)
strcpy(handle->lgh_name, name);
else
GOTO(out, rc = -ENOMEM);
rc = 0;
} else {
GOTO(out, rc);
}
} else {
lustre_build_llog_lvfs_oid(&handle->lgh_id,
handle->lgh_file->f_dentry->d_inode->i_ino,
handle->lgh_file->f_dentry->d_inode->i_generation);
}
} else {
LASSERTF(open_param == LLOG_OPEN_NEW, "%#x\n", open_param);
handle->lgh_file = NULL;
}
/* No new llog is expected but doesn't exist */
if (open_param != LLOG_OPEN_NEW && handle->lgh_file == NULL)
GOTO(out_name, rc = -ENOENT);
return 0;
out_name:
if (handle->lgh_name != NULL)
OBD_FREE(handle->lgh_name, strlen(name) + 1);
out:
return rc;
}
/*
* Set the attributes of an object
*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, oid) called and then assigned
* to a transaction group.
*/
static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_attr *la, struct thandle *handle)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
sa_bulk_attr_t *bulk;
__u64 valid = la->la_valid;
int cnt;
int rc = 0;
ENTRY;
LASSERT(handle != NULL);
LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(obj));
LASSERT(obj->oo_sa_hdl);
oh = container_of0(handle, struct osd_thandle, ot_super);
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(oh->ot_tx->tx_txg != 0);
/* Only allow set size for regular file */
if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
valid &= ~(LA_SIZE | LA_BLOCKS);
if (valid == 0)
RETURN(0);
OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
if (bulk == NULL)
RETURN(-ENOMEM);
/* do both accounting updates outside oo_attr_lock below */
if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
/* Update user accounting. Failure isn't fatal, but we still
* log an error message */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
la->la_uid, 1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname, la->la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
obj->oo_attr.la_uid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname,
obj->oo_attr.la_uid, rc);
}
if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
/* Update group accounting. Failure isn't fatal, but we still
* log an error message */
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
la->la_gid, 1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname, la->la_gid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
obj->oo_attr.la_gid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to update accounting ZAP for user "
"%d (%d)\n", osd->od_svname,
obj->oo_attr.la_gid, rc);
}
write_lock(&obj->oo_attr_lock);
cnt = 0;
if (valid & LA_ATIME) {
osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
osa->atime, 16);
}
if (valid & LA_MTIME) {
osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
osa->mtime, 16);
}
if (valid & LA_CTIME) {
osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
osa->ctime, 16);
}
if (valid & LA_MODE) {
/* mode is stored along with type, so read it first */
obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
(la->la_mode & ~S_IFMT);
osa->mode = obj->oo_attr.la_mode;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
&osa->mode, 8);
}
if (valid & LA_SIZE) {
osa->size = obj->oo_attr.la_size = la->la_size;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
&osa->size, 8);
}
if (valid & LA_NLINK) {
osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
&osa->nlink, 8);
}
if (valid & LA_RDEV) {
osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
&osa->rdev, 8);
}
if (valid & LA_FLAGS) {
osa->flags = attrs_fs2zfs(la->la_flags);
/* many flags are not supported by zfs, so ensure a good cached
* copy */
obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
&osa->flags, 8);
}
if (valid & LA_UID) {
osa->uid = obj->oo_attr.la_uid = la->la_uid;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
&osa->uid, 8);
}
if (valid & LA_GID) {
osa->gid = obj->oo_attr.la_gid = la->la_gid;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
&osa->gid, 8);
}
obj->oo_attr.la_valid |= valid;
write_unlock(&obj->oo_attr_lock);
rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
RETURN(rc);
}
/*
* Retrieve the attributes of a DMU object
*/
int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
struct osd_object *obj, struct lu_attr *la)
{
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
sa_handle_t *sa_hdl;
sa_bulk_attr_t *bulk;
int cnt = 0;
int rc;
ENTRY;
LASSERT(obj->oo_db != NULL);
rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
SA_HDL_PRIVATE, &sa_hdl);
if (rc)
RETURN(rc);
OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
if (bulk == NULL)
GOTO(out_sa, rc = -ENOMEM);
la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
if (rc)
GOTO(out_bulk, rc);
la->la_atime = osa->atime[0];
la->la_mtime = osa->mtime[0];
la->la_ctime = osa->ctime[0];
la->la_mode = osa->mode;
la->la_uid = osa->uid;
la->la_gid = osa->gid;
la->la_nlink = osa->nlink;
la->la_flags = attrs_zfs2fs(osa->flags);
la->la_size = osa->size;
if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
if (rc)
GOTO(out_bulk, rc);
la->la_rdev = osa->rdev;
la->la_valid |= LA_RDEV;
}
out_bulk:
OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
out_sa:
sa_handle_destroy(sa_hdl);
RETURN(rc);
}
/*
* we allocate the requested pages atomically.
*/
int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc)
{
wait_queue_t waitlink;
unsigned long this_idle = -1;
cfs_time_t tick = 0;
long now;
int p_idx, g_idx;
int i;
LASSERT(desc->bd_iov_count > 0);
LASSERT(desc->bd_iov_count <= page_pools.epp_max_pages);
/* resent bulk, enc iov might have been allocated previously */
if (desc->bd_enc_iov != NULL)
return 0;
OBD_ALLOC(desc->bd_enc_iov,
desc->bd_iov_count * sizeof(*desc->bd_enc_iov));
if (desc->bd_enc_iov == NULL)
return -ENOMEM;
spin_lock(&page_pools.epp_lock);
page_pools.epp_st_access++;
again:
if (unlikely(page_pools.epp_free_pages < desc->bd_iov_count)) {
if (tick == 0)
tick = cfs_time_current();
now = cfs_time_current_sec();
page_pools.epp_st_missings++;
page_pools.epp_pages_short += desc->bd_iov_count;
if (enc_pools_should_grow(desc->bd_iov_count, now)) {
page_pools.epp_growing = 1;
spin_unlock(&page_pools.epp_lock);
enc_pools_add_pages(page_pools.epp_pages_short / 2);
spin_lock(&page_pools.epp_lock);
page_pools.epp_growing = 0;
enc_pools_wakeup();
} else {
if (++page_pools.epp_waitqlen >
page_pools.epp_st_max_wqlen)
page_pools.epp_st_max_wqlen =
page_pools.epp_waitqlen;
set_current_state(TASK_UNINTERRUPTIBLE);
init_waitqueue_entry_current(&waitlink);
add_wait_queue(&page_pools.epp_waitq, &waitlink);
spin_unlock(&page_pools.epp_lock);
waitq_wait(&waitlink, TASK_UNINTERRUPTIBLE);
remove_wait_queue(&page_pools.epp_waitq, &waitlink);
LASSERT(page_pools.epp_waitqlen > 0);
spin_lock(&page_pools.epp_lock);
page_pools.epp_waitqlen--;
}
LASSERT(page_pools.epp_pages_short >= desc->bd_iov_count);
page_pools.epp_pages_short -= desc->bd_iov_count;
this_idle = 0;
goto again;
}
/* record max wait time */
if (unlikely(tick != 0)) {
tick = cfs_time_current() - tick;
if (tick > page_pools.epp_st_max_wait)
page_pools.epp_st_max_wait = tick;
}
/* proceed with rest of allocation */
page_pools.epp_free_pages -= desc->bd_iov_count;
p_idx = page_pools.epp_free_pages / PAGES_PER_POOL;
g_idx = page_pools.epp_free_pages % PAGES_PER_POOL;
for (i = 0; i < desc->bd_iov_count; i++) {
LASSERT(page_pools.epp_pools[p_idx][g_idx] != NULL);
desc->bd_enc_iov[i].kiov_page =
page_pools.epp_pools[p_idx][g_idx];
page_pools.epp_pools[p_idx][g_idx] = NULL;
if (++g_idx == PAGES_PER_POOL) {
p_idx++;
g_idx = 0;
}
}
if (page_pools.epp_free_pages < page_pools.epp_st_lowfree)
page_pools.epp_st_lowfree = page_pools.epp_free_pages;
/*
* new idle index = (old * weight + new) / (weight + 1)
*/
if (this_idle == -1) {
this_idle = page_pools.epp_free_pages * IDLE_IDX_MAX /
page_pools.epp_total_pages;
}
page_pools.epp_idle_idx = (page_pools.epp_idle_idx * IDLE_IDX_WEIGHT +
this_idle) /
(IDLE_IDX_WEIGHT + 1);
page_pools.epp_last_access = cfs_time_current_sec();
spin_unlock(&page_pools.epp_lock);
return 0;
}
/**
* Implementation of the llog_operations::lop_open
*
* This function opens the llog by its logid or by name, it may open also
* non existent llog and assing then new id to it.
* The llog_open/llog_close pair works similar to lu_object_find/put,
* the object may not exist prior open. The result of open is just dt_object
* in the llog header.
*
* \param[in] env execution environment
* \param[in] handle llog handle of the current llog
* \param[in] logid logid of llog to open (nameless llog)
* \param[in] name name of llog to open (named llog)
* \param[in] open_param
* LLOG_OPEN_NEW - new llog, may not exist
* LLOG_OPEN_EXIST - old llog, must exist
*
* \retval 0 on successful open, llog_handle::lgh_obj
* contains the dt_object of the llog.
* \retval negative value on error
*/
static int llog_osd_open(const struct lu_env *env, struct llog_handle *handle,
struct llog_logid *logid, char *name,
enum llog_open_param open_param)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_ctxt *ctxt = handle->lgh_ctxt;
struct dt_object *o;
struct dt_device *dt;
struct ls_device *ls;
struct local_oid_storage *los;
int rc = 0;
ENTRY;
LASSERT(env);
LASSERT(ctxt);
LASSERT(ctxt->loc_exp);
LASSERT(ctxt->loc_exp->exp_obd);
dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt;
LASSERT(dt);
ls = ls_device_get(dt);
if (IS_ERR(ls))
RETURN(PTR_ERR(ls));
mutex_lock(&ls->ls_los_mutex);
los = dt_los_find(ls, name != NULL ? FID_SEQ_LLOG_NAME : FID_SEQ_LLOG);
mutex_unlock(&ls->ls_los_mutex);
LASSERT(los);
ls_device_put(env, ls);
LASSERT(handle);
if (logid != NULL) {
logid_to_fid(logid, &lgi->lgi_fid);
} else if (name) {
struct dt_object *llog_dir;
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out, rc = PTR_ERR(llog_dir));
dt_read_lock(env, llog_dir, 0);
rc = dt_lookup_dir(env, llog_dir, name, &lgi->lgi_fid);
dt_read_unlock(env, llog_dir);
lu_object_put(env, &llog_dir->do_lu);
if (rc == -ENOENT && open_param == LLOG_OPEN_NEW) {
/* generate fid for new llog */
rc = local_object_fid_generate(env, los,
&lgi->lgi_fid);
}
if (rc < 0)
GOTO(out, rc);
OBD_ALLOC(handle->lgh_name, strlen(name) + 1);
if (handle->lgh_name)
strcpy(handle->lgh_name, name);
else
GOTO(out, rc = -ENOMEM);
} else {
LASSERTF(open_param & LLOG_OPEN_NEW, "%#x\n", open_param);
/* generate fid for new llog */
rc = local_object_fid_generate(env, los, &lgi->lgi_fid);
if (rc < 0)
GOTO(out, rc);
}
o = ls_locate(env, ls, &lgi->lgi_fid, NULL);
if (IS_ERR(o))
GOTO(out_name, rc = PTR_ERR(o));
/* No new llog is expected but doesn't exist */
if (open_param != LLOG_OPEN_NEW && !dt_object_exists(o))
GOTO(out_put, rc = -ENOENT);
fid_to_logid(&lgi->lgi_fid, &handle->lgh_id);
handle->lgh_obj = o;
handle->private_data = los;
LASSERT(handle->lgh_ctxt);
RETURN(rc);
out_put:
lu_object_put(env, &o->do_lu);
out_name:
if (handle->lgh_name != NULL)
OBD_FREE(handle->lgh_name, strlen(name) + 1);
out:
dt_los_put(los);
RETURN(rc);
}
static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
struct lu_fid *fid)
{
struct link_ea_header *leh;
struct link_ea_entry *lee;
struct lu_buf buf;
int rc;
ENTRY;
buf.lb_buf = osd_oti_get(env)->oti_buf;
buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
if (rc == -ERANGE) {
rc = osd_xattr_get(env, o, &LU_BUF_NULL,
XATTR_NAME_LINK, BYPASS_CAPA);
if (rc < 0)
RETURN(rc);
LASSERT(rc > 0);
OBD_ALLOC(buf.lb_buf, rc);
if (buf.lb_buf == NULL)
RETURN(-ENOMEM);
buf.lb_len = rc;
rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
}
if (rc < 0)
GOTO(out, rc);
if (rc < sizeof(*leh) + sizeof(*lee))
GOTO(out, rc = -EINVAL);
leh = buf.lb_buf;
if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
leh->leh_magic = LINK_EA_MAGIC;
leh->leh_reccount = __swab32(leh->leh_reccount);
leh->leh_len = __swab64(leh->leh_len);
}
if (leh->leh_magic != LINK_EA_MAGIC)
GOTO(out, rc = -EINVAL);
if (leh->leh_reccount == 0)
GOTO(out, rc = -ENODATA);
lee = (struct link_ea_entry *)(leh + 1);
fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
rc = 0;
out:
if (buf.lb_buf != osd_oti_get(env)->oti_buf)
OBD_FREE(buf.lb_buf, buf.lb_len);
#if 0
/* this block can be enabled for additional verification
* it's trying to match FID from LinkEA vs. FID from LMA */
if (rc == 0) {
struct lu_fid fid2;
int rc2;
rc2 = osd_find_parent_by_dnode(env, o, &fid2);
if (rc2 == 0)
if (lu_fid_eq(fid, &fid2) == 0)
CERROR("wrong parent: "DFID" != "DFID"\n",
PFID(fid), PFID(&fid2));
}
#endif
/* no LinkEA is found, let's try to find the fid in parent's LMA */
if (unlikely(rc != 0))
rc = osd_find_parent_by_dnode(env, o, fid);
RETURN(rc);
}
/** Set up a mgc obd to process startup logs
*
* \param sb [in] super block of the mgc obd
*
* \retval 0 success, otherwise error code
*/
int lustre_start_mgc(struct super_block *sb)
{
struct obd_connect_data *data = NULL;
struct lustre_sb_info *lsi = s2lsi(sb);
struct obd_device *obd;
struct obd_export *exp;
struct obd_uuid *uuid;
class_uuid_t uuidc;
lnet_nid_t nid;
char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
char *ptr;
int rc = 0, i = 0, j, len;
LASSERT(lsi->lsi_lmd);
/* Find the first non-lo MGS nid for our MGC name */
if (IS_SERVER(lsi)) {
/* mount -o mgsnode=nid */
ptr = lsi->lsi_lmd->lmd_mgs;
if (lsi->lsi_lmd->lmd_mgs &&
(class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
i++;
} else if (IS_MGS(lsi)) {
lnet_process_id_t id;
while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
continue;
nid = id.nid;
i++;
break;
}
}
} else { /* client */
/* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
ptr = lsi->lsi_lmd->lmd_dev;
if (class_parse_nid(ptr, &nid, &ptr) == 0)
i++;
}
if (i == 0) {
CERROR("No valid MGS nids found.\n");
return -EINVAL;
}
mutex_lock(&mgc_start_lock);
len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
OBD_ALLOC(mgcname, len);
OBD_ALLOC(niduuid, len + 2);
if (!mgcname || !niduuid)
GOTO(out_free, rc = -ENOMEM);
sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
OBD_ALLOC_PTR(data);
if (data == NULL)
GOTO(out_free, rc = -ENOMEM);
obd = class_name2obd(mgcname);
if (obd && !obd->obd_stopping) {
int recov_bk;
rc = obd_set_info_async(NULL, obd->obd_self_export,
strlen(KEY_MGSSEC), KEY_MGSSEC,
strlen(mgssec), mgssec, NULL);
if (rc)
GOTO(out_free, rc);
/* Re-using an existing MGC */
atomic_inc(&obd->u.cli.cl_mgc_refcount);
/* IR compatibility check, only for clients */
if (lmd_is_client(lsi->lsi_lmd)) {
int has_ir;
int vallen = sizeof(*data);
__u32 *flags = &lsi->lsi_lmd->lmd_flags;
rc = obd_get_info(NULL, obd->obd_self_export,
strlen(KEY_CONN_DATA), KEY_CONN_DATA,
&vallen, data, NULL);
LASSERT(rc == 0);
has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
/* LMD_FLG_NOIR is for test purpose only */
LCONSOLE_WARN(
"Trying to mount a client with IR setting "
"not compatible with current mgc. "
"Force to use current mgc setting that is "
"IR %s.\n",
has_ir ? "enabled" : "disabled");
if (has_ir)
*flags &= ~LMD_FLG_NOIR;
else
*flags |= LMD_FLG_NOIR;
}
}
recov_bk = 0;
/* If we are restarting the MGS, don't try to keep the MGC's
old connection, or registration will fail. */
if (IS_MGS(lsi)) {
CDEBUG(D_MOUNT, "New MGS with live MGC\n");
recov_bk = 1;
}
/* Try all connections, but only once (again).
We don't want to block another target from starting
(using its local copy of the log), but we do want to connect
if at all possible. */
recov_bk++;
CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
rc = obd_set_info_async(NULL, obd->obd_self_export,
sizeof(KEY_INIT_RECOV_BACKUP),
KEY_INIT_RECOV_BACKUP,
sizeof(recov_bk), &recov_bk, NULL);
GOTO(out, rc = 0);
}
CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
/* Add the primary nids for the MGS */
i = 0;
sprintf(niduuid, "%s_%x", mgcname, i);
if (IS_SERVER(lsi)) {
ptr = lsi->lsi_lmd->lmd_mgs;
if (IS_MGS(lsi)) {
/* Use local nids (including LO) */
lnet_process_id_t id;
while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
rc = do_lcfg(mgcname, id.nid,
LCFG_ADD_UUID, niduuid,
NULL, NULL, NULL);
}
} else {
/* Use mgsnode= nids */
/* mount -o mgsnode=nid */
if (lsi->lsi_lmd->lmd_mgs) {
ptr = lsi->lsi_lmd->lmd_mgs;
} else if (class_find_param(ptr, PARAM_MGSNODE,
&ptr) != 0) {
CERROR("No MGS nids given.\n");
GOTO(out_free, rc = -EINVAL);
}
while (class_parse_nid(ptr, &nid, &ptr) == 0) {
rc = do_lcfg(mgcname, nid,
LCFG_ADD_UUID, niduuid,
NULL, NULL, NULL);
i++;
}
}
} else { /* client */
/* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
ptr = lsi->lsi_lmd->lmd_dev;
while (class_parse_nid(ptr, &nid, &ptr) == 0) {
rc = do_lcfg(mgcname, nid,
LCFG_ADD_UUID, niduuid, NULL, NULL, NULL);
i++;
/* Stop at the first failover nid */
if (*ptr == ':')
break;
}
}
if (i == 0) {
CERROR("No valid MGS nids found.\n");
GOTO(out_free, rc = -EINVAL);
}
lsi->lsi_lmd->lmd_mgs_failnodes = 1;
/* Random uuid for MGC allows easier reconnects */
OBD_ALLOC_PTR(uuid);
ll_generate_random_uuid(uuidc);
class_uuid_unparse(uuidc, uuid);
/* Start the MGC */
rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
(char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
niduuid, NULL, NULL);
OBD_FREE_PTR(uuid);
if (rc)
GOTO(out_free, rc);
/* Add any failover MGS nids */
i = 1;
while (ptr && ((*ptr == ':' ||
class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
/* New failover node */
sprintf(niduuid, "%s_%x", mgcname, i);
j = 0;
while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
j++;
rc = do_lcfg(mgcname, nid,
LCFG_ADD_UUID, niduuid, NULL, NULL, NULL);
if (*ptr == ':')
break;
}
if (j > 0) {
rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
niduuid, NULL, NULL, NULL);
i++;
} else {
/* at ":/fsname" */
break;
}
}
lsi->lsi_lmd->lmd_mgs_failnodes = i;
obd = class_name2obd(mgcname);
if (!obd) {
CERROR("Can't find mgcobd %s\n", mgcname);
GOTO(out_free, rc = -ENOTCONN);
}
rc = obd_set_info_async(NULL, obd->obd_self_export,
strlen(KEY_MGSSEC), KEY_MGSSEC,
strlen(mgssec), mgssec, NULL);
if (rc)
GOTO(out_free, rc);
/* Keep a refcount of servers/clients who started with "mount",
so we know when we can get rid of the mgc. */
atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
/* We connect to the MGS at setup, and don't disconnect until cleanup */
data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
OBD_CONNECT_LVB_TYPE;
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
#else
#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
#endif
if (lmd_is_client(lsi->lsi_lmd) &&
lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
data->ocd_version = LUSTRE_VERSION_CODE;
rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
if (rc) {
CERROR("connect failed %d\n", rc);
GOTO(out, rc);
}
obd->u.cli.cl_mgc_mgsexp = exp;
out:
/* Keep the mgc info in the sb. Note that many lsi's can point
to the same mgc.*/
lsi->lsi_mgc = obd;
out_free:
mutex_unlock(&mgc_start_lock);
if (data)
OBD_FREE_PTR(data);
if (mgcname)
OBD_FREE(mgcname, len);
if (niduuid)
OBD_FREE(niduuid, len + 2);
return rc;
}
static int llog_process_thread(void *arg)
{
struct llog_process_info *lpi = arg;
struct llog_handle *loghandle = lpi->lpi_loghandle;
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = lpi->lpi_catdata;
char *buf;
__u64 cur_offset = LLOG_CHUNK_SIZE;
__u64 last_offset;
int rc = 0, index = 1, last_index;
int saved_index = 0;
int last_called_index = 0;
LASSERT(llh);
OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
if (!buf) {
lpi->lpi_rc = -ENOMEM;
return 0;
}
if (cd != NULL) {
last_called_index = cd->lpcd_first_idx;
index = cd->lpcd_first_idx + 1;
}
if (cd != NULL && cd->lpcd_last_idx)
last_index = cd->lpcd_last_idx;
else
last_index = LLOG_BITMAP_BYTES * 8 - 1;
while (rc == 0) {
struct llog_rec_hdr *rec;
/* skip records not set in bitmap */
while (index <= last_index &&
!ext2_test_bit(index, llh->llh_bitmap))
++index;
LASSERT(index <= last_index + 1);
if (index == last_index + 1)
break;
repeat:
CDEBUG(D_OTHER, "index: %d last_index %d\n",
index, last_index);
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, LLOG_CHUNK_SIZE);
last_offset = cur_offset;
rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
index, &cur_offset, buf, LLOG_CHUNK_SIZE);
if (rc)
goto out;
/* NB: when rec->lrh_len is accessed it is already swabbed
* since it is used at the "end" of the loop and the rec
* swabbing is done at the beginning of the loop. */
for (rec = (struct llog_rec_hdr *)buf;
(char *)rec < buf + LLOG_CHUNK_SIZE;
rec = (struct llog_rec_hdr *)((char *)rec + rec->lrh_len)){
CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
rec, rec->lrh_type);
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
rec->lrh_type, rec->lrh_index);
if (rec->lrh_index == 0) {
/* probably another rec just got added? */
rc = 0;
if (index <= loghandle->lgh_last_idx)
goto repeat;
goto out; /* no more records */
}
if (rec->lrh_len == 0 ||
rec->lrh_len > LLOG_CHUNK_SIZE) {
CWARN("invalid length %d in llog record for index %d/%d\n",
rec->lrh_len,
rec->lrh_index, index);
rc = -EINVAL;
goto out;
}
if (rec->lrh_index < index) {
CDEBUG(D_OTHER, "skipping lrh_index %d\n",
rec->lrh_index);
continue;
}
CDEBUG(D_OTHER,
"lrh_index: %d lrh_len: %d (%d remains)\n",
rec->lrh_index, rec->lrh_len,
(int)(buf + LLOG_CHUNK_SIZE - (char *)rec));
loghandle->lgh_cur_idx = rec->lrh_index;
loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
last_offset;
/* if set, process the callback on this record */
if (ext2_test_bit(index, llh->llh_bitmap)) {
rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
lpi->lpi_cbdata);
last_called_index = index;
if (rc == LLOG_PROC_BREAK) {
goto out;
} else if (rc == LLOG_DEL_RECORD) {
llog_cancel_rec(lpi->lpi_env,
loghandle,
rec->lrh_index);
rc = 0;
}
if (rc)
goto out;
} else {
CDEBUG(D_OTHER, "Skipped index %d\n", index);
}
/* next record, still in buffer? */
++index;
if (index > last_index) {
rc = 0;
goto out;
}
}
}
out:
if (cd != NULL)
cd->lpcd_last_idx = last_called_index;
OBD_FREE(buf, LLOG_CHUNK_SIZE);
lpi->lpi_rc = rc;
return 0;
}
int llog_origin_handle_next_block(struct ptlrpc_request *req)
{
struct obd_export *exp = req->rq_export;
struct obd_device *obd = exp->exp_obd;
struct obd_device *disk_obd;
struct llog_handle *loghandle;
struct llogd_body *body;
struct llogd_body *repbody;
struct lvfs_run_ctxt saved;
struct llog_ctxt *ctxt;
__u32 flags;
__u8 *buf;
void *ptr;
int rc, rc2;
ENTRY;
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (body == NULL)
RETURN(-EFAULT);
OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
if (!buf)
RETURN(-ENOMEM);
ctxt = llog_get_context(obd, body->lgd_ctxt_idx);
if (ctxt == NULL)
GOTO(out_free, rc = -ENODEV);
disk_obd = ctxt->loc_exp->exp_obd;
push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
rc = llog_create(ctxt, &loghandle, &body->lgd_logid, NULL);
if (rc)
GOTO(out_pop, rc);
flags = body->lgd_llh_flags;
rc = llog_init_handle(loghandle, flags, NULL);
if (rc)
GOTO(out_close, rc);
memset(buf, 0, LLOG_CHUNK_SIZE);
rc = llog_next_block(loghandle, &body->lgd_saved_index,
body->lgd_index,
&body->lgd_cur_offset, buf, LLOG_CHUNK_SIZE);
if (rc)
GOTO(out_close, rc);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER,
LLOG_CHUNK_SIZE);
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
GOTO(out_close, rc = -ENOMEM);
repbody = req_capsule_server_get(&req->rq_pill, &RMF_LLOGD_BODY);
*repbody = *body;
ptr = req_capsule_server_get(&req->rq_pill, &RMF_EADATA);
memcpy(ptr, buf, LLOG_CHUNK_SIZE);
GOTO(out_close, rc);
out_close:
rc2 = llog_close(loghandle);
if (!rc)
rc = rc2;
out_pop:
pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
llog_ctxt_put(ctxt);
out_free:
OBD_FREE(buf, LLOG_CHUNK_SIZE);
return rc;
}
/**
* Callback handler for receiving incoming completion ASTs.
*
* This only can happen on client side.
*/
static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int lvb_len;
LIST_HEAD(ast_list);
int rc = 0;
LDLM_DEBUG(lock, "client completion callback handler START");
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
int to = cfs_time_seconds(1);
while (to > 0) {
schedule_timeout_and_set_state(
TASK_INTERRUPTIBLE, to);
if (lock->l_granted_mode == lock->l_req_mode ||
lock->l_flags & LDLM_FL_DESTROYED)
break;
}
}
lvb_len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB, RCL_CLIENT);
if (lvb_len < 0) {
LDLM_ERROR(lock, "Fail to get lvb_len, rc = %d", lvb_len);
GOTO(out, rc = lvb_len);
} else if (lvb_len > 0) {
if (lock->l_lvb_len > 0) {
/* for extent lock, lvb contains ost_lvb{}. */
LASSERT(lock->l_lvb_data != NULL);
if (unlikely(lock->l_lvb_len < lvb_len)) {
LDLM_ERROR(lock, "Replied LVB is larger than "
"expectation, expected = %d, "
"replied = %d",
lock->l_lvb_len, lvb_len);
GOTO(out, rc = -EINVAL);
}
} else if (ldlm_has_layout(lock)) { /* for layout lock, lvb has
* variable length */
void *lvb_data;
OBD_ALLOC(lvb_data, lvb_len);
if (lvb_data == NULL) {
LDLM_ERROR(lock, "No memory: %d.\n", lvb_len);
GOTO(out, rc = -ENOMEM);
}
lock_res_and_lock(lock);
LASSERT(lock->l_lvb_data == NULL);
lock->l_lvb_data = lvb_data;
lock->l_lvb_len = lvb_len;
unlock_res_and_lock(lock);
}
}
lock_res_and_lock(lock);
if ((lock->l_flags & LDLM_FL_DESTROYED) ||
lock->l_granted_mode == lock->l_req_mode) {
/* bug 11300: the lock has already been granted */
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "Double grant race happened");
GOTO(out, rc = 0);
}
/* If we receive the completion AST before the actual enqueue returned,
* then we might need to switch lock modes, resources, or extents. */
if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
LDLM_DEBUG(lock, "completion AST, new lock mode");
}
if (lock->l_resource->lr_type != LDLM_PLAIN) {
ldlm_convert_policy_to_local(req->rq_export,
dlm_req->lock_desc.l_resource.lr_type,
&dlm_req->lock_desc.l_policy_data,
&lock->l_policy_data);
LDLM_DEBUG(lock, "completion AST, new policy data");
}
ldlm_resource_unlink_lock(lock);
if (memcmp(&dlm_req->lock_desc.l_resource.lr_name,
&lock->l_resource->lr_name,
sizeof(lock->l_resource->lr_name)) != 0) {
unlock_res_and_lock(lock);
rc = ldlm_lock_change_resource(ns, lock,
&dlm_req->lock_desc.l_resource.lr_name);
if (rc < 0) {
LDLM_ERROR(lock, "Failed to allocate resource");
GOTO(out, rc);
}
LDLM_DEBUG(lock, "completion AST, new resource");
CERROR("change resource!\n");
lock_res_and_lock(lock);
}
if (dlm_req->lock_flags & LDLM_FL_AST_SENT) {
/* BL_AST locks are not needed in LRU.
* Let ldlm_cancel_lru() be fast. */
ldlm_lock_remove_from_lru(lock);
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST;
LDLM_DEBUG(lock, "completion AST includes blocking AST");
}
if (lock->l_lvb_len > 0) {
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_CLIENT,
lock->l_lvb_data, lvb_len);
if (rc < 0) {
unlock_res_and_lock(lock);
GOTO(out, rc);
}
}
ldlm_grant_lock(lock, &ast_list);
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
/* Let Enqueue to call osc_lock_upcall() and initialize
* l_ast_data */
OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 2);
ldlm_run_ast_work(ns, &ast_list, LDLM_WORK_CP_AST);
LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
lock);
GOTO(out, rc);
out:
if (rc < 0) {
lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_FAILED;
unlock_res_and_lock(lock);
wake_up(&lock->l_waitq);
}
LDLM_LOCK_RELEASE(lock);
}
